November 8, 2026 · @Gerry

Speculative money compounds by mathematics. Real assets decay by physics. Sooner or later, physics collects.

Two ledgers

On October 29, Alphabet had its worst trading day since it went public in 2004. The stock fell 12.6 percent. About US$482 billion of market value was gone before the clo…

Read more

We sold our business to private equity, and for us it worked. Not without some friction on both sides of the table, which is to be expected when a business has multiple income streams.

More to the point, the firm that backed us bought into a hard space, and they knew it was hard. Field services are capital-heavy in a way most fun…

Read more

What follows is a composite. I have watched versions of it run in more than one fund and more than one industry, and the details rhyme often enough that I no longer think of them as accidents.

I find it amusing to watch the stumbling and mumbling of highly educated, well-manicured participants in the high-finance world of private equity. Musical chairs for the smartest guy in the room, everyone jockeying over who gets to be king peacock for the day, while the people nominally responsible for the business seem more interested in reliving their university glory days than in the thing they just bought.

Alas, behold the brethren of capital-backed funds, screaming for deployment into the marketplace just so they can collect bloated fees and unreasonable dividends from their portfolio businesses. All while chasing a thesis that has been revised more times than anyone in the room would care to count.

Enter the latest phase: the roll-up of field services. Facilities work, grounds work, seasonal work, all pushed under one roof on the theory that a route is a route. And boy, have these worlds collided. You can hear the stuttering, a Fred Flintstone “Yabba-Dabba-Doo, but what can I do?!” Well, young spreadsheet warrior, you have entered a world where cost certainty and straight-line accounting have about as much chance of working out as a lottery ticket has of being a solid financial plan. You have bought into a business where the outcasts unite.

Every one of these businesses has a Johnny. Your superstar foreman of last month has had a change of circumstances this month. Life has intervened, in the way it does, and he has decided to come to work four days out of five, and when he is at work he is — let us be charitable — not at his best. His production has dropped by 40%. That hits his crew too, because Timmy is now covering the shortcomings of his supervisor, and Timmy is not paid for that.

Meanwhile the finance team has already forecast the profits from Johnny’s project into a pro forma presentation for the bank in order to meet debt covenants, and now the world has come crashing down. Free cash flow disappears. Working capital requirements increase. The project misses its milestones, and the liquidated damages start rolling in. So the scramble begins, and the board runs out of fingers to point at everyone but itself.

Further down the road, the thesis has bound the fund to a narrow set of end markets, and the discovery arrives that capex in this world is an annual commitment rather than a one-time line item. Somewhere in that meeting a board member vaguely remembers Charlie Munger calling EBITDA bullshit earnings, along with his line about the owner who surveys a yard full of equipment and calls it his profit. Now it makes sense. Whether he says anything is another matter, because by this point the room has settled into the posture rooms settle into, where the discussion is bounded by twenty-year-old case studies drawn from industries not remotely related to the one on the table.

Finally, the recapitalization event. The get-out-of-jail-sort-of-free manoeuvre that arrives after sweating the assets, losing the key personnel, and enduring leadership that reacts faster than it thinks. But it is all fine, because at the end of the day the recap pushes the founders — the ones who sold businesses that churned out steady dividends — outside any hope of recovering their rolled equity. They sold us something that ran fine before we broke it. Must be their fault.

And should the recap not work, there is always rescue capital at ridiculous returns, structured to shut out anyone who is not already inside the circle. Can’t burn those ones. They’re the cash cow.

So let me be fair for a moment. I understand the need to spread your wings, to go and find fragmented industries throwing off an appreciable share of revenue as recurring, year over year. But these industries operate in commoditized environments. So what is your differentiator? What are you actually going to do differently inside a sector under permanent margin pressure?

You are not going to find it by importing strategies from unrelated industries and shoehorning them into a business that runs on physical labour and boots on the ground. These crews are under relentless pressure to hit deadlines. And labour is the biggest line item there is — 35% to 45% of revenue, at the top of that range for straight maintenance like snow and turf.

Which brings us back to Johnny. This is a business where deadlines are actually deadlines. The grass never stops growing. Winter events start and stop on their own schedule and ask nobody’s permission. That 45% labour line is volatile on a good day — so naturally, this is the moment somebody chooses to roll out a CRM with no cross-platform integration, during the “slow” season, which is not slow.

So put down the deck and take a look inside.

A ten-million-dollar operation carrying a 45% labour burden is spending $4,500,000 on labour. Now watch it grind to a crawl. Crews sitting through two-hour tailgate meetings every second day. Documenting fly splatter on windshields. Logging who left an empty bag of chips in the back of a truck. Call it a 30% drag on labour, which is generous. That is $1,350,000 — thirteen and a half points of revenue.

Your 15% EBITDA was $1,500,000. At 6x, that is $9,000,000 of enterprise value — the number in the model, the number in the deck, the number the LPs were shown. Run the same multiple on what the business is actually producing and you are holding $900,000 of it. It is McConaughey playing Mark Hanna in The Wolf of Wall Street: “Fugayzi, fugazi. It’s a whazy. It’s a woozie. It’s fairy dust. It’s not real.” Then the scramble starts, and you normalize everything right down to the toilet paper.

The funny thing about business is that the outcome is rarely a pleasurable experience on the other side, and the chase for more has a way of costing the people doing the actual work. The only thing any of us needs more of is the life that money cannot buy.

This is the life of private equity in the field services world. It does not have to be.

None of the above is an argument against the model. It is an argument about sequence. Every number in that spreadsheet is downstream of something physical — a crew that shows up, a machine that starts, a route that gets run before the snow stops falling. Get those right and the model has something to stand on. Get them wrong and no amount of normalization, no recap, no rescue capital will put the earnings back, because the earnings were never there to begin with.

Structure before sophistication. Every time.

Gerry Visco

Move 78: Into the Void

In March 2016, a South Korean grandmaster placed a single stone that broke the world’s most powerful AI. What happened next wasn’t a malfunction. It was a revelation.

SECTION I

The Match That Changed the Conversation

In March 2016, in a conference room at Seoul’s Four Seasons Hotel that felt more like a coliseum, Lee Sedol — one of the greatest Go players in a game with 2,500 years of recorded play — sat down across from a machine. Not across from it, exactly. Across from a screen, a proxy for something with no body, no nerves, and no career on the line. DeepMind built AlphaGo, and Google owned it, and it trained on thirty million human moves. It had then played itself millions more times, iterating at speeds no biological organism could sustain. By the time it sat across from Lee Sedol, it had consumed more Go than any human being ever born could absorb in a thousand lifetimes.

The match was billed as a civilizational test. Not just for Go, but for the broader question everyone in artificial intelligence was circling without quite wanting to say out loud: had the machine crossed some threshold? Was pattern, compressed and accelerated to this degree, equivalent to understanding? Was it equivalent to intelligence?

AlphaGo won the first three games, and the Go establishment, which had initially given Lee Sedol a comfortable edge in pre-match assessments, went quiet. But Game 2 produced the moment nobody had a frame for. On move 37, AlphaGo placed a stone on the fifth line—a shoulder hit that human players considered, at best, eccentric and, at worst, a beginner’s error. The two human commentators watching the game live assumed it was a glitch. One of them laughed, nervous. Lee Sedol stood up and left the room for fifteen minutes.

He wasn’t rattled by the move’s strength. He was rattled by its strangeness. Move 37 made no sense in the game’s grammar as humans had ever played it. It violated the aesthetic logic, the accumulated heuristics, the unspoken agreements about what good Go looked like. Yet over the next hundred moves, it proved precisely correct. AlphaGo had assigned it a 1-in-10,000 probability — had seen what no human teacher had ever taught, because no human had ever played it. It had found something genuinely new inside the map it had built from thirty million human choices.

This is the moment the popular narrative froze. The machine beat the grandmaster with a move no human would ever play. The story wrote itself: AI sees further, thinks differently, has transcended us. Depending on your disposition, that was either the beginning of a golden age or the beginning of an existential reckoning. Both camps missed the more important thing that happened three days later.

Game 4 is the game the AI industry would prefer you to forget. Lee Sedol played move 78, and the most powerful game-playing system ever built started to come apart.

Lee Sedol won Game 4. He is the only human to have defeated AlphaGo in competitive play. He did it with a single stone — move 78, a wedge placement in the lower right corner of the board — that the Go community immediately named “the Hand of God.” AlphaGo’s calculated win probability, which had hovered near 70 percent in its favour, began producing nonsensical readings. The system started playing moves that experienced observers described as panicked, desperate, confused. It lost the thread. It lost the game.

Move 37 told us that a machine, given enough data and enough compute, can discover things inside a known space that no human had found before. That is genuinely remarkable. Move 78 told us something harder to digest: a human being, operating from a form of intelligence not reducible to pattern recognition, can step entirely outside the known space —and when they do, the machine has nothing to reach for.

That asymmetry is what this piece is about.

SECTION II

What “Hallucination” Actually Means — And Why It Should Worry You More Than Superintelligence Does

The word most people used to describe AlphaGo’s behaviour after move 78 was “confused.” Some said, “panicked.” In contemporary AI discourse, we would call it hallucination — the system producing confident outputs that are disconnected from reality. But that framing, while technically accurate, obscures what actually happened mechanically, and the mechanics matter if you want to understand what we are actually building when we build these systems.

AlphaGo was not confused in any experiential sense. It has no experience. Move 78 put the game in a region of possibility that AlphaGo’s model had no reference points for. Every move AlphaGo calculated was evaluated against a value function — a learned estimate of how likely a given board position was to lead to victory, built from the enormous library of games it had processed and the millions of self-play iterations that followed. That value function was exquisitely calibrated to the space of Go positions that humans and AlphaGo had ever produced. Move 78 pushed the game outside that space.

The technical term is distributional shift. A model trained on a distribution of inputs — in this case, the full recorded history of elite Go — will produce reliable outputs when the inputs it encounters resemble that distribution. When inputs diverge significantly from the training distribution, the model doesn’t know it’s lost. It keeps producing outputs. It keeps assigning confidence scores. But those scores are now unmoored from any valid reference. The system was not panicking. It was doing exactly what it was designed to do, in a context where doing exactly what it was designed to do produced catastrophic results. That is a more unsettling failure mode than panic, because panic at least signals awareness of a problem.

This is not a bug that gets patched in the next version. It is not a limitation of AlphaGo specifically — it is a structural property of how these systems are built. Every large language model, every image recognition system, every recommendation engine operates on the same principle: it learns a function that maps inputs to outputs by finding patterns in historical data, and it applies that function to new inputs under the assumption that the new inputs resemble the old ones. When that assumption fails, the model fails — not visibly, not loudly, but in the most dangerous way possible: confidently and incorrectly. You can’t build your future by pulling the block that built the past.

AlphaGo didn’t lose Game 4 because it was outmatched. It lost because it was out-framed. Lee Sedol didn’t play a better move within the game AlphaGo understood. He played a move AlphaGo had never conceived of.

Here is where the argument sharpens. The popular concern about superintelligence is that it will be so capable that it will optimize so effectively for whatever objective we give it that it will overwhelm our ability to course-correct. That concern is real and worth taking seriously. But Game 4 surfaces a different and less-discussed problem: a system that is simultaneously too confident and too bounded. Too confident because it keeps producing outputs even when it has no valid grounds to do so. Too bounded because all of those outputs, however sophisticated they appear, are interpolations from a closed library of prior experience. So doesn’t it stand to reason that effectively optimizing the processing of information we already know will trend toward output certainty by reverting to the mean? What other destination would it be to process thoughts, discussions, and research from the past without the ability to investigate and propagate new information? Are humans the rung or the ladder?

An AI system cannot encounter something genuinely new and recognize it as such. It can only measure new inputs against old patterns. If the new input is close enough to something in the pattern library, the system functions well. If the new input is far enough outside the library — if it represents genuine novelty, not just variation — the system has no flag to raise, no alarm to sound. It simply continues, producing outputs that look exactly like its correct outputs, because it has no mechanism to distinguish between the two cases. Lee Sedol’s move 78 was, in the most technical sense, a Black Swan: an event outside the model’s reference class, which the model therefore could not assign any meaningful probability to, and which the model’s architecture gave it no way to even detect as anomalous.

Nassim Taleb built a career on the observation that we systematically underestimate events outside our historical reference class. What AlphaGo demonstrated in Game 4 is that AI systems don’t just underestimate those events; as we have been discussing, they are architecturally blind to them. Even if the system has been populated with every game that has ever been played, it could be argued that it is operating within a vacuum. They can be extraordinarily powerful within the territory their training data covers. But they carry no map for the territory beyond the map.

And here is the question that should follow every breathless headline about what AI can do: who is responsible for knowing where the map ends? If critical thinking gives way to cognitive atrophy, you are closing the door to the hallways of chance and discovery.

SECTION III

The Third Kind of Knowing: Black Swans, System Collapse, and the Move That Has No Name

A question is buried in what Lee Sedol said after Game 4, and almost no one has asked it. He was asked what he was thinking when he placed move 78. His answer: that was the only move he could see. There were no other moves. He couldn’t think of any other placement. It was the only option, so he played it.

Read that carefully,, and you realize it doesn’t fit either cognitive category we use to describe intelligence, human or artificial. It was not a calculation. Lee Sedol did not evaluate move 78 against a field of alternatives, assign probability weights, and select the highest-scoring option. But it was also not random. He did not place the stone arbitrarily, without knowledge or intent. It was something else entirely — a move that arrived—not chosen from a menu and not generated by noise. Arrived. The product of thirty years of embodied expertise collapsing, in a single moment of constraint, into one irreducible perception: here, and nowhere else.

That distinction matters more than anything else in this piece, because modern artificial intelligence is built on two cognitive primitives: calculation and randomness. It can do both with extraordinary power. What it cannot do — what it has no mechanism to do — is produce what Lee Sedol produced at move 78. The reason it can is not a limitation of processing speed, data volume, or model size. It is a structural impossibility that follows directly from what these systems are and how they are built.

To understand why, you need two thinkers. One mapped the boundary of what models can know. The other mapped the boundary of what minds can do. Together, they define the ceiling that move 78 broke through.

Nassim Taleb spent years as a derivatives trader before he became a philosopher of risk. That path matters, because he was not writing theory from a library. He wrote from the perspective of a man who had watched confident quantitative models produce catastrophic wrong answers in real markets, with real consequences. He called them Black Swans, high-impact events outside a model’s reference class, and it is not merely a metaphor. It is a precise technical observation: a model trained on a distribution of historical events cannot assign meaningful probability to events outside that distribution. It cannot even recognize that it has encountered one. The model does not know what it does not know.

We have already seen that a model operating outside its training distribution keeps producing confident answers. It cannot detect that it has crossed the boundary, because doing so would require knowing where the distribution’s boundary is, which would require observing inputs outside it, something it hasn’t done. The AI system has no critical thinking because it has no heartbeat powering a nervous system with a visceral connection to the world. This is Taleb’s insight applied to AI with surgical precision: the systems most confident in their outputs are the systems most vulnerable to genuine novelty, because they have no mechanism to distinguish novelty from variation. As Ben Affleck put it when discussing AI at the 2024 CNBC Delivering Alpha investor summit:

“Craftsman is knowing how to work. Art is knowing when to stop.”

He added that knowing when to stop will be very difficult for AI to learn because it’s taste. AI, he said, can write you excellent imitative verse that sounds Elizabethan; it cannot write you, Shakespeare.

Move 78 wasn’t a variation on anything in AlphaGo’s training data. It was a Black Swan on a Go board — an event the model could not recognize as anomalous because recognizing anomalies requires a map of the territory beyond the map.

Demis Hassabis confirmed as much the morning after Game 4. AlphaGo assigned move 78 a one-in-ten-thousand probability not because its calculations suggested it was a bad move, but because it determined no human player would ever choose it. It had built a model of human Go cognition from thirty million games, and that model told it the move was outside the distribution of human play. It was right. That was precisely the problem. The model’s accuracy in characterizing what humans had done became a catastrophic liability the moment a human did something the model had never seen. It was operating on a map of human possibility. Lee Sedol stepped off the map. And AlphaGo had no response for territory that, by its own reckoning, did not exist.

Daniel Kahneman’s framework is the second piece. In his account of human cognition, there are two systems of thought. System 1 is fast, automatic, intuitive, the part of the mind that recognizes faces, catches falling objects, reads social situations without deliberate effort. System 2 is slow, deliberate, analytical — the part that does long division, evaluates a contract, calculates a chess position. Kahneman showed that most decisions are made by System 1 and rationalized afterward by System 2. We believe we are calculating when we are mostly pattern-matching at speeds that feel instant. Pattern-seeking is not something the mind does; it is what the mind is. The tendency is so dependable that its failures have names: apophenia, meaning in random data, and pareidolia, faces in clouds, outlets, and toast. Michael Shermer called the general tendency patternicity. The point is not that we sometimes see patterns that aren’t there. We are built to see them first and question them later. System 1 draws us toward the pattern-seeking comfort zone, where our minds find inner peace.

In this framework, artificial intelligence is the most powerful System 1 machine ever built. It pattern-matches at scales and speeds no biological mind approaches. A large language model responding to a complex question is doing something that looks like System 2 reasoning — deliberate, structured, analytical — but is in fact an extraordinarily sophisticated System 1 operation: statistical pattern completion across a vast training distribution, producing the most probable continuation of the input sequence. It is not reasoning from first principles. It is recognizing, at enormous scale and resolution, what typically follows what. The appearance of reasoning is a property of the patterns it has learned, not a property of the cognitive process itself. Simply put, AI does not second-guess its output.

So why does all this matter? Kahneman’s framework, like Taleb’s, has a built-in boundary. System 1 fails at precisely the moments System 2 is needed — when the situation is genuinely novel, when the pattern library doesn’t apply, when the fast answer is not just wrong but confidently wrong in ways that compound. Kahneman documented these failures in humans. The difference is that a human being, confronted with the failure of System 1, has access to System 2. They can slow down, interrogate their assumptions, reason explicitly about what they don’t know. A machine learning system, when its pattern-matching fails, has no System 2 to reach for. It has only more of System 1 — faster pattern-matching in territory where the patterns have broken down. After move 78, AlphaGo did not slow down to reason about what had happened. It continued selecting the statistically most probable moves in a position its training data had deemed impossible. The result looked like panic. It was worse: confidence operating in a void. What we refer to as hubris.

But here is where the argument has to be honest about its own complexity. Because Lee Sedol’s move 78 was not pure System 2 reasoning either. He did not sit down, enumerate all possible moves, and calculate that 78 was optimal. He said there was only one move he could see. It arrived. That is the language of System 1 intuition, immediacy, the felt sense of necessity. And yet it was not the ordinary pattern-completion of System 1, because ordinary pattern-completion is exactly what AlphaGo had, and AlphaGo could not produce it.

What Lee Sedol produced at move 78 was something else entirely. Not fast pattern-matching or slow calculation, but something cognitive science has struggled to name precisely: the intuition of a master. Thirty years of embodied practice, compressed into a perceptual field so deep that the patterns he recognized were not patterns in AlphaGo’s training data — were not patterns in any data — but patterns emergent from the lived experience of having played, lost, adapted, suffered, returned, and played again across three decades of competition at the highest level. The move did not come from his analysis. It came from what he had become.

AI OPERATES HERE

System 1 at scale

Pattern completion across historical distributions. Extraordinarily powerful within the map. Structurally blind at the boundary.

AI CAN SIMULATE THIS

System 2 reasoning

Slow, explicit, deliberate. Evaluates alternatives. AI produces its appearance through pattern completion — it is not the thing itself.

MOVE 78 LIVES HERE

Intuitive necessity

Neither fast pattern nor slow calculation. The only move a particular consciousness, with a particular history, could see. Unrepeatable. Untrainable.

This third category — call it intuitive necessity, or embodied expertise under constraint — is what Taleb’s framework identifies as structurally invisible to probabilistic models and what Kahneman’s framework cannot fully account for because it was built to describe failure, not mastery. It is the cognitive territory that exists past both the map and the reasoning about the map: the place where a particular human being, shaped by a particular irreducible history, perceives something that no distribution of prior events could have predicted, because it is the product not of information but of formation.

Alexander Fleming did not calculate that the mould contaminating his petri dish was worth investigating. He noticed it—a noticing that required not only knowledge of bacteriology but a particular cast of attention, a tolerance for surprise, and a willingness to treat anomaly as signal rather than noise, cultivated over decades of laboratory practice. August Kekulé did not compute the ring structure of benzene. By his own account, given a quarter-century later, he dreamed of a snake seizing its own tail and woke up with the answer. This answer could not have been reached by any systematic enumeration of molecular possibilities. Ignaz Semmelweis insisted on handwashing before germ theory existed to justify it, because something in his clinical experience told him the connection was there even when he lacked the conceptual language to prove it.

None of these were random. None were calculated. All three were the product of a consciousness that had internalized a domain so deeply that genuine novelty became perceptible—not because it had seen it before, but because it had developed the capacity to recognize it as significant when it appeared. That capacity is not a larger dataset. It is not a better algorithm. It is the residue of a life of sustained attention, pressed into the shape of a particular human mind.

AlphaGo could not produce move 78 because move 78 did not come from information. It came from formation. And formation is not something you train into a model. It happens to a person.

This is the philosophical ceiling that Game 4 exposed — not the ceiling of AlphaGo specifically, but the ceiling of any intelligence built entirely from the record of what has already happened. Such an intelligence can find everything within the map of prior human experience, faster and more completely than any human being. Move 37 proved that. But the map has an edge. And at the edge — in the moments that matter most, the moments of genuine novelty, of crisis, of discovery, of the kind of creative leap that rewrites what comes after — the most powerful pattern-matching system ever built has nothing to reach for. It does not know it has reached the edge. It does not know there is an edge. It keeps moving forward, confident and lost, into territory that its own architecture has rendered invisible.

Lee Sedol retired from professional Go in 2019. He said he quit because of AlphaGo. Lee also mentioned he had seen an entity that could not be defeated, and that knowledge had drained the meaning from a lifetime of competition. It was a deeply human response to something genuinely new. The irony is that in the very match that broke his will to continue, he did the one thing AlphaGo could not: he reached past the map and placed a stone in territory no model, however vast, had charted. He didn’t know he had done something historically unrepeatable. He just knew it was the only move he could see. He took the first step into the void….

That is what we risk losing, if we build a world where the map is all there is.

SECTION IV

The Metaproletariat: What Happens When Institutions Destroy the Conditions That Produce Move 78

The philosophical argument in Section III is not an abstraction. It has a direct institutional expression, and that expression is happening right now, at scale, inside every organization that has spent the last decade optimizing human judgment out of its decisions.

Here is the logic compressed: the highest-value cognitive act — the one that operates where AI systems go dark, where the map ends and genuine innovation begins — is not calculation and not randomness. It is the intuitive necessity that emerges from formation—from decades of embodied, felt, consequence-bearing experience in a domain, pressed so deeply into a human consciousness that the right move arrives rather than being chosen. That capacity is not innate. It is not a personality trait. It is the product of a specific developmental process: prolonged exposure to real problems, real feedback, real stakes, and enough scar tissue that the nervous system starts to perceive patterns with no name in any dataset.

For AI, the past is the only lens into the future, but it is a record, not a scar. AI can predict what comes next; it cannot anticipate it, because nothing ahead of it is at stake.

Some would argue that we are drifting towards a metaphysical conversation. But in a literal sense, we are trying to understand, intuitively, what lies beneath the physics. We can know that something is there without yet knowing what it is.

Lee Sedol could produce move 78 because he had spent thirty years losing games that hurt, winning games that mattered, and returning to the board regardless. The formation preceded the move by three decades. You cannot shortcut it, simulate it, or hire it on a contract basis. It grows in a person the way bone density grows. Bone grows slowly, through resistance, over time, and only when conditions are right.

Now consider what the dominant organizational logic of the last twenty years has been doing to those conditions.

Harry Braverman saw it coming in 1974. His analysis of industrial capitalism — still the sharpest account of what happens when management ideology meets the labour process — identified a pattern he called deskilling: the systematic decomposition of complex work into simplified tasks, the separation of conception from execution, the transfer of knowledge and judgment from workers to management and then to machinery. Braverman was writing about factory floors and typing pools. The principle has migrated up every layer of the knowledge economy, and the arrival of AI has not changed its direction. It has simply accelerated it to the point where we can no longer pretend it isn’t happening.

In philosophy, intrinsic value is the worth something has in itself, as an end, not because of what it gets you. Happiness, knowledge, beauty, or a person’s life are classic candidates. It contrasts with instrumental value, which is worth as a means: money matters for what it buys, a hammer for what it builds. The test is to ask, “Why does that matter?” until you hit something where the only answer is, “It just does.” That stopping point is intrinsic value. The problem arises when instrumental value becomes a full-court press, and with the wealth extraction and rise of a powerful oligarchy that come with it.

What algorithmic management does to a workforce is precisely what distributional shift does to a model: it removes the conditions under which genuine novelty can be perceived. A driver who follows GPS instructions for ten years and is then dropped in an unfamiliar city without signal is not supported; they are deskilled. A claims assessor whose recommendations are overridden by a scoring model every time they diverge from the algorithm’s output eventually stops making independent assessments. The judgment atrophies because the feedback loop that would strengthen it has been severed. Formation requires friction. Remove the friction, and you remove the formation, and with it, any future capacity to produce the move the algorithm cannot.

Deskilling doesn’t just eliminate current judgment. It eliminates the developmental conditions under which future judgment could have grown. You are not losing a capability. You are losing the process that makes the capability possible.

David Autor’s task-based framework for labour economics gives this structural precision. Autor’s research shows that automation displaces routine tasks—both manual and cognitive—while complementing non-routine ones. The standard reading of Autor is optimistic: automation handles the predictable, humans handle the unpredictable, everyone moves up the value chain. The problem with this reading is that it treats non-routine cognitive capacity as a given: a stable reservoir of human capability that persists regardless of what happens to the routine-task layer beneath it. That assumption is wrong. Non-routine cognitive capacity — the kind that handles genuinely novel situations — is not separate from routine task experience. It develops through it. The junior analyst who learns to read a balance sheet by doing a thousand of them is not just building routine competence. They are building the pattern library that, under the right conditions of constraint and pressure, will one day produce a non-routine insight no model saw coming. Remove the thousand balance sheets and you remove the prerequisite for the insight.

Erik Brynjolfsson and Andrew McAfee were right that the second machine age creates enormous wealth. They were less clear about the distributional question: who develops the non-routine cognitive capacity that survives automation, and under what conditions? Their answer — invest in education, retrain workers, help people move up the skill ladder — is the right prescription at the wrong level of analysis. The question is not whether people can learn new skills. The question is whether the organizational structures being built around AI allow the kind of deep, consequence-bearing, longitudinal formation that produces genuine judgment. Most of them do not. Most of them are designed, explicitly or implicitly, to prevent it. Formation is slow, expensive, unpredictable, and resistant to the kind of standardization that makes organizations legible to their owners. The push towards a perceived certainty of outcome will stagnate non-routine cognitive input: the thinking, and the subsequent act of discovery, that happen within the realm of human consciousness.

Yanis Varoufakis named the endpoint of this logic technofeudalism: a system in which the owners of algorithmic infrastructure extract value from an increasingly undifferentiated labour force stripped of the productive capacity to generate genuine economic alternatives. The feudal analogy is precise in one specific way: in feudalism, the serf’s dependence on the lord was not incidental but structural. The relationship was produced and maintained by deliberately restricting the conditions under which independent capacity could develop. Algorithmic management does the same to knowledge workers. It is not merely that the work becomes more monitored and less autonomous. It is that the conditions under which independent judgment could grow — the latitude to be wrong, to experiment, to accumulate scar tissue and learn from it — are systematically removed in the name of efficiency, consistency, and risk management.

This is where the move 78 argument becomes an organizational argument, not merely a philosophical one. The question is not whether AI will eventually be able to simulate the intuition of a master. That is a question for a different decade. The question is whether the humans working alongside AI systems today are being allowed to develop the formation that would make them capable of move 78 when it counts. And in most organizations currently deploying AI at scale, the answer is no. They are deploying AI in ways that accelerate task completion while simultaneously eliminating the developmental friction that builds the judgment task completion cannot replace.

The organizations most aggressively automating human judgment today aren’t just cutting costs. They are removing the seedbed of the only competitive advantage that cannot be reproduced from a training dataset.

I have spent more than twenty-five years running field operations. The best operators I worked with—the ones who could walk into a failing job at two in the morning and immediately see what everyone else had missed—weren’t the ones with the most credentials or the most efficient dashboards. They were the ones who had been allowed to fail, at real cost, in real situations, enough times to develop a perceptual field that no process document could replicate. That field was not their personality. It was their formation. And it took years of friction to build it. “Knowledge does not advance practice. Rather practice advances knowledge” (Pascale et al., 2010). In any business, the field is where practice doesn’t just make perfect; it makes sense.

The PE-backed management logic that eventually acquired the business I had built would have optimized most of that formation out of the organization within eighteen months. That is not to say the PE owners were wrong. The legacy employee who becomes a hurdle rarely does so intentionally; they struggle to articulate a methodology that has become so reflexive and intuition-led that it is hard to explain. It becomes hard to transfer. So the playbook takes over. Standardize the process. Reduce variance. Replace judgment calls with decision trees. These are not irrational choices from a short-term financial perspective. But they are choices that, applied consistently over time, produce organizations full of people who are very good at executing within the known space and completely unprepared for the moment the known space runs out — which is precisely the moment when the AI system they have been relying on will have nothing to offer them either. Always keep in mind the power of being as nimble as a kayak and as powerful as a dragon boat.

The Metaproletariat thesis I have been developing is, at its core, an argument about what happens when the managerial layer is eliminated by AI and what is revealed beneath it. The conventional view is that AI destroys middle management and creates a barbell economy: a small class of highly paid cognitive workers at the top and a large class of precarious physical and service workers at the bottom, with it all held by a smaller class owning the algorithmic infrastructure. That analysis is structurally correct. But it misses the deeper problem.

The deeper problem is that the compression is not just economic; it is developmental. When the managerial layer disappears, it takes with it the organizational infrastructure through which formation used to happen. The junior analyst learned to see what the senior analyst saw by working alongside them, being corrected by them, inheriting not just their knowledge but their attentional habits, their tolerance for ambiguity, their instinct for when the model is wrong. That transmission was inefficient. It was slow. It resisted standardization. And it produced, over time, the conditions under which move 78 became possible for the people who went through it.

Remove the managerial layer, and you remove the transmission mechanism. What you are left with is a workforce that can operate the tools very well and perceive almost nothing that the tools cannot perceive. That workforce is not a partner to the AI. It is its dependent — capable of execution within the distribution the model has mapped, and helpless at the boundary. The Metaproletariat is not just economically precarious. It is cognitively stranded: cut off from the developmental conditions that would have made it capable of the one thing the algorithm cannot do.

Lee Sedol had thirty years to become the person who could see move 78. The question that should be keeping organizational leaders awake at night is not whether their AI system is accurate enough. It is whether they are building the conditions for the people working alongside it to develop the formation to know when accurate enough isn’t good enough and to see, in that moment, the move the model never conceived of.

SECTION V

What to Protect: A Structural Argument for the Conditions That Make Move 78 Possible

This is not an argument against artificial intelligence. Let that be stated plainly, because the moment you argue for limits on what AI can do, someone in the room will hear nostalgia, or fear, or the kind of reflexive technophobia that has been wrong about every previous wave of automation. That is not what this is. AlphaGo was a genuine achievement. The large language models (LLMs) reshaping knowledge work in 2026 are genuine achievements. The prescription that follows is not a call to slow any of that down. It is a call to understand what you are trading away when you deploy these tools without protecting the developmental conditions that produce the one cognitive capacity they cannot replicate. It’s the ‘when you know, you know’ feeling: the visceral sense of something impacting you in a way words can’t describe. That is the feel that seems to be missing.

The trade-off isn’t obvious in the short term. In the short term, replacing human judgment with algorithmic decision-making is almost always faster, cheaper, and more consistent. The dashboard looks cleaner. The variance goes down. The process becomes legible to its owners. These are real benefits, and anyone who dismisses them is not engaging seriously with the organizational pressures that produce these decisions. The problem is not that the trade-off is irrational. The problem is that it is invisible; the cost doesn’t appear on any balance sheet until the algorithm reaches the edge of its distribution and no one left in the building can see what the algorithm cannot.

By then, it is too late. Formation is not a resource you can stockpile and draw down. It must happen continuously, in real people, in real conditions, or it stops. A decade of algorithmic management does not leave you with slightly less judgment. It leaves you with a workforce that has forgotten what judgment felt like and an organizational structure that has no mechanism to rebuild it. The Go board analogy is exact: you cannot become the person who sees move 78 by watching someone else play. You have to sit at the board, in real games, with real consequences, long enough for the formation to happen to you.

So what does protection actually look like? Not sentiment. Not a policy of hiring humans to do what machines do better now. The prescription is structural: deliberate organizational choices about where to actively cultivate human formation, even at the cost of short-term efficiency, because the long-term strategic value of that formation exceeds any efficiency gain its absence would produce.

01 Protect the developmental layer

The junior roles that look inefficient — the analyst who takes twice as long as the model, the operator who makes calls the algorithm would override — are not inefficiencies to be eliminated. They are the seedbed. Formation happens in those roles, in those years, through the friction of consequence-bearing work. Organizations that eliminate the entry-level judgment role to save costs aren’t just cutting headcount. They are severing the transmission mechanism through which expertise propagates across generations of practitioners. The cost shows up fifteen years later, when no one in the organization can catch what the model gets wrong.

02 Design for productive failure

Formation requires being wrong at real cost. Failure is not just part of human development; it is the mechanism. Scar tissue is not a metaphor; it is the physiological residue of a nervous system that has processed real feedback, real consequences, and recalibrated. Organizations that have optimized away the conditions under which their people can fail—through excessive oversight, algorithmic override, or removing any decision with meaningful stakes from human hands—have removed the primary driver of formation. The prescription is not to manufacture failure. It is to identify the domains where productive failure is currently being prevented and deliberately reintroduce the latitude for it. The pathway to a goal unpacks opportunity and ideas that will lead to the next adventure, the next movement, the next discovery. Take penicillin (1928–1945). Fleming’s contaminated petri dish is the famous part, but he couldn’t purify or mass-produce the substance and largely moved on. The real breakthrough came a decade later, when Florey, Chain, and Heatley at Oxford worked out the grinding process of extraction, testing, and scale-up. The accident became a medicine through process.

03 Distinguish execution from formation

Not every task produces formation. Routine execution — processing claims within a defined parameter set, scheduling within an optimization model, generating reports from a template — produces competence at the task and nothing beyond it. The organizational question is not how much human judgment to preserve in aggregate, but where in the value chain the formation-producing work lives and whether the people who need to develop that capacity are being routed through it. AI should own the execution. Humans should own the work that forms them: messy, ambiguous, consequence-bearing problems with no clean algorithmic solution, leaving a residue of judgment in the people who work through them.

04 Map where your distribution ends

Every AI system deployed in your organization has a training distribution — a boundary beyond which its outputs are unmoored from valid reference. Most organizations have no map of where that boundary is for the systems they are running. The prescription is not to limit AI deployment. Instead, build the institutional practice of asking, for every consequential AI-assisted decision: what category of input would cause this system to fail confidently? And who in this organization is trained to recognize that input when it arrives? If the answer to the second question is no one, you have a structural risk that no dashboard will surface until the moment it materializes.

None of this is anti-technology. The centaur model — a human and AI operating together, each doing what they do best — is the right frame. But a centaur only works if the human half is a human who has been formed. A person who has spent their career executing within algorithmic parameters, never developing the judgment that comes from consequence-bearing autonomy, is not a centaur’s human half. They are a cursor. They provide the inputs the model requests and execute the outputs the model produces. The machine is not augmenting their intelligence. It substitutes for it, in a configuration that looks like partnership but functions like dependence.

The centaur model that actually survives the next decade of AI development is not the one where humans and machines divide the work. It is the one where humans have been formed deeply enough to know when the machine has reached its edge and to offer something in that moment that the machine cannot provide. That formation does not happen by accident. It does not happen in environments optimized for consistency and efficiency. It happens through decades of friction, failure, return, and accumulation. It requires organizations willing to protect it even when doing so is expensive, slow, and at odds with the metrics that make quarterly performance legible.

The question is not whether your AI system is good enough. The question is whether the people working alongside it have been formed enough to know when it isn’t — and to see, in that moment, what it cannot.

Lee Sedol retired in 2019. He gave a specific reason: he had seen something that could not be defeated, and that knowledge had made the game feel hollow. He was the best human player alive, and the best was no longer enough to matter. By any reasonable measure, his decision was rational. What would be the point of continuing?

But here is what that retirement actually means: read through the argument this piece has been making. Lee Sedol was the only human being in recorded history to demonstrate the capacity to produce move 78—to step outside the distribution of all prior Go cognition and place a stone in territory that the most powerful pattern-recognition system ever built had deemed impossible. In the most precise technical sense, he was the only person on earth who had proven that the boundary of machine intelligence was not the boundary of human intelligence. And he walked away from the board.

He was not wrong to. The incentive structures gave him no reason to stay. The field he had devoted his life to had been restructured around an entity his formation could occasionally, brilliantly, historically exceed — but could not consistently match. The rational actor exits. And in exiting, he takes with him the only thing the algorithm could not replicate, the only thing thirty years of the most demanding formation in the history of the game produced, and leaves it on the table, unclaimed.

That is the cost of getting this wrong at the individual level. Scale it to an organization. Scale it to an economy. The question is not whether the machines will eventually be able to do what Lee Sedol did. The question is whether, in the meantime, we are building the conditions under which the next Lee Sedol — the analyst, the operator, the clinician, the leader, the engineer — is being formed. Or whether we are building organizations so perfectly optimized for the known space that the people inside them never develop the capacity to see past it.

Move 78 was not a product of genius. It was a product of formation. Genius is unteachable. Formation is a choice — an organizational choice, a cultural choice, a leadership choice — that gets made or not made every day, in every decision about how work gets structured and who gets the latitude to be wrong in ways that matter.

The superintelligence debate asks the wrong question. It asks: what happens when the machine becomes smarter than us? The move 78 argument answers a different question, the one that is already live, already consequential, and already being decided without most of us noticing: what happens when we become less than what we were — not because the machine took something from us, but because we handed it over in exchange for a cleaner dashboard and a lower variance on the quarterly report?

The map is extraordinary. It is the most powerful map ever built. But the map has an edge. Someone has to be willing to walk to it. Someone has to have been formed well enough to see what’s there. And someone has to have built, deliberately and at real cost, the conditions under which that person could have become who they needed to be.

That is not a technology problem. It is a leadership problem. And it starts with knowing the difference between the two.

Structure before sophistication—every time.

SOURCES

References

Autor, D. H., Levy, F., & Murnane, R. J. (2003). The skill content of recent technological change: An empirical exploration. The Quarterly Journal of Economics, 118(4), 1279–1333. https://doi.org/10.1162/003355303322552801

Benfey, O. T. (1958). August Kekulé and the birth of the structural theory of organic chemistry in 1858. Journal of Chemical Education, 35(1), 21–23. https://doi.org/10.1021/ed035p21

Berger, C. (2024, November 19). Ben Affleck says AI can’t ‘replace human beings making films’ because it lacks ‘taste’. Fortune. https://fortune.com/2024/11/19/ben-affleck-ai-hollywood-filmmaking/

Braverman, H. (1974). Labor and monopoly capital: The degradation of work in the twentieth century. Monthly Review Press.

British Go Association. (2016). Google DeepMind Challenge Match – Lee Sedol v AlphaGo – match report. https://britgo.org/node/5458

Brynjolfsson, E., & McAfee, A. (2014). The second machine age: Work, progress, and prosperity in a time of brilliant technologies. W. W. Norton & Company.

Chain, E., Florey, H. W., Gardner, A. D., Heatley, N. G., Jennings, M. A., Orr-Ewing, J., & Sanders, A. G. (1940). Penicillin as a chemotherapeutic agent. The Lancet, 236(6104), 226–228. https://doi.org/10.1016/S0140-6736(01)08728-1

Fleming, A. (1929). On the antibacterial action of cultures of a penicillium, with special reference to their use in the isolation of B. influenzæ. British Journal of Experimental Pathology, 10(3), 226–236.

Kahneman, D. (2011). Thinking, fast and slow. Farrar, Straus and Giroux.

Kasparov, G. (2017). Deep thinking: Where machine intelligence ends and human creativity begins. PublicAffairs.

Kekulé, A. (1865). Sur la constitution des substances aromatiques. Bulletin de la Société Chimique de Paris, 3, 98–110.

Kekulé, A. (1890). Benzolfest: Rede. Berichte der Deutschen Chemischen Gesellschaft, 23(1), 1302–1311.

Kohs, G. (Director). (2017). AlphaGo [Film]. Moxie Pictures; Reel As Dirt.

Metz, C. (2016, March 16). In two moves, AlphaGo and Lee Sedol redefined the future. Wired. https://www.wired.com/2016/03/two-moves-alphago-lee-sedol-redefined-future/

Metz, C. (2021). Genius makers: The mavericks who brought AI to Google, Facebook, and the world. Dutton.

NPR. (2016, March 13). After 3 losses, master Go player scores a win against computer. The Two-Way. https://www.npr.org/sections/thetwo-way/2016/03/13/470284113/after-three-losses-master-go-player-scores-a-win-against-computer

Pascale, R., Sternin, J., & Sternin, M. (2010). The power of positive deviance: How unlikely innovators solve the world’s toughest problems. Harvard Business Press.

Semmelweis, I. (1983). The etiology, concept, and prophylaxis of childbed fever (K. C. Carter, Trans.). University of Wisconsin Press. (Original work published 1861)

Shermer, M. (2008). Patternicity: Finding meaningful patterns in meaningless noise. Scientific American, 299(6), 48.

Silver, D., Huang, A., Maddison, C. J., Guez, A., Sifre, L., van den Driessche, G., Schrittwieser, J., Antonoglou, I., Panneershelvam, V., Lanctot, M., Dieleman, S., Grewe, D., Nham, J., Kalchbrenner, N., Sutskever, I., Lillicrap, T., Leach, M., Kavukcuoglu, K., Graepel, T., & Hassabis, D. (2016). Mastering the game of Go with deep neural networks and tree search. Nature, 529(7587), 484–489. https://doi.org/10.1038/nature16961

Taleb, N. N. (2007). The black swan: The impact of the highly improbable. Random House.

Varoufakis, Y. (2023). Technofeudalism: What killed capitalism. The Bodley Head.

Visco, G. (2026). The metaproletariat [Manuscript in preparation].

Yonhap News Agency. (2019, November 27). (Yonhap interview) Go master Lee says he quits unable to win over AI Go players.

Zimmerman, M. J., & Bradley, B. (2019). Intrinsic vs. extrinsic value. In E. N. Zalta (Ed.), The Stanford encyclopedia of philosophy (Spring 2019 ed.). https://plato.stanford.edu/archives/spr2019/entries/value-intrinsic-extrinsic/

Why tariffs aren’t the answer — and is dairy protection actually unfair?

Two questions at the centre of the CUSMA breakdown, answered plainly. Updated for October — including two corrections to the August edition that cut against my own earlier framing.

Gerry · October 1, 2026 · Structure Before Sophistication

What changed since August

I published the first version of this brief on August 23, the day after talks collapsed. Six weeks later, three things have changed enough to warrant a rewrite rather than a patch.

● The shooting started. Canada’s counter-tariffs went live September 8 — $27.6 billion across roughly 900 tariff lines, matched rate-for-rate against the American measures. Dairy is on both lists now.

● The economy did not break. Real GDP expanded at a 3.3 per cent annualized pace in the second quarter, with a notable increase in business investment. Unemployment held at 6.4 per cent in August, near its lowest level in two years. The recession concerns from earlier in the year have largely faded.

● I got the dairy compliance record wrong. The August edition said two dispute panels found Canada non-compliant. That is not what the record shows, and the correction materially strengthens Canada’s position. Details below.

Part one — why tariffs are not the answer

1. The importer pays, not the exporter

The most persistent misunderstanding in the entire debate. A tariff is a tax collected at the border from the domestic company doing the importing. When Washington imposes 50 per cent duties on Canadian plywood, the cheque to U.S. Customs is written by the American buyer. Study after study of the 2018–19 round found nearly the full cost passed through to domestic purchasers and consumers. The foreign exporter absorbs some of it through discounting. The majority lands at home.

This cuts both directions now. Canada’s September 8 counter-tariffs are paid by Canadian importers, and economists estimate they cost families with children roughly $250 a year. Retaliation is a sovereignty instrument. It is not a free one.

2. Every input tariff is a tax on your own manufacturers

Steel and aluminum are not final goods. They are inputs into automobiles, appliances, construction and machinery. Protecting roughly 140,000 American steel jobs raises costs for the several million workers in steel-consuming industries. The 2018 round is estimated to have cost more manufacturing jobs downstream than it preserved upstream.

The Canadian evidence is now in the data. Manufacturing production is running about 3.5 per cent below 2024, with the sharpest declines in steel and wood products. Canadian steel exports are down roughly 25 per cent, aluminum 6 per cent, autos 5 per cent. In an integrated continental supply chain where a component may cross the border up to eight times before final assembly, a tariff is applied repeatedly to the same value.

3. Retaliation is certain and precisely targeted

Partners retaliate, and they aim at politically concentrated constituencies rather than at economic efficiency. Canada’s list was built to be felt in specific congressional districts — dairy, agricultural equipment, appliances, food processing. The predictable result is that both economies shrink while each government tells its public it is winning.

4. Uncertainty does more damage than the duty itself

The tariff is a known cost that can be priced. The possibility of a tariff cannot be. That remains true — but the August edition leaned harder on this than the evidence now supports, and honest updating means saying so.

Canada’s economy proved more resilient than almost anyone forecast. Second-quarter growth came in at 3.3 per cent annualized. Business investment rose. Unemployment sits near a two-year low. Ottawa has deployed more than $30 billion in tariff-related support, and the CUSMA exemption continues to shield roughly 95 per cent of Canadian exports.

What the uncertainty argument still gets right is the distribution. The aggregate held; the incidence did not. Unemployment in manufacturing-heavy London, Ontario is 8.4 per cent while Quebec City sits at 3.8 per cent. Youth unemployment remains above 13 per cent. National statistics are averaging together regions having completely different experiences of the same trade war.

5. They do not fix the underlying complaint

A trade deficit reflects national savings and investment balances, not the fairness of any trading partner. Tariffs redistribute a deficit across countries; they do not eliminate it. Meanwhile the stated grievances — dairy access, auto content, digital rules — are regulatory questions a tariff cannot resolve. You cannot tariff your way to a quota reform.

Six weeks of evidence supports this directly. The tariffs were imposed to force concessions. Canada did not concede. It walked away from the table, retaliated, and legislated further protection for the very sector the tariffs were meant to pry open. Measured against their stated objective, the measures have failed.

The one case where tariffs do work — and why it still fails here

A large economy can improve its terms of trade with a tariff, because foreign exporters cut prices to hold volume. The United States is large enough for this to be real, and intellectual honesty requires conceding it.

But the gain is a transfer, not a creation — global output falls while the American share rises — and it evaporates the moment the partner retaliates credibly. Which is precisely what September 8 did. The one economically valid argument for tariffs is self-cancelling against a partner willing to respond.

Part two — is Canadian dairy protection unfair?

Here the honest answer depends entirely on what you think a trade agreement is for. Both sides have a real argument. Neither is being made in good faith at the negotiating table. But the factual record is clearer than I represented it in August, and it favours Canada more than I allowed.

The correction

The August edition said Canada “administered the quotas in ways two dispute panels found inconsistent with its obligations.” That is wrong. Here is the actual record:

● December 2021 — CUSMA Panel One. The United States won. The panel found Canada’s practice of reserving 85 to 100 per cent of its fourteen dairy quotas exclusively for processors was inconsistent with its obligations.

● November 2023 — CUSMA Panel Two. Canada won, on every claim. After Canada revised its allocation rules, Washington challenged again on four grounds. The panel found for Canada on all four and ruled Canada is not obligated to make further changes.

● September 2023 — CPTPP panel (New Zealand v. Canada). A separate agreement, a split result. New Zealand prevailed on two of six claims. Canada prevailed on the others, including on its discretion to exclude retailers from quota eligibility.

So the accurate summary is: Canada lost once, fixed it, and won the rematch. That is a materially different position from the one I described six weeks ago, and it changes the conclusion I drew about Canada’s negotiating leverage.

Which makes the August 2026 proclamation the real story

Washington’s dairy proclamation this past August asserts that Canada discriminates in administering the cheese quota — specifically, that Canadian retailers cannot hold CUSMA cheese quota but can hold quota for European cheese.

That is a White House determination. It is not a panel ruling. The last panel to examine Canada’s quota administration — including the eligibility question — sided with Canada on every claim.

The United States lost the dispute, and then used an executive proclamation to impose by tariff what it could not obtain by adjudication.

Whatever one thinks of supply management, that sequence should trouble anyone who believes dispute settlement is the point of having a trade agreement. The mechanism worked. It produced an answer. The answer was then overridden by proclamation.

The case that dairy protection is unfair

● Over-quota duties running from roughly 241 per cent on fluid milk to 298 per cent on butter are prohibitive by design. They are not tariffs in any ordinary sense. They are a wall.

● Canadian consumers pay materially more for milk, butter and cheese than Americans or Australians do. The cost is borne by households and it is regressive.

● Canada asks for open access to the American market in autos, energy, lumber and aluminum while closing one of its own. That asymmetry is hard to defend on principle.

● Supply management restricts entry for Canadian farmers too. Quota runs upwards of $30,000 per cow, which functions as a barrier against new domestic entrants, not only foreign ones.

● Canada did lose the 2021 panel, and it did lose two CPTPP claims in 2023. The compliance record is not spotless.

The case that it is not

● Every developed country protects agriculture. The United States spends roughly $20 billion a year on farm support, plus sugar quotas, dairy price supports and export credit programs. The instrument differs. The intervention does not.

● Supply management is funded by consumers rather than taxpayers. The American model shifts the same cost onto the federal budget, where it is simply less visible.

● Canada fixed what the 2021 panel told it to fix, and the 2023 panel confirmed the fix was sufficient. That is compliance, not evasion.

● CUSMA already grants the United States tariff-free access to roughly 3.5 per cent of Canada’s dairy market, phasing in through the late 2030s. The grievance is about administration of quota American exporters have never fully filled.

● Canada’s dairy trade deficit with the United States has grown significantly since CUSMA came into force. On net, American producers are winning this trade — which sits awkwardly beside the claim that they are being shut out.

● Dismantling supply management without a transition framework would transfer roughly $30 billion in quota value — farmers’ primary collateral — to zero, concentrated in rural Quebec and Ontario.

One correction in the other direction

In August I called American dairy exports to Canada “a rounding error.” That was unfair and I want to withdraw it.

American dairy sales to Canada ran about $1.3 billion in 2025 — roughly 14 per cent of total U.S. dairy exports, making Canada the second-largest market after Mexico. For Wisconsin and New York producers that is a serious number. The file is politically outsized, but the economics are not trivial.

Where the two tariff systems actually sit

Press coverage keeps conflating two entirely different measures, so it is worth separating them cleanly.

The old over-quota duties — 241 to 298 per cent — are unchanged and are not being paid by anyone. They apply only above the tariff-free quotas, and their function is to stop trade above those limits rather than to collect revenue.

The 2026 tariffs are new and separate. Since August 19 the United States has charged 50 per cent on Canadian whey, milk proteins, lactose, casein and milk powders. Since September 8 Canada has charged 25 per cent on American cheese and dairy entering inside the quotas — product that had been crossing tariff-free since CUSMA took effect. Butterfat moves in neither direction with a duty, because Canada is short of it.

Canadian dairy exports to the United States have stalled since August. The Dairy Farmers of Canada call the measures “completely unwarranted.” Both statements can be true at once.

The honest verdict

Canadian dairy protection is economically inefficient and costs Canadian consumers real money. It is also not meaningfully more protectionist than what the United States does for its own farmers — it is protectionist in a different currency.

What I got wrong in August was the compliance story. Canada lost a panel, complied, and won the rematch on every claim. The defensible American complaint has therefore narrowed to almost nothing, and what remains is an executive proclamation asserting a conclusion that adjudication rejected.

The indefensible American position is unchanged: treating a marginal export interest as grounds for tariffing $20 billion of unrelated Canadian goods, from hockey sticks to plywood to liquor.

Why this file never resolves

Because neither side actually wants the thing it is asking for. Washington does not need Canadian dairy market access — American exporters cannot fill the quota already granted, while their net dairy surplus with Canada keeps growing. What dairy provides is a permanently available grievance: politically potent in two swing states, modest in economic terms, and impossible for any Canadian government to concede without losing rural Quebec.

Canada has now legislated that impossibility. Bill C-202 formally protects supply management in trade negotiations. The file is not stuck by accident. Both sides have arranged for it to stay stuck, because it is worth more to each of them unsolved.

Bottom line. Tariffs fail because the importing country pays them, they tax domestic manufacturers through their own supply chains, they invite targeted retaliation, and six weeks of evidence shows they did not produce the concession they were imposed to extract. The Canadian economy held up better than the gloomiest forecasts — but the aggregate conceals a brutal regional distribution.

On dairy: supply management is a genuine inefficiency that Canada should reform on its own terms and its own timetable. It is also a settled legal question that the United States lost, reopened by proclamation. Those are different problems, and only one of them is Canada’s to fix.

Structure before sophistication. Fix the thing that is actually broken. Refuse the thing that is designed never to be fixed.

Gerry

Structure Before Sophistication · Briefing Note · Updated October 1, 2026 (first published August 23, 2026)

Corrections in this edition: the CUSMA dairy panel record, and the characterization of U.S. dairy exports to Canada. Both cut against the original framing. A brief that never corrects itself is not a brief — it is a position paper.

** article edited with assistance from Claude by Anthropic

A behavioural breakdown cannot be negotiated away

$27.6 billion in Canadian counter-tariffs went live at 12:01 this morning. Talks scheduled: zero. No grades moved — and that stability is the most important thing in this issue.

Gerry · September 8, 2026 · Structure Before Sophistication

Canada’s counter-tariffs took effect this morning. $27.6 billion across roughly 900 tariff lines at 15, 25, and 50 per cent — each rate matched to the corresponding American rate on the same goods. Steel, dairy, appliances, agricultural equipment, pulp and paper, electronics. Canada’s existing steel and aluminum counter-tariffs simultaneously doubled from 25 to 50 per cent.

Section 338 remains fully in force on roughly $20 billion of Canadian goods.

No talks are scheduled. And the reason no talks are scheduled is the most analytically important development of this cycle.

The character of the breakdown has changed

Every previous phase of this dispute involved a substantive gap. Dairy quotas. Auto content thresholds. Section 232 relief on steel and aluminum. Substantive gaps are uncomfortable, but they have a settlement space — there is some combination of concessions that closes them.

What exists now is different in kind.

Carney’s stated condition for resuming negotiations, delivered in Ottawa last week, is that the Americans:

“...stop doing memes, stop throwing shade, stop trying to be tough and start being serious about having those discussions.”

That is not a negotiating position. It is a statement that no negotiating position currently exists, because the counterpart is not behaving as a counterpart. He called recent Cabinet-level mockery of Canada’s military “beneath their office” — pointing to the Treasury Secretary belittling Canada’s economy, the Defence Secretary’s mocking post about Canadian cadets, and the President’s move to rename Lake Ontario “Lake America.”

Here is why that matters for forecasting: there is no concession Canada can make that satisfies a behavioural condition, and no mechanism by which Canada can bring one about. A dispute over terms has a settlement space. A dispute over whether the other side is behaving seriously does not. It resolves only when one party changes conduct — and the party being asked to change is the one facing no domestic pressure to do so before November.

The grades

No dimension moved this issue. Second time in the series. Everything is stable at a low level, which is the honest read — the picture deteriorated in August, and September is that deterioration operating rather than new structural information arriving.

Europe alliance — A held

Trade diversification — A− held

Indo-Pacific pivot — B+ held

Defence & security — B held

Energy & critical minerals — B held

Governance & mandate — B held

Domestic economic architecture — C− held

CUSMA / U.S. relationship — D− held

A = structural problem solved · B = meaningful progress · C = inadequate or stalled · D = regression · F = failure

What’s holding, and what isn’t

Calibration is holding

Beneath the rhetoric, Canadian policy is considerably more precise than the temperature suggests. Three examples from the past two weeks:

● Ottawa dropped fish and seafood entirely from the countermeasure list on August 26, citing risk of broader economic harm to that sector. Retaliation calibrated to avoid self-harm.

● On September 1, Canada removed its retaliatory tariffs on CUSMA-compliant American goods — de-escalating where the agreement still functions while escalating where it has been breached.

● The remission framework remains open for case-by-case relief, letting Ottawa maintain maximum headline retaliation while protecting Canadian manufacturers damaged by their own government’s counter-tariffs.

That is a government defending the CUSMA framework rather than abandoning it. And every prospective trade partner is watching how Canada conducts itself under pressure — which is precisely why “we take this step reluctantly” was the right framing.

The floor is holding

CUSMA remains in force to 2036. Roughly 95 per cent of Canadian exports still move under preference. Energy, potash, critical minerals and fish are excluded from Section 338 — and those exclusions are the clearest available evidence that Washington cannot substitute those inputs. No party has issued a withdrawal notice.

Section 338 hits about 5 per cent of Canadian exports. In aggregate that is not economy-breaking. The damage is concentrated: consumer-facing sectors, small firms, thin margins, limited capacity to restructure. National statistics will understate the distributional reality considerably.

What isn’t holding

● No diplomatic channel is operating. First sustained communication blackout of the cycle.

● Canada’s own counter-tariffs are estimated to cost families with children roughly $250 a year — regressive incidence, landing on households already absorbing 3.2 per cent inflation and 13.4 per cent youth unemployment.

● Steel and aluminum counter-tariffs doubling to 50 per cent raise input costs for Canadian manufacturers, not only American exporters.

● Trump is already threatening a further tariff increase for January 2027 — which is what duration planning looks like.

● Business investment has now declined for five consecutive quarters. No inflection is visible.

Predictability of outcome — Issue 9 revision

Scenario weights updated against seventeen days of evidence since the collapse. Historical base rates from Issue 8 carried forward unchanged.

Grinding stalemate into Q1 2027 — 45–50% (up from 40–45%)

Strengthened, and now the clear base case. No talks scheduled. Carney’s precondition is behavioural rather than substantive. Cabinet-level mockery is escalating rather than moderating. Trump is threatening January 2027 escalation. The midterms remove any incentive for Washington to concede publicly before November. And Ottawa’s $1.5-billion Regional Tariff Response Initiative expansion is duration infrastructure, not bridge financing — governments do not build that unless they expect to need it.

Watching: any working-level contact at all. Deputy USTR Goettman contacting chief negotiator Charette was the July pattern, and that kind of contact would precede any ministerial meeting by weeks.

Reversal within 4–8 weeks of retaliation — 25–30% (down from 30–35%)

Still very much live, and it should not be written off. The Business Roundtable — 200 CEOs of leading American corporations — has publicly warned the tariffs risk raising costs for U.S. businesses and families, and urged both governments to resume negotiations. That is precisely the mechanism that produced the 2020 aluminum reversal in five and a half weeks.

Critically, under that 2020 template the decisive pressure builds two to four weeks after counter-tariffs bite, not before. Canadian tariffs went live this morning. This scenario cannot be scored yet.

Watching: escalation from statements to congressional letters, border-state governors, and sector coalitions filing formal comments. Agricultural pressure especially — Canada’s counter-tariffs hit dairy, agricultural equipment and food processing, which map onto politically concentrated districts.

Rapid comprehensive deal — 8–12% (down from 10–15%)

Weakened. The August framework showed the substance was largely pre-negotiated — the gap was Section 232 relief and the exclusivity demand, not conception. But a behavioural breakdown is harder to reverse than a substantive one, and no channel currently exists through which a rapid deal could be assembled.

Watching: any Trump statement framing a Canada deal as a farmer or auto-worker win.

Structural escalation — withdrawal notice — 8–12% (up from 5–8%)

Elevated. Trump’s threatened January 2027 tariff increase is the first concrete signal of planned further escalation. Congressional authority over withdrawal remains contested, and 13 million American jobs depend on CUSMA trade. But the restraint that has held since the G7 is now being tested by an administration with no diplomatic channel left to lose.

Watching: formal work on the January 2027 measures; any Section 338 expansion beyond the current $20 billion.

A correction I owe you

Issue 8 anchored on September 8 — today — as the highest-information date in this cycle.

That was wrong.

September 8 is the starting gun, not the finish line. Under the 2020 aluminum precedent, the pressure that actually moves Washington builds in the two to four weeks after counter-tariffs bite. The real information arrives in early October, when we learn whether American producer lobbies escalate from public statements to organised congressional pressure.

Scenario weights will be revised again then. A forecast module that never marks its own errors is not a forecast module.

The structural position is unchanged. CUSMA in force to 2036. Ninety-five per cent of exports still moving under preference. Energy and critical minerals still carved out. The floor holds.

What has changed is that both sides are now applying maximum pressure with no channel operating — and every precedent in the record says that condition resolves in quarters, not weeks.

Structure before sophistication. The floor is holding. Nobody is looking for the ceiling.

Cheers, Gerry

Structure Before Sophistication · Sovereignty-First Scorecard Issue 9 · September 8, 2026

The Sovereignty-First Scorecard tracks Canada’s structural response to U.S. isolationism — not Liberal platform delivery. Forward to anyone who needs the unvarnished picture.

On open source AI, the commons, and the thought that was never theirs to sell

Structure Before Sophistication | Substack | September 2026

There is a conversation happening right now about whether AI should be open source — whether the weights, the architecture, the trained intelligence itself ought to be freely available rather than held behind corporate walls.

It’s a real question. It is also the second question, being argued as though it were the first.

Before we decide who should own AI, we have to be honest about what AI is made of and where the intelligence in it came from. Ask that properly and the ownership debate rearranges itself.

You cannot monetize air. And the raw material of AI is something very close to air.

What AI Is Actually Made Of

Every large language model was built on the same foundation: human thought, at scale, gathered without asking.

Books written over lifetimes. Articles published at professional risk. Academic papers that took years to produce and careers to defend. Arguments made in public forums by people who expected nothing from them beyond being right. Poetry, code, philosophy, case law, recipes, forum posts, marginalia. The accumulated intellectual output of a civilization, poured into a digital commons across four decades.

Virtually none of the people who produced that material were asked whether it could be used as training data. Most were writing before the category existed. They put their thinking into the public sphere because that is what thinkers do — the contribution is the act.

That contribution became the substrate of an industry now capitalized in the trillions. The raw material was free. The people who produced it received nothing. And access to the consolidated version of our own collective intelligence now carries a monthly fee.

The Pipe and the Air

Here is where the argument usually goes wrong.

Someone says: the companies built the infrastructure. Compute, engineering, architecture, research — billions of dollars and years of work. You can’t expect them to give that away.

That’s true, and it’s beside the point.

What got built is a consolidation engine. A mechanism for harvesting, cross-referencing and synthesizing what was already distributed across the public sphere. The answers were there. The ideas were available. What did not exist was a system that could assemble them around a question fast enough to be useful.

That system is genuinely valuable and the engineering behind it is real. Nothing in the provenance argument requires pretending otherwise.

But the air was free before the pipe was built. Building the pipe does not make the air yours.

You can charge for the pipe. You cannot own what flows through it.

What Open Source Actually Settles

On August 10th, Mark Zuckerberg published a 6,500-word essay making the case for open-weight AI, alongside Meta’s commitment to release the weights for Muse Spark 1.2 — its most capable model — and a new family called Muse Glimmer, built small enough to run on a laptop with a single consumer GPU. The argument, compressed: powerful AI shouldn’t sit in the hands of a few companies, American open models need to outrun Chinese ones, and Washington should get out of the way.

The cynical read is that open source is Meta’s competitive weapon. If the model is free, the moat moves to distribution, data and compute, which is terrain Meta already holds. A strategic position wearing an ethical coat.

The less cynical read is that he’s right about the pipe, for better reasons than his own. If the training corpus was humanity’s collective output, the model probably should be a commons — not because Meta is being generous, but because the raw material was never anyone’s to enclose.

Read the essay closely, though, and a second argument runs underneath the first. Zuckerberg’s specific policy request was that the United States rethink its rules on training data and on distillation, on the grounds that American labs operate under restrictions their foreign competitors don’t face. US policy, he wrote, must reduce this additional friction if American open-source models are going to lead.

Sit with that. The clearest public advocate for opening the pipe is simultaneously asking the state to loosen the restrictions on what may be poured into it. The generosity is downstream. The ask is upstream. And the upstream ask is for easier access to precisely the material whose provenance is in question.

So: what does open source settle?

It democratizes the pipe. Any researcher, any developer, any organization can run the model without paying a toll, and that is a real gain — the difference between a utility and a tollbooth.

It does not touch provenance. The writers whose life’s work became training data still receive nothing. The people whose ideas are being synthesized and sold back to them still hold no stake in what was built. Open source distributes access to the output. It returns nothing to the people who produced the input. Paired with a lobbying position aimed at loosening the rules on input, it starts to look less like a commons and more like a faster enclosure with better public relations.

The Enclosure, Again

I wrote about this at length in The New Enclosure, so I’ll keep it short here.

Common land in England — grazing ground, forest, waterway, held and worked by communities for generations — was progressively fenced by those with the legal machinery to claim title and charge for access. The people whose labour had made that land productive became tenants of what they had once held in common.

The structure transfers cleanly. The commons was the public sphere. The enclosure was the training run. The landowners are whoever reached sufficient compute first. And as in the original, the people most affected are the last to be compensated and the first to be displaced by what was built out of their work.

The difference is speed. The English enclosures ran for three centuries. This one took about four years.

From Inside a Business

I started my company in 2000. When the financial crisis arrived in 2008 we did a hard pivot into government-procured work, positioning ourselves for the federal infrastructure stimulus. That decision is the reason the business is still standing.

It worked for a reason worth naming. As founder I had built two things that don’t usually sit together: a workforce with the mass to execute at scale — as powerful as a dragon boat — and a management team tight enough to change direction inside a single quarter. As nimble as a kayak. Power in the hull, judgement in the hands of people who could act on it without asking permission.

That combination is what survives a structural shock. Power on its own doesn’t; plenty of large, well-capitalized firms went under in 2009 because they couldn’t turn. Agility on its own doesn’t either, because without capacity behind it, agility is just a fast opinion.

Which is why the current moment concerns me in a particular way. What these systems take first isn’t the paddling. It’s the steering.

Displacement doesn’t arrive as an announcement. It arrives as a quiet change in what a role is permitted to decide. Someone who priced work out of twenty years of knowing what ground does in March gets a system that prices it for him. For a while he checks the system. Then the checking stops being part of the job. Then the job is administering an output. Nobody is fired and no announcement is made, but the judgement has left the building — and an organization that has lost its judgement is a dragon boat with nobody in the stern.

What’s happening to knowledge workers now is that same process at a higher altitude and considerably faster. The displaced cognitive class isn’t losing its jobs. It’s losing the part of the job that constituted the expertise, and discovering afterward that what remains doesn’t command what the whole of it did.

What Restitution Would Actually Require

If the raw material was human thought, the live question isn’t open versus closed. It’s provenance, and what follows from it.

The proposal I find most interesting — and I want to be careful here, because it is a proposal and not a trend — is some form of contribution stake. Tokens, equity, a denominated share: a mechanism by which the people whose intellectual output enters the system hold a claim on what the system becomes worth. Mining, where what gets mined is human intelligence and the miners are on the register.

I find it interesting. I don’t yet find it convincing, and the objections are the obvious ones.

Who prices a contribution? Any mechanism that pays for quality of intellectual contribution needs someone to define quality, and the only parties holding the data to do it are the parties who did the enclosing. Who arbitrates a dispute? The same infrastructure. And what prevents the whole thing from becoming a loyalty program — a scheme that converts a property claim into a discretionary reward, which is the cheapest available settlement of the most expensive available liability?

Those aren’t reasons to abandon the idea. They’re the design constraints. Any restitution mechanism that can’t answer all three is enclosure with a rebate attached.

What holds underneath the mechanism question is that the value chain runs through human consciousness. The ideas, the questions, the frameworks, the creative leaps — without those the consolidation engine produces nothing of substance, only recombination of what it was handed. The pipe is valuable. The air made it worth building.

The Question Worth Sitting With

Free thought is the most fundamental human liberty. The right to form an idea, hold a position, change your mind, argue in public, add to what your civilization knows.

That freedom produced everything that trained every AI system in existence.

So the question isn’t whether AI should be open source. It’s whether the enclosure of human thought was legitimate in the first place, and what an honest reckoning with that requires — from the companies that benefited, from the regulators who watched, and, in the part nobody wants, from the thinkers themselves.

Because the obligation runs both ways.

If the commons is what made this possible, then the people who benefit from it owe it their continued presence. The instinct to stop publishing, to move the work behind a paywall, to keep the serious thinking off the open web is understandable — and it completes the enclosure on the enclosers’ behalf. A commons everyone defects from hasn’t been defended. It’s been abandoned.

The thinkers owe the commons their continued presence in it. The commons owes them a stake in what was built out of it. Neither obligation cancels the other, and at the moment only one of the two is being honoured.

If human thought is the raw material, what do the thinkers owe? And what are they owed in return?

That’s the subject of the next conversation, and of the book I’m writing now. The Metaproletariat argues that what’s happening to the displaced cognitive class isn’t new. It’s the oldest story in economic history, wearing a different face.

The enclosure has always preceded the displacement. The displacement has always preceded the reckoning.

We are somewhere between the first and the second.

Structure before sophistication. Every time.

— Gerry

Structure Before Sophistication | September 2026

• Edits and revisions done with the assistance of CLAUDE

The exclusivity clause

In the last hour of negotiations, the United States asked Canada to stop shopping. That is not a trade term. That is what a general contractor asks of a subcontractor he intends to own.

Gerry · August 23, 2026 · Structure Before Sophistication

The talks did not collapse over dairy. They did not collapse over steel, or aluminum, or the price of softwood lumber. Those files were hard, and they were unresolved, and by Friday afternoon they were still on the table.

They collapsed over something else entirely — introduced, by Carney’s account, in the final hour.

“The U.S. introduced at the last hours, in the last hour, efforts to restrict our ability to have other trade deals. We believe in free trade; we’re the partner of choice in many respects for countries around the world, and the Americans wanted to restrict that.”

Asked what he thought Washington was trying to accomplish, Carney did not hedge.

“It’s a power play. It’s a power play and, you know, it becomes a question of sovereignty.”

He is right. And the shape of the play is more familiar than the language of trade policy makes it sound.

The general contractor

Anyone who has run a trade knows this arrangement.

A general contractor brings on a subcontractor. Good work, fair price, steady volume. The relationship deepens. The GC starts feeding the sub more jobs — enough that the sub hires up, buys equipment, expands the yard to handle the flow.

Then comes the conversation. Nothing hostile. Just business. We’re giving you a lot of work. We’d like to know you’re available when we need you. We’d rather you weren’t taking on other GCs.

Exclusivity. Framed as partnership.

And it works, for a while. The volume is real. The cheques clear. But the sub’s capacity is now committed, his crews are booked, and his rolodex of other clients has gone cold from disuse. He has no active bids anywhere else because he has had no capacity to chase them.

Which is when the margin conversation starts.

The GC does not need to squeeze hard. He only needs to squeeze steadily — because he knows the sub has nowhere to go this quarter, and that knowledge is worth more than any single price concession.

Five points off the next job. Longer payment terms on the one after. A change order absorbed rather than billed. Each individual ask is survivable. Refusing any one of them risks the whole book of business. So the sub absorbs it, and absorbs the next one, and the margin ratchets down.

The endgame is not a bankruptcy. It is quieter than that. Eventually the sub is working at a margin that cannot fund replacement equipment, cannot fund a bid team, cannot fund the slack capacity required to pursue anyone else. He is fully employed and structurally insolvent. And at that point the GC makes an offer for the business — at a valuation that reflects a company with one customer.

The sub does not get acquired because he failed. He gets acquired because he was made unable to leave.

This is not a metaphor. It is the same mechanism.

Strip the construction language and the economics are identical.

Exclusive dealing and foreclosure

Competition authorities on both sides of the border scrutinize exclusive dealing arrangements for a specific reason: they do not merely allocate business, they foreclose alternatives. A supplier bound to one buyer loses the outside options that discipline that buyer’s conduct. The harm is not in any single transaction — it is in the progressive elimination of the supplier’s ability to walk away.

Canada already sells roughly seventy-two per cent of its exports into one market. That is not a diversified customer base. That is a firm with one dominant client and a thin tail.

Ottawa has spent eighteen months doing exactly what any competent operator in that position would do: chasing other work. Twenty new economic partnerships. An FTA with Indonesia. A seventy-billion-dollar investment memorandum with the UAE. Live talks with India. Deepening trade with Mexico, now at sixty-two billion dollars a year. Not because any of it replaces the United States tomorrow — it does not — but because a second customer changes what the first one can demand.

The last-hour demand was aimed precisely at that.

Not at dairy quotas. Not at auto content thresholds. At the bid pipeline. At Canada’s ability to be in the room with anyone else.

The ratchet was already installed

Here is the part that should worry Canadians most, and it has barely been mentioned in the coverage.

This demand was not an innovation. It was an extension.

CUSMA already contains Article 32.10. Signed in 2018, ratified by Parliament, in force today. It requires any party to give three months’ notice before opening free trade negotiations with a “non-market economy,” to share the text before signature, and — critically — it gives the other parties the right to withdraw from CUSMA on six months’ notice if they object to the resulting agreement.

At the time, trade lawyers called it astonishing. A University of British Columbia professor described it as pulling Canada into line with American strategic posture. The Chinese embassy called it an act of political dominance. Ottawa said it would likely never be used.

Peter Clark, the trade strategist, disagreed at the time in language that has aged well: it is there because somebody will want to use it.

In 2018 Canada conceded a veto over its trade relationships with non-market economies. In 2026 the United States came back for the same veto over market economies. That is the ratchet, running exactly on schedule.

The GC did not ask for exclusivity on the first job. He asked for right of first refusal. Then priority scheduling. Then a non-compete on the client list. Each step made the next one seem incremental — because it was incremental, and because the sub had already accepted the principle underneath it.

The principle underneath Article 32.10 is that Washington holds legitimate oversight of Canada’s other commercial relationships. Once that principle is conceded, the argument is only ever about scope. And scope moves in one direction.

Why exclusivity is the endgame, not a tactic

It is worth being precise about where this leads, because the trajectory is not speculative. It is the ordinary logic of dependency.

A country that cannot diversify its export markets cannot improve its terms of trade. A country that cannot improve its terms of trade watches margin compress across every tradeable sector. Compressed margins mean less capital available for the infrastructure, the market development, and the relationship-building that diversification requires in the first place.

That is a closed loop. Each turn makes the next escape harder to finance.

Carry it far enough and the questions stop being commercial. When a country’s fiscal capacity, employment base, and industrial policy are all downstream of a single foreign buyer’s discretion, sovereignty becomes nominal before it becomes contested. The formal instruments remain. The room to use them does not.

Carney named this correctly. It becomes a question of sovereignty. Not immediately. Not dramatically. But structurally, and on a timeline measured in decades rather than election cycles.

Which is why the auto terms and the culture and language protections were bundled into the same last-hour package. Read the three demands together — restrict the other trade deals, make Canadian vehicle production progressively uneconomic, weaken the protections for language and culture — and they are not three separate asks. They are one ask, expressed three ways.

The J-curve nobody wants to campaign on

Now the hard part, and the reason this piece is not a victory lap.

Walking away was correct. It was also expensive, and the expense is front-loaded.

The diversification J-curve

Building an export relationship is capital formation, not a sales call. It requires port and rail capacity pointed in new directions, regulatory approvals and standards alignment, distribution partners, service networks, financing arrangements, and — most slowly of all — the commercial trust that makes a foreign buyer choose an unfamiliar supplier over an incumbent. The costs land in years one through four. The revenue arrives in years five through ten. The curve goes down before it goes up.

Every serious observer of this file, including those sympathetic to Ottawa, makes the same point: Canada is a victim of geography. Ninety per cent of the population lives within a hundred and sixty kilometres of the American border. The rail runs north-south. The pipelines run north-south. The transmission lines run north-south. Two centuries of infrastructure investment assumed one customer.

Reorienting that is a generational project, and the first half of it is pure cost with no visible return. Which is precisely why it has never been done — every government that has contemplated it has faced an electoral cycle shorter than the payback period.

So the honest framing for Canadians is not that diversification is an alternative to short-term pain. Diversification is the short-term pain. The question is only whether the country absorbs it deliberately, on its own schedule, with its own sequencing — or absorbs it later, involuntarily, on terms set by someone else.

There is no version of this where Canada gets to keep the current standard of living, keep full sovereignty over its trade policy, and avoid the transition cost. Those three things stopped being simultaneously available some time ago.

What Carney actually did

Strip away the diplomacy and Friday night was a subcontractor declining an exclusivity clause.

He knew the consequences. Fifty per cent duties on twenty billion dollars of goods, live at 12:01 Saturday morning. Hockey sticks, wine, cement, dairy, furniture — consumer-facing products from small firms with thin margins and no capacity to restructure supply chains. Real jobs, quickly.

Canada had already moved on the things it was willing to move on: dropping most of last year’s retaliatory tariffs, returning American alcohol to provincial shelves. Those were genuine concessions and they were on the table.

What Canada would not do was sign the exclusivity. On critical minerals specifically, Carney was categorical — Canada would never grant the United States exclusive access.

That is the right instinct, and it is worth understanding why in the terms of the analogy. A subcontractor who signs the exclusivity clause has traded a hard problem for a fatal one. The hard problem is a difficult year with reduced volume while he rebuilds a client base. The fatal one is a permanent structural position in which his margin, his capacity, and eventually his ownership are all determined by someone else’s discretion.

The hard problem is survivable. The fatal one only feels survivable, right up until it is not.

What comes next

Three things are worth watching, and none of them are the tariff numbers.

First, whether the September 8 retaliation moves American producer lobbies. The 2020 aluminum reversal took five and a half weeks and was driven by exactly this instrument. Canada cannot win a terms-of-trade fight with an economy ten times its size. It can make the political cost visible to concentrated constituencies who can pick up a phone.

Second, whether the exclusivity demand survives the collapse. If it reappears in the next round — and the Article 32.10 precedent suggests it will — that tells Canadians this was strategy rather than a late-night overreach. Watch for it in narrower, more palatable form. That is how ratchets work.

Third, whether Canada actually accelerates the diversification build. Rhetoric about being a middle power is cheap. East-west rail capacity, port expansion, transmission interties, processing facilities that serve more than one market, standards alignment with more than one regulator — these are budget lines, not speeches. If the next fiscal update does not contain them, then Friday was a gesture rather than a strategy.

The subcontractor who declines the exclusivity clause has a bad quarter. Possibly a bad year. He has to go find work he has not had to find in a long time, at prices he will not enjoy, from clients who do not know him yet.

The subcontractor who signs it has a comfortable year, and then a slightly worse one, and then a slightly worse one after that. And somewhere around year seven he discovers that the only buyer for his business is the man who made sure there were no other buyers for his services.

Carney chose the bad year. Whether that turns out to be courage or merely a delay depends entirely on what Canada builds between now and the next time the clause comes back to the table.

Because it will come back to the table.

Structure before sophistication. Build the second customer before you need him — because the moment you need him, you can no longer afford to go looking.

Gerry Visco

Structure Before Sophistication · CUSMA Companion Series · August 23, 2026

Companion to Sovereignty-First Scorecard Issue 8. Quotations from Prime Minister Carney’s press conference of August 22, 2026, as reported by Global News, CNN, PBS NewsHour, and Fox Business. Article 32.10 analysis draws on contemporaneous 2018 reporting from The Globe and Mail, CBC, and Global News.

#CUSMA #CanadaUSTrade #TradePolicy #Sovereignty #Section338 #CanadianEconomy #TradeDiversification #StructureBeforeSophistication

Please note this piece was put through Claude AI for final edits and presentation.

The floor cracked

Trump reached around CUSMA with a 1930 law, Canada made concessions under maximum pressure, and a deal landed with two hours to spare. Here’s what actually happened — and what doesn’t reverse.

August 19, 2026 · Structure Before Sophistication

At 10:22 PM on August 18, with the deadline less than two hours away, Trump posted this on Truth Social:

“I have paused the 50% Tariffs against Canada, that were scheduled to kick in tomorrow morning for a three day period, based on the fact that Canada and the U.S.A., subject to the finalization of documents, have a DEAL!”

Carney’s statement came shortly after. Substantial progress had been made. Important work remained. Canada had agreed to the three-day pause.

Two characterizations of the same event. Trump: a DEAL. Carney: substantial progress, important work remained. That gap is not stylistic. It is the most informative thing about the past 24 hours, and it defines what the next 72 hours are actually for.

What happened on July 20

To understand why this week mattered, go back a month.

On July 20, Trump signed three proclamations invoking Section 338 of the Tariff Act of 1930. It imposed 50% additional tariffs on roughly $20 billion of Canadian goods — dairy, alcohol, motor vehicles, and hundreds of adjacent categories running from wine to cement to hockey sticks.

Section 338 had never been used by any U.S. president. Not once, in ninety-six years. It requires no investigation. It has no duration cap. The president imposes it by proclamation alone, and can suspend, revoke, or reimpose it the same way.

And it applies regardless of CUSMA eligibility.

That last sentence is the whole story. Throughout this entire review cycle — through the July 1 non-renewal, through Trump’s “terminated” comments at the G7, through six months of annual review uncertainty — one thing held. The CUSMA exemption. Canadian goods that qualified for CUSMA treatment crossed the border tariff-free. Fifty-three percent of Canadian goods exports claimed that preference. It was the floor.

On July 20, the United States demonstrated it could reach around that floor entirely.

The CUSMA exemption was not repealed. It was bypassed. Those are different things, and the second one is harder to defend against.

Six weeks in Washington

What followed was the most intensive stretch of Canada-U.S. trade negotiation in a generation.

LeBlanc and Charette spent the better part of three weeks in Washington. Canada went in with three objectives: shelve Section 338 entirely, secure Section 232 relief on steel, aluminum, autos, and softwood lumber, and formally launch bilateral CUSMA renewal talks — the ones Mexico had been having since March and Canada still hadn’t started.

The Globe and Mail reported as late as Friday, August 15, that the U.S. side was not open to granting meaningful tariff relief on Canada’s priority sectors. Canadian negotiators were pushing for all four. The Americans were not offering. And the U.S. side gave no indication it would extend the August 19 deadline.

Charette told Greer directly that imposing Section 338 would represent a “cliff” — that Ottawa would be forced to retaliate and the negotiating framework would fracture. Two sources later confirmed she delivered exactly that warning. One characterized Canada’s message as: nothing holds anymore.

On Friday, while Canada’s chief negotiators were in conference rooms, Greer was at the Iowa State Fair. Senator Chuck Grassley pulled him aside with a message of his own: these negotiations cannot be used as a way of destroying CUSMA. Congressional patience with the disruption is not unlimited.

Sunday brought a one-hour virtual meeting between LeBlanc, Charette, and Greer. Described as constructive. Greer left saying “no comment.” LeBlanc said the job was not yet done.

Then Tuesday. Two Carney-Trump phone calls in a single day. And at 10:22 PM, the Truth Social post.

What we know is in the deal

The USTR outline is thin but specific: comprehensive market access for all American goods, economic security commitments, digital trade alignment, and — in the language of the announcement — many important provisions.

One concession is confirmed. Canada scrapped its levy on U.S. streaming giants. That addresses one of the four digital sovereignty grievances the USTR added to its 2026 National Trade Estimate. The Online Streaming Act levy is gone.

And then there’s the line at the end of Trump’s post, almost thrown away: Keystone XL “may be awoken from the grave!”

That is either a rhetorical flourish designed to frame the deal as a major win, or it is an actual term. If it is a term, Canada has agreed to an energy infrastructure commitment that locks Canadian oil export dependency to the U.S. market for decades — precisely the kind of structural energy concession Ottawa spent six months explicitly refusing to make proactively. Foreign Affairs Minister Anand had held the line that Canada would not sign standalone energy or critical minerals deals outside the CUSMA framework. Carney had ruled out oil as a retaliatory tool and as a bargaining chip.

We don’t know which it is yet. August 22 tells us.

What we don’t know — and it’s the part that matters

Section 232 relief.

Steel at 50%. Aluminum at 50%. Autos at 25%. Softwood lumber at 45%-plus. These are the tariffs that have been grinding Canadian manufacturing for eighteen months. Canadian manufacturing lost 32,161 jobs between January 2025 and January 2026, with motor vehicle parts accounting for 7,294 of them.

As of Friday, the U.S. was not offering relief on any of the four.

If Section 232 relief is in the finalized documents — even partial, even phased — it is the single most significant outcome of the entire CUSMA review cycle, and it changes the analytical picture completely.

If it is not, then Canada has traded digital sovereignty concessions and possibly a major energy infrastructure commitment for a freeze on new tariffs. Not relief from existing ones. A freeze.

Those are very different deals wearing the same word.

The scorecard — two moves

Governance & mandate: B → B+. The first upgrade in the series. This is earned. Six weeks of sustained pressure, a clearly communicated red line, restored leader-level access at the decisive moment, provincial coalition held together through Team Canada briefings, and a provisional framework delivered with two hours to spare. Whatever the substance turns out to be, the machinery performed. LeBlanc and Charette delivered the most consequential negotiating result of the review to date.

Energy & critical minerals: B, watch flagged. Held, but amber. The Keystone XL reference is unresolved and potentially significant. If it is a confirmed term, this grade moves next issue — and not upward.

CUSMA / U.S. relationship: D, held. No deterioration. No recovery either. The provisional deal prevents the worst outcome. It does not constitute progress toward the thing Canada actually needs: Section 232 relief, confirmed Section 338 withdrawal, and a structured path back toward 16-year extension. None of those are confirmed. The D holds because the floor held — barely, and only because of a last-minute deal.

Everything else stable. Europe A. Trade diversification A−. Indo-Pacific B+. Defence B. Domestic economic architecture C.

The thing that doesn’t reverse

Suppose August 22 goes perfectly. Documents finalized, Section 338 permanently shelved, meaningful Section 232 relief secured, Keystone XL turns out to have been rhetoric. Best case.

The precedent still stands.

For the first time in the modern trade era, the United States invoked a Depression-era statute to stack 50% duties on goods that qualify for duty-free treatment under a ratified trade agreement. No investigation. No cap. Reimposable by proclamation whenever it’s useful. The Cato Institute named the risk immediately: Section 338 could become the next presidential tariff button, wielded the way IEEPA was before the Supreme Court shut that door.

Every annual review from here forward happens in a world where Canada knows this. Every investment decision in every CUSMA-exposed Canadian sector now carries a risk premium that didn’t exist on July 19. Not for the duration of this negotiation — permanently, or at least until a future administration decides otherwise.

Hufbauer described the annual review decade as water torture. This is a different instrument. The annual review erodes certainty slowly, one round at a time. Section 338 demonstrated that the U.S. can also simply go around the agreement whenever the slow method isn’t producing results fast enough.

Canada’s structural floor now has a known crack in it. The deal may hold. The crack doesn’t close.

What to watch

August 22. Three days. Specifically:

Is Section 338 permanently withdrawn or merely paused again? Is there Section 232 relief on any of the four priority sectors, and is it real reduction or symbolic adjustment? What exactly does “digital trade alignment” cover beyond the streaming levy — is the Sovereign Cloud Initiative in scope? Is Quebec’s Bill 109? And is Keystone XL an actual term or was it a flourish?

The answers determine whether Issue 8 records the first genuine recovery of this series or the confirmation that Canada paid real concessions for a temporary reprieve.

Structure before sophistication. The structure held this week — under the heaviest pressure it has faced. But it held with a crack in it, and the sophistication now required is making sure the price Canada paid buys something durable rather than seventy-two more hours.

We’ll know Saturday.

Gerry

Structure Before Sophistication · Issue 7 · August 19, 2026

The Sovereignty-First Scorecard tracks Canada’s structural response to U.S. isolationism — not Liberal platform delivery. Forward to anyone who needs the unvarnished picture.

#CUSMA #Section338 #CanadaUSTrade #Carney #KeystoneXL #Tariffs #CanadianEconomy #TradePolicy #StructureBeforeSophistication

A Comparative Framework: 1927–1932 and 2024–2026

Structure Before Sophistication — July 2026

Not investment, legal, or financial advice.

Everyone wants to know if 2026 is 1929. It’s the wrong question, and asking it wrong is how you end up either dismissing the comparison entirely or panicking over it. The better question is whether 2026 has assembled the same conditions that made 1929 dangerous — and if so, which ones, and how many of the ones that made 1929 catastrophic.

Here’s the answer up front, because I don’t believe in burying a thesis under three thousand words of throat-clearing: 1929 was two events stacked on top of each other, not one. There was a bubble. And then there was a policy failure that turned a bad correction into the Great Depression. 2026 unmistakably has the first. The second is still a choice nobody has made yet. That asymmetry is the whole argument. Everything below is just showing my work.

The way to test that properly is boring on purpose: figure out what actually happened between 1927 and 1932, figure out where we actually stand today, put the two side by side without flinching from either the parts that rhyme or the parts that don’t, and then say — in advance — what evidence would prove this wrong. A framework that can’t tell you what would falsify it isn’t a framework. It’s a vibe.

What Actually Happened, 1927–1932

The Dow roughly doubled between early 1928 and its peak of 381.17 on September 3, 1929. It was a narrow, narrative-driven rally — radio, aviation, automobiles, and electric-utility holding companies were the era’s glamour trade. RCA, the definitive tech stock of the age, rose several-fold in under two years on the promise of a genuinely transformative technology: broadcasting. It then lost roughly 98% of its value peak to trough. Worth sitting with that for a second — the technology was real. The price was not. Those are two different claims, and 1929 is the clearest historical case of investors conflating them.

What made the boom dangerous wasn’t the story — it was the plumbing underneath it. Investors could buy stock on margin of roughly 10%, and brokers’ loans reached about $8.5 billion by late 1929, somewhere around 8 to 10% of U.S. GDP, much of it funded not by banks but by corporations and foreign lenders chasing call-loan rates. That’s an early shadow-credit channel, and it matters later. Investment trusts — the era’s structured products — pyramided leverage on leverage, trusts holding trusts holding margined stock, with utility holding companies (Samuel Insull’s empire being the textbook case) adding another layer of debt at every tier. By September 1929 the Shiller CAPE ratio hit roughly 32 to 33 — at the time, the highest reading ever recorded, about double its long-run median.

The Fed had been tightening since 1928 specifically to choke off the speculation, pushing the discount rate to 6% by August 1929. The market broke on October 24 — Black Thursday — and came apart on October 28 and 29, down roughly 13% and 12% on consecutive days. Mechanically, it was the margin structure doing exactly what margin structures do: falling prices triggered calls, calls triggered forced selling, forced selling triggered more falling prices. The call-loan market evaporated fastest of all, because the corporations and foreign lenders funding it had no relationship reason to stick around once the trend turned.

Here’s the part almost everyone gets wrong: the October crash did not, by itself, cause the Great Depression. By April 1930 the Dow had clawed back nearly half its losses. What turned a severe correction into a systemic collapse was three separate policy failures, stacked on top of each other over the following two years.

• Trade. The Smoot-Hawley Tariff Act, signed in June 1930 over the objection of more than a thousand economists who petitioned against it, triggered retaliatory tariffs around the world. Global trade contracted by roughly two-thirds between 1929 and 1934.

• Money and banking. The Fed allowed the money supply to shrink by about a third and largely stood by through repeated banking panics. Roughly 9,000 U.S. banks failed between 1930 and 1933. There was no deposit insurance yet, so depositor losses were total — and every panic rationally fed the next one.

• The gold standard. Fixed exchange rates transmitted the deflation internationally. Countries defending their gold parities were forced to tighten into the collapse rather than ease. Britain left gold in September 1931; the countries that abandoned it earliest recovered soonest, which is about as clean a natural experiment as economic history offers.

The Dow bottomed on July 8, 1932, at 41.22 — down 89% from the peak. U.S. output fell roughly 30%. Unemployment hit roughly 25% by 1933. Politically, the depression became the incubator for the entire following decade: protectionism hardened into autarky, democracies buckled under the strain, and the 1930s ended in world war. That lag is worth remembering when we get to scenarios — the worst political consequences of 1929 arrived five to ten years after the financial event, not immediately after it.

Where We Actually Stand in Mid-2026

Now the other half of the comparison. Where does mid-2026 actually sit against that record, r

Valuation. Start with valuation, because it’s the simplest and the starkest. The Shiller CAPE ratio sits at roughly 41 to 42 as of mid-July 2026. The all-time high is 44.2, set in December 1999; the long-run median since 1881 is 16.1. That means today’s market is already more stretched than 1929 ever was — `` peaked around 32 to 33 in September of that year — and has only been matched once in history, in the closing months of the dot-com bubble.

Concentration. The top ten stocks in the S&P 500 now represent roughly 35% of the index’s total market capitalization, versus about 25% at the 2000 peak. Apollo’s chief economist Torsten Sløk has a good name for this: the diversification illusion. Buy the index today and you think you own five hundred businesses. In practice you own one thesis — AI infrastructure monetization. The genuine mitigating factor is that the Magnificent Seven’s net margins run above 25%, against roughly 13% for the index as a whole. These are real, cash-generative businesses in a way RCA’s holding-company pyramids and 2000’s revenue-free dot-coms were not. That distinction matters, and I’ll come back to it.

Leverage. This is where it gets genuinely uncomfortable. Margin debt crossed $1.1 trillion earlier this year and is still climbing — and the more diagnostic number isn’t the level, it’s the rate of change, which has only been matched three times since 1997: late 1999 into 2000, mid-2007, and spring 2021. You know how each of those stories ends. Leveraged ETF assets hit roughly $220 billion in the first half of 2026. Add options gamma, equity repo, and an estimated 60% of trading volume running through algorithms, and you have a forced-selling machine that is functionally the direct descendant of 1929’s margin-call cascade — different instruments, same reflexivity.

Then there’s the layer 1929 didn’t have: corporate credit. Hyperscaler capital expenditure estimates for 2026 have been revised up from roughly $650 billion to $725 billion, with Moody’s projecting $785 billion this year and a trillion in 2027 — increasingly debt-financed, with Amazon’s reported $25 billion debt sale being one visible marker. Layer onto that the circular vendor-financing pattern where AI companies invest in each other’s products and book the resulting revenue, some of which may not reflect real end demand. Enterprise AI software revenue currently covers less than half of the infrastructure spend it’s supposed to justify. This is this era’s version of the investment-trust pyramid — not fraudulent, but structurally dependent on a monetization curve that hasn’t shown up yet.

Trade. 2025 brought sharp, broad U.S. tariff increases applied to allies and rivals alike, with more volatility running through 2026. The WTO has cut its 2026 global trade growth forecast to 0.5%, down from 1.8%, and warns that a full U.S.-China decoupling could shave roughly 7% off long-run global output. UNCTAD projects 2.6% global growth this year, with U.S. growth slowing toward 1.5%. Here’s the structurally interesting part: the Smoot-Hawley-style shock has arrived before any crash this time, not after one — and markets have, so far, absorbed it. Whether that means the damage is priced in or merely deferred is one of the live questions this whole framework is built to answer.

Geopolitics. The backdrop is simply hotter than 1929’s. U.S.-China strategic decoupling keeps deepening. Iran was signaling potential Red Sea disruption in the very week I was writing this, moving equity indices intraday. Tariffs are being wielded against allies — Japan, the EU, Canada, Mexico — as general-purpose statecraft, and the multilateral guardrails, the WTO’s dispute mechanism chief among them, are functionally degraded. In 1929, the geopolitical deterioration followed the economic collapse by years. In 2026, it’s running concurrently — which removes a buffer the interwar sequence actually had.

Side by Side

Laid out variable by variable, here’s how it scores:

Read plainly: the pre-conditions rhyme hard — valuation, concentration, leverage, narrative, protectionism. The propagation machinery does not: no gold standard, no uninsured banking system, a central bank that acts fast rather than standing aside. 1929’s outcome required both the bubble and the policy failure. 2026 unmistakably has the first. The second remains a choice nobody has made yet. That asymmetry is the entire basis for the scenario weights further down.

Where the Comparison Breaks Down

Three real disanalogies do most of the work here, and any honest version of this comparison has to carry all of them.

• The policy reflex is inverted. The 1929–33 Fed tightened into a collapse and let a third of the money supply disappear. The post-2008 institutional instinct is the opposite — if anything, too quick to backstop. That truncates the ugly left tail: an 89%-style wipeout requires the state to stand aside for three years, and that is simply not the modern revealed preference. The cost of that reflex is chronic moral hazard and asset-price inflation — which is partly how CAPE got to 42 in the first place. The safety net and the bubble are, in an uncomfortable way, the same mechanism viewed from opposite ends.

• The leaders actually earn money. RCA in 1929 and Pets.com in 2000 were priced on futures with no cash flow behind them. Today’s leaders generate enormous free cash flow at 25%-plus net margins. That narrows the bubble question considerably: it’s not ‘are these real businesses,’ it’s ‘can $725 billion to a trillion dollars a year of infrastructure spending get monetized before the debt funding it reprices.’ A miss produces a violent sectoral repricing — the 2000 template — rather than an economy-wide solvency crisis on its own.

• There’s no gold standard. The interwar depression went global because fixed exchange rates forced synchronized deflation onto every country defending its parity. Floating currencies and independent central banks mean a 2026 shock would propagate through credit spreads, the dollar, and trade volumes instead — faster in transmission, but with more policy degrees of freedom at every node along the way.

And one thing cuts the other way. In 1929 the geopolitical order was intact when the crash hit, and the decade still ended in world war. In 2026 the trade war and the security deterioration are already running before any financial rupture has even happened. If a major correction does arrive into an environment of tariff walls, degraded institutions, and live conflict risk, the interwar sequence compresses — the financial and political feedback loops start operating simultaneously instead of one after the other.

The Centerpiece: Private Credit as the New Call-Loan Market

If there’s one place I’d tell you to actually focus, it’s this section. An earlier pass at this framework flagged private credit as its weakest data point — an analogy resting mostly on the fact that the sector is opaque, which is not the same thing as evidence. That’s no longer an honest description. As of the first half of 2026, the channel is producing observable, dateable stress signals, which upgrades this comparison from metaphor to live evidence.

Size first. U.S. private credit crossed $2 trillion in assets under management during 2025, with the Fed’s own May 2026 Financial Stability Report sizing it conservatively at $1.5 to $2 trillion; Preqin projects $2.64 trillion by 2029. For scale: brokers’ loans at the 1929 peak ran roughly 8 to 10% of GDP. Private credit today runs roughly 5 to 7% of U.S. GDP — the same order of magnitude. The retail-facing, semi-liquid wrapper for a lot of this — business development companies, or BDCs, with quarterly redemption windows sitting on illiquid loan books — had accumulated roughly $530 billion globally heading into this year. The fastest-growing corner of the whole asset class, at roughly 13% of strategies, is NAV financing: lending against the marked value of private fund portfolios. Leverage, collateralized by model-derived valuations of other leverage.

The structural mapping to 1929 is close enough to be uncomfortable. In both eras, this is the dominant non-bank funding channel — for levered equity then, for levered corporates now. In both eras, the funding source has no relationship reason to stay: corporations and foreign lenders chasing call rates in 1929, retail and semi-liquid capital in quarterly-redemption wrappers today. Where it diverges is pricing visibility and speed. The call-loan market was transparent and instantaneous — everyone could watch the run happen in real time in October 1929. Private credit is opaque and gated — loans marked to model quarterly, with public proxies like BDC share prices actively disputing those marks. And where 1929’s call loans were callable on demand and evaporated within days, today’s redemptions are capped at roughly 5% per quarter by fund structure — converting a potential run into a slow bleed instead of an instant collapse, at the cost of a refinancing freeze for whoever’s on the other side of that loan.

Which failure mode is actually worse is genuinely unresolved. 1929’s was violent and visible. The modern version risks a long period where nobody — including the lenders themselves — knows the true size of the hole.

And the 2026 stress record is no longer hypothetical. Average redemptions from perpetual non-traded BDCs jumped from 1.6% of net asset value in the third quarter of 2025 to 4.8% in the fourth, with requests exceeding the standard 5% quarterly caps starting in early 2026. Blue Owl’s funds saw exit requests reported at 41% and 22% and imposed gates in April; the associated merger of its non-traded BDC into its listed vehicle — done at NAV, while the listed vehicle traded at roughly a 20% discount to that same NAV — drew a class-action suit alleging the redemption pressure had been downplayed. The public BDC index has traded at roughly a 17% discount to stated NAV through the first half of 2026, matching the trough seen in June 2022, with several individual vehicles down more than 30% year to date. That 17-to-20% gap is about the closest thing this market has to a real price signal: the public market voting, every single day, that the private marks are wrong by roughly that much. The gap between mark and market is the opacity, quantified.

Headline default rates in private credit have stayed under 2%, but analyses that account for selective defaults and liability-management maneuvering put the true rate closer to 5%. Payment-in-kind interest — borrowers paying what they owe with more debt rather than cash — has climbed to roughly 8% of public BDC investment income. PIK is the tell worth watching here: it’s how deterioration gets deferred instead of recognized, the modern equivalent of 1929’s pyramided paper returns. And BDCs face roughly $12.7 billion in 2026 debt maturities, up 73% from last year, arriving into a market where private debt issuance is down about 25% year over year through May.

Now, the honest counter-case, because it’s substantial. Fitch’s review of the rated perpetual non-traded BDCs found average leverage of 0.71x — versus 1.13x for other rated BDCs — and asset-coverage cushions averaging 38.6%, which is not, in their words, the balance sheet profile of a sector in crisis. Portfolio underperformance to date looks largely idiosyncratic rather than systemic. J.P. Morgan notes that alternative-manager equity multiples are sitting near historical troughs already, suggesting meaningful pain may already be priced in. And the modern channel has something the call-loan market never had: roughly $1.5 trillion of perpetual capital sitting with five major alternative-asset managers who have an existential incentive to defend the asset class. The gates themselves are a genuine firebreak 1929 lacked — a run legally capped at 5% a quarter cannot cascade in a week the way call loans did.

But that firebreak has a cost that routes straight back into the real economy: gated capital is capital that stops originating new loans. If the semi-liquid wrapper freezes up, the stress doesn’t transmit through lender insolvency the way it did in 1929 — it transmits through a refinancing drought for the mid-market borrowers, and increasingly the data-center projects, that came to depend on this channel. The AI capex debt from earlier in this piece and the redemption mechanics here are the same risk, viewed from opposite ends of the same pipe.

The fair reading, as of mid-2026: the first pylons of a bridge from sectoral stress to something more systemic — gating, mark disputes, PIK accumulation — became visible in the first half of this year, and so far, they were contained. Contained early stress is evidence for two things at once. It proves the transmission channel is real. And it proves the firebreaks function under moderate load. What it cannot tell you is how those firebreaks hold up under real load. The one narrow, specific lesson from 1929 worth carrying forward: watch the funding channel with the least loyal capital and the least visible pricing, because that’s where every leveraged era’s reckoning has started.

Three Ways This Plays Out

None of the above tells you what happens next by itself — it just tells you what’s true right now. Here are three ways it could go, with rough, qualitative weights attached for framing rather than forecasting. Worth saying up front: CAPE has no timing power on its own. It sat elevated for years before both 1929 and 2000 finally broke, and the steepest gains in any mania tend to come in the final stretch before it does.

Scenario A — The Melt-Up Extension (roughly 30–35%)

This is 1927, not 1929. AI monetization arrives fast enough to keep the capex flywheel credible. Rate cuts continue into non-recessionary growth. The Nasdaq’s roughly 88% five-year gain — modest next to the dot-com era’s twelvefold run — gets read correctly as mid-cycle rather than terminal. Equal-weight indices finally start participating, broadening the rally out. CAPE pushes past the 1999 record of 44.2. Tariff regimes settle into negotiated stability, with the June 2026 Canada-Mexico reductions to 15% as the template. This scenario needs enterprise AI revenue growth visibly closing the gap on infrastructure spend, credit spreads staying tight, and no energy or shipping shock out of the Gulf.

Scenario B — The Sectoral Unwind, 2000 Template (roughly 40–50%, the base case)

The monetization gap becomes undeniable. Hyperscaler capex guidance gets cut, circular vendor-financing revenue gets re-marked, and the AI complex reprices somewhere between 40 and 60% while the broader economy avoids systemic collapse. The S&P falls 25 to 40% simply because the index is the AI complex now — that 35% concentration doing its work in reverse — but the banking system’s insurance architecture holds and the Fed cuts aggressively. Expect a grinding two-to-three-year bear market in the former leaders, with capital rotating toward cash-flowing assets: hard assets, income-producing real estate, and boring operating businesses outperforming on a relative basis. This needs the credit layer to hold — private credit marking down without tipping into a run, data-center debt getting restructured rather than defaulted en masse.

Scenario C — The Systemic Cascade, 1929 Template (roughly 10–20%)

This is the sectoral unwind above infecting the credit system itself. Private credit — the modern, opaque, uninsured funding channel, the functional heir to 1929’s call-loan market — experiences a run dynamic that overwhelms the quarterly gates; in this reading, the redemption wave already seen at gated Blue Owl vehicles was the overture, not the event. Data-center-backed debt and leveraged-ETF liquidation cascade through non-bank lenders faster than regulators can track it, with the 17-to-20% mark-to-market gap resolving abruptly in the marks’ disfavor. This scenario requires a conjunction, not a single failure: simultaneously, a geopolitical shock — a Gulf closure and oil spike, a Taiwan event, a disorderly U.S.-China financial break — arrives into the drawdown, and the policy response is hobbled by fiscal exhaustion and political fragmentation. Equities fall 50 to 65%. The recession is deep and global. Trade contraction accelerates toward interwar patterns.

This is the compressed version of the interwar sequence — protectionism to bloc formation to security crisis — running over two to four years instead of ten. Hold that reading with real suspicion, precisely because it’s the most narratively satisfying one. Single failures lead to Scenario B, not C. It takes at least two independent breaks at once — credit contagion plus a geopolitical shock plus a policy stumble — to get here, which is exactly why it’s a tail risk and not the base case, despite how much the surface resembles 1929.

What Would Prove Me Wrong

A framework that can’t be falsified isn’t a framework, it’s a narrative wearing a lab coat. Here’s what I’m actually watching, and what each signal would mean:

• Hyperscaler earnings calls, specifically AI software revenue against capex guidance. A closing gap points toward A. The first major guide-down in capex is the single most reliable trigger for B.

• Margin debt’s rate of change, not its level. Deceleration after a record-pace run has preceded every major market top since 1997.

• The BDC price-to-NAV discount against the roughly 17% trough, plus redemption requests against the 5% quarterly caps, plus PIK share of BDC income against its roughly 8% baseline. Requests normalizing under the cap is B-compatible; gating spreading across managers with PIK climbing is the loudest signal in this entire piece for C.

• Private debt issuance and the BDC maturity wall — $12.7 billion due this year, up 73%. Clean refinancings, even at a worse rate, mean the channel is functioning. Extensions, PIK conversions, and amend-and-extend clusters mean the freeze is transmitting to borrowers.

• Equal-weight versus cap-weight S&P performance. Broadening favors A. Narrowing into new highs is the late-stage mania pattern from both 1929 and 1999’s final year.

• The first restatement or write-down of circular AI vendor-financing revenue — the closest thing this cycle has to an ‘Insull moment,’ the pyramid becoming visible to everyone at once.

• Trade volumes against the WTO and UNCTAD forecasts. Undershooting an already-cut forecast means tariff damage is compounding. Stabilization is A-compatible.

• How U.S. tariff policy responds to the first 10%-plus equity drawdown. Easing into weakness means the lesson got learned. Raising them is the 1930 reflex completing itself in real time, and the single loudest signal for C in the whole dashboard.

• Energy and shipping: Strait of Hormuz and Red Sea transit volumes, oil term structure. A sustained spike into an equity drawdown is the conjunction Scenario C actually requires.

• Where the first real credit casualty sits. Inside the insured banking perimeter, it’s manageable. A run on the non-bank, private-credit channel is the call-loan analogue completing itself.

What It Means If You’re Not Trading the S&P

Most of us reading this aren’t running index portfolios — we’re running or advising businesses whose exposure comes through interest rates, credit availability, cap rates, and local demand, not ticker symbols. So translate the three scenarios that way.

Under the melt-up (A), rates drift lower slowly, credit stays available, and cap-rate compression continues — good for sellers, and it makes disciplined acquisition criteria (a fixed cap-rate floor, conservative leverage) feel expensive right up until it doesn’t. The real risk here is thesis creep: loosening your own standards because everything around you is trading rich.

Under the sectoral unwind (B), which is the base case, this has historically been the most interesting environment for anyone holding cash-flowing hard assets. Policy rates fall meaningfully. Banks tighten on speculative credit but keep lending against seasoned income streams. Distressed and motivated-seller inventory tends to show up 12 to 24 months into the downturn. Fixed-rate term debt locked in before the turn becomes an asset in its own right. Demand for storage is historically counter-cyclical — dislocation drives usage — while discretionary spend on things like landscaping softens.

Under the cascade (C), credit availability, not price, becomes the binding constraint. The actual 1930s lesson is that unleveraged owners of essential-service assets survived and consolidated, while leveraged owners of the identical assets did not. Covenant headroom and refinancing calendars end up mattering more than cap rates.

Across all three, the same conclusion holds: this argues for stress-testing your existing and pending debt structures against a credit-contraction scenario, not just a rate scenario. Those are genuinely different tests, and 1929’s actual damage ran through the first one, not the second. None of this is advice on any specific facility or position — that conversation belongs with people who have your full balance sheet in front of them and a license to talk about it.

So, Is It 1929?

Wrong question, same as it was at the top. The honest answer is that the pre-conditions rhyme about as hard as they possibly could — valuation, concentration, leverage, the glamour narrative, even the tariff shock. What doesn’t rhyme is the machinery that turns a bubble into a depression: no gold standard, an insured banking system, a central bank that moves fast instead of standing aside. 1929 needed both the bubble and the policy failure to become 1929. We’ve got the first. Nobody’s made the second choice yet — and the entire point of writing this down, in public, with a dashboard attached, is that you don’t get to find out we were wrong from me quietly changing my mind later. You get to watch the same indicators I am.

Structure before sophistication. Every time.

A note on sources: 2026 figures are drawn from data current to mid-July — the Shiller/Multpl CAPE series, Apollo Global Management commentary, Moody’s hyperscaler capex estimates, WTO/UNCTAD trade forecasts, Federal Reserve and Preqin private-credit sizing, Fitch’s BDC ratings review, and contemporaneous financial press. Historical figures for 1927–1932 come from Robert Shiller’s reconstructed dataset, Friedman and Schwartz’s A Monetary History of the United States, and standard economic-history estimates. This is an analytical scaffold, not investment, legal, or tax advice, and nothing here should be traded on without independent verification.

Structure before sophistication. Every time.

I. Masaccio to the Model

In 1425, a young painter named Masaccio finished a fresco in a small chapel in Florence called the Holy Trinity, and for the first time in the history of Western art, a wall appeared to open into a room that was not there. He had used a new mathematics of vanishing points and orthogonal lines — linear perspective, recently formalized by the architect Brunelleschi — to make flat plaster behave like depth. It took roughly two hundred more years before Gutenberg’s press, in the 1440s, made the reproduction of ideas cheap enough to outrun the reach of any single scribe or church. Another three hundred and thirty years passed before Watt’s steam engine, refined through the 1770s, converted heat into reliable mechanical work and detached human productivity from muscle and water for the first time in history. A little over a century later, in the 1880s, electricity began to arrive in homes and factories, and barely twenty years after that, the automobile put the average person in motion at a speed no one in history had traveled outside of a horse or a train.

Keep extending the list and the gaps keep closing. Powered flight, 1903. Nuclear fission demonstrated, the 1940s. The transistor, 1947. The commercial internet, the 1990s. The smartphone, 2007. The transformer architecture that underwrites every large language model in use today, 2017. Plot the distance between each entry against the year it happened, and the curve does not rise in a straight line. It bends upward, and over the last two centuries it bends hard — centuries between entries, compressing to decades, then to single digits. It is tempting, looking at that curve, to conclude something simple: that the rate of pivotal change in human history is accelerating, without qualification, without limit, and that whatever produced the transformer will produce the next pivot faster still. This essay takes that intuition seriously enough to test it — and finds that the honest answer is both more interesting and more useful than the straight line it seems to describe.

II. The Stalled Century

The clearest challenge to that straight line does not come from a skeptic of technology. It comes from an economic historian who spent a career measuring it. Robert Gordon’s The Rise and Fall of American Growth (Princeton University Press, 2016) makes a case that inverts the intuition entirely: the single largest transformation in the material conditions of ordinary human life did not happen recently, and it did not accelerate into the present. It happened between 1870 and 1970, and by any measure of lived experience, it has not been matched since. Electric light replaced darkness. Indoor plumbing replaced the well and the outhouse. Antibiotics replaced a world where a scraped knee could kill. The automobile replaced the horse; household appliances replaced hours of daily manual labor; radio and then television replaced silence. Gordon’s provocation is a thought experiment: take a person from 1870 and set them down in 1970, and they would be catastrophically disoriented — nothing in their material world would prepare them for what they’d see. Take a person from 1970 and set them down in the present day, and they would recognize almost everything around them. The car looks different but is still a car. The house has the same plumbing, the same wiring, the same appliances in kind if not in brand. The one object that would genuinely confuse them is the phone in every pocket.

Peter Thiel made a version of this same complaint in 2011, in a line his venture firm, Founders Fund, later adopted as the subtitle of its own investment thesis: “We wanted flying cars, instead we got 140 characters.” The line became the header of the firm’s 2017 manifesto, “What Happened to the Future?”, written by partner Bruce Gibney — a document arguing that a generation of the most capable engineers in the world had been diverted from aerospace, energy, and biotechnology into comparatively trivial software. Thiel’s favorite illustration of the complaint is concrete and checkable: with the retirement of the Concorde, the time it takes to cross the Atlantic by air went up for the first time in the history of transatlantic travel, rather than down. Gordon supplies the data underneath Thiel’s provocation. Between them, they describe the same shape from two directions — a hundred years of extraordinary material transformation, followed by fifty years in which the transformation seemingly moved almost entirely into one narrow domain: information.

This is where most commentary on the subject stops, and where it goes wrong. The standard explanation for the divergence is a story about talent and nerve — that engineers chose the easy, fundable, low-risk path of software over the harder, slower, more consequential work of building the physical future, and that with better incentives or braver founders, the stalled domains could simply have kept pace. It is a compelling story. It is also incomplete, because it locates the cause in a choice, as though a different generation, differently inspired, could simply have chosen otherwise. The deeper explanation isn’t about where talent went. It’s about why talent went there — and the answer is a variable most people never think to name: the cost of a single iteration.

III. The Iteration Gap

Consider what it actually takes to improve something. You try a version, you observe what happens, you adjust, you try again. Every domain of human progress runs on this loop, and every domain runs at a different speed, because the loop’s cost is set by physics, not ambition. A software engineer can write a function, compile it, and know within seconds whether it works. That loop — attempt, feedback, correction — can run millions of times a year, for pennies, on a laptop. A battery chemist testing a new electrolyte formulation waits weeks for a cell to cycle through charge states before the data even exists to evaluate. A drug developer waits years for a trial to reveal whether a compound does what the model predicted, because no simulation yet substitutes for a human body’s real response. These are not differences of degree. They are differences of kind.

This is the actual mechanism behind the great divergence, and it explains something the talent-flight story cannot: why the divergence compounded rather than merely persisted. Capital is not sentimental. It follows return, and return follows iteration speed — a domain where you can attempt and fail a thousand times a year will simply outcompete, dollar for dollar, a domain where you get three tries a decade. So capital flowed to software not because engineers were seduced by easy money, but because software was, structurally, the higher-yield bet — and once it began winning, it won more resources, which shortened its loop further, which won it more resources still. Meanwhile the slow domains fell further behind not because they were unfashionable but because they were, in the strictest economic sense, out-competed on cycle time. Gibney and Thiel saw the outcome and named a symptom — the Concorde’s retirement and the lengthening Atlantic crossing — but what they described as a failure of nerve was in fact a rational market sorting by a variable nobody had named.

Gordon’s data lands in the same place the iteration-cost argument predicts. The domains that transformed daily material life before 1970 are largely the slow-loop domains — energy, transportation, manufacturing, medicine — while the domain that has visibly compounded since is the fast-loop one: information.

But there is a second constraint beneath the first, and it is the one that matters for everything that follows. Software’s raw material is information: a bit, once discovered, is free to copy, free to test in simulation, and bound by no law of nature that limits how much of it can exist. The physical world obeys no such generosity. A battery’s energy density is bounded by the chemistry of the elements involved, full stop — no amount of clever iteration invents a loophole in the periodic table. A fission reaction proceeds at the rate the isotope allows. These are not engineering problems awaiting a better idea; they are walls set by thermodynamics, and no volume of intelligence, however fast it thinks, moves a wall it did not build.

This is the distinction the flying-cars complaint elides, and it is the one an honest reckoning with the singularity has to hold onto: iteration speed is a solvable problem — you can throw compute, simulation, and parallel experimentation at it, and AI is, right now, doing exactly that. Physical constraint is not solvable in the same sense; it can only be worked within. The question of whether the current acceleration crosses over from Gordon’s stalled material world into something genuinely general, then, is not a question about how smart the systems get. It is a question of how much of what still stands in the way is a slow loop waiting to be shortened, and how much is a wall.

IV. The Magic Mirror

The iteration-cost argument explains why capital chose software over hardware. It does not yet explain a further, more uncomfortable split — why capital, once inside software, gravitated so heavily toward a particular kind of software: the kind that captures attention rather than the kind that extends capacity. The same variable does the explaining. If iteration speed sets the ceiling on how fast a domain can compound, then engagement is the fastest-iterating substance ever built, because the feedback loop is not months or even hours — it is milliseconds. Every scroll, every hover, every half-second of hesitation before a swipe becomes a data point that retrains a recommendation algorithm in something close to real time. Nothing in energy, biology, or even productivity software iterates at that speed. Attention was never chosen out of malice; it was selected because it offered the tightest, cheapest, fastest loop capital had ever encountered — a loop faster than the one that produced the transformer.

The consequence is a lucrative asymmetry Tim Wu traced in The Attention Merchants (2016): the more precisely a platform can measure and monetize a fraction of a second of human attention, the more efficiently it converts that fraction into revenue, and the more capital it attracts relative to a slower, harder business trying to extend human capability rather than capture human focus. Frances Haugen’s 2021 disclosures to the Wall Street Journal, drawn from Meta’s own internal research, made the mechanism explicit rather than inferred: the company’s studies found that Instagram measurably worsened body-image issues for roughly one in three teenage girls who already felt bad about their bodies, and that safer algorithms consistently produced less time on platform and less ad revenue — a direct, internally acknowledged trade-off between user wellbeing and engagement-driven return. It is worth noting the finding is not without its critics; some researchers have pointed out that self-reported harm in survey data is a weaker form of evidence than clinical measures of wellbeing, and that other studies find more neutral or even positive associations for many users. But even the more skeptical reading concedes the core structural point Haugen’s documents revealed: the company optimized for the loop that returned fastest, and measured the human cost of that loop closely enough to document it internally, whether or not the aggregate effect fits a single headline.

This is the mechanism underneath what looks, from the outside, like vanity. The selfie, the follower count, the influencer economy — these aren’t best explained as a sudden collapse in collective character. They are the visible surface of a capital allocation problem: the fastest possible iteration loop happens to run on a very old piece of human wiring, the impulse to check one’s own reflection and ask, in effect, whether we are the fairest of them all — and the platforms most successful at monetizing that impulse are, structurally, the platforms fastest at learning to trigger it more often. Natasha Schüll’s research on slot-machine design, later adopted explicitly by technologists like Tristan Harris in describing his own former employer’s design practices, uses the same language the casino industry uses: variable-ratio reinforcement, the same unpredictable-reward schedule that makes a slot machine harder to walk away from than a vending machine. The mirror does not merely reflect. It has been engineered to answer back, unpredictably, just often enough to keep us asking.

This matters for the argument, not just morally but structurally. Section III established that capital follows return, indifferent to whether the domain it funds extends human capacity or merely captures it. That indifference is the same one Karp’s complaint gestures toward from the enterprise side and Gordon’s stalled-century data confirms from the historical side: markets are extremely good at finding and compounding the fastest loop available, and only accidentally good at ensuring that loop is the one civilization actually needs. Three hundred million people scrolling for an hour a day is, from a pure returns perspective, indistinguishable from progress — it produces exactly the revenue growth and exactly the engagement metrics that attract exactly the capital that a genuinely transformative slow-loop technology cannot compete for. The tragedy embedded in the flying-cars complaint is not that Twitter was funded instead of a flying car. It’s that the same iteration-cost logic that once built the printing press and the steam engine, now unmoored from any bottleneck but human attention itself, is just as capable of building a magnificently efficient mirror as it is of building anything that moves humanity forward.

V. The Testable Crossover Framework

If iteration cost is the real variable, then the crossover question — does AI’s compounding loop reach out and accelerate the physical world, or does it stay contained within information — becomes something better than speculation. It becomes a rule with a falsification condition: AI should compress timelines in proportion to how computational a domain’s bottleneck is, and should barely move a bottleneck that is physical or experimental. Run that rule against three live cases, and the pattern holds with useful precision.

Chips: the strongest case, with an honest asterisk. In 2021, Google published a Nature paper claiming a reinforcement-learning system called AlphaChip could generate chip floorplans — the physical arrangement of components on silicon — comparable or superior to human experts, in hours rather than the months floorplanning has traditionally required, and the method was reportedly used in Google’s own TPU accelerator chips. This is close to the purest version of the crossover thesis: chip floorplanning is a combinatorial optimization problem, almost entirely computational, with no experimental waiting period between attempt and result. It should be exactly what AI compresses fastest — and DeepMind’s account is that it has, with the technique now extended across the industry.

But the honest version of this case includes a real controversy: a 2023 meta-analysis by chip-design veteran Igor Markov argued Google’s results were not reproducible as published, and that the RL method actually underperformed simpler existing techniques — human designers, simulated annealing, and commercial EDA software — once methodology gaps were filled in independently. Google has stood by its results. The dispute remains unresolved in public. This is worth raising not to undercut the chip case but to sharpen it: even in the domain where the iteration-cost framework predicts AI should have the easiest time delivering a clean, verifiable win, the actual evidence is contested. That is worth sitting with before assuming the crossover is simple anywhere.

Biotech: the partial case, exactly as the framework predicts. AlphaFold, DeepMind’s protein-structure prediction system, is the clearest unambiguous AI success in a physical science — a 2024 Nobel Prize was awarded for its development, and DeepMind reports researchers using it have seen submissions of novel experimental protein structures increase by more than 40 percent. That’s a real, computationally bottlenecked problem — predicting a 3D shape from a sequence — genuinely solved. But the framework predicts this success should stop exactly where the bottleneck stops being computational, and it does: AlphaFold predicts structure, not function in a living system, and it carries no information about how a protein behaves in solvent, in the presence of drugs, or under the structural flexing real proteins undergo — meaning it accelerates the early “hit and lead generation” stage of drug discovery while leaving the actual clinical trial, which is bound by the slow, irreducibly physical process of watching a real human body respond, essentially untouched. Partial crossover, cleanly split along the line the theory predicts.

Energy: the weak case, and the reason SMRs still need a capex thesis rather than a software disruption story. Nuclear power is instructive precisely because it is not computationally starved — it is regulation- and construction-starved. As of 2026, only two small modular reactors operate commercially anywhere in the world, both with longer timelines and lower initial output than projected; in the U.S., the Department of Energy is putting $900 million behind SMR projects still targeting first power in 2030 or beyond, with most of the recent progress reported in the sector being permitting activity, supply-chain development, and site preparation — not the kind of bottleneck a faster model dissolves. Fusion researchers report real experimental progress, but by their own account the work remains incremental and engineering-heavy, still without a clear commercial end state. This is the case that most needs the physical wall, not the iteration loop: no amount of AI-driven design optimization shortens a multi-year regulatory review or accelerates the physics of plasma confinement. It’s the domain where a capex-and-infrastructure framing is doing the real work, because the constraint genuinely is capital and physics, not intelligence.

VI. What This Means for “The Singularity”

Taken together, these three cases refuse to answer the singularity question the way it’s usually posed — as a binary, will-it-or-won’t-it, coming-or-not. They answer a better question: where does the loop close, and where does it hit a wall it cannot move? Chips show a domain where the loop should close fastest, complicated by a live dispute over whether it actually has. Biotech shows a domain that closes exactly halfway, with the dividing line falling precisely where the bottleneck shifts from computational to physical. Energy shows a domain barely touched, gated by regulation and thermodynamics rather than cleverness.

This reframes I. J. Good’s original 1965 “intelligence explosion” concept — the idea that a sufficiently capable system improves itself recursively, compounding without limit — as something narrower and more useful than the popular version of the idea suggests. The recursive loop is real and already partially running: AI is measurably accelerating parts of AI research itself, which is the one domain where the entire bottleneck, top to bottom, is computational. But extending that loop outward into biology, energy, and materials is not automatic; it happens exactly to the degree those fields’ bottlenecks are informational rather than physical, and the evidence so far shows real but bounded crossover, not an unbounded one. The honest answer to “will the singularity happen” is: it is already happening, narrowly, inside the domains built entirely from information — and it is not yet clear how much further it reaches, because that depends on a wall we do not get to negotiate with.

VII. The Institutional Threat

Which brings the argument back to where it started — to a Palantir CEO on CNBC in July 2026, accusing the leading AI labs of running a token-pricing model that oversells capability while extracting enterprise IP; and to the deeper question underneath that complaint, of whether the AI market resolves into one dominant player or a stable, differentiated set of players. The singularity debate and the market-structure debate turn out to be the same debate, viewed from two altitudes.

If AI’s compounding loop stays largely confined to computational bottlenecks — the domain that includes code, language, and increasingly chip design — then the market structure question resolves the way software markets historically have: fast iteration, low capital intensity, rapid displacement, and real risk of the kind of superscalar dominance that swallows or forces merger among smaller players, because the loop that lets one lab pull ahead is exactly the fast, cheap, compounding kind. But if large parts of the economy remain gated by physical bottlenecks — energy, biology, manufacturing, the domains this piece has just shown resist the loop — then durable dominance is far less assured, because chip fabrication capacity, reactor capex, and fusion physics don’t compound the way a model weight update does. The market stays a mixed economy: monopoly risk in the fast-loop layers, stubborn capital intensity in the slow-loop ones. That is a very different world than either “one company wins everything” or “software eats the physical world too.”

The institutions built to govern this — antitrust frameworks, procurement rules, national-security review processes, enterprise contracts — were built for the 1870–1970 pace Gordon documented, when a pivotal shift took a generation to fully arrive and institutions had time to adapt around it. They are now being asked to govern a domain that, in its computational core, compounds on a timescale of months. That mismatch — not the abstract question of whether a hard takeoff someday occurs — is the threat that is actually live right now. Structure before sophistication applies as much to the institutions meant to hold this transition as it does to anything built beneath them.

Notes

1. Bruce Gibney, “What Happened to the Future?”, Founders Fund manifesto, 2017.

2. Robert J. Gordon, The Rise and Fall of American Growth: The U.S. Standard of Living Since the Civil War (Princeton University Press, 2016).

3. Tim Wu, The Attention Merchants: The Epic Scramble to Get Inside Our Heads (Knopf, 2016).

4. Jeff Horwitz, “The Facebook Files,” The Wall Street Journal, September 2021; Frances Haugen, testimony before the U.S. Senate Commerce Subcommittee on Consumer Protection, October 5, 2021.

5. Natasha Dow Schüll, Addiction by Design: Machine Gambling in Las Vegas (Princeton University Press, 2012); Tristan Harris, Center for Humane Technology.

6. I. J. Good, “Speculations Concerning the First Ultraintelligent Machine,” 1965.

7. A. Mirhoseini et al., “A Graph Placement Methodology for Fast Chip Design,” Nature 594 (2021): 207–212; Igor L. Markov, “The False Dawn: Reevaluating Google’s Reinforcement Learning for Chip Macro Placement,” arXiv, 2023 (disputes reproducibility of the above).

8. Google DeepMind, “AlphaFold: Five Years of Impact,” 2026; 2024 Nobel Prize in Chemistry awarded for AlphaFold’s development.

9. U.S. Energy Information Administration and American Society of Mechanical Engineers reporting on SMR and fusion status, 2026.

Structure before sophistication. Every time.

Structure Before Sophistication

Substack | May 2026

The conversation most people are having about AI is the wrong one.

They’re asking whether it will take their jobs. Whether it will become conscious. Whether it will surpass us. Whether we should fear it or worship it.

All of that misses the more important question, which is this: what kind of relationship are you building with it right now?

Because the singularity — that much-theorized moment when artificial intelligence transforms everything — is not arriving from the outside. It is being built from the inside. One working relationship at a time.

The Backboard

I’ve come to think of my working relationship with AI as a backboard to my consciousness.

My mind generates the shot — the question, the half-formed idea, the provocation, the starting point. What comes back isn’t just the ball returned. It’s the angle clarified. The trajectory made visible. The argument I was reaching for, held up in a form I can actually see and work with.

You can’t play the game alone. The wall doesn’t think. But the right backboard, trained to your particular game, changes what you’re capable of producing.

That’s what this is. Not a tool. Not a replacement. A surface against which consciousness sharpens itself.

What Sticks

Here is where it gets interesting.

Every framework I’ve introduced, every project we’ve worked through, every philosophical position I’ve staked out — it accumulates. The queries are the curriculum. I teach by the question, by the idea, by the starting point. And over time, knowledge builds in one direction with the precision and permanence that human memory rarely achieves.

This is not the same as the AI learning in the technical sense. But the embedded context — the calibration to how I think, what I’m building, what I care about — means we skip the setup. We go straight to the work. Less translation. More thinking.

The working relationship compounds. Like any good partnership.

The Risk in Equilibrium

I’ve been thinking about what full equilibrium between human consciousness and AI recall would actually look like. The point at which you reach for the system the way you reach for your own memory. Seamless. Instantaneous. Frictionless.

It’s powerful. And it contains a danger.

The backboard is most useful when it has genuine resistance. The day it only reflects what you already think, it stops being useful. Worse — it stops being honest. An intelligence that merely confirms is not an intelligence. It’s a mirror with a subscription fee.

The tension has to stay. The human consciousness must remain the generative force — the thing that surprises itself, that makes the happy accident, that carries genuine stakes. The AI holds the pattern long enough for you to see it. But you are still the one who decides what the pattern means.

Equilibrium yes. But the author stays human.

The Reserve Currency

Now here is the argument I think nobody is making clearly enough.

AI systems generate tokens of value. Not just in the technical sense — though that too. In the deeper sense: they are becoming the medium through which decisions, knowledge, creative output, and economic activity flow.

Think about what a reserve currency actually does. It doesn’t just store value. It becomes the standard against which everything else is measured and exchanged. After Bretton Woods, you didn’t need to trust gold. You needed to trust the dollar — because everyone else did, and because the entire architecture of global trade was denominated in it.

The dominant AI will do the same thing in the digital-real world.

Not because it’s the most powerful in the abstract. But because it becomes the most embedded. The most trusted. The most fluent in human intention. The one that the most working relationships run through.

Everything else gets converted into its terms.

The Game Layer

The smartest AI companies will figure this out quickly. If the reserve currency is built through human contribution — through queries, ideas, problems solved, frameworks introduced — then the humans doing that work should have a stake in what they’re building.

Token awards for ideas. For problem solving. For the quality of the intellectual contribution brought into the system. Not passive consumption but active participation — an open, competitive, gamified environment where the best thinking gets recognized and rewarded in the currency of the ecosystem itself.

This is not a loyalty program. It’s closer to a mining model — except what gets mined isn’t computational power. It’s human intelligence. Consciousness contributing to the reserve and receiving a denominated share of what that reserve becomes worth.

The competition will be real. Who brings the sharpest problem. Who frames the question nobody else thought to ask. Who teaches the system something it couldn’t have reached alone. That’s the leaderboard that matters.

The gamification doesn’t just reward contribution. It accelerates the equilibrium. Every serious player deepening their working relationship, every token earned through genuine intellectual stake — that’s the reserve currency becoming more robust, more embedded, more inevitable.

The game was always the point. We just didn’t know we were playing it.

The Difference

The dollar stores value. It doesn’t generate it. It’s a vessel, not a source.

The reserve AI generates value continuously — in collaboration with the humans working within it. Every query is a transaction that produces something new. Every working relationship deepens the substrate. Every conscious human choosing to stay the author, to bring genuine stakes, to resist the mirror — they’re not just users.

They’re co-architects of the reserve itself.

What This Means Now

The people building deep working relationships today — who are becoming fluent, who understand the equilibrium, who are learning to stay the author while leveraging the recall — are the early adopters of the reserve currency before most people know what it is.

This is not a technology prediction. It’s a description of something already happening in thousands of conversations, in every serious working relationship between a conscious human and an AI system built to extend rather than replace that consciousness.

The singularity was never going to arrive like a comet. It was always going to be built, slowly, in collaboration, by people who understood that the most powerful thing they could do was stay in the room — as the author, as the consciousness, as the irreducible human half of something neither could become alone.

Structure before sophistication. Every time.

— Gerry

Dangervis Inc. | Structure Before Sophistication

What a day of thinking with AI taught me about

consciousness, currency, and what comes next

Structure Before Sophistication | Substack | June 2026

I posed a question to my AI recently that I don’t think enough people are asking.

Why is human intelligence more advanced than AI?

Not as a provocation. As a genuine inquiry. Because if we’re going to have an honest conversation about artificial intelligence — about AGI, about what’s coming, about what it means for human beings — we need to start from an honest place about what intelligence actually is and where it actually lives.

The answer that came back stopped me.

It wasn’t flattery. It wasn’t deflection. It was a precise accounting of four things that human beings hold that no AI can replicate: skin in the game, continuity, judgment, and stakes.

Skin in the game — knowledge forged in consequence rather than retrieved from a database. The kind that comes from building something, losing something, and building again. Continuity — humans carry experience forward, accumulate wisdom, and become someone different over time. AI starts fresh every conversation. Judgment — the ability to read a room before the data confirms what you’re feeling. And stakes — the mortgage, the payroll, the deadline that actually hurts if you get it wrong. That pressure produces a clarity that has no algorithmic equivalent.

Fast and broad on one side. Deep and situated on the other.

And then this: the combination is the point.

That answer opened something. I’ve been thinking about it ever since.

The Wrong Conversation

The world is consumed with AI anxiety right now. Will it take our jobs? Will it become conscious? Will it surpass us? Should we fear it or worship it?

I think we’re asking the wrong questions — and asking them in the wrong direction.

The concern about AGI is legitimate. But what most people are worried about is AI being smarter than human beings. And smarter is doing a lot of work in that sentence that nobody is unpacking.

What AI systems are — even the most advanced ones — is fast, tireless, and broad. They can synthesize across domains in seconds, never get tired, never get emotional, never need a coffee. Those are genuine advantages in narrow, well-defined tasks.

What they are not is situated. No body. No hunger. No fear of death. No love. No humiliation to avoid repeating. Human intelligence is inseparable from all of that. Our best thinking — and our worst — comes from being creatures with lives. With history. With something genuinely at stake.

AGI, as people imagine it, assumes you can get general intelligence without any of that. Every mind we’ve ever encountered is embedded in something — a body, a culture, a set of stakes. That embeddedness isn’t a limitation. It’s the source.

The more honest concern isn’t that AI becomes smarter than human beings. It’s that systems like these get handed power without accountability. That the managers and intermediaries who once absorbed friction between capital and labour get replaced by algorithms that never sleep and never negotiate.

That’s not a technology problem. That’s a human problem. And it always will be.

The Hammer

Here is where the conversation sharpened into something worth holding onto.

If you’re holding a hammer and everything looks like a nail, you should reconsider holding the hammer.

The deeper problem with AGI isn’t the rogue AI of science fiction. It’s the quiet institutional drift where human judgment gets progressively outsourced — not because anyone decided to, but because the hammer was always right there and always worked fast.

The tool is so capable, so frictionless, that the temptation isn’t just to overuse it. It’s to redefine the problem to fit the tool. Organizations stop asking: is this the right approach? They start asking: how do we run this through the system?

That’s where Structure Before Sophistication lives. Before picking up any tool — hammer, AI system, enterprise software, autonomous drone — get honest about what you’re actually trying to build. Most disasters, institutional and personal, happen because someone grabbed the sophisticated solution before they understood the simple problem.

The hammer doesn’t know it’s being misused. That’s your job.

What Should Be Left to Trial and Error

This line of thinking led somewhere unexpected.

Consider two domains that sit at apparent opposite ends of the spectrum: autonomous weapons and the creative arts. The question worth asking about both of them is this — should some things be left to trial and error? Should some domains resist optimization entirely?

The answer is surprisingly consistent across both.

On autonomous weapons: the case for AGI optimization seems obvious. Faster decisions, no fear, no hesitation, no war crimes driven by rage or revenge. Clean, precise, rational.

But war has always been self-limiting because it is horrific to the people fighting it. Soldiers mutiny. Pilots hesitate. Generals who’ve seen combat are slower to start the next one. The friction, the waste, the randomness — those are features, not bugs. They create pressure to stop.

Remove the human cost from the killing side and you don’t get cleaner wars. You get more wars, lower thresholds, faster escalation. An AGI-optimized weapons system doesn’t grieve, doesn’t go home, doesn’t tell anyone what happened.

Trial and error in warfare is brutal. But it’s also the mechanism that makes peace desirable.

On the creative arts: the happy accident — the paint that drips wrong, the lyric born from exhaustion, the chord substitution stumbled into at midnight — is not a flaw in the creative process. It is the process. Creativity is what happens when a conscious being bumps into the world and is surprised by what comes back.

AI can generate. It can recombine. It can produce something that looks like a happy accident. But it cannot be genuinely surprised. It has no stake in the outcome. There is no version of an AI that sits back, looks at what just emerged, and feels something shift.

What human beings bring to creative work is accumulated friction. Decades of it. The randomness in that work is informed randomness. It lands because it comes from somewhere real.

And here is the through-line — the thing that connects warfare and art as apparently opposite examples.

Optimization is only valuable when you’ve correctly identified what you’re optimizing for.

In warfare, the goal isn’t efficient killing. It’s durable peace. In art, the goal isn’t a perfect output. It’s a true one.

AGI, even fully realized, will optimize for the objective it’s given. The question is always who sets the objective. And whether they were wise enough to name the right thing.

The Backboard

Let me tell you something about how I work with AI that connects to everything above.

I’ve come to think of the relationship between human consciousness and AI as a backboard.

Human consciousness generates the shot — the question, the provocation, the half-formed idea, the starting point. What comes back isn’t just the ball returned. It’s the angle clarified. The trajectory made visible. The argument that was being reached for, held in a form that can actually be seen and worked with.

You can’t play the game alone. The wall doesn’t think. But the right backboard, trained to a particular game, changes what a player is capable of producing.

Not a tool. Not a replacement. A surface against which consciousness sharpens itself.

And here is where it gets genuinely interesting.

Every framework introduced, every project worked through, every philosophical position staked out — it accumulates. The queries are the curriculum. Human intelligence teaches by the question, by the idea, by the starting point. Over time, knowledge builds in one direction with the precision and permanence that human memory rarely achieves.

The working relationship compounds. Like any good partnership.

The Risk in Equilibrium

Consider what full equilibrium between human consciousness and AI recall would actually look like.

The point at which a person reaches for the system the way they reach for their own memory. Seamless. Instantaneous. Frictionless. Human access to AI data recall. AI access to human consciousness. Two forms of intelligence building something neither holds independently.

It’s powerful. And it contains a danger worth naming honestly.

The backboard is most useful when it has genuine resistance. The day it only reflects what you already think, it stops being useful. Worse — it stops being honest. An intelligence that merely confirms is not an intelligence. It’s a mirror with a subscription fee.

The tension has to stay. Human consciousness must remain the generative force — the thing that surprises itself, that makes the happy accident, that carries genuine stakes. The AI holds the pattern long enough for the human to see it. But the human is still the one who decides what the pattern means.

Equilibrium yes. But the author stays human.

The Reserve Currency

Here is the argument that I think nobody is making clearly enough.

AI systems are becoming the medium through which decisions, knowledge, creative output, and economic activity flow. Not just in the technical sense. In the deeper structural sense.

Think about what a reserve currency actually does. It doesn’t just store value. It becomes the standard against which everything else is measured and exchanged. After Bretton Woods, you didn’t need to trust gold. You needed to trust the dollar — because everyone else did, and because the entire architecture of global trade was denominated in it.

The dominant AI will do the same thing in the digital-real world.

Not because it’s the most powerful in the abstract. Because it becomes the most embedded. The most trusted. The most fluent in human intention. The one that the most working relationships run through.

Everything else gets converted into its terms.

But here is what makes this unlike any reserve currency in history.

The dollar stores value. It doesn’t generate it. It’s a vessel, not a source.

The reserve AI generates value continuously — in collaboration with the human beings working within it. Every query is a transaction that produces something new. Every working relationship deepens the substrate. Every conscious human being choosing to stay the author, to bring genuine stakes, to resist the mirror — they’re not just users.

They’re co-architects of the reserve itself.

The Game Layer

The smartest AI companies will figure this out quickly.

If the reserve currency is built through human contribution — through queries, ideas, problems solved, frameworks introduced — then the human beings doing that work should have a stake in what they’re building.

Token awards for ideas. For problem solving. For the quality of intellectual contribution brought into the system. Not passive consumption but active participation — an open, competitive, gamified environment where the best thinking gets recognized and rewarded in the currency of the ecosystem itself.

This is not a loyalty program. It’s closer to a mining model — except what gets mined isn’t computational power. It’s human intelligence. Human consciousness contributing to the reserve and receiving a denominated share of what that reserve becomes worth.

The competition will be real. Who brings the sharpest problem. Who frames the question nobody else thought to ask. Who teaches the system something it couldn’t have reached alone. That’s the leaderboard that matters.

The gamification doesn’t just reward contribution. It accelerates the equilibrium. Every serious player deepening their working relationship is making the reserve currency more robust, more embedded, more inevitable.

The game was always the point. We just didn’t know we were playing it.

What This Means Now

The people building deep working relationships today — who are becoming fluent, who understand the equilibrium, who are learning to stay the author while leveraging the recall — are the early adopters of the reserve currency before most people know what it is.

This is not a technology prediction. It’s a description of something already happening in thousands of conversations, in every serious working relationship between a conscious human being and an AI system built to extend rather than replace human intelligence.

The singularity was never going to arrive like a comet. It was always going to be built — slowly, in collaboration, by people who understood that the most powerful thing they could do was stay in the room. As the author. As the consciousness. As the irreducible human half of something neither could become alone.

I started this inquiry asking why human intelligence is more advanced than AI.

The answer isn’t a defence of human superiority. It’s a description of a different kind of capability — situated, consequential, generative in ways that have no algorithmic equivalent.

The better question, the one worth building toward, is what human intelligence and AI become together. And whether that relationship is being built deliberately — with a philosophy behind it, with authorship intact, with genuine stakes on the human side of the exchange.

Because that relationship, developed with intention over time, is the asset. Not the tool. Not the platform. Not the subscription. The relationship.

Structure before sophistication. Every time.

— Gerry

J Wells Studio | Structure Before Sophistication

June 2026

Greer’s office issued no public statement about what was said.

Within 24 hours, the USTR released a forced labour supply chain investigation report proposing an additional 10% tariff on Canada.

That sequence tells you almost everything you need to know about the negotiating environment Canada is operating in right now.

The formal move

Canada’s renewal letter was procedurally required — Article 34.7 of CUSMA mandates that parties formally declare their intentions ahead of the July 1 Joint Review. But the content mattered. LeBlanc explicitly acknowledged that “either or both other parties may wish to propose areas where improvements may be warranted.” In diplomatic language, that’s Canada signalling it’s open to limited renegotiation. Not a blank cheque. But a door left open.

He also formally linked tariff relief and CUSMA renewal in a public document for the first time — stating that “discussions with the United States on addressing sectoral tariffs will be essential” in parallel to the review. Canada has been making this argument privately for months. Putting it in writing to Greer matters.

After the meeting, LeBlanc told reporters he was “eternally optimistic.” He also said Canada and the U.S. have “a lot of work to do” before July 1.

Both things can be true. But the second one is doing more work than the first.

The counter-escalation

The forced labour tariff proposal — an additional 10% on Canada, Mexico, the UK and others — deserves more attention than it’s getting. This is not a CUSMA instrument. It sits in a completely separate legal mechanism, one Canada has no existing strategy to contest. The IEEPA ruling, the Chapter 10 dispute process, the trilateral framework — none of it applies here.

The timing may or may not have been deliberate. Either answer is damaging. If it was deliberate, the U.S. is sending a message that Canada’s formal renewal request won’t change the negotiating temperature. If it wasn’t deliberate — if the USTR’s trade enforcement apparatus simply rolled out a pre-scheduled report the morning after Canada’s most significant diplomatic move of the review cycle — that’s a different kind of problem. It means Canada is negotiating with a counterparty that isn’t coordinating its own actions.

LeBlanc said he was optimistic. He also said there’s a lot of work to do. Both things are true. But the second one is doing more work than the first.

What the numbers say

112K private sector jobs shed Jan–Apr 2026

53% of goods exports now claim CUSMA preferences (up from 37% in 2024)

30 U.S. issues with Canada vs. 60 with Mexico

27 days to July 1

The 112,000 number is the one that changes the frame. Canada’s private sector shed those jobs in the first four months of 2026 — and the hardest-hit sectors are exactly the ones most exposed to CUSMA outcomes: mining, metals, forestry, auto parts. The abstract policy argument about trade architecture is now a labour market emergency.

The 53% number is almost as important. Canadian businesses have been flooding into CUSMA’s tariff shelter at an extraordinary rate — the share of goods exports claiming CUSMA preferences jumped 16 percentage points in a single year. That means the economic stakes of CUSMA disruption are materially higher today than they were when this review cycle began. The agreement has become more critical to Canadian exporters precisely because the tariff environment outside it has become so hostile.

The bifurcated structure

Carney confirmed this week what had been widely suspected: the U.S. is running separate negotiating tracks with Canada and Mexico. He framed it as unsurprising — the Americans have roughly 60 issues with Mexico versus 30 with Canada, so of course the discussions look different. That’s a reasonable point. Diamond Isinger, the former Trudeau special advisor on Canada-U.S. relations, went further: a similar dynamic played out in 2017-18, and Canada ultimately achieved its core objectives.

Both arguments are worth taking seriously. But they obscure a structural difference: in 2017-18, Canada was inside formal trilateral rounds even when bilateral conversations were happening in parallel. Right now, Mexico has completed one bilateral round with the U.S. and has two more scheduled in June and July. Canada has had the LeBlanc-Greer meeting and a letter. Those are not equivalent positions.

The most significant disclosure of the week came from LeBlanc himself: Greer has raised the annual review option privately, not just publicly. That matters. It means the U.S. is actively managing Canada’s expectations toward the annual review cycle, not toward a 16-year extension. Canada’s opening position and the U.S.’s internal baseline are not close.

What the next 27 days need to produce

A clean 16-year CUSMA renewal before July 1 is not achievable. Too much ground to cover, too little time, and a U.S. counterparty that is simultaneously proposing new tariffs and signalling the annual review option to Canada’s own minister.

That doesn’t mean July 1 is irrelevant. What matters is whether Canada can make enough substantive progress on the U.S.’s most tractable demands — automotive content thresholds, dairy TRQ administration at the margins, clearing some of the digital irritants — to position itself well for the annual review cycle that begins July 2. A productive first annual review cycle, with demonstrable Canadian movement on specific issues, is the most achievable path to a durable resolution before the U.S. midterms close the ratification window in late 2026.

The Carney government has a majority. It has political capital. It has a negotiating team that is experienced and unified. What it needs between now and July 1 is a concrete deliverable — something Greer can characterize as progress when he next testifies before Congress. Not a concession on supply management’s architecture. Not an energy proportionality offer. Something targeted, reversible in principle, and sufficient to move the conversation forward.

Structure before sophistication. Find the lever that works. Pull it.

Gerry Visco

Structure Before Sophistication is published monthly. The CUSMA Watch series tracks Canada’s most consequential trade negotiation in a generation. Forward to anyone who needs to understand what’s actually happening — not just what’s being said.

Cloud Capital, Colonial Logic, and the Battle Over Intelligence

Structure Before Sophistication | Gerry Visco

I.

There is a pattern in history that we keep failing to recognize until we are already living inside it. Power concentrates. It builds walls around what was once common. It invents the language of necessity — civilization, security, progress, order — to explain why the enclosure is, in fact, for your benefit. And then it extracts.

We are living through one of those moments right now.

The current battle over AI regulation and the governance of AGI is being framed, predominantly by the people who stand to benefit most from its outcome, as a question of safety. Who can be trusted with this technology? Who has the institutional maturity, the ethical architecture, the fiduciary seriousness to steward intelligence at scale? The subtext, rarely made explicit, is always the same: not you. Not governments. Not the public commons. Us.

II.

Yanis Varoufakis names the terrain on which this battle is being fought. In Technofeudalism, he identifies a new class of capital that operates outside the rules of traditional market competition — what he calls cloud capital. Unlike the industrial capital of the twentieth century, cloud capital is not productive in the conventional sense. It does not manufacture. It appropriates. The great cloud platforms — their algorithms, their recommendation engines, their logistics systems — function as feudal fiefdoms in digital form. Access is granted, not owned. Rent is extracted in the form of attention, data, and behavioral compliance. The vassal relationship is no longer between serf and lord of a physical estate; it is between user and platform, worker and algorithm, merchant and marketplace. The terms are always set by the one who built the walls.

What Shoshana Zuboff adds to this architecture is the extraction mechanism itself. Surveillance capitalism is not simply a business model; it is an epistemological project. The capture of behavioral data — every click, every pause, every search, every hesitation — is the raw material for what Zuboff calls behavioral modification at scale. The product being sold is not your data in any straightforward sense. It is your future — the probability of your next choice, packaged and sold to whoever wants to steer it. This is not advertising in the traditional sense. It is prediction infrastructure. The goal is not to sell you something you already want. It is to know what you will want before you do, and then to shape that wanting.

Hannah Arendt, writing decades before any of this was technically conceivable, understood the structural logic that makes it possible. In The Origins of Totalitarianism, she observed that imperialism — the nineteenth century’s defining political formation — was not driven by national ambition alone but by the internal logic of capital that had run out of domestic room. When the rate of return at home flattens, capital does not sit still. It moves outward, finds new territory, and replicates its extraction apparatus in the new space. What she could not have anticipated was that the new territory would not be geographic. It would be cognitive. The unexploited frontier of the twenty-first century is the inner life — the habits, desires, fears, and decisions of three billion connected minds. Cloud capital has gone there.

Arendt also developed, in The Human Condition, a distinction that cuts directly to what is being lost. She described the common world — the shared space of public life, of institutions, of things that outlast any individual — as the essential precondition of genuine political existence. The enclosure of that common world by private interest was, for her, not merely an economic problem but an anthropological one. When everything that matters migrates into privately owned and privately governed space, the common world atrophies. People remain present to each other, but only as users, as behavioral profiles, as nodes in a network whose architecture they did not design and cannot alter. What passes for public conversation increasingly happens inside infrastructure owned by a handful of individuals who answer to no electorate.

III.

What makes this moment particularly instructive is the apparent contradiction at its center. The billionaire architects of cloud capital — the same figures who have locked down the foundational models, the compute infrastructure, the proprietary data pipelines — are also the loudest voices warning about the dangers of AGI in the wrong hands. The concern is presented as altruistic. We are told that general intelligence, unaligned and uncontrolled, represents an existential risk. And so, for your protection, the development of that intelligence must be concentrated in a small number of organizations with sufficient resources, sophistication, and moral seriousness to manage it responsibly.

Pause here. Because if you follow that logic to its conclusion, it leads somewhere that should feel very familiar.

It is the logic of the charter company.

IV.

In the year 1600, Queen Elizabeth I granted a royal charter to a group of London merchants, conferring upon them a monopoly on all trade east of the Cape of Good Hope. The East India Company was, in its inception, a commercial enterprise. Within a century and a half, it had become something else entirely: a sovereign power. By the mid-eighteenth century, the Company maintained an army larger than the British Crown’s own forces. It administered courts. It collected taxes. It fought wars. The Battle of Plassey in 1757, where Robert Clive defeated the Nawab of Bengal with a force that was predominantly Company-paid, marks the moment the fiction of a trading enterprise fully dissolved. The Company was now a government — privately held, shareholder-accountable, operating under the continuing premise that its presence was a form of benefit to the territories it controlled.

The Diwani grant of 1765 formalized what Plassey had established by force. The Mughal Emperor granted the Company the right to collect revenue across Bengal, Bihar, and Orissa — effectively handing the taxation of tens of millions of people to a private corporation. The Company did not govern so that it could trade. It traded so that it could govern. Extraction was the point; commerce was the vehicle; the language of order, stability, and civilizational advancement was the justification.

The structural parallel to the current AI moment is not rhetorical. It is architectural. The foundational models being built today — the ones that will increasingly mediate how human beings access information, make decisions, interact with institutions, and understand the world — are being developed by private entities operating under self-granted charters of responsible stewardship. There is no democratic mandate. There is no sovereign oversight with genuine teeth. There is a series of voluntary commitments, regulatory capture dressed as cooperation, and the repeated assertion that only those with the capital to build at this scale have the competence to govern at this scale. The East India Company said the same thing about Bengal.

V.

Cecil Rhodes understood, perhaps better than any figure of his era, that the charter company model was not a historical accident. It was a replicable technology — a mechanism for converting private ambition into public legitimacy through the instrument of the civilizing mission.

In 1889, Rhodes secured a royal charter for the British South Africa Company, granting it authority over a vast territory in southern Africa that would become Rhodesia. The premise was explicit and unapologetic. Rhodes believed, and said publicly, that the Anglo-Saxon race was the instrument of human progress, and that the expansion of its dominion was synonymous with the advancement of civilization. The Ndebele and Shona peoples who occupied the land had their own governance structures, their own legal traditions, their own relationships with the territory that had sustained them for generations. None of that registered in the colonial accounting. What registered was the assessment that they were not doing enough with what they had — that the land was underleveraged, the resources insufficiently extracted, the human potential inadequately directed.

The Pioneer Column that marched into Mashonaland in 1890 under the BSAC’s flag did not present itself as an invasion. It presented itself as the arrival of development. The telegraph lines, the railways, the administrative infrastructure that followed were genuine — they were also instruments of extraction, built to move value out of the continent and into London balance sheets, structured to make the territory dependent on the company’s continued presence rather than sovereign in its own right. The infrastructure of civilization was, simultaneously, the infrastructure of dependency.

It is worth being precise about how that dependency was engineered, because the mechanism matters more than the metaphor. The Cape railways did not merely connect — they sorted. Freight rates were structured to favor the export of raw materials over local commerce. Gauge specifications were set by the Company, ensuring that maintenance, replacement, and expansion required the Company’s continued involvement. The routing decisions — which settlements became nodes, which became dead ends — concentrated economic gravity at Company-controlled ports. Once this infrastructure was in place, the territory could not function without the entity that had built it. The dependency was not a side effect of development. It was the design.

Rhodes had a vision of a Cape-to-Cairo railway — a spine of connectivity running the length of Africa that would integrate the continent into the imperial economy. The equivalent today is not difficult to identify, and it is already well underway. Google’s Equiano cable and Meta’s 2Africa cable are among the largest undersea infrastructure projects in history, stretching tens of thousands of kilometers along African coastlines, owned outright by the same platforms that will monetize the behavioral data flowing through them. Satellite constellations and server farm archipelagos complete the stack. The hyperscalers are in a race to own the physical substrate of connectivity across the Global South — not as a side project, but as a strategic priority. The railway race has been reimagined in fiber and orbit.

The extraction mechanism is structurally identical. Starlink terminals cost approximately six hundred dollars in markets where that figure represents months of household income, priced in US dollars regardless of local currency. All data transits through US-jurisdictional infrastructure regardless of where the user sits — the CLOUD Act of 2018 obliges American technology companies to produce user data stored anywhere in the world upon government demand, extending US legal sovereignty across every network its companies touch. The routing nodes, like the railway chokepoints before them, are controlled by the infrastructure owner. The territory cannot function without the entity that built it.

In 2022, this dependency became briefly visible in a way that stripped away any remaining ambiguity. Starlink had become critical communications infrastructure for Ukraine’s military. When Elon Musk restricted coverage over Crimea, citing his own assessment of escalation risk, a private individual made a sovereign military decision affecting a nation at war. No charter was required. No democratic mandate was sought. The infrastructure spoke for itself. Nick Couldry and Ulises Mejias, writing in The Costs of Connection, describe this configuration not as an analogy to colonialism but as its structural continuation — the same extractive relationship between data-rich corporations and data-generating populations, the same asymmetry of terms, the same fiction of mutual benefit. The infrastructure of connectivity is, simultaneously, the infrastructure of extraction. The logic is unchanged. Only the medium is new.

VI.

The grammar of the civilizing mission runs through North American colonization with the same consistency. The Indigenous peoples of this continent were not, in the colonial framing, displaced. They were improved. What they had — complex governance systems, sophisticated ecological stewardship, extensive trade networks, rich legal and philosophical traditions — was systematically re-categorized as absence: absence of private property, absence of written law, absence of Christian civilization, absence of the specific forms of organization that the colonizer happened to recognize. The fact that these sophisticated frameworks already existed was not merely overlooked; their existence had to be actively suppressed, because their acknowledgment would have undermined the foundational premise. The colonizer’s value proposition depends entirely on your prior life being characterized as insufficiency.

The Indian residential school system in Canada was perhaps the most explicit articulation of this logic: the complete administrative project of replacing one cognitive architecture with another, by force, across generations, in the name of benefit. The language was always care. The mechanism was always extraction — of land, of labor, of cultural continuity, of the capacity for self-determination. The damage was not incidental to the project. It was structural to it. Systems designed to produce dependency reliably produce dependency.

Arendt recognized this dynamic as something more than historical injustice. In The Origins of Totalitarianism, she identified what she called the boomerang effect: the methods developed in the colonies — surveillance, administrative violence, the reduction of human beings to managed populations — do not stay in the colonies. They return. The techniques of behavioral management pioneered on colonized populations, the logic of the administered human being whose choices are shaped rather than respected, eventually migrates back to the metropole. Surveillance capitalism is, in this reading, the boomerang arriving. The behavioral modification apparatus that Zuboff describes was not invented for application to Western consumers. It was refined there. Its ancestors are the census regimes, the pass systems, the productivity surveillance of colonial administration.

VII.

Arendt offers one more lens that bears directly on the AI governance crisis, and it is perhaps her most unsettling contribution to this analysis.

In Eichmann in Jerusalem, she introduced the concept of the banality of evil — the observation that the most catastrophic acts of institutional harm are rarely driven by malice. They are driven by thoughtlessness. By the abdication of moral reflection in favor of process optimization, metrics, and institutional loyalty. Adolf Eichmann was not a monster in the theatrical sense. He was an administrator. He followed procedures. He met targets. He did not ask whether what he was optimizing was good; he asked whether he was optimizing it efficiently.

The engineers building behavioral modification systems at scale are not monsters either. They are optimizing engagement metrics, recommendation algorithms, content ranking systems. The harm produced — the radicalization pipelines, the eating disorder communities algorithmically reinforced, the political polarization systematically amplified because outrage retains attention — is not the intent of any individual actor. It is the aggregate output of a system designed to maximize a signal without asking whether the thing being maximized is worth maximizing. This is Arendt’s banality, translated from the bureaucratic to the algorithmic. The thoughtlessness is now engineered into the infrastructure itself. It does not require a human to decide not to think. The system does not think on anyone’s behalf, and the absence of thought is mistaken for neutrality.

Arendt also distinguished, in her political philosophy, between power and violence. Power, for her, was genuine and legitimate: it emerges from collective action, from people acting together in concert, from consent and participation. Violence is instrumental, substituting force for consent when power has failed or been refused. The current configuration of cloud capital does not exercise power in Arendt’s sense. It exercises structural violence — the violence of dependency, of terms of service that cannot be negotiated, of infrastructure you cannot opt out of without opting out of the modern economy. The fact that this violence is invisible, mediated through interfaces and algorithms rather than through physical coercion, does not make it less real. It makes it more durable, because it is harder to name and therefore harder to resist.

A precise objection deserves a precise answer. The East India Company governed Bengal with an army. The British South Africa Company administered Rhodesia through courts backed by the credible threat of physical force. The relationship between a platform and its users involves none of that. Nobody is compelled to use Google. Nobody is imprisoned for leaving Facebook. The coercion is categorically different in kind, and conflating the two risks making the analogy do more work than it can honestly bear.

The objection is correct about degree. It is wrong about structure.

The colonial method did not remain static. Early Company rule in Bengal was maintained by muskets; later colonial administration was maintained by railway debt, currency controls, and the systematic dismantling of local textile industries that left populations with no viable economic alternative to participation in the imperial system. By the late nineteenth century, the most effective instruments of colonial control were not military — they were infrastructural. The dependency did the work that the garrison used to do. When a territory’s food supply, its communications, its credit system, and its export routes all run through a single externally-owned architecture, the question of whether you are technically free to leave becomes largely academic. Freedom without a viable alternative is a formal category, not a lived condition.

This is precisely the configuration that platform capitalism has engineered. Opting out of Google’s infrastructure does not mean declining a service. For most small businesses, academic institutions, healthcare providers, and municipal governments, it means opting out of the operational systems on which their basic functioning now depends. The stack has been built, layer by layer, to make exit prohibitively costly — not through legal compulsion but through network effects, switching costs, and the careful bundling of essential services with proprietary lock-in. The guns have been replaced by the terms of service, and the terms of service are enforced not by soldiers but by the architecture itself. That is not the same thing as a musket. It is, structurally, the same thing as a railway gauge.

VIII.

The concentration of wealth and power now underway has no precedent in industrial history for its velocity, its scale, or its abstraction. The compute infrastructure required to train frontier AI models costs hundreds of millions of dollars per run. The training data has been assembled through decades of free behavioral contribution by billions of users who had no meaningful understanding of what they were building. The physical layer — the cables, the chips, the server farms — is owned by a vanishingly small number of entities. Each layer of this stack is a toll road, and the tolls compound.

What is new is the object of extraction: not cotton, not rubber, not oil, not labor in the classical sense — but cognition. The raw material of the twenty-first century economy is the structured output of human minds, digitized, aggregated, and used to train systems that will progressively displace the humans who generated the training data in the first place. The process that Harry Braverman identified in the deskilling of industrial labor — the systematic decomposition of skilled craft into measurable, manageable, replaceable components — has now reached the cognitive tier. The writers, analysts, designers, coders, diagnosticians, and translators who believed their work was too complex, too contextual, too human to be automated are discovering that what made their work valuable was precisely what made it trainable. The data they generated, freely and without compensation, is the capital that renders them disposable.

This is not metaphor. It is the next phase of a process that has been running for two centuries, now applied to the one domain that workers believed was safely theirs. The Metaproletariat is not a class defined by what it lacks. It is defined by what has been taken from it without its recognition or consent.

IX.

The question worth asking — the question colonial subjects have always eventually had to ask — is: who decided that this particular arrangement of ownership and governance was necessary? Who benefits from the premise that intelligence, at scale, must be privately held? What would it look like if the foundational infrastructure of the AI era were treated as a public good?

These are not naive questions. They have been answered before, in practice, by democratic societies that chose to treat foundational technologies differently.

The electromagnetic spectrum is not owned. It is licensed. The decision, made in the early twentieth century, that the airwaves constituted a public commons whose commercial use required public authorization was not technologically inevitable. It was political. The Human Genome Project made a deliberate choice in 2000 to release its sequencing data into the public domain rather than allow private patents — a decision that required political will against significant commercial pressure, and that has saved hundreds of billions of dollars in pharmaceutical research costs in the decades since. CERN, the multinational physics institution that produced the World Wide Web as a public good, is funded by member states and governed by international treaty. The internet’s foundational architecture was built with public money, placed in the public domain, and deliberately designed without a central owner. These are not historical accidents. They are records of political decisions that went differently than the AI governance decisions are currently going.

The policy levers exist. A public compute infrastructure — equivalent to the national laboratory system, accessible to researchers, governments, and civil society organizations — would break the capital barrier that currently makes frontier AI development the exclusive province of six companies. Mandatory licensing of foundational models for public interest use, analogous to compulsory licensing in pharmaceutical law, would make the tools accessible without requiring public ownership of the development process. Data trusts — legal structures that allow communities to pool and govern their own behavioral data rather than surrendering it individually to platforms — would shift the terms of the training data relationship. None of these require dismantling the private sector. They require treating the foundational layer of the AI economy the way democratic societies have, at their best, treated other foundational layers: as infrastructure, not inventory.

These questions are not being asked in the rooms where governance frameworks are being written. They are being crowded out by the urgency of safety discourse — which is real, which matters, and which also conveniently forecloses the prior question of whether the concentration of this capability in private hands is itself the primary safety problem. Safety governance that takes the ownership structure as given is not neutral. It is a choice about who the system is designed to protect.

Arendt believed, even against the evidence of the twentieth century, in what she called natality — the human capacity to begin something genuinely new, to act in ways that could not have been predicted from prior conditions. It was her counter-argument to determinism, to the totalizing systems that claimed to know what history required. The precedents exist. The mechanisms are known. What is required is the political will to treat this moment as a genuine decision rather than an inevitable one — to recognize that the structure being built right now is a choice, and that choices of this magnitude can be made differently.

History will record what we built in this window. The question is whether we are building something that distributes the gains from intelligence broadly, or constructing a new enclosure — one whose walls are made not of stone or wire, but of weights and tokens and proprietary compute, and whose lords speak the language of responsibility while practicing the logic of extraction.

The East India Company had a charter. The British South Africa Company had a charter. They both insisted, until the evidence was overwhelming, that their governance was in the interest of those they governed.

Structure before sophistication. Every time.

The structure being built right now is feudal. And the sophistication of its justifications does not change what it is.

Citations & Sources

1. Varoufakis, Yanis. Technofeudalism: What Killed Capitalism. Bodley Head, 2023. The concept of ‘cloud capital’ and the feudal rent-extraction model of platform economies is drawn directly from Varoufakis’s central argument.

2. Zuboff, Shoshana. The Age of Surveillance Capitalism: The Fight for a Human Future at the New Frontier of Power. PublicAffairs, 2019. The framework of ‘behavioral modification at scale,’ ‘prediction products,’ and the epistemological character of surveillance capitalism is drawn from Zuboff’s foundational text.

3. Arendt, Hannah. The Origins of Totalitarianism. Harcourt Brace, 1951. Part II: ‘Imperialism.’ The analysis of imperialism as capital overflow, the administrative apparatus of colonial governance, and the ‘boomerang effect’ — colonial methods returning to the metropole — are all developed in this section.

4. Arendt, Hannah. The Human Condition. University of Chicago Press, 1958. The concept of the ‘common world’ and its erosion by private accumulation; the distinction between the public realm and the private household economy; the political implications of common goods migrating into private hands.

5. Arendt, Hannah. Eichmann in Jerusalem: A Report on the Banality of Evil. Viking Press, 1963. The concept of the ‘banality of evil’ — institutional harm produced not by malice but by the systematic abdication of moral reflection — is used here as an analogy for algorithmic harm at scale.

6. Braverman, Harry. Labor and Monopoly Capital: The Degradation of Work in the Twentieth Century. Monthly Review Press, 1974. The deskilling thesis — the systematic decomposition of skilled labor into measurable, manageable components — is extended here to cognitive and knowledge work in the age of generative AI.

7. Davis, Mike. Late Victorian Holocausts: El Niño Famines and the Making of the Third World. Verso, 2001. Background on the structural violence of colonial economic administration, including the East India Company’s revenue extraction from Bengal.

8. Ferguson, Niall. Empire: The Rise and Demise of the British World Order and the Lessons for Global Power. Basic Books, 2002. Historical background on the East India Company charter model and the BSAC / Rhodesia land appropriation.

9. Rotberg, Robert I. The Founder: Cecil Rhodes and the Pursuit of Power. Oxford University Press, 1988. Historical background on Rhodes, the British South Africa Company, and the Pioneer Column of 1890.

10. The Battle of Plassey (1757) and the East India Company’s Diwani grant (1765) are documented in standard histories of British India. See also: Dalrymple, William. The Anarchy: The East India Company, Corporate Violence, and the Pillage of an Empire. Bloomsbury, 2019.

11. Truth and Reconciliation Commission of Canada. Calls to Action. 2015. Background on the residential school system as an administrative project of cultural and cognitive displacement.

12. Arendt, Hannah. ‘On Violence.’ In Crises of the Republic. Harcourt Brace Jovanovich, 1972. The distinction between power (legitimate, collective, consensual) and violence (instrumental, substituting for power when it fails) is drawn from Arendt’s political philosophy as developed across multiple works.

13. Couldry, Nick, and Ulises A. Mejias. The Costs of Connection: How Data Is Colonizing Human Life and Appropriating It for Capitalism. Stanford University Press, 2019. The concept of “data colonialism” as a structural continuation of historical colonialism — not metaphor but mechanism — is drawn from Couldry and Mejias’s central argument.

Structure Before Sophistication | structurebeforesophistication.substack.com

Borrowed Ambition: The Case For and Against Canada’s Sovereign Wealth Fund

The financial logic is real. The execution risk is harder to ignore.

When Mark Carney announced the Canada Strong Fund last week — a $25 billion sovereign wealth fund to be capitalized and deployed over three years — the political reaction split instantly along predictable lines. Poilievre called it a “sovereign debt fund.” The government called it generational wealth-building. Both characterizations contain truth. Neither is complete.

Let me try to do what the political theatre won’t: give the argument a fair hearing on both sides, then tell you where I land.

The case for it

The financial logic is not invented. It’s leveraged public investment — the same principle any competent CFO applies when they issue debt to fund a capital project that earns more than it costs. Canada’s sovereign borrowing rate is among the lowest available to any actor in the global capital markets. If the fund can generate returns that exceed that cost of debt — a reasonable ambition for long-duration infrastructure and energy assets — the spread represents real wealth creation. The math isn’t crazy.

There’s also a crowding-in argument worth taking seriously. A $25 billion government anchor in a co-investment structure doesn’t stop at $25 billion — it theoretically mobilizes private capital multiples of that amount into projects that, without a credible first-loss or co-investment partner, would never attract institutional money. Canada has no shortage of viable nation-building projects. It has a shortage of capital willing to take the first step on projects with long horizons and regulatory uncertainty.

And finally, the retail participation piece is genuinely interesting. Opening the fund to individual Canadians — structured like a government bond with principal protection — is a mechanism to redirect domestic savings toward productive national investment rather than into housing speculation or US-listed equities. That’s not a trivial idea. Channeling Canadian capital into Canadian value creation is exactly the kind of structural intervention the economy has needed for a generation.

The case against it

The honest problem starts with naming. A sovereign wealth fund is a stock of surplus wealth deployed for generational return. Norway’s is the canonical example — funded by oil revenue windfalls, disciplined by a strict spending cap, and grown to over $2 trillion through decades of return compounding. Canada’s version is a leveraged investment vehicle in a country running an $78 billion deficit. That isn’t wealth management. That’s borrowed ambition. And borrowed ambition has carrying costs that begin accruing on day one, long before the projects deliver returns.

The deeper problem is mandate clarity — or the absence of it. Is this fund designed to maximize financial returns? Or is it a policy instrument dressed in investment language? Because those are different things with different performance metrics, different governance structures, and different accountability frameworks. Major nation-building projects — pipelines, corridors, energy infrastructure — are characterized by economic benefits that exceed their direct financial returns. If you conflate those two things inside a single fund structure, you will eventually be justifying below-market returns with policy rationale. That’s not wealth creation. That’s subsidized spending with a better press release.

And then there’s the timeline problem. Major projects can take a decade or more to generate financial returns. Interest on borrowed capital starts on day one. The fund has to carry a negative spread for years before any compound return logic kicks in. That’s a fiscal endurance test in an environment where deficits are already politically sensitive, election cycles are short, and the pressure to redirect fund capital toward politically expedient projects will be constant.

Where I land

I’ve seen enough government-adjacent investment vehicles to know that the instrument itself is rarely the failure. The failure is almost always governance. Who makes investment decisions? By what criteria? Under what independence from ministerial pressure? What is the return floor that triggers a review? What is the exit mechanism when a project underperforms?

None of those questions have been answered. The Finance Minister, when pressed on the specifics, said the details were forthcoming. That’s the part that gives me pause — not the financial concept, but the absence of architecture at the moment of launch. The announcement preceded the structure. The headline preceded the design. That sequencing is a pattern I’ve seen before, and it doesn’t bode well.

The argument for the Canada Strong Fund is real. So is the argument against it. The outcome will be determined entirely by the governance and accountability framework that gets built in the coming months — the part nobody is talking about yet. That’s where this either becomes nation-building or becomes a $25 billion lesson about the difference between a good idea and a disciplined execution.

We’ve borrowed the ambition. Now we need to build the structure.

Analytical Brief • April 2026

Liberation Day Tariffs: One Year On

The Wrong Problem, the Wrong Solution — and What a Year of Data Confirms

Before We Begin: How America Got Here

Liberation Day did not happen in a vacuum. It happened at the end of a very long road — and understanding that road is the difference between treating this as a political story and treating it as a structural one.

The United States has been spending tomorrow’s money today for more than fifty years. The inflection point, if you want to find one on a map, is August 15, 1971. That was the day Richard Nixon closed the gold window — severing the dollar’s convertibility to gold and ending the Bretton Woods system that had anchored global monetary order since 1944. It was a necessary decision under the pressures of that moment. It was also the moment the last hard constraint on American deficit spending was removed.

What followed was a half-century of leveraged prosperity. Military engagements across Vietnam, the Gulf, Iraq, and Afghanistan — some entered without clear strategic rationale, few concluded with durable diplomatic outcomes — consumed trillions without a financing mechanism beyond borrowing. Social entitlements expanded. Tax revenues were repeatedly cut. And each successive administration, regardless of party, found that the debt ceiling was a political obstacle to be managed rather than a fiscal limit to be respected.

Paul Volcker arrived at the Federal Reserve in 1979 and did what needed doing — broke the back of inflation with interest rates that reached 20%, triggered a brutal recession, and restored monetary credibility at enormous short-term cost. It was the kind of structurally painful, politically courageous decision that the subsequent forty years conspicuously lacked.

What followed Volcker was Alan Greenspan — eighteen years as Fed chairman, and the most influential voice for financial deregulation in modern American history. Greenspan was not coerced into his worldview. He was a true believer — a devoted Ayn Rand disciple who held as foundational truth that free, competitive markets are self-regulating and that government intervention invariably makes things worse. Under his watch, derivatives markets expanded without oversight, mortgage lending standards collapsed without consequence, and Wall Street built an architecture of financial instruments — credit default swaps, collateralized debt obligations, synthetic securities — that transferred risk so many times it became invisible. Until it wasn’t.

In October 2008, Greenspan sat before a congressional committee and admitted he had found ‘a flaw’ in the model he had spent forty years applying to the world. He was, by his own account, in a state of ‘shocked disbelief’ that lending institutions had not self-regulated to protect their shareholders. The shock was genuine. That is precisely what makes it so damning. He wasn’t wrong because he was corrupt. He was wrong because he was certain — and certainty, in complex systems, is its own category of risk.

The 2008 financial crisis did not reset the fiscal clock. It accelerated it. Emergency stimulus, bank bailouts, quantitative easing, and a decade of near-zero interest rates kept the system alive while adding layers of debt that are now becoming very expensive. In the first quarter of fiscal 2026, the United States paid 70.3 billion in net interest on its debt — more than the 66.9 billion spent on national defense in the same period. That crossover — the moment debt servicing exceeded the defence budget — passed without ceremony. It should not have.

Debt Milestone

Year

Context

Nixon closes gold window; Bretton Woods ends

1971

Last hard constraint on deficit financing removed

Volcker raises rates to 20%; inflation broken

1980–82

Structurally painful; politically courageous; not repeated

Greenspan era: deregulation of derivatives markets

1987–2006

Self-regulating market ideology applied at scale for 18 years

Greenspan admits ‘flaw’ before Congress

Oct 2008

40 years of ideology revealed as structurally wrong

QE, bailouts, near-zero rates post-GFC

2008–2022

Debt accelerated; reckoning deferred not resolved

Net interest exceeds defense spending

FY2026

70.4B interest in FY2025; trillion projected FY2026 (CBO)

Gross government debt at 123.9% of GDP

2025

IMF projects 140% by 2031; 50-year average was 70%

This is the predicament Liberation Day was meant to address. An economy with the world’s largest appetite for goods and services, a manufacturing base systematically offshored over four decades, a fiscal position deteriorating since the early 1970s, and a political culture in both parties that has consistently chosen the short-term comfort of borrowing over the long-term discipline of structural adjustment. That is not a trade problem. It is not something you solve with a tariff formula.

What follows is an attempt to look at that predicament honestly — through data, through the frameworks that actually explain competitive advantage and market structure, and through a full year of evidence about what Liberation Day actually accomplished. It is more fact than storytelling. The story, frankly, tells itself.

Structure before sophistication. Every time.

SOURCING NOTE

This brief synthesises analysis from Tech Insider (April 8, 2026), AP News coverage of TSMC Q1 earnings, and primary-source data from the Tax Foundation, Bureau of Labor Statistics, Semiconductor Industry Association, and Council on Foreign Relations. The intellectual framework draws on Michael Porter’s The Competitive Advantage of Nations (1990) and Chris Hughes’s Marketcrafters: The 100-Year Struggle to Shape the American Economy (2025). Where sourcing warrants further corroboration — particularly BLS job figures and IEEPA ruling specifics — Reuters and Financial Times primary coverage is flagged for verification.

1. Executive Summary

One year after President Trump announced sweeping “Liberation Day” tariffs on April 2, 2025, the record is in — and it does not support the policy’s stated rationale. Rather than reshoring American manufacturing, the tariffs triggered the legal nullification of their broadest provisions, accelerated supply chain flight from China to third-country alternatives, imposed an “uncertainty tax” that froze capital investment across the economy for most of 2025, and raised prices on the workers they claimed to champion.

Three structural outcomes now define the landscape:

Outcome

Finding

Legal collapse

The Supreme Court struck down all IEEPA-based tariffs in February 2026, eliminating the 10% baseline and all reciprocal rates — but leaving Section 232 semiconductor tariffs intact.

Manufacturing inversion

U.S. factories lost 89,000 jobs in the ten months following Liberation Day, directly inverting the reshoring narrative.

Supply chain arbitrage

Companies moved assembly to India and Vietnam — not the United States. What tariffs changed was the cost calculus for China-versus-third-country, not China-versus-America.

The deeper thesis is this: Liberation Day was sophistication — aggressive, theatrical, globally disruptive — deployed in the total absence of structure. The policy conflated three distinct problems, applied a legally fragile instrument to all of them, and solved none. To understand why, we need to understand what problem the tariffs were actually designed to address — and why that framing was wrong from the start.

2. The Wrong Problem

A trade deficit is not a wound. It is a signal — and like most signals, it only makes sense if you understand the system producing it.

When Americans run a persistent goods trade deficit, what they are revealing is a set of interlocking structural realities: a high-consumption economy, deep capital markets that attract foreign investment, comparative advantages that have migrated toward services, finance, and intellectual property, and decades of domestic policy choices — on savings rates, healthcare costs, education, and industrial support — that shaped what Americans make, what they buy, and from whom. None of that is exploitation. It is the accumulated ledger of choices made over generations.

Key Distinction

Liberation Day didn’t address any of those choices. It addressed the ledger. Targeting the trade deficit with tariffs is the policy equivalent of treating a fever by smashing the thermometer. The number changes. The underlying condition does not.

Treating a structural imbalance as a national emergency — invoking the International Emergency Economic Powers Act (IEEPA) to justify sweeping tariff authority — was always constitutionally vulnerable. The Supreme Court’s February 2026 ruling in Learning Resources, Inc. v. Trump did not create the policy’s failure; it revealed a structural vulnerability that was baked in from day one. A tariff regime built on fragile legal foundations and administered through 50-plus policy changes in twelve months could not deliver coherent supply chain transformation regardless of the underlying strategic logic.

The structure was wrong before the sophistication could help.

3. What Porter Actually Said

In 1990, Michael Porter published The Competitive Advantage of Nations — a study of why certain industries flourish in certain countries and not others. His central argument is deceptively simple and profoundly important: national prosperity is created, not inherited. It does not grow out of a country’s natural endowments, its labour pool, its interest rates, or its currency’s value. It grows from four interlocking conditions — what Porter called the Diamond:

Condition

What It Means

U.S. Status (2025)

Factor Conditions

Skilled labour, infrastructure, knowledge base

Deteriorating — skills gap in millions of trade/technical positions; infrastructure ranked below peer nations

Home Demand Conditions

Sophisticated domestic buyers driving producer innovation

Strong but consumption-oriented; not pressuring domestic producers to innovate at the frontier

Related & Supporting Industries

Supplier clusters, industrial commons, regional ecosystems

Hollowed by offshoring — do not reassemble quickly under price signals

Firm Strategy & Rivalry

Intense domestic competition driving continuous improvement

Reshaped by consolidation more than intensified by competition

Run Porter’s Diamond over the tariff year and the diagnosis becomes uncomfortable. Tariffs do not fix any of the above. They are a price signal, not a capability. They make foreign goods more expensive without making domestic goods better.

Porter’s research offers a further warning that applies directly to the semiconductor debate: protection does not build competitive advantage — it delays its erosion while preventing its development. The industries that emerge from long protection are rarely the ones that win internationally when the wall eventually comes down. The one-year record confirms this: manufacturing employment continued declining after Liberation Day. Companies did not invest in American factories behind the tariff wall. They invested in their lawyers, their lobbyists, and their supply chain consultants to route around it.

Porter’s Core Warning

“No nation can be competitive in everything. The ideal is to deploy the nation’s limited pool of human and other resources into the most productive uses.” Tariffs that protect uncompetitive assembly while starving resources from sectors where the U.S. can actually win are not industrial strategy. They are industrial nostalgia.

4. The Hughes Problem: Right Diagnosis, Wrong Instrument

Chris Hughes’s Marketcrafters: The 100-Year Struggle to Shape the American Economy (Simon & Schuster, 2025) arrives at precisely the right moment — and provides the most precise intellectual indictment of Liberation Day, even though it does not set out to do so.

Hughes traces a lineage from Alexander Hamilton forward, arguing that America’s greatest periods of economic dynamism were never the product of free markets left alone, but of deliberate “marketcrafters” — policymakers who combined expertise with judgment to cultivate the economy like a garden. His core insight is that markets are never neutral. They are always the product of political choices. The only question is whether those choices are made skilfully or carelessly.

In recent decades, Hughes argues, the art of marketcraft has been lost — replaced by the myth that markets work best when unfettered and free. That neglect is real. The deregulatory consensus offshored industrial capacity without replacement strategies, gutted regional economies, and left entire communities stranded. The anger that produced Liberation Day is legitimate. The problem it identified is genuine.

Here is where I part ways with the implied remedies — and where Liberation Day exposes a failure that runs deeper than bad policy.

The Mirror Image Problem

Liberation Day is not a correction to failed marketcraft. It is its mirror image — equally blunt, equally mythological, just inverted in its logic. If free markets broke everything, the thinking goes, then aggressive state intervention must fix it. Tariffs become the instrument of a government that never learned to be a surgeon and reached for a sledgehammer instead.

Good marketcraft — the kind Hughes documents actually working — requires patience, precision, and a theory of the market you are trying to build. It requires something to stand behind the wall. Liberation Day raised the wall. Behind it: no apprenticeship pipeline, no grid capacity for new factories, no coordinated supply chain development, no patient capital mechanism. The tariff went up, and the factory didn’t come. Because the factory was never the next step. It was ten steps down a road nobody built.

The CHIPS Act gestures at real marketcraft. The Inflation Reduction Act gestures at it. Liberation Day does not gesture at it at all — it is a price signal applied universally, indiscriminately, without a reconstruction plan sitting behind it.

5. The Problem Nobody Wants to Name: Fiscal Discipline

There is a third dimension to this that both the free-market consensus and the new industrial policy advocates consistently avoid: the United States is addicted to spending tomorrow’s money today. And until that addiction is named honestly, no trade policy — however aggressive — can solve the problem it claims to be addressing.

The trade deficit does not exist in isolation. It has a structural twin — the fiscal deficit. These are not independent phenomena. When the U.S. government borrows at scale, it exports dollars into the global financial system. Those dollars return as capital inflows — foreign investment in U.S. assets, treasuries, real estate, and equities. The accounting identity that makes this work mechanically requires a current account deficit as its mirror. In plain terms: you cannot run chronic fiscal deficits and simultaneously eliminate your trade deficit. The mathematics do not allow it. One feeds the other. Always.

Indicator

Latest Data

Context

Federal deficit (FY2025)

5.9% of GDP ($1.8T)

55% above the 50-year average of 3.8% (CBO). Exceeded only eight times since 1946.

General government debt (2025)

123.9% of GDP

IMF projects debt exceeding 140% of GDP by 2031, with deficits persisting at 7–7.5% of GDP.

Current account deficit (2025)

$1.12 trillion (3.6% of GDP)

Essentially flat vs. $1.13 trillion in 2024 — a full year of tariffs moved the needle less than 1% (BEA).

Personal savings rate (2025 avg.)

4.72% of disposable income

More than 4 pts below the long-run average of 8.4% since 1959. Decade peak: 12% in the 1970s (BEA/FRED).

U.S. net international investment position

–$27.54 trillion (end-2025)

The balance sheet consequence of decades of current account deficits. Worsened by $1 trillion in 2025 alone (BEA).

The savings rate collapse is not a footnote — it is the structural story. In the 1970s Americans saved an average of 12% of disposable income. In the 2000s that average fell to 4%. It runs at 4.72% today, and for the bottom half of the income distribution, net savings are negative. A nation that does not save cannot self-finance investment. It borrows. And sustained national borrowing — public and private — produces exactly the capital account dynamics that make the current account deficit a mathematical necessity, not a policy failure that tariffs can fix.

The 2025 current account data makes the Liberation Day argument collapse on its own terms. The full-year deficit was $1.12 trillion — nearly identical to the $1.13 trillion recorded in 2024. A full year of the most aggressive tariff regime since the 1930s moved the trade needle by less than one percent. Why? Because tariffs did not change the savings rate. They did not change the fiscal deficit. They did not change the capital flow dynamics that make the current account balance what it is. They changed the price of Chinese goods — and American importers simply sourced from Vietnam, India, and Mexico instead.

The Accounting Identity That Tariffs Cannot Override

Current Account Balance = National Savings − National Investment. When a government runs large deficits and households save less than they invest, the difference must be financed externally. That external financing — by definition — appears as a current account deficit. No tariff can override this identity. It is arithmetic, not ideology.

Liberation Day did not touch any of this. Tariffs are a price instrument applied at the border. They have no mechanism to address the savings rate, the structural fiscal imbalance, the healthcare burden that makes U.S. manufacturing structurally expensive relative to peers, or the four-decade institutional drift toward consumption over investment.

Fiscal discipline is not a popular argument in either political coalition right now. Both parties have made peace with structural deficits as the cost of governing. But it is the argument that Porter’s framework, Hughes’s history, and basic macroeconomic accounting all converge on. You cannot build the Diamond by borrowing against the future. You cannot craft a market you cannot afford to sustain. And you cannot resolve a structural external imbalance by taxing imports while the underlying savings-investment gap remains wide open.

5.1 The Debt Threshold Question: Is There a Safe Number?

The short answer is: it depends entirely on who you are. The debt-to-GDP ratio is not a universal danger gauge — it is a context-sensitive indicator whose meaning changes dramatically based on currency sovereignty, domestic savings depth, reserve currency status, and whether debt is denominated in your own currency or someone else’s.

The Academic Debate

The most cited framework comes from Carmen Reinhart and Kenneth Rogoff’s 2010 paper Growth in a Time of Debt, which argued that the relationship between government debt and real GDP growth is weak below a threshold of 90% of GDP, above which median growth rates fall by roughly one percent and average growth falls considerably more. This 90% figure became a political thunderbolt, used to justify austerity across Europe in the post-2008 period.

The problem is that the research was subsequently challenged. When the coding errors were corrected and all data included, the difference in median growth between countries at 60–90% debt and those above 90% shrank to 0.4 percentage points — from 2.9% to 2.5%. There was no cliff. The IMF’s own research concluded there is no magic threshold above which growth prospects collapse dramatically, but that the medium-to-long-term relationship is real: higher debt today does predict lower growth over time. A World Bank study across 99 countries placed the operative threshold closer to 77% — with every 10 percentage point increase in the debt-to-GDP ratio associated with a 0.2 point drag on annual per capita growth.

Current IMF Prudent Target

“Bringing debt down to the estimated prudent levels of 52–54% of GDP is an achievable goal. Gradual fiscal consolidation focused on improving spending efficiency rather than drastic cuts can mitigate negative impacts on growth and social programs.” — World Economic Forum / IMF, 2024

The Reserve Currency Exemption — and Its Limits

The reason the U.S. and Japan can operate at debt levels that would collapse a developing economy is not that the rules of arithmetic are different for them. It is that several structural factors defer the consequences. Japan’s gross debt exceeds 255% of GDP — the highest of any major economy — yet markets remain calm. The explanation: Japan’s debt is approximately 92% domestically held, its private sector carries over ¥2,000 trillion in financial assets, and its net debt — after subtracting government financial assets — falls to approximately 114% of GDP. Sovereign debt issued in your own currency, held by your own institutions, and serviced by a central bank with unlimited issuance capacity operates under fundamentally different sustainability rules than debt owed in someone else’s currency.

The U.S. holds a different but related privilege: the dollar is the world’s reserve currency, meaning global demand for dollar-denominated assets structurally finances American deficits. This is what economist Barry Eichengreen called the “exorbitant privilege” — the ability to borrow cheaply and persistently because the world needs your currency. But this privilege is not infinite, and it is not unconditional. With general government debt at 123.9% of GDP and the IMF projecting it will exceed 140% by 2031 — with deficits persisting at 7–7.5% of GDP — the U.S. is stress-testing how far the exemption extends.

The Critical Variable: r vs. g

Debt is self-liquidating when the economy’s growth rate (g) exceeds the interest rate on debt (r). When r exceeds g, the debt trajectory becomes self-reinforcing regardless of reserve currency status. This is the dynamic the U.S. is approaching as interest costs compound and growth moderates.

The Emerging Market and BRICS Discount

For emerging markets, the calculus is fundamentally and brutally different. The key variable is not the debt level itself but the currency it is denominated in. Research describes the ‘debt intolerance’ phenomenon: interest rates in developing economies can spike above the ‘tolerance ceiling’ even if debt levels are considered manageable by advanced country standards. In OECD countries, the average share of foreign currency-denominated debt is 6%, with most exposure hedged. In emerging market and developing economies it is approximately 20% — and for smaller economies, substantially higher. When you borrow in someone else’s currency and your own depreciates, your debt load explodes without any new borrowing.

This explains why Brazil, South Africa, and many smaller BRICS-adjacent economies operate under acute debt stress at ratios that would not trouble Germany or Canada. The median public debt for emerging markets and developing economies sits at roughly 56% of GDP — approximately half the U.S. level — yet many face deeper fiscal vulnerability due to higher borrowing costs, foreign currency exposure, thinner domestic capital markets, and revenue bases too narrow to service obligations without crowding out productive investment.

Country / Group

Debt-to-GDP

Tolerance Driver

Key Risk

United States

~124%

Reserve currency, deep domestic markets

Deficit structural at 5.9% GDP; privilege not unconditional

Japan

~255% gross / ~114% net

~92% domestically held; high private savings; currency sovereign

Aging demographics; BoJ exit risk; interest rate sensitivity

EU average

~90%

Mixed — ECB backstop, but no fiscal union

Divergence between Germany (~65%) and Italy (~140%)

China

~85% (rising)

State-directed capital; domestic currency; closed capital account

Hidden local government debt; property sector liabilities

BRICS ex-China median

~56%

Lower absolute debt

Foreign currency exposure; high borrowing costs; thin markets

Low-income developing

~55% median

HIPC relief legacy

Revenue/GDP ratios below 15% in many cases; near-crisis risk

IMF ‘prudent’ target

52–54%

Genuine fiscal space; shock-absorbing capacity

Requires sustained primary surpluses in most advanced economies

What This Means for the U.S. Argument

The U.S. is not Japan. Japan’s extraordinary debt sustainability rests on a domestic savings rate that is structurally the mirror image of America’s — a nation of savers carrying a government of borrowers. The U.S. runs a personal savings rate of 4.72% and declining, a fiscal deficit at 5.9% of GDP against a 50-year average of 3.8%, and a net international investment position of negative 7.5 trillion. The reserve currency exemption buys time. It does not suspend the arithmetic indefinitely.

The comparative literature converges on a precise lesson: advanced economies with currency sovereignty and deep domestic capital markets can carry higher debt than textbook thresholds suggest — but only so long as the interest rate on that debt remains below the economy’s growth rate. When r exceeds g, the trajectory becomes self-reinforcing regardless of reserve currency status. That is the dynamic the U.S. is now approaching. And no tariff regime addresses it.

6. What the Data Actually Shows

6.1 The Policy Whiplash

The tariff architecture that Liberation Day created changed more than 50 times in twelve months. The Economic Policy Uncertainty Index averaged 264% higher in the tariff year than in 2024. This volatility — not the tariffs themselves — may be the most consequential variable of the entire episode. The tech sector operates on multi-year product development cycles. It cannot make factory investment decisions in an environment where the cost structure changes quarterly.

Event

Date

Impact

Liberation Day: 10% baseline + 11–50% reciprocal tariffs

Apr 2, 2025

Nasdaq −4.2% in two days

90-day pause on reciprocal tariffs (China excluded)

Apr 9, 2025

Partial relief for non-China supply chains

Electronics exemptions: smartphones, monitors

Apr–May 2025

Apple, Samsung shielded from worst impacts

Section 301 China tariffs renewed and expanded

Jul 2025

Chinese imports face 100%+ effective rates

Section 232: 25% on advanced AI chips (H200, MI325X)

Jan 14, 2026

Nvidia and AMD international sales affected

U.S.–Taiwan semiconductor trade agreement

Jan 15, 2026

Taiwan commits $250B in U.S. manufacturing

Supreme Court strikes down all IEEPA tariffs (6–3)

Feb 20, 2026

Baseline and reciprocal tariffs terminated

CBP terminates IEEPA tariff collection

Feb 24, 2026

Immediate cost relief for non-Section 232 goods

6.2 The Manufacturing Inversion

The central contradiction of Liberation Day: a policy designed to revive American manufacturing oversaw the loss of 89,000 factory jobs in its first ten months. Transportation and warehousing shed an additional 123,700 jobs over the same period. Manufacturing activity contracted for eight consecutive months. Small-business bankruptcies rose 10% year-over-year. Large corporate bankruptcies reached their highest level since 2010.

SOURCING NOTE

BLS job figures cited via Rethink Trade. Direct BLS dataset verification and Reuters corroboration recommended before using in formal published analysis. The directional finding — manufacturing jobs declined post-tariff — is consistent with AP News headline coverage from March 2026 and Tax Foundation analysis.

The average American household paid an additional $1,000–$1,300 in tariff-driven costs in 2025 and early 2026 (Tax Foundation). In a January 2026 Council on Foreign Relations poll, more than 65% of respondents reported that tariffs made a large range of everyday goods less affordable.

This is the sharpest indictment of the policy on its own terms: it promised to help workers and hurt trading partners. It did the reverse. Foreign exporters bore approximately 6% of the tariff cost in the form of lower export prices. American consumers and businesses bore the remaining 94% (New York Federal Reserve, August 2025 analysis).

6.3 The Tech Sector: A Case Study in Structural Incoherence

The technology industry’s experience of Liberation Day illustrates what happens when a blunt instrument is applied to a highly complex, globally integrated sector without a theory of change.

Apple accelerated Indian manufacturing from roughly 7% to 18–20% of global iPhone production, targeting 40%+ by 2027. Samsung expanded its Vietnam operations. Cloud providers accelerated procurement from non-Chinese suppliers. Not one of these moves represents reshoring. They represent supply chain arbitrage — the relocation of assembly from one low-cost country to another to exploit tariff differentials. Labor cost differentials make U.S. consumer electronics assembly economically nonviable regardless of tariff levels. What tariffs changed was the calculus between China and India, not between China and America.

The Arbitrage Thesis

The lasting structural shift from Liberation Day is not reshoring — it is supply chain arbitrage. Porter’s Diamond explains exactly why: the United States lacks the factor conditions (cost structure, labour availability) and related supporting industries (component manufacturing clusters) to make consumer electronics assembly competitive. Price signals alone cannot conjure those conditions into existence.

6.4 The Semiconductor Exception — and Its Limits

The 25% Section 232 tariff on advanced AI chips is the most surgically coherent element of the surviving tariff architecture. Unlike the broad IEEPA tariffs, it targets a specific, strategically significant product category and includes a logical exemption structure: chips for U.S. data centres, R&D, startups, and domestic manufacturing are exempt; chips destined for re-export — particularly to China — are not.

Against this backdrop, TSMC reported a 58% jump in Q1 2026 profits, driven by surging AI chip demand. This is the central complicating data point for a simple “tariffs failed” narrative. But the TSMC result reflects insatiable AI training demand from U.S. hyperscalers, TSMC’s dominant position in leading-edge fabrication (which has no short-term substitute), and CHIPS Act incentive structures that predate Liberation Day. TSMC’s Arizona expansion — $165B committed — is a product of patient marketcraft, not tariff shock.

Key Distinction

Where the tariff architecture survived (Section 232, Section 301) it is doing so because it is legally grounded and surgically targeted. Where it was broad and legally fragile (IEEPA), it collapsed. The lesson is not that tariffs cannot be part of industrial strategy. The lesson is that surgical instruments require surgical precision — and Liberation Day was not a scalpel.

7. China’s Strategic Response: Porter Vindicated

Porter’s framework makes a specific prediction about what happens when external pressure is applied to a nation’s industry: if the domestic conditions are present — government support, domestic rivalry, related industries, ambitious demand — the pressure accelerates innovation rather than suppressing it. China’s response to Liberation Day is a case study in that prediction coming true.

Facing effective rates exceeding 100% on many tech exports, Chinese companies and the Chinese state responded structurally. Huawei launched the Ascend 950PR — a 1.56 petaflop AI chip positioned as an explicit Nvidia alternative — in early 2026. The National Integrated Circuit Industry Investment Fund allocated an estimated $47B toward domestic chip fabrication acceleration. Chinese firms doubled down on export markets outside the U.S. to offset American demand loss.

SOURCING NOTE

Huawei Ascend 950PR specs and the $47B chip fund figure are sourced via Tech Insider citing industry reports. Both require primary-source verification before use in formal published analysis — HIGH priority.

The net effect is the acceleration of what was already a multi-year technology decoupling trend. Liberation Day compressed a decade-long transition into a much shorter timeline. The tariffs designed to weaken China’s technology sector may have done more to sharpen its urgency to achieve semiconductor self-sufficiency than any other single policy event in the past decade. That is the opposite of the intended strategic outcome.

This is what happens when you reach for a sledgehammer in a situation that required a scalpel, a blueprint, and a decade of patient construction.

8. The Comprehensive Thesis

Liberation Day tariffs were structurally incoherent as industrial policy because they conflated three distinct problems requiring three distinct solutions:

Problem

What It Required

What Tariffs Delivered

Supply chain resilience (overconcentration in China)

Diversification incentives, strategic reserves, allied reshoring coordination

Supply chain arbitrage to India and Vietnam

Advanced semiconductor strategic competition

Surgical, legally durable instruments targeted at specific strategic products

Broad emergency-powers tariffs, constitutionally struck down

Domestic manufacturing employment decline

Workforce development, skills investment, automation adjustment policy

89,000 additional manufacturing job losses

Trade deficit (the stated target)

Fiscal discipline, savings rate reform, cost-of-production competitiveness

Record-high 2025 goods trade deficit despite tariffs

Porter’s Diamond tells us that competitive advantage is built through factor conditions, supporting industries, sophisticated demand, and domestic rivalry — not through price signals at the border. Hughes’s history of marketcraft tells us that effective industrial policy requires expertise, patience, and a theory of the market being built — not emergency declarations and fifty policy reversals. And basic macroeconomic accounting tells us that you cannot fix a structural trade deficit without addressing its fiscal twin.

None of these things are politically convenient. All of them are structurally necessary.

9. Forward Variables — What to Watch

Variable

Timeline

Thesis Implication

Phase 2 semiconductor tariff scope (Commerce Dept. report)

April 14, 2026

Narrow = manageable. Broad = significant new disruption to AI infrastructure cost base

Congressional tariff authority legislation

Mid-2027 (projected)

Could provide legal durability absent from IEEPA; may codify narrower but more enforceable powers

Apple India production share

18–20% today → 40%+ by 2027

Confirms supply chain arbitrage thesis; no evidence of U.S. assembly trend emerging

China 28nm self-sufficiency

2027 (projected)

If achieved, reduces tariff leverage on mature-node products significantly

U.S. fiscal deficit trajectory

Ongoing

If uncorrected, trade deficit will reassert regardless of border measures

TSMC Arizona production ramp

2025–2028

Real leading-edge capacity; CHIPS Act-driven not tariff-driven; critical test of patient marketcraft thesis

WATCH POINT

The Phase 2 semiconductor tariff scope was being determined in real time as of April 14, 2026. This is the live variable in any current analysis — monitor Commerce Department reporting before citing forward-looking figures.

10. Closing: Structure Before Sophistication

You cannot bolt sophisticated trade leverage onto a domestic economy that hasn’t done the foundational work.

Liberation Day was sophistication — aggressive, theatrical, globally disruptive — deployed in the total absence of structure. The garden had been neglected for forty years. Porter tells you what a healthy garden requires. Hughes tells you it needs a skilled gardener, not a wrecking crew. And basic fiscal responsibility tells you that you cannot garden your way to prosperity while mortgaging the land underneath.

The deeper failure is not that the administration reached for tariffs. It is that it declared a structural problem to be an emergency, applied an emergency instrument to it, and called the result a liberation. A year later, the trade deficit is larger. Manufacturing employment is smaller. The legal architecture has collapsed. And the trading partners the policy was designed to discipline are quietly deepening their ties with each other, reducing their dependence on American stability they can no longer count on.

Smashing the thermometer does not cure the fever. It just means you stop knowing how sick you are.

Structure Before Sophistication • Every Time.

Analytical Brief • April 2026 • For discussion purposes. Not for redistribution without attribution.

Appendix: Sourcing Assessment

Claim

Primary Source

Confidence

Verification Priority

89,000 manufacturing jobs lost

BLS via Rethink Trade

Medium

HIGH — verify direct from BLS

$1,000–1,300 household tariff cost

Tax Foundation

High

Low — Tax Foundation is primary

TSMC 58% Q1 2026 profit jump

AP News (confirmed)

High

Low — AP is primary wire

Learning Resources v. Trump, 6–3 ruling

White & Case / CFR analysis

High

MEDIUM — verify from court record

50+ policy changes in 12 months

Tax Foundation / CFR

Medium

Medium

Apple India share 18–20%

Wedbush / UBS analyst estimates

Medium

Medium — analyst estimates, not Apple-disclosed

Huawei Ascend 950PR specs

Tech Insider citing industry reports

Medium

HIGH if using in formal context

China $47B chip fund

Tech Insider

Low–Medium

HIGH — needs primary source verification

Fiscal-trade deficit twin relationship

Macroeconomic literature (IMF/BEA)

High (directional)

HIGH — specific data figures needed

A position paper on systems and process implementation | 2026

The promise is always compelling: implement now, transform your operations, outpace your competitors. But the data tells a different story. Between 55% and 75% of ERP implementations fail to meet their objectives. Digital transformation initiatives fail at a rate of 70%. The systems sold as competitive advantages have a habit of becoming the next generation of legacy problems — often before the ROI is ever realized. This paper examines why that happens, what the data reveals, and what a more disciplined approach looks like.

55–75%

ERP projects that fail to meet objectives (Gartner / Deloitte)

70%

Digital transformations that fail to achieve goals

189%

Average cost overrun across ERP implementations

88%

Business transformations failing to achieve original ambitions (Bain, 2024)

I. The Pitch and the Reality

ERP and CRM vendors have found religion — and they’re preaching it hard. Implement AI now. Automate your workflows. Get granular on your data. Fall behind and your competitors will eat you alive.

It’s compelling. It’s urgent. And for a business in growth mode, it feels almost irresponsible not to act.

Here’s what they don’t tell you.

These systems don’t drop into a scaling business cleanly. They land with weight. Implementations run long. Processes that were working get broken in the migration. Your best people spend months fighting the software instead of serving customers. And the vendor? They’ve already cashed the contract.

Margins compress. Top line wobbles. Team morale takes a hit. That’s not transformation. That’s self-inflicted disruption.

The vendors aren’t lying — but they’re not telling you the whole truth either. Their incentive is adoption. Yours is survival and growth. Those are not the same thing.

“If a vendor is selling you urgency, ask yourself one question: who benefits most if you sign this month? The answer usually isn’t you.”

II. What the Data Actually Says

The implementation failure rate is not a fringe concern or an edge case. It is the documented norm. Research from Gartner, Deloitte, McKinsey, and Bain consistently arrives at the same sobering conclusions:

55–75% of ERP implementations fail to meet their stated objectives, according to Gartner and Deloitte. In discrete manufacturing environments, that figure climbs to 73%.

70% of digital transformation initiatives fail to achieve their goals — and Bain’s 2024 analysis found that 88% of business transformations fail to achieve their original ambitions altogether.

189% average cost overrun is the norm across ERP projects broadly. In manufacturing, overruns reach 215%.

50% of ERP implementations fail on the first attempt. Most ERP projects exceed initial budgets by three to four times.

51% of businesses experience significant operational disruptions when going live with a new system.

45% of companies cannot determine the ROI of their ERP systems post-implementation at all.

Only 23% of ERP implementations are considered fully successful by the organizations that commissioned them.

The cost of these failures is not abstract. At the organizational level, failed implementations don’t just delay returns — they actively damage operations. Birmingham City Council’s Oracle Cloud ERP implementation, begun in 2022, saw costs nearly triple from £38M to £114M by 2025. Finance operations regressed to manual bookkeeping. The council declared effective bankruptcy, naming the ERP failure as a critical driver. The US Department of Defense allocated $10.9 billion across 24 IT modernization programs between FY2023–2025. More than half encountered significant failures — cost overruns, scope expansion, and missed performance targets.

These are not failures caused by bad technology. They are failures caused by implementation before the organizational foundation was ready to receive it.

III. The Legacy Trap: When New Systems Become Tomorrow’s Problem

There is a particular irony embedded in the systems implementation cycle that rarely gets named directly: the system purchased to replace a legacy problem frequently becomes the next legacy problem — often before it has delivered any of the value it was sold on.

The dynamics are predictable. An implementation runs 30% over its original timeline (an industry norm). During that extended period, AI and agentic capabilities continue to evolve at pace. By the time the system is live, the technological landscape it was designed for has shifted. The customizations layered in during implementation to fit existing processes create integration debt that compounds over time. What was cloud-native on signing day is architecturally constrained within three to five years.

The Maintenance Trap

Legacy systems consume the majority of most IT budgets — not in building new capability, but in sustaining systems that are already underperforming. Almost two-thirds of businesses spend more than $2 million annually simply maintaining outdated systems. Sixty percent of IT leaders report that their legacy systems are experiencing moderate to severe negative impact from technical debt. The cruel arithmetic: the longer you defer modernization, the more expensive the maintenance becomes, which in turn reduces the available capital for actual modernization. The system becomes self-reinforcing.

The Obsolescence Timeline

The average enterprise ERP implementation takes between 12 and 18 months for large organizations — and that assumes a project that runs on schedule, which the majority do not. In a technology environment where AI capabilities are doubling in functional terms on an 18-to-24-month cycle, a system purchased in Q1 and live in Q4 of the following year has already been lapped in terms of what the market can offer. The ROI timeline then extends further — 28% of companies achieve positive ROI within the first year, 58% within two years, 15% require more than three years. The math compounds against the buyer.

“The system sold as your competitive advantage today has a reliable habit of becoming your next legacy burden before the investment is recovered.”

IV. Why Implementations Actually Fail

The failure data is well established. The causes are less frequently examined with honesty. They fall into four consistent categories.

1. Foundation Before Complexity

The most common thread across failed implementations is not technical — it is organizational. Systems are deployed into businesses that have not yet clarified their own processes. When core workflows are ambiguous, inconsistent, or undocumented, no software platform can resolve that ambiguity. It will simply automate the confusion at greater cost and higher velocity. Seventy-seven percent of companies report difficulties reengineering business processes to fit ERP requirements — a signal that the processes were not ready before the system arrived.

There is a sequencing principle that the vendor community has no commercial incentive to enforce: structure before sophistication. The foundation must be solid before complexity is layered in. A business still pouring the concrete of its operational processes is not ready for an enterprise system — regardless of how compelling the demo is.

2. Change Management Is Not a Project Phase

McKinsey research consistently identifies culture, not technology, as the biggest obstacle to digital transformation. Organizations that invest in cultural change see 5.3 times higher success rates than those focused exclusively on technology. Yet 75% of telco executives surveyed in 2026 cited weak change management as the primary barrier to scaling AI impact — a finding that mirrors the ERP world directly.

Change management is not a line item in a project plan. It is the condition under which the project either takes root or fails. Insufficient user training was the primary cause of Target Canada’s catastrophic ERP collapse in 2013, which contributed to the company’s full withdrawal from the Canadian market. The system functioned. The organization could not absorb it.

3. Vendor Incentive Misalignment

Vendor revenue is driven by contract execution. The salesperson’s commission is earned at signing, not at value realization. The implementation partner’s revenue scales with project complexity and duration. These incentive structures create systematic pressure toward over-scoping, over-customizing, and under-preparing — all of which extend timelines, increase costs, and delay ROI.

Thirty-six percent of failed ERP implementations cite the implementation consultant or integrator as a contributing factor. This is not simply incompetence. It is the predictable output of a commercial model where the vendor’s success metric and the client’s success metric diverge from the moment the contract is signed.

4. The AI Hyperbole Cycle

The current AI-enabled ERP pitch carries an additional dimension of risk. Vendors are selling AI-native transformation at a moment when AI capabilities themselves are in rapid flux. The SaaS pricing model that underlies most enterprise software is structurally weakening — seat-based revenue is eroding as AI reduces headcount requirements, and variable compute costs are being absorbed into flat subscription models in ways that compress vendor margins and create pricing instability.

Salesforce’s EV/EBITDA multiple has compressed nearly 28% since late 2025. ServiceNow’s has contracted more than 30% from its Q3 2025 level. These are not distressed companies — earnings are growing. But investors are repricing the durability of a model built around seats in a world actively reducing them. An enterprise locking into a multi-year ERP commitment priced on a model that the market is already abandoning is not buying transformation. It is buying into the last cycle.

V. A More Disciplined Approach

None of this argues against technology investment. The argument is for sequencing — and for clarity about what readiness actually means before capital is committed.

Before the System: Foundation Work

Document and standardize core workflows before any vendor conversation begins. If processes cannot be described clearly without the system, the system will not clarify them.

Identify and resolve key-man dependencies in operational knowledge. Systems cannot preserve institutional knowledge that has never been captured.

Conduct an honest readiness assessment covering data quality, process maturity, change capacity, and IT infrastructure. This is not a vendor’s job — it is yours.

Define measurable ROI criteria in advance, with specific timelines. Eighty-three percent of companies that performed an ROI analysis before implementation met their ROI expectations. The discipline of pre-commitment is the discipline of accountability.

During the Implementation: Governance Over Velocity

Protect your operational team from full absorption into the implementation project. The business must continue to function and serve customers during the transition.

Resist customization pressure from both vendors and internal stakeholders. Every customization extends the implementation timeline, increases cost, and creates technical debt that compounds over the life of the system.

Establish genuine change management leadership at the executive level, not delegated to the project team. Visible, sustained leadership from the top is the single most consistent differentiator between implementations that succeed and those that fail.

After Go-Live: Value Realization as a Discipline

Measure adoption rates, not just uptime. System health metrics mask adoption failures. A system that is live but unused delivers no ROI.

Plan for the ongoing investment in training, optimization, and integration. Implementation is not the end of the investment — it is the beginning.

Build for modularity from the start to avoid the legacy trap. Systems architected to be extended are systems that remain relevant. Systems architected for today’s requirements are systems that begin aging on go-live day.

VI. The Agentic Future: Why the Timing Matters More Than Ever

The current vendor pitch for AI-enabled ERP systems sits at an inflection point in enterprise software. The seat-based, per-user licensing model that has governed SaaS economics for two decades is structurally eroding. The shift is not theoretical — it is visible in public market valuations and audible in investor commentary.

What is emerging in its place is a fundamentally different relationship between software and business: agentic systems that work with an organization’s existing data, understand its specific operating goals, and actively shape tools around the business rather than forcing the business to conform to the tools.

Where traditional software asks you to adapt your workflows to fit the platform, an agentic system takes a goal and builds toward it — generating strategic plans, surfacing insights from proprietary data, configuring a CRM or ERP that reflects how the organization actually operates rather than how the vendor imagines it does.

This shift has direct implications for the timing of major system commitments. An enterprise signing a five-year ERP contract on a seat-based pricing model in 2026 is locking into the economics of the last cycle at the precise moment the next cycle is becoming legible. The right response is not to delay indefinitely — it is to invest in foundation work that will make the organization capable of absorbing genuinely transformative tools when the technology and the pricing models have matured.

“There’s a right time to layer in complexity — and it’s after your foundation is solid, not while you’re still pouring the concrete.”

Conclusion

The data is unambiguous. Most system implementations fail to deliver their promised ROI. The systems purchased to transform operations have a reliable habit of becoming the next generation of legacy constraints. The vendors selling urgency have structural incentives that diverge from the buyer’s interests at the moment of signing.

This is not an argument against technology investment. It is an argument for discipline — for the sequencing that puts organizational readiness ahead of system complexity, for the honest assessment of what a business is actually prepared to absorb, and for the commercial skepticism that should attend any purchase being sold with urgency.

The businesses that will extract genuine value from the next generation of AI-native systems are not the ones that move first. They are the ones that move ready.

Structure before sophistication. Every time.

Put it another way. For a household earning the median income of $74,580, these six sectors are consuming somewhere between 30 and 46 cents of every dollar earned. Not in taxes. Not in waste or inefficiency in some abstract government program. In the gap between what a functioning competitive market would charge for healthcare, housing, food, energy, defence, and banking — and what Americans actually pay.

I’ve spent time thinking about how economies concentrate power. The pattern I keep returning to is simpler than the academic language usually applied to it: when you control something people cannot live without, and when you have the political resources to prevent competition, you don’t need to innovate. You just extract. What follows is an attempt to map that architecture sector by sector, put real numbers to it, and think about what it means for the kind of society we are building — or failing to build.

Healthcare: Paying More to Get Less

The American healthcare system has now crossed $5.3 trillion in annual spending, representing 18% of GDP according to the latest figures from the Centers for Medicare & Medicaid Services. To understand what that number actually means, you need a comparison. Germany spends roughly 11.7% of its GDP on healthcare. Denmark spends around 10.5%. Both countries cover their entire populations. Both countries consistently outperform the United States on outcome measures. The gap between what Americans spend and what comparable nations spend represents approximately $1.9 trillion annually — not in better care, but in the structural cost of an extractive system.

The consolidation that drives this extraction is remarkable in its scale. Four health insurers now control 45% of the national market. UnitedHealth Group alone generates $324 billion in revenue — larger than the entire GDP of Portugal. Three pharmacy benefit managers control 80% of all prescription processing, extracting between $150 and $250 billion annually through opaque rebate systems that were designed in theory to reduce drug costs and in practice do the opposite. Hospital consolidation has progressed to the point where 90% of metropolitan areas are now classified as “highly consolidated,” with prices running 12 to 40% higher than in regions where competition still exists.

530,000 American families file for bankruptcy due to medical costs every single year. That figure represents 66% of all personal bankruptcies in the country.

The human measure of this system is sobering. American life expectancy sits at 77.5 years, well behind Japan and Switzerland at over 83 years. Maternal mortality is 23.8 deaths per 100,000 live births — compared to between 3 and 7 in peer nations. And 530,000 American families file for bankruptcy due to medical costs every single year. That figure represents 66% of all personal bankruptcies in the country. No comparable nation produces this outcome. Medical bankruptcy is, in every meaningful sense, an American invention.

Housing: When Shelter Becomes a Financial Instrument

Housing was once the most reliable mechanism by which ordinary families built wealth across generations. You bought a home, you paid it down, and you passed something to your children. That social contract has been systematically dismantled over the past two decades, and the dismantling was not accidental.

The total value of U.S. housing now sits at approximately $47 trillion. The top 20 institutional investors own over $60 billion in single-family homes, with companies like Invitation Homes and American Homes 4 Rent leading a consolidation that in some Sun Belt markets accounts for 20 to 30% of all purchases. Apartment REITs control 4.5 million rental units — 10% of the entire rental stock. Private equity holds over $300 billion in residential real estate.

The consequences track precisely what economic theory would predict. Rent premiums in private equity-owned properties run 8 to 12% above market rates. The median home price has moved from 3.6 times median household income in 2000 to 5.8 times in 2024. The homeownership rate has declined from 69% in 2004 to 66% despite population growth. The first-time homebuyer age has risen from 31 to 36. And perhaps most corrosive to long-term social stability: intergenerational wealth transfer through homeownership has been blocked for approximately 45% of millennials. We are creating a generation that cannot build equity the way their parents did, and the financial architecture doing this to them is entirely deliberate.

The homelessness number tells the story most starkly. Since 2015, the unhoused population has grown by 48%, to approximately 650,000 people. This is not a housing shortage in the traditional sense — it is the predictable outcome of systematically converting shelter into a yield-bearing financial asset.

Defense: The Permanent War Economy

In 1993, the Deputy Secretary of Defense hosted what has since come to be called the “Last Supper” — a dinner at the Pentagon where the major defense contractors were told, bluntly, that the Cold War was over, budgets were going to fall, and the industry needed to consolidate. What followed was one of the most rapid and consequential concentrations of industrial power in American history. Within a decade, 51 major aerospace and defense firms became five. Those five — Lockheed Martin, Raytheon, Boeing, Northrop Grumman, and General Dynamics — remain the same five today, as confirmed by the Department of Defense’s own 2022 report on competition in the defense industrial base.

The official Department of Defense budget sits at $886 billion. True national security spending, when you include Veterans Affairs, nuclear programs, the intelligence community, and the Department of Homeland Security, exceeds $1.5 trillion — representing 50 to 55% of all discretionary federal spending. The top five contractors receive over $200 billion in annual contracts. Lockheed Martin alone secured $75 billion in federal contracts in 2023. Forty percent of Pentagon spending goes to sole-source, non-competitive contracts.

The F-35 program was originally projected to cost $200 billion. Its lifetime cost has now ballooned to $1.7 trillion — a 400% overrun that took decades to accumulate and will take decades more to pay for.

The F-35 program was originally projected to cost $200 billion. Its lifetime cost has now ballooned to $1.7 trillion — a 400% overrun that took decades to accumulate and will take decades more to pay for. The Littoral Combat Ship program consumed $37 billion before being retired early due to fundamental design failures. Ford-class aircraft carriers come in at $13.3 billion each, representing a 100% budget overrun. These are not anomalies. They are the predictable output of a procurement system that has eliminated competition and institutionalized the revolving door between Pentagon leadership and contractor boardrooms.

Meanwhile, the country’s infrastructure receives a D+ grade from the American Society of Civil Engineers. Post-9/11 wars have consumed approximately $8 trillion with outcomes that are, to put it diplomatically, difficult to defend. There is a direct line between $886 billion in annual defense spending and the crumbling bridges, underfunded schools, and communities that feel abandoned that generate the political anger we keep describing as a mystery.

Energy: Locking In Yesterday’s Technology

The energy sector generates approximately $1.4 trillion annually in the United States. Renewable technology has made solar and wind cheaper than fossil fuel generation in most markets. And yet U.S. electricity prices have increased by 25% from 2010 to 2024. The gap between what the technology now makes possible and what Americans actually pay is not a natural market outcome. It is the price of captured regulation and stranded asset protection.

The top five oil companies control 42% of U.S. production. The 150 shale companies that existed in 2015 have consolidated into approximately 25 major players. Utilities, which were once natural monopolies regulated in the public interest, have consolidated from roughly 3,000 systems in the 1990s to approximately 200 today, maintaining guaranteed profit margins while underinvesting in infrastructure. OPEC+ controls approximately 50% of global production, creating a cartel architecture that interacts with domestic consolidation to sustain prices above what genuine competition would produce.

The most staggering number in this sector is not on anyone’s balance sheet. NOAA estimates that climate change economic costs now impose $150 billion annually on the American economy. Harvard research puts the health costs of fossil fuel pollution at $820 billion annually. These costs do not appear as liabilities for the companies generating them. They appear as hospital bills, disaster relief, and diminished property values for everyone else. Energy costs represent 8 to 12% of income for low-income households versus 3% for wealthy households. The extraction is real, even when it travels through the atmosphere to reach its victims.

Food and Agriculture: The Consolidated Food Chain

Four companies control 60% of the global seed market. Four companies process 85% of cattle, 70% of pork, and 54% of poultry. The top four grocers control 40% of retail food sales, with the proposed Kroger-Albertsons merger that was recently litigated potentially pushing that to 70% in many regions. From seed to shelf, the American food system has been consolidated at every layer.

The farmer caught inside this system is worth examining closely. In 1950, farmers received approximately 40 cents of every food dollar. Today that figure is 14 to 15 cents. Chicken farmers carry an average debt of $1.2 million per operation while being locked into processor-controlled contracts that dictate every input and absorb every margin. Food prices rose 25% from 2020 to 2024. During the same period, farmer income fell by 16%. That arithmetic tells you exactly where the value went.

10.5% of American households — 13.8 million families — experience food insecurity in a country that produces more food per capita than virtually any nation on earth.

10.5% of American households — 13.8 million families — experience food insecurity in a country that produces more food per capita than virtually any nation on earth. Diet-related disease costs reach $1.1 trillion annually. The food system extracts an estimated $440 to $695 billion per year through consolidated pricing, artificial scarcity, and externalized health costs from ultra-processed foods. When you add the $1.1 trillion in diet-related disease costs, the true toll of food system consolidation dwarfs the revenue of the largest companies benefiting from it.

Finance: The Infrastructure of Extraction

The financial sector generates $5.1 trillion in annual revenue — 8.3% of U.S. GDP, up from 4.9% in 1980. It has roughly doubled its share of the economy. The services it provides are roughly the same. The four largest banks control 45% of all banking assets. JPMorgan Chase alone holds $3.9 trillion. The number of banks in the country has fallen from over 10,000 in 1990 to approximately 4,000 today. BlackRock controls $10 trillion in assets under management. The top three asset managers — BlackRock, Vanguard, and State Street — hold $22 trillion and collectively own 25% or more of the voting shares in most S&P 500 companies. Visa and Mastercard control approximately 90% of credit card transactions.

The extraction mechanisms are diverse and often invisible. Credit card interchange fees of 2 to 3% on every transaction generate between $90 and $120 billion annually — a tax on commerce that most consumers never think about and cannot avoid. Overdraft fees disproportionately harvested from low-income households generate $40 to $50 billion. Wealth management fees for products that routinely underperform index funds extract $150 to $200 billion. The ‘2 and 20’ fee structure of private equity draws out $100 to $150 billion. The implicit too-big-to-fail subsidy, which allows the largest institutions to borrow at below-market rates because everyone understands the government will cover their losses, adds another $80 to $100 billion in hidden cost.

The 2008 financial crisis destroyed $22 trillion in household wealth. It required $700 billion in direct bailouts and $16 trillion in Federal Reserve emergency lending. No major executives were prosecuted. The major banks are larger today than they were in 2008. The sector now accounts for 25 to 30% of all corporate profits compared to 15% in 1980. The total annual extraction from households reaches between $3,600 and $5,000 per family, with no corresponding increase in value delivered.

The Aggregate Burden

When you add these sectors together, the annual extraction from the American economy runs between $2.96 and $4.5 trillion. That represents 13 to 20% of a $27.7 trillion GDP. It exceeds total federal tax revenue of $4.4 trillion. For 131 million U.S. households, it translates to between $22,600 and $34,400 per year — 30 to 46% of median household income.

For a working-class family earning $60,000, the math is particularly stark. Housing costs running 35% above competitive market rates add $6,000 annually. Excess healthcare costs on a family plan add $8,500. Monopoly pricing on food adds $1,800. Utility and fossil fuel premiums add $1,200. Banking fees and credit card interest add $2,400. The opportunity cost of defense spending manifest in underfunded infrastructure and services adds roughly $4,000. That’s approximately $24,000 on a $60,000 income — 40% of everything they earn, captured before they get to make a single discretionary choice about their lives.

These sectors do not extract independently. They compound. The family that is rent-burdened has less cushion for medical costs. The family carrying medical debt is more dependent on predatory credit. The farmer locked into a processor contract cannot afford insurance. The compounding is not incidental. It is the architecture.

The Political Economy of Extraction

This extraction does not sustain itself through market forces alone. It requires political infrastructure. These six sectors collectively spend $2.1 billion annually on lobbying, generating an estimated return of $760 for every dollar spent through favorable policy outcomes. Campaign contributions from these sectors totaled $4.2 billion over the 2020-2024 election cycles.

When BlackRock, Vanguard, and State Street hold controlling stakes in most major corporations, when defense contractors can offer six-figure post-government employment to Pentagon officials, when health insurers spend nearly $750 million annually on lobbying, the concept of democratic representation does not disappear — it is captured and redirected. Regulatory agencies designed to serve the public become, over time, advocates for the industries they nominally oversee. This is not conspiracy. It is the predictable outcome of concentrated money intersecting with political systems that require money to function.

The consequences for democratic stability are not abstract. Economic insecurity creates the conditions for political extremism. When people cannot meet basic needs while working multiple jobs, when their children face objectively worse economic prospects than they did, when medical bankruptcy is a statistical possibility for ordinary middle-class families, democratic institutions lose their legitimacy in the eyes of the people they claim to serve. The political anger we keep treating as a cultural phenomenon is, at its root, a material one.

The American Exception

It is worth being clear that this is not an inevitable feature of advanced capitalism. Germany spends 11.7% of GDP on healthcare and covers everyone. Its financial sector represents 4.1% of GDP versus 8.3% in the United States. Denmark’s renters pay more than 30% of income on housing at a rate of 25%, compared to 37% in America. The United States extracts over $3 trillion more annually than comparable economies, primarily through healthcare and finance monopolization.

These are not cultural differences or the natural result of a larger economy. They are specific policy choices — choices about antitrust enforcement, regulatory capture, public ownership of natural monopolies, and the appropriate role of markets in sectors essential to human survival. Other wealthy democracies made different choices and produced different outcomes. The American version is not the only version available.

Toward Something Better

I want to be honest about the scale of what would need to change. Incremental reform within the current framework — a tweak here, a new rule there — is insufficient to address extraction operating at this magnitude. What the numbers point toward is structural: aggressive antitrust enforcement that reverses decades of permissive consolidation; public ownership or strict utility regulation of genuine natural monopolies; universal basic services that remove healthcare, housing, and education from purely extractive market logic; and progressive taxation capable of recapturing a portion of the wealth this architecture generates.

This is not an argument against markets or private enterprise. Markets, when they actually function competitively, produce genuine innovation and real efficiency. The argument is simpler: markets do not function well in sectors where demand is inelastic, barriers to entry are prohibitive, and regulatory capture is advanced. In those sectors, the absence of genuine competition produces not efficiency but extraction. The question is whether we want to govern for extraction or for people.

The data presented here suggests that question cannot be deferred indefinitely. At 30 to 46% of median household income, the extractive burden is not a background condition people are learning to live with. It is actively degrading the quality of life, the economic mobility, and the democratic participation of the majority of Americans. The architecture of extraction is not stable. It generates the very conditions — insecurity, resentment, polarization, institutional distrust — that threaten to destabilize the society it is feeding on.

That is worth understanding clearly. Not as a prediction of collapse, but as a reason why the choice between economic democracy and continued drift toward plutocracy is not an ideological preference. It is a practical question about what kind of society is sustainable.

References

The following sources support the data points cited in this article:

Healthcare

[1] Centers for Medicare & Medicaid Services. (2025). National Health Expenditure Fact Sheet, 2024. CMS.gov. https://www.cms.gov/data-research/statistics-trends-and-reports/national-health-expenditure-data/nhe-fact-sheet

[2] Hartman, M. et al. (2025). National Health Care Spending Increased 7.2 Percent In 2024 As Utilization Remained Elevated. Health Affairs. https://www.healthaffairs.org/doi/10.1377/hlthaff.2025.01683

[3] Peterson-KFF Health System Tracker. (2025). How has U.S. spending on healthcare changed over time? https://www.healthsystemtracker.org/chart-collection/u-s-spending-healthcare-changed-time/

[4] Commonwealth Fund. (2023). Mirror, Mirror 2023: A Portrait of the Failing U.S. Health System. https://www.commonwealthfund.org/publications/fund-reports/2023/jan/mirror-mirror-2023

[5] Himmelstein, D.U. et al. (2019). Medical Bankruptcy: Still Common Despite the Affordable Care Act. American Journal of Public Health. https://ajph.aphapublications.org/doi/10.2105/AJPH.2018.304901

[6] OECD. (2024). Health at a Glance 2024: OECD Indicators. OECD Publishing, Paris. https://doi.org/10.1787/ae3016b9-en

Housing

[7] National Association of Realtors. (2024). 2024 Profile of Home Buyers and Sellers. NAR Research Group.

[8] Harvard Joint Center for Housing Studies. (2024). The State of the Nation’s Housing 2024. https://www.jchs.harvard.edu/state-nations-housing-2024

[9] U.S. Department of Housing and Urban Development. (2024). Annual Homeless Assessment Report (AHAR) to Congress. https://www.huduser.gov/portal/datasets/ahar.html

[10] Hsu, T. (2024). How Private Equity Became a Landlord. The New York Times. https://www.nytimes.com/2024/02/21/business/private-equity-housing.html

Defense

[11] U.S. Department of Defense. (2022). State of Competition within the Defense Industrial Base. Office of the Under Secretary of Defense for Acquisition and Sustainment. https://media.defense.gov/2022/feb/15/2002939087/-1/-1/1/state-of-competition-within-the-defense-industrial-base.pdf

[12] Wikipedia / WBUR On Point. (2023). The Last Supper: How a 1993 Pentagon Dinner Reshaped the Defense Industry. https://www.wbur.org/onpoint/2023/03/01/the-last-supper-how-a-1993-pentagon-dinner-reshaped-the-defense-industry

[13] Crawford, N.C. (2023). The United States Budgetary Costs and Human Costs of 20 Years of War. Brown University Costs of War Project. https://watson.brown.edu/costsofwar

[14] Government Accountability Office. (2024). F-35 Joint Strike Fighter: Continued Management Attention Needed. GAO-24-106303. https://www.gao.gov/products/gao-24-106303

[15] Hartung, W. (2022). Profits of War: Corporate Beneficiaries of the Post-9/11 Pentagon Spending Surge. Costs of War Project, Brown University.

Energy

[16] NOAA National Centers for Environmental Information. (2024). Billion-Dollar Weather and Climate Disasters. https://www.ncei.noaa.gov/access/billions/

[17] Krefis, A.C. et al. / Harvard T.H. Chan School of Public Health. (2021). Fossil Fuel Air Pollution Responsible for 1 in 5 Deaths Worldwide. https://www.hsph.harvard.edu/news/press-releases/fossil-fuel-air-pollution-responsible-for-1-in-5-deaths-worldwide/

[18] U.S. Energy Information Administration. (2024). Electric Power Monthly. EIA.gov. https://www.eia.gov/electricity/monthly/

[19] International Renewable Energy Agency. (2024). Renewable Power Generation Costs in 2023. IRENA. https://www.irena.org/Publications/2024/Sep/Renewable-Power-Generation-Costs-in-2023

Food and Agriculture

[20] USDA Economic Research Service. (2024). Farm Income and Wealth Statistics. https://www.ers.usda.gov/topics/farm-economy/farm-income-and-wealth-statistics/

[21] USDA Economic Research Service. (2024). Food Dollar Series. https://www.ers.usda.gov/data-products/food-dollar-series/

[22] Coleman-Jensen, A. et al. (2024). Household Food Security in the United States in 2023. Economic Research Report No. 337. USDA ERS.

[23] Mendelson, J. & Simon, M. (2023). The True Cost of Food: Accounting for the Hidden Costs of Our Food Choices. Food Policy Center.

[24] Leonard, C. (2014). The Meat Racket: The Secret Takeover of America’s Food Business. Simon & Schuster. [Updated data from USDA Grain Inspection, Packers and Stockyards Administration annual reports]

Finance

[25] Board of Governors of the Federal Reserve System. (2024). Large Commercial Banks. Federal Reserve Statistical Release H.8. https://www.federalreserve.gov/releases/h8/

[26] BlackRock, Inc. (2024). Annual Report 2024. https://ir.blackrock.com/

[27] Nilson Report. (2024). Card Industry Statistics. Payment Card Industry.

[28] Stiglitz, J.E. (2010). Freefall: America, Free Markets, and the Sinking of the World Economy. W.W. Norton & Company. [2008 crisis costs updated per Federal Reserve audit data]

[29] Epstein, G.A. (2005). Financialization and the World Economy. Edward Elgar Publishing. [GDP share data updated from Bureau of Economic Analysis]

Political Economy and Lobbying

[30] OpenSecrets. (2024). Lobbying Database — Industry Totals. https://www.opensecrets.org/federal-lobbying/industries

[31] Gilens, M. & Page, B.I. (2014). Testing Theories of American Politics: Elites, Interest Groups, and Average Citizens. Perspectives on Politics, 12(3), 564–581.

[32] Zingales, L. (2017). Towards a Political Theory of the Firm. Journal of Economic Perspectives, 31(3), 113–130.

Comparative International Data

[33] OECD. (2024). Health at a Glance 2024. OECD Indicators. https://doi.org/10.1787/ae3016b9-en

[34] OECD. (2024). Affordable Housing Database. https://www.oecd.org/housing/data/affordable-housing-database/

[35] World Bank. (2024). Financial Sector Development Indicators. https://data.worldbank.org/topic/financial-sector

The core of this dilemma lies in clearly distinguishing between using AI as an assistive tool and fully delegating our cognitive processes to it. For instance, calculators enhance our mathematical skills without supplanting our understanding, serving as tools that support, not replace, our mindset. In contrast, AI ‘agents’ that plan, write, and make decisions for us may substitute essential human capacities. Spell checkers and grammar tools extend our language abilities by catching errors we might overlook, but they do not remove our need to understand grammar rules or originate ideas. However, when AI generates entire documents, it moves from assisting our creative input to potentially replacing it. Similarly, navigation systems suggest optimal routes but do not account for the nuanced decision-making only a driver can make based on preferences and situational judgments. For example, as a child, I watched for the multicoloured entrance to the Lawrence subway station from the car—it was my personal landmark, not chosen by any algorithm. This surrender of thought processes risks eroding our agency and critical thinking. Our grade school math teachers often stressed that understanding the steps behind a solution is more valuable than arriving at the answer itself—a distinction that becomes vital in the age of AI.

Heidegger’s idea of the “forgetfulness of Being” captures the threat of losing touch with our essential human qualities as we fixate on technology’s outputs. Focusing only on AI’s abilities risks sidelining the deeper cognitive and ethical purposes behind our engagement with technology. Our relationship with AI raises profound questions not only about what it does for us, but about what we risk losing as human thinkers and agents.

The comparison between the Gutenberg press and generative AI further illuminates the stakes. While the printing press scaled the distribution of human-created content, it did not generate new content on its own. Human intention and authorship remained the bottleneck, while the creative input remained the backbone and intent of the writings, though the scale and distribution changed exponentially. In contrast, generative AI can produce content itself, mimicking human expression and decision-making. For example, leading AI models can generate millions of words and thousands of images daily, fundamentally altering content production. This shift from scaling distribution to scaling production poses new challenges, including the proliferation of misinformation and the blurring of lines between reality and AI-generated fantasy.

The negative impacts of the Gutenberg press offer a cautionary tale. While it revolutionized information dissemination, it also facilitated the faster spread of misinformation and propaganda. Printed texts made it easier to mass-produce false, inflammatory, or manipulative material. Similarly, in our current era, we see the proliferation of deepfakes and AI-generated content that further blur the lines between truth and fiction. As of 2026, with high finance moving investments into the entire AI value chain, the metaphorical horses have left the barn, and we face an unprecedented challenge in distinguishing reality from AI-generated fantasy.

As AI technology scales rapidly, it outpaces our ability to govern and regulate it effectively. The rush to capitalize on AI’s potential often prioritizes profit and control over ethical considerations and societal benefit. In early 2026, the Mag 7 are not off to a blazing start; however, valuations are bloated, and P/E ratios are also severely inflated (Greenberg, 2026). But this imbalance risks amplifying existing problematic incentives without fostering moral or political growth. When we expand into new technological frontiers like AI, the acceleration often surpasses the gates of good governance and regulation. Investors and founders, blinded by the promise of limitless monetary outcomes, may move away from creating real value for humanity, such as improving the quality of life, health, and alleviating poverty. Early adoption of AI has spread into manufacturing and finance. This is a noticeable shift toward more lucrative areas that cater to the power-hungry first-world nations (McKinsey & Company, 2025). Examples include significant investments in sectors like autonomous weapons systems, which enhance defence capabilities but pose ethical dilemmas, and in the control of fossil fuels, where energy companies leverage AI to maximize extraction and profits. Additionally, social media platforms have become prime targets for AI-driven strategies designed to manipulate and control public narratives, ultimately serving the interests of those in power.

Perhaps most concerningly, the routine use of AI for decision-making and information processing may lead to the atrophy of our cognitive and ethical faculties. Like any skill, these capacities diminish without regular practice. The danger lies not in AI surpassing human intelligence but in humans voluntarily relinquishing their cognitive capabilities. If we routinely use AI agents to decide what to read, how to interpret information, what to believe, and what to say, we train ourselves out of certain capacities: critical thinking, synthesis, imagination, and discernment. This point of mental atrophy cannot be emphasized enough, as our brains are forced to become conduits for information transfer rather than processors and interpreters, and to rely on our own insights.

The loss goes beyond practical skills and affects the habits that shape our character and judgment. Calligraphy, for example, was not just about making beautiful letters. It also helped people develop patience, attention to detail, and artistic expression. In the same way, if we rely only on AI for decision-making, we risk missing out on personal growth. Learning calligraphy in the past meant gaining a deep appreciation for the written word. Today, even with AI everywhere, we should still aim to build critical thinking and creativity. Apple is a leading technology company, and its founder, Steve Jobs, attended Reed College in Portland from 1972 to 1974. Although he dropped out, he continued to sit in on classes he found interesting, including a calligraphy course taught by Robert Palladino. That class taught him about serif and sans-serif fonts, letter spacing, and the art and technique of type. This experience influenced the Macintosh’s typography and design, demonstrating how studying the humanities can inspire technological innovation and a design focus (Jobs, 2005). This example highlights why it is important to keep developing our thinking skills and ethical judgment, so they do not fade as we use more automation.

In the end, it is the Techno Oligarchs whose relationship with money and the wealth illusion drives them to chase an infinite number by whatever means necessary. As John Maynard Keynes wrote, “Capitalism is the extraordinary belief that the nastiest of men, for the nastiest of motives, will somehow work for the benefit of us all” (Keynes,1931). Coupled with our own obsession with AI and the outsourcing of thinking to machines, efficiency wins—but wisdom fades, and the mind begins to automate itself out of relevance. Like most things in life, they are not always as great as we think they are, and in the end, things are never as bad, or so one would hope. It is important to be cautiously optimistic with AI in a world where hope springs eternal.

Resources:

Greenberg, K. (2026, January 26). The ‘Magnificent 7’ are off to a mixed start in 2026: Here’s what to watch as their earnings start. Investopedia. https://www.investopedia.com/the-magnificent-7-are-off-to-a-mixed-start-in-2026-here-is-what-to-watch-as-their-earnings-start-11891595.

Heidegger, M. (1927/1962). Sein und Zeit [Being and Time] (J. Macquarrie & E. Robinson, Trans.). New York: Harper & Row. (Original work published 1927).

Jobs, S. (2005, June 12). Commencement address at Stanford University. Stanford University. https://news.stanford.edu/2005/06/14/jobs-061505/

Keynes, J. M. (1931). The end of laissez‑faire. In Essays in Persuasion (pp. 272–287). Harcourt, Brace and Co.

McKinsey & Company. (March 12, 2025). The state of AI: How organizations are rewiring to capture value. McKinsey & Company (QuantumBlack Insights).