On July 1, a 927-page financial disclosure landed with the Office of Government Ethics. Buried in it: the president made at least $1.4 billion from crypto last year. More than any publicly traded American crypto company. The number is large enough to be a headline and vague enough to hide the mechanism. The mechanism is the story.

The Cleanest Transfer

He called it a scam once. In June 2021, well after his first term, he told Fox Business that Bitcoin seemed like a scam and a disaster waiting to happen, arguing he wanted the dollar to remain the world’s currency. That instinct was correct on the merits, crypto had no underlying asset, no cash flow, nothing to underwrite the price. What changed by 2024 was not the analysis. What changed was seeing the mechanics from the inside, where the person issuing the token collects regardless of what happens to the people buying it. He did not become a believer in crypto. He became a believer in being the house.

He has run that playbook before, literally. His Atlantic City casinos went through four bankruptcies between 1991 and 2004, more than any other major American business according to a Temple University study of the period. Investors and bondholders absorbed the losses each time. Trump did not. As CEO from 2001 to 2005 he drew about $3.2 million a year, and his base salary rose after the 2004 bankruptcy, not despite it. The New York Times, reviewing decades of filings, found he put up little of his own money, shifted personal debts onto the casino companies, and collected salary, bonuses, and fees while the businesses hemorrhaged value under him. The casino always kept its cut. So did he. Crypto just removed the building.

Start with the memecoin, because it is the easy case. In January 2025, days before the inauguration, a token called $TRUMP launched. It hit a market cap north of $15 billion within a day. It trades today near $1.70, down roughly 98 percent from that peak.

A memecoin has no product. No revenue. No claim on anything. Its price is sentiment, full stop. So when a fan buys $TRUMP, the dollar does not go toward a company building something. It goes to whoever is on the other side of the trade, and it generates a fee that flows back to Trump-affiliated entities every single time the token changes hands. He collects on volume. Not on the token going up. On the token moving at all.

That detail matters more than the price chart. It means his incentive was never for the coin to succeed as an investment. It was for people to keep trading it. Reuters reported in June that the Trump family realized at least $2.3 billion across four crypto projects while more than a million investors sit on roughly $2.3 billion in combined losses on the memecoin alone.

What the Coin Actually Buys

Some buyers were not investing at all. Last May, the top 220 holders of $TRUMP were invited to a black-tie dinner with the president. The twenty-five biggest got a private reception. People were paying for proximity, not upside, and at least some of them got exactly the access they paid for.

That is a defensible transaction if you know what you are buying. The problem is most buyers were not shown that framing at purchase. They bought during a euphoric launch spike believing they held a speculative asset with a chance of appreciation. What they actually held was a fee-generating instrument for someone else, dressed as an investment. The loyalty was real. The transfer was real. The understanding, for most of them, was not.

Same Pattern, More Rooms

World Liberty Financial runs the identical extraction with more rooms to get lost in. Where the memecoin takes a fan’s money directly at the point of sale, WLFI adds two layers before the public ever shows up: a foreign government paying for access, and a company borrowing real money against tokens it created for free. By the time a retail buyer enters, the insiders have already been paid twice.

Free Tokens, Foreign Cash

WLFI minted 100 billion tokens at founding. The Trump family and its affiliates received 22.5 billion of them, at zero cost, through an entity that Trump majority-owns. Locked, initially. But issued for nothing.

Four days before the second inauguration, a separate and undisclosed deal closed. An investment vehicle tied to Sheikh Tahnoon bin Zayed Al Nahyan, the UAE’s national security advisor, paid $500 million for 49 percent of the company itself, not the token. At least $187 million of that flowed directly to Trump family entities, with roughly $31 million more to entities tied to Steve Witkoff, whose son co-founded the venture. Two of the sheikh’s associates quietly joined WLF’s board. None of this was disclosed to the public at the time. The company also raised $550 million from private token sale rounds before the token ever traded openly.

So before a single American fan bought a single public token, insiders and one foreign government had already moved close to three-quarters of a billion dollars into Trump-linked pockets. The public had not yet been invited to the table.

The Public Enters, Late and High

The token went live in September 2025, peaked near $0.31 to $0.46, and within a day fell nearly half. Early presale wallets, holding tokens bought at $0.015 to $0.05, moved millions to exchanges within hours of being able to claim them. By April 2026, the price sat near $0.08, down more than 75 percent from launch.

The company’s own terms are explicit that the token carries no dividend, no equity, no share of revenue. It is a governance vote and nothing else. A retail buyer who thought they were buying a piece of a business bought a right to vote on proposals in a company that does not have to pay them anything, ever, no matter how much money the business makes.

Borrowing Against Nothing

In April 2026, WLF’s treasury deposited five billion of its own governance tokens, the same tokens created from nothing, into a lending platform called Dolomite and borrowed roughly $75 million in stablecoins. Over $40 million of it moved to Coinbase Prime, a venue typically used to convert crypto into cash. Dolomite’s co-founder is also WLF’s chief technology officer.

This is precisely the kind of transaction a vigilant, non-conflicted regulator exists to stop before it happens, not report on after. An insider pledging a token it created for free, and controls, to extract real and spendable money, while ordinary depositors on the lending platform found their funds locked up at near-full utilization.

That is a related-party transaction with a captive counterparty and no independent board to say no. In any other regulated financial institution, an examiner would have flagged the self-dealing and the concentration risk before the money moved, not after Justin Sun called it treating users as a personal ATM.

The token fell to a new low within a day. Less than a week later, WLF proposed unlocking 62 billion more tokens for founders, team, and advisors, opening exits for insiders who previously had none, while the price sat near its floor. The oversight that would normally catch this sits inside the same administration collecting the fees.

The Fee Never Loses

Trace the money and a pattern holds across both ventures, memecoin and governance token alike. The Trump family collects on fees, on token sales, on foreign equity checks, on treasury borrowing, regardless of which direction the price moves. The UAE-linked investor bought equity in the company, not exposure to the retail token’s swings. The company extracts liquidity by borrowing against tokens that cost it nothing to create. Every one of those flows is structured to be indifferent to whether the public ever gets its money back.

The only party in this story whose outcome depends on the price going up is the retail buyer. Everyone else built a position that pays regardless. That asymmetry is not a bug in either venture. It is the design, repeated twice, in two different costumes, for two different audiences, neither of whom were told which side of the trade they were standing on.

James Allen Ramsey authors J Allen Insights on the intersection of technology, geopolitics, and human development. He is a retired tax attorney and consultant whose career included GM, PwC, Shaw Pittman, EY and Bionic System Analytics.

Twelve ways to fall on purpose. Tamara Gassner demonstrates them in under two minutes on a studio floor. Every robotics lab in the world has spent millions trying to solve what she teaches on Tuesday afternoons.

The video is a tutorial. Modern dance. Floor transitions. Twelve clean, continuous ways to move your body from standing to the ground and back again without breaking rhythm or losing control. The Knee Turn. The Attitude Turn. The Front Slide. Each one a different negotiation between gravity, momentum, and a human body that knows how to lose its footing gracefully.

Nobody in the humanoid robotics industry has knocked on that door yet. They should be running.

The Data Problem Nobody Has Solved

The humanoid robot deployment numbers for mid-2026 are real and they are impressive. Figure AI is producing more than one robot per hour at its BotQ factory. Boston Dynamics has committed its entire 2026 Atlas production to Hyundai and Google DeepMind. Agility’s Digit has moved more than 100,000 totes in active warehouse operations. The sector has crossed from prototype to production.

But every one of those deployments shares the same constraint. The robots are doing narrow, repetitive tasks in structured environments. Sheet metal loading. Tote transfer. Component sorting. The physical world those robots inhabit has been engineered to minimize variability, because variability is what breaks them.

The reason is not hardware. The actuators are improving fast. The reason is data. A humanoid robot learns to move the way any system learns: by processing enormous volumes of examples. And the examples the industry has been generating are not good enough for what comes next.

The dominant approach today is teleoperation. A human operator guides the robot through a task. The system records the motion. That recording becomes training material. It works for narrow tasks in controlled settings. It does not scale. The throughput ceiling is the number of human operators you can hire and the hours they can log. You cannot teleoperate your way to a general-purpose robot.

Simulation is the other path. Build a digital environment, train the robot in physics models, transfer the learned behavior to the real world. NVIDIA’s GR00T pipeline has shown meaningful gains using synthetic motion data. But simulation breaks on contact physics. The robot that folds a shirt perfectly in a virtual environment fails on the wet towel, the sliding cup, the surface it has never encountered. The gap between simulated physics and actual physics remains a live problem.

Voice commands are the third approach, and the most limited. Converting verbal instruction into motor behavior runs into a wall that language itself built. “Pick up the cup” contains almost no information about grip angle, wrist rotation, approach vector, or force calibration. Human language was never designed to describe physical motion with the precision a robot requires. A child learns to pick up a cup through thousands of repetitions of embodied experience. You cannot compress that into a sentence.

The industry knows it has a data problem. It is looking in the wrong places.

What the Studio Already Has

The training corpus that solves humanoid movement already exists. It was filmed over decades. It lives in archives that have not been valued yet. And it is richer, more varied, and more physically instructive than anything a warehouse or factory floor has ever produced.

It is called dance.

Video is the right medium because it captures what language cannot. Wrist angle. Body weight shift. How a human hand adjusts mid-grasp when an object shifts unexpectedly. The anticipatory lean before a step. The micro-correction that prevents a stumble. All of that information is implicit in the footage. None of it survives translation into words.

The scale is unprecedented. Decades of filmed rehearsals, performances, and tutorials represent a corpus of human movement that dwarfs anything a teleoperation program could generate. And the movement it contains is not narrow and repetitive. It is the full physical envelope of what a human body can do.

Modern dance specifically solves problems that industrial footage cannot approach.

Three Techniques, Three Unsolved Problems

Gassner’s tutorial is not art criticism. It is engineering documentation that nobody in a robotics lab has read yet.

The Knee Turn and Cross Turn train rotational energy dissipation across multiple joints simultaneously. The torso rotates to absorb gravity on descent, spreading the load across the kinetic chain rather than concentrating it at a single point of impact. Current humanoid fall protocols are essentially emergency stops. The joints lock, the system detects instability, and it either catches itself rigidly or goes down hard. What this technique encodes is something entirely different: a continuous negotiation with gravity that keeps the system in control through the entire arc of descent. That is the physics a robot needs to survive a stumble on an uneven surface without destroying its actuators.

The Attitude Turn encodes anticipatory load balancing. One leg sweeps behind in a curved arc, acting as a counterweight to control the rate of descent before the descent is fully committed. The body makes a decision about where it is going before it arrives there. That predictive quality is precisely what Vision Language Action models are attempting to learn from static task footage and struggling to extract. A system trained on Attitude Turn data learns to model its own trajectory and manage it through the arc. That is proprioceptive intelligence embedded in visual data.

The Knee Slide and Front Slide are the most important techniques for the argument. They train the robot to use momentum and physics rather than muscular resistance. Current humanoid locomotion is almost entirely resistive. The system braces against gravity at every step, fighting the physics rather than working with them. A slide transition teaches something different: the body commits to momentum and manages it through yield, not force. That is a fundamentally different and more efficient control paradigm, and no warehouse deployment has ever required it.

Twelve ways to fall on purpose. Each one a lesson the industry has not yet paid tuition for.

The Archive Nobody Has Priced

Contact improvisation may be the single richest dataset in existence for humanoid training. Two bodies negotiating weight sharing, floor transitions, and unpredictable partner responses in real time. That is the closest human movement gets to operating in an unstructured environment with dynamic obstacles. No script. No fixed sequence. Pure generalization.

Release technique trains controlled collapse and recovery. The Alvin Ailey archive contains decades of it. The Paul Taylor archive. The repertory of every major contemporary dance company that ever put a camera in a rehearsal room. Filmed rehearsals are better than performances for training purposes because they show correction, repetition, and the moment a movement fails and recovers. That failure data is exactly what a robot needs.

Martial arts kata has been discussed in robotics circles. But contemporary dance is richer because it does not reduce to a fixed sequence. Improvisation forces generalization in a way that scripted movement never can.

The rights question is where the investment thesis sharpens and where the negotiation has not happened yet.

Dance companies are not sitting on assets they know are valuable. The licensing cost of a major contemporary dance archive right now is a rounding error compared to what a robotics company would pay for equivalent teleoperation hours. The conversation has not happened yet because nobody on the robotics side has walked into the right building. When they do, they will move fast. That is rational. Speed is the moat.

The problem is not that the robotics company moves first. The problem is that the dance company negotiates blind. A one-time licensing fee for an archive that will train a system worth billions in deployment contracts is not a negotiation. It is a clearance sale conducted by a seller who does not know what is on the shelf.

The deal that gets signed before the dance world understands what it is selling will be announced as something unremarkable. A content partnership. An AI training agreement. The press release will not say what was actually purchased. And once the corpus is ingested, trained into the weights, and deployed across a commercial fleet, the moment to negotiate has passed. The value has transferred. It will not transfer back.

YouTube is where this dynamic is most exposed and most invisible. Tamara Gassner posted a two-minute tutorial and it is now engineering documentation worth more than anyone in the dance world has been told. She is not alone. Thousands of dance educators, choreographers, and practitioners have built movement libraries on YouTube that represent exactly the training corpus the humanoid robotics industry needs. They posted for students, for community, for the love of the craft. None of them posted a terms-of-service agreement covering AI ingestion at commercial scale. None of them will receive a phone call before the scraper runs.

YouTube creators in particular have no institutional protection. No archive manager. No legal counsel. No collective bargaining structure. A major dance company at least has a board and an attorney. A teacher with a camera and a channel has a Terms of Service agreement she did not write and a platform that has already licensed her content for AI training purposes in ways she may not have read. The rip does not require a conversation. It has already been authorized by a checkbox she clicked when she uploaded.

What the Packing Tape Tells You

The fall survivability question is the manufacturing moat hiding in plain sight. A humanoid robot that can recover from an unplanned floor transition in a real unstructured environment, a wet kitchen floor, a child’s toy left on the stairs, a surface the simulation never modeled, is a categorically different product from one that operates only in a controlled cell.

The Beijing half-marathon was a closed course. Honor’s Lightning finished faster than any human world record holder. That is a demo. The test that matters is what happens when the surface is wrong and the robot was not expecting it.

A system trained on Tamara Gassner’s twelve transitions handles that test differently than a system trained on sheet metal loading. It has seen what controlled descent looks like from the inside. It has processed the rotational absorption, the anticipatory counterweight, the yield to momentum rather than resistance against it. It knows how to fall on purpose.

The lab that hired a choreographer instead of a tenth teleoperation operator is ahead. Nobody has announced that hire yet.

That is the tell. But the tell only matters if someone is keeping score on both sides of the ledger.

Who Gets Paid When the Robot Learns to Dance

The robotics company that cracks the movement data problem first will be worth an extraordinary amount of money. The dancers whose movement trained the system will not be in the room when that valuation is set. That is not an accusation. It is the default outcome of every technology transition that has preceded this one.

ERP systems absorbed the tacit knowledge of accountants and process managers and returned nothing to the people who built that knowledge. Offshore platforms absorbed the institutional expertise of American manufacturing workers and distributed the gains to shareholders and consumers. Shared services models extracted the procedural intelligence of entire departments and handed the savings to the enterprise. The pattern is consistent across four deskilling waves. The contributor is consumed. The builder captures. The gap between those two outcomes is treated as the natural order rather than a choice that could be made differently.

The Citizens Capital Right does not ask the robotics company to slow down. It asks the dance world to negotiate with open eyes before the deal closes rather than discover afterward what it gave away.

The Alvin Ailey archive is worth more today than it was three years ago. Not because the performances changed. Because the context around them changed. The same movement corpus that was previously valuable only as artistic heritage is now engineering documentation for a multi-billion dollar deployment market. That revaluation happened without a conversation, without notice, and without any adjustment to what the archive’s stewards understand themselves to be holding.

Open eyes means knowing that before the term sheet arrives. It means understanding that a one-time licensing fee and an ongoing stake in a system trained on your movement are categorically different instruments. It means having counsel in the room who has read the AI training literature, not just the standard content licensing precedents. It means asking what the corpus will be worth in five years inside a deployed commercial fleet, not what it is worth today as an archival curiosity.

For YouTube creators that conversation is already over. The checkbox was clicked. The terms authorized the ingestion. Tamara Gassner filmed twelve transitions for her students and the platform she trusted to distribute them has already made decisions about downstream use that she was not party to. The Citizens Capital Right framework treats that outcome as a property claim that was settled without the owner’s informed consent, not as a freely negotiated transaction. The difference matters legally and it matters morally.

The robotics company should move fast. That is not the problem. The problem is a dance world that does not yet know it is sitting on something the fastest movers in technology need. The window to negotiate from strength is open now. It will not stay open.

That is the only tell that matters.

James Allen Ramsey with J Allen Insights covers AI economics, labor displacement, and the political economy of technological change. The Citizen’s Capital Right series, Parts 1 through 3, is available in the archive for readers who want the full property-rights argument.

Gallup’s June 2026 patriotism survey, released ahead of the country’s 250th , found 70 percent of Republicans and 14 percent of Democrats describe themselves as “extremely proud” to be American. A 56 point gap, the widest Gallup has recorded since it began asking the question in 2001. Widen the measure to include “very proud” and the numbers move to 93 percent and 27 percent. Independents, for what it’s worth, sit closer to the Democratic number than the Republican one.

The response from the right has been swift and, within its own frame, coherent. Rep. Harriet Hageman took to the House floor to argue the numbers reflect a deliberate project: the left has “weaponized our history with slavery to indict our nation as irredeemable,” replacing shared pride with what she called “self-flagellation.” The implicit charge is not just that Democrats feel less patriotic. It is that a specific faction, Democratic Socialists most pointedly, have made insufficient love of country into something closer to a political program. Given a summer where Mamdani-aligned candidates were sweeping Democratic primaries in New York, the timing gave the argument a live target.

Take the argument on its own terms before answering it, because it deserves that. There is a real pattern underneath the outrage. A YouGov survey conducted the same month found that half of Republicans define patriotism as supporting one’s country unconditionally, compared to 11 percent of Democrats. Ninety six percent of Republicans report being proud to be American against 58 percent of Democrats. The gap is not a polling artifact. Something has genuinely diverged in how the two coalitions relate to the nation, and pretending otherwise does not serve anyone.

But notice what the argument assumes before it even gets to the data. It assumes that the highest form of patriotism is unconditional, that pride ought to hold steady regardless of what the government does, and that any softening of that pride in response to the government’s conduct is itself the defect to be diagnosed. That assumption is doing all the work in Hageman’s speech, and in the broader charge against the Democratic Socialist left. It is worth asking, before accepting it, whether unconditional loyalty to a nation state is actually the virtue this argument takes it to be, or whether it is something closer to a category error, mistaking the state for a claim on the self that no state, of any party, in any century, was ever owed.

That is where a much older argument becomes useful.

Two Definitions Wearing One Word

Political psychologists have a name for the split the YouGov numbers are actually measuring, and it predates this election cycle by a quarter century. Robert Schatz, Ervin Staub, and Howard Lavine, writing in 1999, distinguished between blind patriotism and constructive patriotism. Blind patriotism is rigid and uncritical: national symbols and policy alike are placed beyond question, and dissent reads as disloyalty. Constructive patriotism is attachment paired with a willingness to criticize, on the theory that loyalty to what the country claims to stand for sometimes requires opposing what the country is currently doing.

Once that distinction is on the table, the Gallup and YouGov numbers stop measuring a single thing called patriotism and start measuring two different commitments that happen to share a word. Half of Republicans defining patriotism as unconditional support is not a footnote to the pride gap. It is close to the whole explanation for it. A person who holds the blind definition will report high pride almost regardless of current events, because current events are not the input the definition asks about. A person who holds the constructive definition will report pride that moves with the government’s conduct, because conduct is precisely the input their definition is built to track. Ask two groups holding two different definitions the same question and a large gap is not evidence one side loves the country more. It is evidence they are answering different questions while using the same four words.

This matters for how the charge against the Democratic Socialist left should be read. When Hageman describes the left’s relationship to national history as “self-flagellation,” she is not simply wrong that something has changed. She is applying the blind definition and treating any departure from it as a symptom rather than an alternative. A critic operating from the constructive definition would put it differently: refusing to look away from slavery, or from a strike that killed 156 people, most of them schoolchildren, in a country where the government’s own investigators found the strike was likely American, is not a failure of love for the country. It is what love for the country’s stated ideals actually requires when the country falls short of them. The alternative, pride that holds steady no matter what the government does, is not a stronger patriotism. It is a patriotism that has stopped listening to its own object.

None of this settles which definition is correct. That is a genuine dispute, not a rhetorical trick, and a fair reading has to admit that constructive patriotism, taken far enough, can curdle into something closer to alienation than love, an objection worth taking as seriously as the one aimed at the blind version. But it does mean the debate currently being fought over a pride gap is actually a debate over which definition of loyalty deserves to be called patriotism at all. That is not a question polling can answer. It is a question with a very long history behind it, considerably older than Gallup, and considerably older than the republic being asked about.

What the Tradition Already Says About Unconditional Loyalty

The dispute over which definition of patriotism deserves the name is not new, and it does not start with Gallup. It is close to the oldest political argument in the Hebrew Bible, and it does not go the way the blind-patriotism position would predict.

In 1 Samuel 8, the elders of Israel come to the aging prophet with a request: give us a king, so that we can be like all the other nations. Samuel takes the request personally, but God tells him something sharper. “They have not rejected you, but they have rejected me, that I should not reign over them.” What follows is not a simple yes. Samuel is instructed to warn the people exactly what a king will cost them: sons taken for war and labor, daughters taken for the palace, fields and vineyards confiscated, “and you shall be his slaves.” The people hear the warning in full and ask for the king anyway. God grants it, under protest, and the warning proves accurate almost immediately. Saul and then Solomon deliver, point by point, everything the text predicted.

Read plainly, this is a story about a people trading a direct, accountable relationship for a centralized human authority they were told in advance would extract from them without limit. It is also, notably, not a story where unconditional support for the new arrangement is treated as a virtue. The text’s sympathy runs the other way. Wanting a king “like the other nations” is the thing being criticized, not the thing being praised.

The New Testament takes up the same tension from the opposite direction, not a people asking for a king, but a people already living under one, and being tested on how far their loyalty to him extends. In Mark 12, Pharisees and Herodians, an unlikely pair, ask Jesus whether it is lawful to pay the imperial tax to Caesar. It is a trap built with real teeth. Say yes, and he is siding with an occupying power against a nationalist audience that despised the tax as a mark of subjugation. Say no, and he is advocating sedition against Rome. He asks for a coin, one that would have carried Tiberius’s image and an inscription naming him son of the divine Augustus, and asks whose image is on it. Caesar’s, they answer. “Then render to Caesar the things that are Caesar’s, and to God the things that are God’s.”

The line gets quoted as a tidy division of church and state, render unto each its own sphere, but that reading undersells what the second half is actually doing. If the coin belongs to Caesar because it bears his image, the question the text leaves hanging is obvious: whose image does a human being bear? Genesis answers that one a chapter into scripture. “So God created man in his own image, in the image of God created he him.” The line is not a balanced split between two equal claims. It is a demotion. Caesar gets the coin. God gets the person entire, which means Caesar’s claim, however real, sits inside a much larger one that limits it. Loyalty to the state is bounded from the very moment it is affirmed.

Neither passage tells a reader not to pay taxes, not to acknowledge civil authority, not to live within the laws of the nation one happens to be born into. Paul is direct about that elsewhere, be subject to governing authorities, they are not instituted for nothing. What neither passage does, anywhere, is treat unconditional loyalty to that authority as the marker of faithfulness. If anything, both texts single out the demand for unconditional allegiance to an earthly power as the thing worth warning against, not the thing worth defending.

Which is worth holding next to a concrete case rather than an abstract one. In March 2026, a US strike hit the Shajareh Tayyebeh school in Minab, Iran, killing 156 people, most of them schoolgirls. A preliminary US military investigation found the strike was likely American, the result of outdated targeting data. The president, on Air Force One, blamed Iran. That response, denying a finding one’s own investigators reached, is what unconditional loyalty actually requires in practice. It is not a hypothetical failure mode. It is what the blind-patriotism position produces when the government it is loyal to does something indefensible: not silence, but denial. Caesar’s coin does not ask for that. It asks for what is owed, not for a story about what did not happen.

Where True Loyalty Actually Sit

If unconditional loyalty to a nation is the category error the tradition warns against, the natural next question is where loyalty of that kind, total and unconditional, actually belongs. The tradition has an answer, and it lays out in three tiers, each one closer to earth than the last:

- Heavenly citizenship. The only claim entitled to be unconditional. Everything else is owed something less than everything.

- Global citizenship. What loyalty looks like in practice, this side of heaven, given that the biggest threats a person faces today do not stop at a border.

- National citizenship. Real, and owed real things, taxes, obedience to just laws, but bounded and never total.

Heavenly citizenship comes first. Paul, writing to a congregation in Philippi, a Roman colony whose residents took real pride in their civic status as citizens of Rome, uses the technical term for that status and redirects it: “our citizenship is in heaven.” The author of Hebrews describes the patriarchs the same way, as people who understood themselves to be strangers and exiles on the earth, seeking a homeland they had not yet reached, “a better country, that is, a heavenly one.” Early Christians kept using this language for themselves, sojourners, resident aliens, in whatever empire or nation they happened to occupy. This is the one claim in the whole tradition that is treated as entitled to be unconditional. Not because earthly citizenship does not matter, but because it is not ultimate, and only the ultimate claim gets to ask for everything.

That top tier does the real work, and it is also, notably, not a claim that requires importing anything from outside the tradition itself. A Catholic reader does not need Kant, or systems theory, or a contested twentieth century cosmology to arrive at bounded national loyalty. Samuel and Mark and Philippians get there on their own, and arguably get there more clearly than most modern political theory manages.

Global citizenship comes second. A first tier that only tells you what loyalty is not owed to does not, by itself, tell you what a rightly ordered loyalty looks like in practice, this side of heaven, in a body that still has to live in a country, pay its taxes, and decide what it owes the people around it. That is what the second tier is for, and it starts from a simple observation: consider the actual shape of the threats a person born in 2026 faces. None of them stop at a border. A pathogen does not check a passport. An asteroid does not consult a treaty. A warming or cooling atmosphere does not distinguish Ohio’s air from Ontario’s. Carl Sagan’s reflection on the pale blue dot, a single pixel of light photographed from the edge of the solar system, makes the same point without needing a single verse of scripture: everyone who has ever lived did so on that pixel, and every rivalry, every border dispute, every claim of national superiority was fought over a fraction of a fraction of it. The scale problem is not a rhetorical device. It is a fact about where humanity actually lives, in a thin, breathable layer around a rock moving through a hostile and indifferent cosmos, with no backup and no rescue.

Contemporary existential risk scholarship, most rigorously catalogued in Toby Ord’s work at Oxford, treats coordination failure, not any single hostile actor, as the dominant multiplier across nearly every serious threat to the species: pandemic, engineered or natural, climate destabilization, nuclear escalation, unaligned artificial intelligence. Fragmented sovereignty is bad at solving problems that do not respect the lines sovereignty draws. This is not a call for world government, and it should not be read as one. It is closer to what political theorists call functionalism, authority pooled narrowly, by function, only where a problem is genuinely planetary, the model already quietly operating behind the World Health Organization and the International Atomic Energy Agency, rather than authority pooled broadly, by sovereignty, in the way a world government would require. Global citizenship is not a mystical claim about human unity. It is a practical recognition that some obligations are owed to the whole crew of the vessel, because some risks threaten the whole vessel, and no single cabin can address them alone.

National citizenship comes third, which leaves it where it belongs, last rather than first. National citizenship is real. Paul tells the Roman church to be subject to governing authorities, and means it. Taxes are owed. Laws bind. A nation is not nothing, and nothing in the first two tiers licenses treating it as such. But a nation is bounded and functional, owed the coin, not the soul, and the moment a nation’s claim on its people is asked to become unconditional, total, immune to correction by what the nation actually does, it has stepped into a place in the hierarchy that was never assigned to it. Not by the framers. By the tradition itself, considerably earlier, and on considerably higher authority.

The Gap Reconsidered

Return, one last time, to the fourteen and the seventy.

The instinct behind the Gallup numbers, and behind Hageman’s floor speech, is that the fourteen percent represents a failure, a coalition that has somehow lost or forfeited its love of country. Nothing in the argument above requires denying that something real separates the two groups. It does require asking a different question about what that something is.

A person operating from the first tier, ultimate loyalty reserved for what is actually ultimate, will not report unconditional pride in any nation, because unconditional pride in a nation is not a category their loyalty has room for. That is not indifference to country. It is the same instinct that made the elders’ demand for a king look, in the text, like a rejection rather than a request. It is the same instinct behind a coin held up and a line drawn around what it can and cannot buy. A drop in “extremely proud” is not obviously a defect once the alternative on offer is unconditional loyalty to a state, any state, and once the tradition being invoked to defend that loyalty is the same tradition that has, since Samuel, treated the demand for it as the thing to watch for.

None of this means the fourteen percent is beyond criticism, or that constructive patriotism cannot tip into something closer to contempt than love, a real risk worth naming rather than waving off. But the poll was never actually measuring love of country in some neutral, theory-free way. It was measuring agreement with one particular, contestable definition of what loyalty to a country ought to look like, a definition the tradition its loudest defenders claim to represent has been quietly arguing against since before the country existed.

James Allen Ramsey creates J Allen Insights on the intersection of technology, geopolitics, and human development. He is a retired tax attorney and consultant whose career included GM, PwC, Shaw Pittman, EY and Bionic System Analytics.

There is a thought experiment worth taking seriously.

Imagine you are an engineer assigned to the International Space Station. You wake up in your sleeping quarters, eat breakfast from the commons, run your diagnostics, and begin your shift. Nobody hands you a bill for the oxygen. Nobody charges you for the meal. Nobody invoices you for the berth you slept in. You are crew. The ship provides what crew requires. In return, you contribute what the mission requires. That exchange is the entire economy of the ISS. It works. It has worked continuously since November 2000. 1

Now scale that thought experiment to eight billion people and one planet.

That is not a fantasy. It is a design problem.

The Ship Is Already Flying

Buckminster Fuller called it Spaceship Earth in 1968. He meant it precisely. A closed system. Finite resources. No resupply from outside. A life support architecture that requires active maintenance or it fails. Passengers who did not choose to board but who are nonetheless aboard. 2

Fuller was right about the diagnosis. What he did not have was the operational technology to run the allocation system at planetary scale. We may now have it.

The Earth provides, within known parameters, everything eight billion crew members require. The calories exist. The fresh water exists. The building materials exist. The medical knowledge exists. The energy potential from sun and wind exceeds current consumption by orders of magnitude. The scarcity that produces poverty is not a natural scarcity. It is an allocation failure. The ship has enough. The distribution system does not deliver it.

This is a solvable engineering problem. It is not a solved moral problem. That distinction matters.

What the ISS Actually Proves

The International Space Station is operated by five space agencies across geopolitical lines that produce genuine conflict on the ground. Russian cosmonauts and American astronauts have maintained continuous cooperative habitation since 2000 through the Iraq War, the invasion of Ukraine, and every other rupture in the relationship between their governments. They do this because the mission requires it and because the alternative is death. 3

Nobody on the ISS uses money. There is nothing to buy. There is no market to clear. The allocation problem is solved upstream by mission logistics. What remains is a pure contribution economy. You perform your crew function or you do not. Everyone aboard knows which. The recognition is immediate, transparent, and consequential in the most direct possible way.

The station has rank. The commander commands. Mission specialists carry different levels of responsibility. But the rank differential does not express itself in compensation because compensation does not exist in that environment. It expresses itself in responsibility, in recognition, and in the deference the crew extends to earned expertise. The environmental systems technician is not ranked below the flight engineer in human worth. They are ranked within their own domain of contribution. The ship needs both. The ship knows it needs both.

This is not utopia. It is engineering.

Born Into Crew

Here is the framing that resolves a problem philosophers have argued about for centuries.

Social contract theory asks how you consent to a system you were born into. You cannot consent before birth. You cannot meaningfully opt out after. John Locke’s answer in his Second Treatise of Government (1689) was tacit consent: remaining within a territory implies acceptance of its governing framework. It is the weakest part of an otherwise powerful framework, and political philosophers from David Hume forward have never stopped arguing about it. 4

The crew framing sidesteps this entirely. You did not consent to being born on this planet any more than a crew member consents to the physics of the vessel they serve on. You are aboard. The ship has requirements. Crew status comes with both provision and contribution. That is not ideology. It is systems logic.

From that premise, certain things follow directly.

Every crew member, by virtue of being crew, has a berth. A place to sleep that is theirs. Every crew member has access to the commons. Food sufficient for health and energy. Every crew member has access to the ship’s medical systems for the duration of the voyage, which is their lifetime. Every crew member has access to the ship’s knowledge systems, education that does not end at eighteen or twenty-two but continues as long as the crew member is curious and the ship is flying. Every crew member has access to recreation, to the spaces and time that make the voyage worth making.

These are not benefits. They are crew provisions. The distinction matters. A benefit is a gift from a system that could withhold it. A crew provision is what the ship requires its crew to have in order to function. The ship provides them not out of generosity but out of operational necessity. A crew member who is hungry, sick, homeless, or uneducated is a crew member operating below capacity. The ship cannot afford that.

Rank Without Worth Hierarchy

Here is where the model does something that most social arrangements have never managed.

The janitor is not ranked below the engineer. Read that again.

They are crew members in different functional domains. Within the sanitation and environmental maintenance domain there is a rank ladder. Junior crew member, senior crew member, specialist, Captain of Sanitation. Within the engineering domain there is a parallel ladder. The ladders are lateral to each other, not stacked. The Captain of Sanitation has achieved the highest recognition within their contribution domain. That rank commands the same crew respect and the same material recognition as the Captain of Engineering.

The criteria for advancing within each domain are published, visible, and consistently applied. You know what is required. You pursue it or you do not. The system does not ask who you know or whether the person evaluating you finds you socially comfortable. It asks what you have demonstrated within your domain. An AI tracking contribution does not have a personal relationship with the candidate. It does not find the quiet crew member awkward and unconsciously discount their record. The record speaks.

Better housing comes with rank. Not with domain. The Captain of Sanitation and the Captain of Engineering live equivalently well. The junior crew member in either domain lives at the junior crew level. The material incentive is to advance within your chosen domain. Not to choose your domain based on material outcome.

This dissolves a false choice that the current system forces on nearly everyone. Right now you choose your occupation under two simultaneous pressures. What can I do well and find meaningful. What will support me and my family materially. Those two questions have different answers for most people. The gap between them is a source of chronic low-grade suffering that the culture has normalized. The crew model removes the second question from the occupational choice entirely. You choose your domain based on fit, interest, and genuine capacity. Material outcome follows rank achieved within whatever domain you choose.

The Recognition Economy

If there is no money, what motivates contribution beyond the minimum?

The same things that motivated the ISS crew member who spent six hours chasing an anomaly in the CO2 scrubber data even though their shift ended three hours ago. Mission clarity. Peer recognition. The intrinsic satisfaction of competence deployed toward something that matters.

Edward Deci and Richard Ryan spent decades studying human motivation empirically. Their self-determination theory, developed across forty years of research and summarized in their foundational 2000 paper in Psychological Bulletin, finds consistently that autonomy, competence, and relatedness drive sustained high performance. 5 Money works as a motivator only when its absence produces anxiety. Once the floor is secure, additional money adds surprisingly little motivational energy. It can actually crowd out the intrinsic drivers it was supposed to supplement.

The crew model removes the anxiety floor. The intrinsic drivers reassert themselves.

What replaces money as the social expression of achievement is recognition. Not flattery. Recognition. The distinction requires that the criteria be visible. Honor requires that the community understand what was done and why it matters. When the crew can read the contribution record, when the advancement criteria are public, when the domain expertise is legible to everyone aboard, the recognition that follows is genuine. The crew knows what the Captain of Sanitation had to demonstrate to earn that designation. The recognition reflects that knowledge.

This is not a new idea. It is what every functional craft tradition, every genuine professional community, every religious order organized around contribution rather than extraction has always understood. The market replaced it with price signals because price signals scale and craft recognition does not. AI changes that calculus. A system capable of tracking contribution at planetary scale makes the recognition economy legible in ways that were not previously possible.

Every Sector Is Critical

No compartment on a functioning spacecraft is expendable.

A pressure failure in the least glamorous section of the ISS kills the crew as certainly as one in the command module. The current global economy treats entire continental regions as compartments that can be allowed to degrade. The Amazon watershed is the ship’s oxygen and water regulation system. The Arctic is the ship’s temperature control system. The Sahel is a solar collection array operating under stress. The permafrost zones are frozen carbon storage that, if breached, cascade through every other system on the ship.

The communities living in and maintaining those zones are performing critical crew functions. The current accounting system records their contribution as zero because no market transaction occurs. An AI running the spaceship allocation layer would value those contributions accurately because it optimizes for ship function rather than for extractable surplus.

Every geographic crew sector has a contribution. No sector is peripheral. No location is expendable. The indigenous community maintaining a rainforest watershed is not behind a development curve that ends somewhere in northern Europe. They are operating a critical life support system that the rest of the ship depends on. The recognition economy extends to geography. The crew sector managing the Congo Basin watershed commands the same respect as the crew sector managing the semiconductor fabrication zone. Different functions. Equivalent dignity.

This is subsidiarity at planetary scale. Decisions are made at the level closest to the system being managed, by the crew members with the most direct knowledge and the most direct accountability. The coordination layer does not direct. It enables and resources. The center’s job is logistics, not command. 6

The Full Crew Manifest

The ship needs engineers. It also needs the quiet contemplative who will sit with a problem for three days before speaking and then say the thing that reframes everything.

The ship needs surgeons. It also needs the autistic systems monitor whose hyperfocus catches the anomaly that cascades into a crisis if undetected.

The ship needs administrators. It also needs the poet who names what the crew is experiencing before anyone else has the language for it.

The current system runs an extraordinarily narrow selection filter. It rewards verbal fluency, competitive drive, tolerance for hierarchy, comfort with self-promotion, and the ability to perform confidence. Everyone else is filtered out, misallocated, or ground down performing a profile that is not theirs. The autistic crew member who cannot work a room but who has logged three hundred accurate anomaly detections is not a social deficit with a technical skill attached. They are a critical systems asset with a communication difference. The frame determines the value assessment entirely.

The advancement criteria system handles this structurally. When criteria are behavioral and outcome based rather than trait based, when the record speaks rather than requiring the crew member to perform legibility to an evaluator’s social comfort, the full range of human cognitive profile becomes visible as asset rather than deviation.

A ship that systematically filters out everyone who does not match the profile of the current leadership is a ship running below its actual capacity. That is not a moral argument in the first instance. It is an engineering argument.

Socialism Without the Historical Ending

This framework will draw a label. That label may be socialism.

The label is not entirely wrong and not entirely right. What this model shares with socialism is the premise that the baseline conditions for a dignified human life should not depend on your position in a market hierarchy. Food, shelter, healthcare, education. These are crew provisions, not market commodities.

What this model does not share with the historical socialist endings is the authoritarian coordination mechanism. The Soviet model replaced market allocation with central planning, which required a planning apparatus, which required enforcement, which required the political infrastructure of control that produced the terror. The failure was not the premise that people deserve a dignified floor. The failure was the coordination mechanism chosen to deliver it.

The ISS does not have a commissar. It has a mission and a logistics system. The crew knows what the ship requires. The allocation system delivers it. Contribution is tracked and recognized. Nobody is coerced because the mission logic is visible to everyone aboard and the alternative to cooperation is sufficiently obvious that coercion is unnecessary.

Friedrich Hayek’s most durable insight, developed in The Use of Knowledge in Society (1945), was that price signals aggregate distributed information that no central planner could collect. 7 He was right. Central planning failed partly because it could not replicate that information aggregation without the price mechanism. But Hayek wrote before the age of planetary-scale AI. The coordination problem that defeated central planning is a different problem when the coordination layer is an AI optimizing for ship function rather than a planning bureau optimizing for political survival or a market optimizing for capital return.

This is not socialism as the twentieth century understood it. It is something the twentieth century did not have the technology to attempt.

The Floor and the Ladder

The architecture has two tiers and it is important not to collapse them.

The floor is the crew provision. Berth, food, healthcare, education, recreation. This does not vary by domain or by rank. It is what the ship provides every crew member by virtue of being crew. It is not means-tested. It is not conditional on contribution above minimum. It is the operational requirement for a crew member functioning at any level.

The ladder is the recognition economy. Rank within domain. Better housing. Greater responsibility. Peer recognition calibrated to demonstrated contribution. This varies enormously. The Captain of anything has earned something the junior crew member has not yet earned. That differential is real and it matters. The ladder gives the floor meaning by making advancement possible and visible.

Collapse the two tiers and you get either a flat system with no incentive structure above minimum contribution, or a system where the floor becomes conditional on performance and you are back to the anxiety economy the model is designed to replace. Keep them separate and you have something that has never quite existed at scale. A system where every human being has enough and where the definition of enough is not the ceiling.

What AI Actually Does Here

The market has performed one genuine function that any replacement must account for. Price signals aggregate distributed information about supply, demand, and relative value across millions of participants who cannot directly observe each other. Hayek was right about that specific point. Central planning failed partly because it could not replicate that information aggregation without the price mechanism.

AI changes the terms of that argument. A system capable of tracking contribution, monitoring resource flows, modeling allocation needs, and optimizing distribution across planetary scale is doing what price signals do. Without requiring profit as the motivating signal. Without requiring capital accumulation as the coordination mechanism. Without producing the inequality and externalized costs that market coordination generates as structural byproducts.

The market was the best available technology for the information problem for roughly three centuries. It is not obvious it remains the best available technology when the alternative is a sufficiently capable AI operating as the ship’s navigation and logistics system.

The political problem is real and must be named honestly. An AI running planetary resource allocation is an AI that every faction with power will attempt to capture and steer. The vendor concentration at the political layer is not a minor implementation detail. It is the central vulnerability of the model. A ship whose navigation system can be seized by a mutinous faction is a ship in permanent danger of going off course.

The answer is not to abandon the model. It is to design the governance of the AI allocation layer with the same seriousness that the model brings to crew provision and rank structure. Open architecture. Distributed oversight. Multiple redundant accountability mechanisms. No single point of political capture. That is a hard engineering problem. It is not an impossible one.

The Voyage

You did not choose to be born. You did not choose this planet or this moment in its history. You arrived as crew on a ship already in flight, with systems already running, with other crew members already at their stations.

The ship has enough for everyone aboard. It has always had enough. The question has never been whether the resources exist. The question has been whether the allocation system delivers them to every crew member or whether it routes them to some crew members and leaves others to manage without.

We have the technology to ask that question differently now. Not as a moral appeal to the conscience of the crew members currently controlling the allocation. As an engineering problem with a known solution set and a coordination layer capable of implementing it.

Every crew member has a berth. Every crew member eats. Every crew member has access to the ship’s medical systems and the ship’s knowledge systems for the duration of the voyage. Every crew member can advance within their chosen domain through criteria that are visible, consistent, and free of the political and social filters that the current system embeds in every evaluation.

No compartment is expendable. No crew member is surplus. No location is peripheral.

The ship is flying. The question is whether we are going to run it like a ship or continue arguing about who owns the oxygen.

A Framework, Not a Blueprint

This piece does not have all the answers to questions that have occupied philosophers, economists, and engineers for centuries. It is not meant to.

The AI governance layer, the contribution measurement architecture, the transition path from the current system to something better. Those are problems for the greatest technical and policy minds on the planet. The purpose here is to state the goal clearly enough that the people capable of solving the hard problems know what they are solving toward.

The alternative is worth naming directly. An AI-accelerated concentration of wealth in an elite that has structurally decoupled from the billions it left behind is not an optimization target. It is not an acceptable outcome dressed in the language of innovation. Creating trillionaires while billions of crew members are left in the hold is not a feature of the system. It is a design failure of the first order.

The ship can be run better than that. This framework is an invitation to the work of figuring out how. Comments, challenges, and serious engagement from anyone who has thought harder about any piece of this are exactly what is needed. The goal is worth the effort. The crew requires it.

This Is What the Bill Was For

Over the past several months this newsletter built the case for a Global Citizen’s Capital Right. The argument was not charity. It was not redistribution. It was a bill presented to the institutions and capital systems that have extracted value from human labor, human attention, human data, and human creative capacity across four successive waves of automation, without adequate return to the people who generated that value.

The CCR series called it restitution. Not a safety net. Not welfare. A property claim. Every worker whose skill was automated, every community whose industrial base was offshored, every household whose domestic data became a training set for systems that now compete with their children’s employment prospects. They are owed. The bill exists whether or not the debtors acknowledge it.

The spaceship is what you build when the bill is paid.

The floor, the berth, the commons, the medical system, the knowledge system, the recognition economy, the transparent rank ladder, the lateral respect between domains, the geographic equity between crew sectors. None of this is possible at scale while the allocation system routes surplus to capital rather than to crew. The CCR establishes the ownership claim that makes the floor fundable. The spaceship establishes what the floor is for.

This is the destination the earlier work was pointing toward without fully naming. The CCR presented the bill. The spaceship is what we buy with it. Not comfort for its own sake. Not consumption without limit. A vessel properly crewed, properly provisioned, and properly governed, flying toward something worth the voyage.

Every crew member aboard. No one left in the hold.

James A Ramsey in J Allen Insights covers AI economics, labor displacement, and the political economy of technological change. James Allen Ramsey is a retired tax attorney and consultant whose career spanned GM, PwC, Shaw Pittman, and EY. The Global Citizen’s Capital Right series ran April 5 through April 21.

Iran does not need the ceasefire to hold. It needs the ceasefire to wobble.

The weekend exchange of fire looked, from the outside, like escalation threatening the peace. Iran struck a Qatari-flagged tanker in the Strait of Hormuz. The US retaliated against Iranian coastal installations. Iran hit US military facilities in Kuwait and Bahrain. Trump threatened that Iran “will no longer exist.” Araghchi threatened a complete halt to negotiations.

Markets read chaos. Analysts read fragility. Both are wrong about the direction of Iranian incentives.

Look at what Iran is actually selling, at what price, and to whom, and the tit-for-tat pattern resolves into something colder and more deliberate: a managed friction strategy that is generating the best oil economics Iran has seen in a generation.

The Toll Booth

When the US and Israel struck Iran on February 28, the IRGC did not simply close the Strait of Hormuz. It converted the strait into a checkpoint. Iranian vessels kept moving. A select list of nations, China, Russia, India, Iraq, Pakistan, received transit permissions. Everyone affiliated with the US or Israel was blocked. The IRGC published its own routing map, pushing non-Iranian traffic through Iranian territorial waters past Larak Island, where the Revolutionary Guards Navy could board, inspect, and in some cases levy fees.

The official IMO shipping lane was abandoned. Iran became the operating authority of the world’s most important energy chokepoint, collecting passage rights on traffic it permitted and eliminating traffic it chose not to permit.

This is not a blockade. A blockade stops all movement. This is a toll booth.

The Competitor Problem, Solved

Before the war, Iran competed for Asian crude buyers against Saudi Arabia, Iraq, Kuwait, and the UAE. Those competitors priced aggressively, had reliable shipping infrastructure, and faced no sanctions risk. Iran moved roughly 1.5 to 2 million barrels per day, almost entirely to Chinese teapot refineries at a $10 to $12 discount to Brent. It was a constrained, discounted market share built on evasion.

The war changed the competitive landscape entirely. Saudi Arabia and the UAE have pipeline bypasses. The East-West pipeline to Yanbu on the Red Sea carries 5 million barrels per day at capacity. The Abu Dhabi to Fujairah line handles 1.5 million. But those two alternatives, running at full capacity, cover only about a quarter of what normally transits Hormuz. Saudi and Emirati volume collapsed anyway. The Saudis shut wells. Iraq, with no bypass and no clout, lost its export capacity almost entirely. Gulf Cooperation Council oil production fell by a reported 10 million barrels per day by mid-March.

Iran’s chief competitors were sidelined. Iran kept moving product.

The Sanctions Inversion

Here is the number that should be in every headline it is not in: on June 22, the US Treasury issued General License X, a 60-day blanket waiver authorizing the production, delivery, and sale of Iranian crude oil, petrochemical products, and petroleum products in US dollars through August 21. This is the first time Iran has been permitted to conduct dollar-denominated oil transactions in more than four decades.

The maximum pressure campaign that began in February 2025, which sought to drive Iranian oil exports to zero, is now the instrument of Iranian sanctions relief. Trump signed it. Treasury issued it. Iran is cashing it.

The pricing consequence is direct. Iranian crude, which previously sold at a $10 to $12 discount to Brent because buyers had to absorb sanctions risk, can now command market rates. Analysts noted that Iranian crude could shift to a premium above Brent given demand pressure. Competitors are constrained. Iran is authorized. The discount evaporates.

UANI, which tracks Iranian oil movements with hostility toward Tehran, confirmed that since June 14, 31 tankers departed the Gulf of Oman carrying approximately 41 million barrels of Iranian oil, generating an estimated $3.5 billion in revenue for the IRGC. The same IRGC running drones into Bahrain this weekend.

What the Attacks Actually Do

Each round of tit-for-tat fire does something specific to the shipping market: it reprices war risk insurance.

Insurance is the mechanism that controls non-Iranian volume more than any military action does. When fighting resumes, actuaries revise. War risk premiums spike. Shipowners hold cargo. Asian importers searching for Saudi or Iraqi crude find that the cost to move it, the insurance premium alone, makes the economics worse than buying Iranian crude under General License X with no war risk because Iran controls the lane.

Every attack keeps that calculus in place. A full ceasefire, with genuine Strait normalization, restores Saudi and Emirati volume to Asian buyers who have spent months paying premium prices for whatever they could source. That supply restoration collapses the price back toward pre-war levels. Saudi Aramco resumed loading at Ras Tanura last week. Brent immediately fell toward four-month lows, dropping more than 10 percent in the prior week before the weekend escalation reversed part of that decline.

Iran does not benefit from Brent at $72 and a normalized strait with Saudi competition restored. Iran benefits from Brent at $85 to $95, Iranian lanes open, Saudi lanes constrained, and war risk insurance keeping non-Iranian tankers at anchor.

The optimal position is not peace. It is controlled friction.

The Sovereignty Argument

Araghchi’s insistence that Iran must govern the Strait of Hormuz is not rhetorical posturing. It is the economic architecture he is protecting.

MOU Article 5 states that Iran will “make arrangements using its best efforts for the safe passage of commercial vessels” during the 60-day window. It does not say Iran cedes control. It does not say the IRGC stands down from the strait. Every US attempt to establish alternative routing or multinational maritime administration is, from Tehran’s perspective, an attempt to dismantle the toll booth that is generating $3.5 billion in two weeks.

The IRGC struck the Qatari tanker, Kiku, last week not randomly. Qatar is a key US mediator and one of the parties the MOU uses to hold unfrozen Iranian funds in escrow. Qatar’s oil moving through the strait on non-Iranian terms is a direct challenge to Iranian operational control of that corridor. The strike said: this lane is ours.

Who Blinks First, Restated

The framework question was always: who absorbs more pain from the current state. The answer, updated for today, is not Iran.

Iran is selling oil legally, in dollars, at improving prices, through a corridor it controls, while its primary competitors sit at anchor. The IRGC is funded. The negotiating position is that any final deal must preserve Iranian sovereignty over the strait, which preserves the toll booth.

The US is being asked to guarantee a deal that requires it to either give Iran permanent control of the world’s most important oil lane or maintain a military presence in the strait indefinitely while managing a restive Congress, a midterm election cycle, and allies in Bahrain and Kuwait who are absorbing drone strikes from the party the US just handed sanctions relief.

The ceasefire wobbles. Iran calls it a “complete halt” to negotiations and then shows up in Doha on Tuesday.

That is not a contradiction. That is the strategy.

James Allen Ramsey of J Allen Insights covers the political economy of AI displacement, technological change, and the institutional failures that precede them. The Iran war is the largest economic shock since 2008, and its labor and capital implications are not being reported as such.

RAISE US launched today. The people behind it are serious. The money is real. The problem is that the architecture assumes something that is no longer true: that there are enough destination jobs to train people toward, and that the educational system can identify them before they disappear.

Gina Raimondo and Eric Holcomb announced RAISE US this morning, a nonpartisan nonprofit backed by more than two dozen major employers, with a target of $1 billion in commitments and initial state partnerships in Arkansas, Connecticut, Maryland, and Utah. The advisory board reads like a who’s who of serious people: David Autor from MIT, Erik Brynjolfsson from Stanford, Raj Chetty from Opportunity Insights, Liz Shuler from the AFL-CIO. The Rockefeller Foundation is in. So are OpenAI and Anthropic. So is Amazon, which just finished laying off 16,000 corporate workers.

The goal is to retrain displaced workers and help them land good jobs in the AI economy.

The problem is not the effort. The problem is the math. And underneath the math is a structural failure in American education that nobody in the coalition is positioned to fix. The foundation was cracked long before AI arrived.

Problem One: There Is No Destination

Every retraining program in history has assumed a destination. You train people toward something. The question RAISE US cannot answer at scale is: toward what?

The standard answer is “AI-adjacent roles.” AI engineers, machine learning operations specialists, AI safety researchers, prompt engineers, data infrastructure architects. These are real jobs. They are also jobs that require years of accumulated technical context, not an 18-month community college certificate or an AI-powered career navigation platform in Arkansas. They are not entry points. They are destinations that presuppose a career path that itself no longer exists.

To be precise: destinations do exist, but they are fewer, more hybrid, and poorly signposted. Many emerging roles combine domain expertise with AI tools: healthcare operations, logistics optimization, energy systems, construction tech. The labor market is shifting from ladder-based progression, where each rung leads predictably to the next, to something more like a lattice, where movement is lateral, contextual, and hard to credential. The problem is not the complete absence of destinations. The problem is that retraining programs are still designed for ladders, and the people who need them most have no map for the lattice.

Here is what is happening to that career path. Employment for software developers aged 22 to 25 fell nearly 20 percent from its 2022 peak, while employment for developers over 26 grew. AI tools now handle the boilerplate coding, routine testing, and scripted debugging that junior roles were built around. The entry-level on-ramp into the one industry everyone assumed would absorb the displaced is collapsing. The ladder is being pulled up from inside the building RAISE US is trying to send people into.

For every AI system deployed, dozens of roles are replaced. A handful of new jobs are created. The new jobs require PhDs or the equivalent in accumulated experience. Neither RAISE US nor anyone else has a credible plan to manufacture those at scale.

Problem Two: The IT Floor Is Gone

Prior disruptions had a floor. When manufacturing left, there was still IT. When call centers offshored, there was still software development. There was always a knowledge-work tier that could absorb people willing to retrain and move up the skill ladder.

That floor is gone now.

Tech layoffs in 2026 are running at 1,115 job cuts per day. The companies conducting those layoffs are not struggling. Oracle, Meta, Amazon, Intuit, Cisco: all profitable, all cutting, all citing AI in the same breath they cite efficiency. Meta plans to cut 8,000 employees and freeze 6,000 open roles it had planned to fill. Coinbase is restructuring around “AI-native pods” and experimenting with one-person teams where a single engineer handles what previously required a designer, a product manager, and a developer.

The prior disruptions each had a sector that was net absorptive at scale. This one does not. Skilled trades, healthcare, and energy infrastructure are absorbing some AI-augmented labor, and those sectors matter. But they cannot absorb millions of displaced knowledge workers, and the sector that was supposed to be the primary destination is contracting its own entry door at the same time it is being asked to serve as the landing zone for everyone else.

RAISE US is training people to climb a ladder whose bottom rungs are being removed.

Problem Three: You Cannot Reskill Your Way to San Francisco

Problems one and two are about jobs that do not exist. Problem three is about geography, and it is the one nobody in the coalition will say, because half of them are headquartered in the cities that caused it.

Prior economic disruptions worked, imperfectly, through geographic mobility. A steelworker in Pittsburgh could move to where the new work was. The geographic arbitrage existed. Workers could follow the jobs, at real personal cost, but it was possible.

That option is structurally broken now.

The remaining human roles in AI-proximate work are concentrated in a handful of metros. San Francisco. Seattle. New York. Boston. The housing cost structure in those cities makes relocation economically irrational for anyone without existing capital. A 50-year-old logistics coordinator in Youngstown who completes a data analytics certificate does not have a viable path to a $4,000-per-month one-bedroom in the Bay Area. The math does not close.

When textile work left the American South for Asia, workers could not follow it across an ocean. That was recognized as a structural failure, not a training problem. This is the same structure, compressed into the domestic economy. The jobs are in three zip codes. The displaced workers are everywhere else. Arkansas LAUNCH, the AI-powered career navigation platform RAISE US is piloting in its first state partnership, will help people find their way toward that gap. It will not close it.

Problem Four: The School System Is Still Teaching BASIC

Problem four runs deeper than the first three.

The retraining conversation assumes the educational pipeline is aimed in roughly the right direction and just needs to move faster. It is not.

American education has a structural lag problem that predates AI by decades. In the 1990s, grades 3 through 12 across much of the country were learning BASIC programming. Not because BASIC was wrong as a concept, but because the curriculum designers were working from a vision of the future that was already obsolete. The actual skills the workforce needed were Microsoft Office, relational databases, and networked communication tools. Those were being deployed in every office in America while children were writing BASIC loops in computer lab. The school system was preparing students for a world that had already moved on.

That lag did not correct itself. It institutionalized.

Today, despite genuine legislative momentum, only 43 percent of career and technical education teachers say their schools use job growth projections to make programming decisions. California passed a law in 2024 requiring AI literacy integration into curriculum frameworks. Virginia issued a statewide request for AI educator training in 2025. Maryland passed the AI Ready Schools Act in 2026. Oklahoma’s most recent legislation requires districts to adopt a written AI policy before the 2027-28 school year. These are real steps. They are also policy responses to a deployment curve that is moving in months, not legislative cycles.

The deeper problem is that curriculum design requires knowing what skills will be relevant when today’s third-grader enters the workforce in 2038. Nobody in the AI field is confident about what work looks like in 2038, because the models that will define that landscape have not been built yet. An educational system debating AI policy for 2027-28 is not equipped to solve a labor market problem that is restructuring quarterly.

Problem Five: The Floor Was Never Level

The lag problem would be serious enough on its own. The equity problem makes it irreversible for millions of children before the disruption even arrives.

American public education is funded primarily through local property taxes. That single structural fact means the quality of a child’s education is determined largely by the wealth of the neighborhood they were born into. Predominantly white school districts, which tend to be smaller and wealthier, receive $23 billion more in aggregate funding than districts serving predominantly students of color. The spending gap between the richest and poorest districts runs to more than 50 percent per student. Reform efforts over three decades have narrowed some of these gaps within states, and the picture is genuinely mixed. But the inequality has never been closed, and the effort to close it is now running directly into federal funding instability at exactly the wrong moment.

This is not a background condition. It is the primary condition that determines whether the RAISE US vision is even theoretically reachable for the workers who need it most.

The children coming out of underfunded districts are not arriving at the workforce as blank slates ready to absorb new training. They are arriving with the compounded disadvantage of smaller vocabularies, weaker analytical foundations, less exposure to technology, fewer networks, and the specific learned belief, reinforced by years of institutional neglect, that the educational system was not built for them. That belief is accurate. It is also the hardest thing to retrain.

Here is the cruelty of the timing. The moment when AI could genuinely democratize education has arrived simultaneously with the moment when AI is eliminating the jobs that education was supposed to lead to. Adaptive AI tutoring platforms can now deliver genuinely personalized instruction at scale, adjusting to each student’s pace and learning style in real time, providing immediate feedback, meeting students where they are. The technology to give every child in a poor district the equivalent of a patient, knowledgeable tutor available at any hour exists today. It is being deployed right now, mostly in wealthy districts and private schools that can afford the implementation.

The tools that could close the educational gap are being distributed along the same property-tax fault lines that created the gap in the first place.

What the AI era actually demands from education is not a fixed set of technical skills. Technical skills have a shorter half-life than ever. What it demands is something harder to credential and harder to test: the capacity for free thinking, adaptive reasoning, and the foundational understanding that learning is not a phase of life but the condition of participating in the economy at all. A student who leaves school knowing how to learn, how to ask good questions, how to evaluate evidence and adapt when the ground shifts, has something that retains value regardless of which specific tools dominate in 2032 or 2038. That is the education the wealthiest districts have always provided, through smaller class sizes, richer curricula, teachers who are not managing crisis conditions, and the simple dignity of institutional attention.

The RAISE US coalition is trying to patch a workforce gap that starts in third grade. The patch is being applied at the moment of job displacement, thirty years after the foundation cracked.

The Same Pitch, Again

This is not the first time American institutions have responded to AI-driven labor disruption with a retraining program. It is the fourth.

ERP systems displaced back-office workers in the 1990s. The answer was retraining. Offshoring displaced shared services workers in the 2000s. The answer was retraining. Robotic process automation displaced data-entry and reporting roles in the 2010s. The answer was retraining. Now AI agents are displacing knowledge workers across every sector. The answer, again, is retraining.

Each time, the workers who were displaced did not land in the new jobs the technology created. The new jobs went to a smaller, more credentialed cohort. The displaced workers aged out, cycled through lower-wage work, or left the labor force. The aggregate statistics looked fine. The distributional reality did not.

RAISE US is serious money and serious people delivering the same pitch into a disruption that is faster, broader, and more structurally complete than any of the prior waves.

Raimondo herself called past retraining efforts “ineffective.” Then she launched another one.

What the Math Actually Requires

The retraining model has a hidden assumption: that the productivity gains from AI will eventually generate enough new demand to re-employ the workers who were displaced. That assumption may be true in aggregate, over a long enough horizon, at a macroeconomic level. It is not a plan for the specific worker in the specific city who is displaced in 2026 and cannot wait for the macro equilibrium to arrive.

To be fair to RAISE US: retraining is not useless. It reduces friction for workers who can transition. It tightens the connection between employers and training providers. It shortens the gap between displacement and re-employment for the people it reaches. The bucket matters locally even when it cannot stop the flood.

The failure mode is not that programs like this do nothing. The failure mode is that they succeed marginally and are then mistaken for a system-level solution. The problem is not that RAISE US is wrong. The problem is that it is operating on the wrong layer. Retraining addresses the symptom. It does not address where the value is going.

The productivity gains are real. They are accruing now. They are flowing to capital: to the companies doing the deploying, to the shareholders holding the equity, to the infrastructure providers collecting the compute revenue. That is not a moral complaint. It is an accounting observation. The value is being created, and it is being booked somewhere. The question is whether the people whose accumulated knowledge, creative output, and labor history made those AI systems possible have any recognized claim on what is being created.

That is the argument for a Citizen’s Capital Right.

Not a safety net. Not a retraining stipend. Not a grant from a philanthropically-funded policy lab that keeps its recommendations at arm’s length from its corporate sponsors. A property right. A recognized claim on the collectively generated knowledge that trained the systems now displacing the people who generated it.

The CCR starts at the citizenship level: every American holds a dividend stake in the AI productivity gains that rest on publicly funded research, educational infrastructure, and the language and culture and creative work that trained every frontier model in production today. That is not redistribution. That is a bill coming due.

The CCR is only the beginning. As AI capabilities extend across borders, as the extraction of human-generated knowledge runs through content moderation workers in Nairobi and data labelers in Manila and warehouse pickers in every logistics hub on earth, the logic extends to a Global Citizen’s Capital Right: participation in the gains proportional to contribution to the substrate those gains were built on.

RAISE US is a bucket. The flood requires different infrastructure.

The bucket is not nothing. The people carrying it are not cynical. But a billion dollars spread across pilot programs in four states, aimed at a job market with no scalable destination for displaced workers, in an IT sector that is eliminating its own entry door, in cities where displaced workers cannot afford to live, taught by an educational system still debating whether to require AI policy by 2027, delivered to a workforce that was underserved by that system from the third grade forward, is not commensurate with the scale of what is arriving.

The CCR is commensurate. It does not ask whether there are enough jobs. It asks a different question: given that the system is generating enormous value from a substrate that everyone contributed to, what is the fair claim of the people who built that substrate?

That question has an answer. RAISE US is not asking i

J Allen Insights covers AI economics, labor displacement, and the political economy of technological change. The Citizen’s Capital Right series, Parts 1 through 3, is available in the archive for readers who want the full property-rights argument.

The Week of Partial Recognition

In the first week of June 2026, three of the most consequential people in the AI economy used the word ownership in the same breath as the public for the first time. Bernie Sanders introduced the American AI Sovereign Wealth Fund Act, which would impose a one-time 50 percent tax on each major AI company’s stock, paid in shares, depositing the equity into a public fund that gives ordinary Americans voting rights, board representation, and eventually a check in the mail. Sam Altman requested a private meeting with Sanders and told him he also wants the public to have equity in AI companies. Donald Trump, speaking to reporters on Air Force One, described a potential partnership where the American people can benefit from the success of AI and called it a beautiful thing.

This has not happened before. For the first time in the AI era, the people with the most power over the outcome are acknowledging the ownership claim rather than redirecting it toward retraining programs and welfare transfers. That is a genuine shift and it deserves to be named as one.

It is also insufficient. Not because the instinct is wrong. Because the ledger they are reading from is missing most of its pages.

Sanders names two entries correctly. The large language models were trained on the accumulated written output of human civilization without compensation. The infrastructure that made them possible, the internet, the semiconductor supply chain, GPS, the fiber networks, the research universities, was built on public investment. He is right on both counts. Those entries belong in the ledger.

But the ledger does not begin with the New Deal. It does not begin with the internet. It does not begin with any moment that fits inside an American policy conversation. It begins much earlier, it is much larger, and until it is read in full, every framework built on a partial reading will reproduce the problem it claims to solve.

What the Ledger Actually Contains

Capitalism did not begin with Adam Smith in 1776. It began with the Portuguese forts on the West African coast in the 1440s, with the Genoese bankers who financed Columbus, with the Spanish silver mines at Potosí where between four and eight million Indigenous and African workers died over three centuries to fund the European price revolution.

The standard textbook story has capitalism emerging from European cleverness, the Protestant work ethic, the enclosure movement, the steam engine. The honest story has capitalism emerging from a triangle: African bodies, American land, European capital, with the bodies and the land providing the value that the capital then claimed as its own productivity.

Eric Williams made this argument in 1944 in Capitalism and Slavery and was ignored for fifty years because the argument was correct and inconvenient. He showed that the British industrial revolution was capitalized in significant part by sugar profits, and that sugar profits were a function of slave labor, and that the cotton mills of Manchester ran on cotton picked by enslaved people in Mississippi. The factory that defined industrial modernity was downstream of the plantation that financed it. The wage worker and the slave were not two different systems. They were two ends of one supply chain.

The accounting becomes harder to ignore when you trace what happened to the wealth. The cotton picked by enslaved hands in Alabama in 1850 became cloth in Manchester, became capital in London, became investments in railways in Argentina, became loans to governments in Egypt, became the basis of the financial system that the City of London still operates today. The descendants of the enslaved received nothing. The descendants of the capital holders received compounding returns for six generations.

When Britain abolished slavery in 1833, it paid twenty million pounds in compensation, equivalent to forty percent of the Treasury’s annual income, and it paid that compensation to the slaveholders, not to the enslaved. The British taxpayer finished paying off that loan in 2015. Every British citizen alive in this century has been servicing the debt owed by the state to slaveholders, while the debt owed to the enslaved has never been calculated, let alone paid.

The American case is the same logic in a different jurisdiction. The Thirteenth Amendment freed four million people who had been worked for centuries and gave them nothing. The forty acres and a mule of Sherman’s Field Order 15 was rescinded within months. The Freedmen’s Bureau was starved of resources and dismantled. Sharecropping reconstituted the plantation under a different legal name. Convict leasing under the Thirteenth Amendment’s loophole produced a second slavery that ran until the Second World War.

The redlining of the Federal Housing Administration in the 1930s and 1940s denied Black families the single largest mechanism of postwar wealth accumulation, the subsidized mortgage, which was the foundation of the white middle class. The wealth gap between Black and white American families today is not the residue of historical injustice. It is the precisely calibrated outcome of policies that ran from 1619 to the present, each of which moved value from one column of the ledger to another, and none of which was ever reversed by payment.

This is the same ledger we are now arguing about with respect to the robot. The Bracero who picked oranges in 1955 and was sent home with nothing is on it. The garment worker in Dhaka is on it. The Congolese cobalt miner is on it. The enslaved African in 1750 is on it. The Indigenous worker who died in the silver mines of Potosí in 1610 is on it. The ledger has been open for half a millennium. The debtor has changed names: the Portuguese Crown, the Dutch East India Company, the British Empire, the American Republic, the multinational corporation, the platform monopoly, the foundation model lab.

The function is continuous. Extract value from labor whose political voice has been engineered down to zero, capture the value at the top of the chain, normalize the arrangement by writing history books in which the value is attributed to the capturer’s cleverness rather than the laborer’s contribution. The robot is the latest mechanism. The pattern is older than any nation currently presenting itself as the rightful holder of the gains.

The Shares That Were Never Issued

In a properly structured corporate formation, the founders who contribute labor and know-how receive shares for that contribution. The sweat equity hits the capital ledger as paid-in contribution. It becomes a permanent claim on the entity. The founder owns part of what the company will become because the work that built the company was treated as capital input, not as expense.

For more than five hundred years, a large portion of the labor and know-how that built the modern economy was harvested without ever hitting the capital ledger. The contributors were real. The contributions were real. The shares were never issued. Some contributions were voluntary and some not.

Every one of them put in what a founder puts in. None of them got what a founder gets. The shares were issued only to whoever held the title. And it continues today with elite contributors.

In May 2026, Meta workers circulated a petition after Mark Zuckerberg announced the Model Capability Initiative, a program logging their keystrokes, mouse movements, and periodic screenshots to train the next system. The petition used precise language: the company is “non-consensually extracting their data for the purposes of AI training.” That sentence is the first sentence of a much longer bill. It is in the same handwriting as the petition the slaves did not get to file, the indentured did not get to file, the children in the mills did not get to file, the miners with destroyed lungs did not get to file. The book is being closed on the labor account. Every share that was never issued is becoming visible at once, because there is no more new contribution on the other side to keep the absence hidden.

The American Trap

At an investor event in Hong Kong, Standard Chartered CEO Bill Winters said the bank’s AI-driven restructuring is “not cost-cutting” but “replacing, in some cases, lower-value human capital with the financial capital and the investment capital,” the goal being to eliminate 7,800 back-office positions by 2030 across HR hubs in India, Malaysia, Poland, and China.

Bill Winters did not misspeak. He used the precise language of an accounting convention that has governed the treatment of labor for five centuries. Lower-value human capital. Not workers. Not people. Not contributors. Capital with a valuation already assigned, and the valuation is lower, which means it gets replaced by capital with a higher return. The transaction is described entirely from the perspective of the entity doing the replacing. The replaced party has no standing in the sentence. They are an input being substituted, not a contributor being exited.

He walked it back the next day. The memo said the quiet part loud, got the backlash, and reached for the familiar instrument. “Where roles do fall away, it reflects changes in the work, not the value of our people.” The words changed. The 7,800 jobs did not change. The accounting convention did not change. The memo is the deferral mechanism in miniature: correct the language, proceed with the transaction, move on.

What Winters said out loud in Hong Kong is what the ledger has been saying in silence since the Portuguese built their first fort on the West African coast in the 1440s. The enslaved African was lower-value human capital compared to the mill owner. The indentured servant was lower-value human capital compared to the colonial merchant. The Potosí miner was lower-value human capital compared to the Spanish Crown. The Kenyan content moderator paid $1.50 an hour to make Western AI products safe is lower-value human capital compared to the San Francisco engineer whose equity vests next quarter. The convention is not new. The press conference is new.

Now comes the trap.

Sanders is proposing an American AI Sovereign Wealth Fund. Trump is proposing an American partnership in AI. The word American is doing the same work in both proposals that lower-value is doing in Winters’ sentence. It is the valuation differential. It is the basis on which some contributors receive a return on their capital and others do not.

The Congolese miner whose cobalt is in the data center that trained the model contributed to the inheritance. He did not choose to contribute. The contribution was extracted under conditions that have not materially changed since the Belgian Crown ran the Congo as a private estate. His contribution is in the ledger. His citizenship is not American. Under both the Sanders proposal and the Trump framing, he receives nothing.

The Kenyan content moderator who spent her working hours reading the most violent and degrading content the internet produces so that the model would learn not to reproduce it contributed to the inheritance. Her labor made the product safe for the Western consumer. Her citizenship is not American. She receives nothing.

The Bangladeshi garment worker whose displacement from textile manufacturing into lower-wage work was the direct result of the automation wave that generated the capital that built the data centers contributed to the inheritance through the labor arbitrage that funded it. Her citizenship is not American. She receives nothing.

The Indian developer whose Stack Overflow answers, whose GitHub repositories, whose technical documentation written in English to reach a global audience became part of the training corpus, contributed directly to the knowledge base the model runs on. His contribution is measurable. His citizenship is not American. He receives nothing.

A Citizen’s Capital Right that stops at the American border is not a settlement of the 500-year ledger. It is the latest entry in it. It takes the extraction pattern, adds a domestic dividend, and calls it justice. The people who were most thoroughly extracted from receive nothing. The people who happened to be born in the country that did the extracting receive a check. The accounting convention does not change. The beneficiary of the convention gets a new name. The function stays continuous.

This is not an argument against an American CCR. It is an argument about what an American CCR without a global framework actually is. It is the latest iteration of the same maneuver the ledger has been recording since Potosí. Extract value from labor whose political voice has been engineered down to zero. Capture the value at the top of the chain. Normalize the arrangement by writing the policy in language that sounds like justice while leaving the foundational entries unaddressed.

The debtor has changed names before. The Portuguese Crown. The Dutch East India Company. The British Empire. The American Republic. The multinational corporation. The platform monopoly. The foundation model lab. And now, potentially, the American AI Sovereign Wealth Fund. Each name change is presented as a new arrangement. The ledger knows the difference between a new arrangement and a new name for the same one.

The basis for the CCR claim is contribution, not citizenship. The labor and knowledge capital investment that justifies the ownership stake was made by every person who contributed to the corpus, to the infrastructure, to the resource base, to the supply chain that built the system. That contribution was made across every inhabited continent, across five centuries, willingly and by force, compensated and uncompensated, documented and invisible. The fact of contribution is the basis for the claim. The accident of citizenship is not.

A framework that honors the claim must be built on that basis or it is not a framework. It is a dividend paid by the latest version of the debtor to the latest version of the preferred creditor, while the foundational account stays closed.

The Winters memo corrected the language and proceeded with the transaction. That is what a domestic-only CCR does to the global ledger. It corrects the language. It calls itself a Citizen’s Capital Right. And it proceeds with the transaction.

The View From High Enough

There is a vantage point from which the accounting becomes impossible to avoid.

The Artemis II crew returned from lunar orbit earlier this year. The coverage was brief. The public attention was thinner than it should have been for the first humans to travel that far from Earth in more than fifty years. But they saw what every astronaut who has made that journey reports in remarkably similar language. The blue marble hanging in darkness. A thin layer of atmosphere separating everything we have ever known from the void. Oceans, continents, rivers, mountain ranges, all without the lines the maps say should be there. One thing. Clearly finite. Clearly the only home any of us has.

The borders that define which contributors receive a return on their capital and which do not are not visible from that altitude. They were never visible from that altitude. They are a convention maintained by the people who benefit from them, legible only at the scale where the fiction can be sustained.

The Iran war has been conducting a tutorial on what the world actually looks like without that fiction. The Strait of Hormuz is twelve miles wide at its narrowest point. Through it passes roughly twenty percent of the world’s oil and a significant portion of the liquefied natural gas that heats European homes and powers Asian factories. When the strait tightens, fertilizer prices move in Brazil. Wheat futures spike in Egypt. Corn prices shift in Iowa. The farmer in Kansas who has never heard of the Hormuz toll strategy discovers that his input costs are determined by a body of water he cannot locate without help. The interdependence was always there. The crisis made it legible.

The supply chain that built the AI systems is the same map at a different scale. The cobalt in the data center was mined in the DRC. The rare earth elements in the chips were extracted from supply chains running through some of the most exploited labor markets on the planet. The fiber carrying the training data was laid across a dozen jurisdictions. The content moderators who made the product safe enough to sell to Western consumers worked in Nairobi and Manila. The developers whose code is in the training corpus worked in Bangalore and Warsaw and São Paulo. The oral traditions and non-English languages that make the model functional across cultures were contributed by people who will never see a terms of service agreement, let alone an equity grant.

This is not one country’s inheritance. It is the inheritance of the species, assembled from contributions made across every inhabited landmass, over five centuries, by people who had no idea they were building something that would one day be worth trillions of dollars to whoever happened to hold the title when the accounting finally came due.

The astronaut looking down sees one system. The ledger, read in full, describes one system. The Strait of Hormuz, the cobalt mine, the content moderator, the training corpus, the wheat price in Egypt: these are not separate stories. They are the same story told from different coordinates on the same sphere. A framework adequate to the claim has to be built at the scale of the system it is describing. Anything smaller is not a settlement. It is a boundary drawn inside the problem.

What the Full Ledger Requires

What the full ledger requires is a Global Citizen’s Capital Right. The GCCR, framed as a return on uncompensated labor and knowledge capital investment, is not redistribution. It is not welfare. It is a property rights claim. The contributors made a capital investment. The investment has been compounding for five centuries inside every technology built on top of it. The AI wave is the moment when the returns become large enough and the extraction visible enough that the claim can finally be asserted with the precision it requires.

But the instrument has to match the scale of the claim. Three things follow from reading the full ledger.

First, a global capital account. The contributions were global. The ownership claim is global. Any settlement that stops at national borders is not reading the full ledger. A framework worthy of the claim recognizes contribution as the basis for ownership across every jurisdiction where that contribution was made. This is not a call for a world government. It is a call for an accounting convention that matches the actual transaction.

Second, prospective yield rather than retrospective debt. The past contributions are not calculable in dollar terms and the attempt produces either impossibly large numbers or politically dismissable ones. The right instrument is a recognized ownership stake in the ongoing productivity of the systems built on uncompensated capital. The yield is forward-looking. The justification is backward-looking. The ledger establishes the claim. The dividend mechanism delivers the return.

Third, the distinction between the GCCR and redistribution must be held at every level and at every border. The claim is not that wealthy nations owe poor nations a transfer. The claim is that contributors were not issued shares for their contributions, and the entity that benefited owes them equity, not charity. That bill survives the redistribution objection because it is not redistribution. It is a property rights settlement five centuries in arrears.

Pope Leo XIII in Rerum Novarum in 1891 said workers have a natural right to the fruits of their contributed labor. He was right. He was also reading from a ledger that did not include the colonial extraction funding the industrial economy he was analyzing. The encyclical was radical for its moment and blind to the scale of what was already underway.

Pope Leo XIV signed Magnifica Humanitas on May 15, 2026, the 135th anniversary of Rerum Novarum, to signal that he understood the continuity. He named the DRC cobalt miner working for less than two dollars a day. He named the content moderator, young, predominantly female, reading the most violent material the internet produces for minimal wages. He wrote that “the bodies of these people are scarred, injured and worn down so that computational flow may continue uninterruptedly.” He called for AI to be disarmed from logics of domination, exclusion, and death. He presented the encyclical alongside a co-founder of Anthropic, in a gesture that said plainly: the Church is not opposed to the technology. It is opposed to the accounting convention that has governed the technology’s relationship to labor since Potosí.

Leo XIV is not completing Leo XIII’s argument. He is reading from a larger ledger than Leo XIII had access to. The moral logic is the same. The full scope of the transaction is now visible in a way it was not in 1891. Both Leos are pointing at the same thing. Neither has yet named the mechanism by which the claim gets structural form. That is what the GCCR is.

Teilhard de Chardin spent his life describing the convergence of human intelligence into a planetary mind, the Noosphere building toward an Omega Point where the accumulated knowledge of the species becomes something the species has never been before. That convergence is not a Western project. Every contribution to the corpus, from every culture, every language, every generation, every person who ever solved a problem and wrote it down or taught it to another, is equity in the planetary mind being assembled right now. The GCCR is the structural expression of what it means to take that seriously. Not as a sentiment. As an accounting entry.

To Commence the Accounting

The ledger has been open for half a millennium. The debtor has changed names. The function has stayed continuous.

What is new is not the debt. What is new is the moment. The AI wave is the first time in history that the mechanism of extraction has become large enough, fast enough, and visible enough that the full bill can be read in a single human lifetime. The cotton-to-Manchester-to-London chain took generations to trace. The cobalt-to-data-center-to-model chain is traceable in an afternoon. The training corpus that contains every language, every culture, every tradition of human knowledge is documented. The content moderators in Nairobi are on the record. The Stack Overflow contributors are in the commit history. The ledger is not hidden. It is being ignored by people with the political incentive to read only the most convenient page.

Sanders is reading one page. Trump is reading a different page. Both pages are from the same document. Neither has opened to the front.

The full ledger requires a reckoning that no current proposal is large enough to deliver. That does not make the current proposals worthless. A domestic CCR that recognizes American contributors is better than a retraining voucher. A stock tax that captures some equity is better than a partnership announcement. Every partial reading of the ledger is better than no reading at all.

But partial readings carry a cost. They normalize the convention while appearing to challenge it. They say: we have answered the ownership question. And the people whose entries are on the pages that were not read have no standing to object, because the people doing the reading have already declared the account settled.

The account is not settled. It has never been settled.

The robot is the latest mechanism. The pattern is older than any nation currently presenting itself as the rightful holder of the gains.

To commence the accounting.

James Allen Ramsey writes J Allen Insights on the intersection of technology, geopolitics, and human development. He is a retired tax attorney and consultant whose career included GM, PwC, Shaw Pittman, and EY.

Early Ghost Automation

Years before anyone was talking about AI in the workplace, there was an accounting staffer at Buick who had figured it out.

He was an Excel wizard. Macros, mostly. He had quietly automated the bulk of his monthly job: the extractions, the reconciliations, the formatted reports that landed on the right desks at the right times. If you walked past his cubicle during the workday, you would sometimes see his screen running on its own. Columns populating. Files generating. The ghost in the machine working while the man himself was at a long lunch, out on the golf course, or wandering the floor catching up with colleagues.

His output was reliable. His deliverables arrived on time. Nobody looked too hard at the hours.

He had discovered, through his own ingenuity and entirely without organizational permission, what every productivity researcher is now confirming: that a significant share of knowledge work can be automated, that the worker who figures this out first captures the recovered time personally, and that the rational move is to tell no one.

That was thirty years ago. The tool was Excel. The skill required was rare enough to keep the pattern contained. But it proved the point that automation was a viable option.

Consultant Based Automation Came Next

A department had a staff person whose job was to pull sales data from several versions of Vertex, serious enterprise tax software, and reconcile it against a legacy accounting system. The work took three weeks every cycle. Three weeks of extraction, cleaning, mapping, and formatting before the actual sales tax preparation could begin. The person who did it was meticulous. The output was reliable. The role made sense on the org chart. Unfortunately, he gave his retirement notice. The tax department was scrambling to replace his work.

My analytics company replaced the process with an Azure SQL platform. The data pulled automatically. The reconciliation ran in hours. The audit files and filing source documents generated themselves, formatted exactly as the team had always received them. And my company received a monthly income stream.

The sales tax prep person didn’t notice anything had changed. The files were there. The formats were right. The audit trail was clean.

The staff person who had spent three weeks building those files every cycle retired just as his position was automated away.

Nobody replaced him.

There was no headcount reduction decision. A frightening cost-cutting mandate was not announced. And no meeting was held to verify that the role was now redundant. The role simply did not need to be posted again. The files kept appearing. The team kept working. One person stopped showing up and the system didn’t miss them.

That is the quiet version of displacement. No announcement. No severance. No WARN Act notice. Just an open requisition that never gets approved.

Now Agentic Automation Arrives at Scale

Microsoft just put agentic Copilot into full commercial release across Word, Excel, and PowerPoint. The headlines called it a productivity release. A feature announcement. A step forward in the AI-powered workplace.

It is none of those things. It is the Buick staffer’s discovery, handed to every knowledge worker in every organization running Microsoft 365. At once. With no technical barrier. And still with no organizational framework for what happens next.

That is a workforce event. Almost nobody is watching it that way.

The first compression already happened in that sales tax department: a three-week job reduced to hours, a role quietly not replaced. Now a second compression arrives on top of it. The agentic tools available today, the same category of capability that Copilot just pushed into full commercial release across every Office installation, can eliminate roughly 120 hours of monthly work from a role like the one that remains.

One hundred and twenty hours. Three full work weeks. Gone.

The question nobody is asking inside that organization is: what happens to those hours?

The honest answer, in most organizations, is nothing intentional. The hours get absorbed. The preparer gets a larger scope. They get told they are a valued member of the team. They get thanked for their adaptability. And they get handed more work to fill the recovered time, because the alternative, acknowledging that 120 hours of monthly effort just became unnecessary, would require a conversation nobody has been trained to have.

This is Not a Hypothetical Pattern. It is the Dominant Pattern.

Recent research confirms what practitioners already know. AI tends to intensify work rather than reduce it in knowledge-worker roles. Productivity rises. Hours do not fall. Organizations optimize the worker by giving them tools, then leave the work structure entirely untouched. Time saved on one task gets backfilled with more tasks, more monitoring, more informal requests, more of whatever the manager can find to justify the headcount.

The worker who figures this out early often makes a rational choice: hide the automation. Use the tool. Do the job in a fraction of the time. Don’t tell anyone. Fill the recovered hours with visible busywork so the performance looks normal.

This is not laziness. It is a rational response to an irrational incentive structure. The worker who discloses that they automated 10 hours of weekly work does not receive 10 hours of autonomy, or a raise reflecting the productivity contribution, or a formal redesign of their role toward higher-value work. They receive more tasks. So they stay quiet. They fill the day. And the organization captures none of the higher-order value those 10 hours could have produced.

Everyone loses except the output number.

Consistent Clerical and Supervisor Push Back

In my consulting work I watched a two-layer suppression system operate with remarkable consistency. When automation arrived and work came in cleaner, on time, with less remediation required, the passive resistance did not come from the clerks alone. It came from their supervisors too. Both roles felt the same threat from the same direction, and their interests aligned perfectly against disclosure.

Consider the supervisor managing ten clerks. That span of control is not incidental. It is the basis of their compensation grade, their title, their seat in the planning meeting, their claim on a dedicated office. Reduce the function to two or three people and you do not have a department anymore. You have a small team that does not require a manager at that level. The supervisor understands this arithmetic without anyone explaining it to them.

So the supervisor who receives cleaner output and faster turnaround does not celebrate. They find reasons to question the new process. They identify edge cases that require human review. They add back steps that restore complexity and justify the headcount. The passive resistance is not sabotage in any dramatic sense. It is a series of reasonable-sounding objections that collectively ensure the automation never reaches its full potential, because full potential means a smaller empire.

The worker hides the recovered time to avoid getting more tasks. The supervisor buries the productivity gain to protect organizational standing. Neither is acting irrationally. Both are responding correctly to the incentive structure they actually live inside.

This is why the AI dividend conversation almost never happens organically. Capturing that dividend requires someone to voluntarily report that their own function, or their team’s function, has been partially automated. No rational actor in that structure raises their hand first. The worker stays quiet. The supervisor adds friction. The gain flows to the output number and stops there, never reaching the conversation about what the organization could do with the recovered capacity.

Copilot does not change this dynamic. It amplifies it. The compressions are now faster, wider, and harder to attribute to any single project or decision. The passive resistance will scale to match.

The full commercial release is not a feature update in this context. It is a scale event.

The compression in that sales tax department required someone who knew how to architect an Azure SQL platform. It required a consultant, a project, a deployment, a decision to invest. The Buick staffer’s macros required years of self-taught Excel mastery most colleagues couldn’t replicate. Both barriers were real. They limited how many workers could do this and how fast the pattern spread.

Copilot removes the barrier entirely. The same category of capability, task completion rather than chat assistance, agentic execution across the documents and workflows knowledge workers actually use, is now available to every organization running Microsoft 365. No project required. No consultant. No Azure architecture. No macro expertise. Just the tool already sitting inside the software the team opened this morning.

The three-week job that became hours in that sales tax department? The remaining preparer’s 120 monthly hours? Those compressions now happen without anyone needing to design them. The worker opens Excel. The agent completes the task. The hours disappear.

The long lunch just went enterprise-wide.

And the organization has no framework for what happens next.

This is a Common Issue

The research on this is striking in its consistency. Companies introduce AI as an efficiency tool without answering the fundamental question behind it: what do we want humans to do more of now?

A few forward-looking organizations are starting to name this. They talk about an “AI dividend”: committing some fraction of recovered time to learning, experimentation, or higher-order work rather than allowing all of it to be absorbed into throughput. They are redesigning jobs explicitly around what the agent handles and what the human handles. This makes time reallocation a performance question rather than leaving it to improvisation.

They are the exception. By a wide margin.

The default is improvisation. Manager and worker both scrambling to fill the recovered time with something that looks like productivity, because neither has been given a framework for what productivity means when the artifact that used to take three weeks now takes three hours.

There is also the failure mode running in the opposite direction. Some organizations are discovering that agents generate low-quality output that humans then spend additional time fixing. The workday doesn’t shrink. It fills with remediation instead of creation. The net time savings disappear into correction cycles. This is showing up in enough organizations that it has acquired its own name: workslop. The agent produces. The human cleans up. The hours that were supposed to be freed become hours of quality control nobody budgeted for.

Either way, absorbed into more tasks or consumed by remediation, the worker does not receive the dividend.

Here is what is happening right now, today, inside organizations that just received Copilot full commercial access.

Someone is running a task that used to take them a full day. The agent completes it in forty minutes. They sit with the recovered hours and feel something between relief and anxiety, because there is no policy, no guidance, no manager conversation, no role redesign that tells them what those hours are for.

So they fill them. They answer emails. They attend the optional meeting they usually skip. They help a colleague with something that isn’t really their job. They look busy. They do not raise their hand and say: I just found six hours. Here is what I propose we do with them.

Because the last person who raised their hand got six more hours of work assigned.

The sales tax preparer who eliminated 120 monthly hours will not announce it. The knowledge worker who automated the weekly report will not document it. The analyst who cut the budget close from two days to two hours will tell no one.

The hours will disappear into the gap between what the job description says and what the job actually requires now. The gap will widen every month. And when those workers eventually leave, for any reason, at any time, their roles will not be backfilled, because nobody will be able to articulate exactly what those roles were still doing.

The quiet attrition channel. No announcement. No severance. No statistics that capture it cleanly.

Just open requisitions that never get approved.

The Same Old Lie Prevails

There is an official story about all of this. You have heard it in every vendor presentation, every HR all-hands, every leadership communication about the AI transformation journey. Automation will free workers from repetitive tasks. It will elevate them to higher-value work. It will make their jobs more creative, more strategic, more fulfilling. The technology handles the routine. The human handles everything that matters.

It is a compelling story. It is also, in most organizations, not what is happening.

The research is consistent and the practitioner experience confirms it. Workers freed from repetitive tasks are not being elevated to strategic work. They are being given more repetitive tasks, because the organization has no framework for anything else. Or they are being quietly not replaced when they leave. Or they are hiding the automation entirely and filling the recovered hours with visible busywork, because disclosure is a trap.

The promise and the reality are not the same thing. The gap between them is where the extracted value lives, and right now that value has only one default destination: the organization’s margin.

What are the Options

It does not have to work that way. There are exactly three places the recovered hours can go, and only one of them is currently receiving them by default.

The first destination is the firm. Recovered hours become higher output from the same headcount. Margin expands. The productivity gain is captured entirely by the organization, with no corresponding adjustment to compensation, workload quality, or job design. This is the default and what happens when nobody makes a different decision.

The second destination is the worker. Recovered hours become investment capacity. The employee uses reclaimed time for learning, for deeper client work, for the creative and strategic contributions that cannot be automated. This requires explicit role redesign, organizational permission, and a leadership culture willing to measure value shift rather than pure volume. It is the right answer and it is currently the exception.

The third destination is time itself. Recovered hours become a shorter work week. The productivity dividend gets returned directly to the people who generated it, not as more interesting work but as more life. A four-day week. An earlier close on Friday. A genuine reduction in the hours the job requires. This option is almost never discussed inside organizations deploying AI tools, even though it is the most direct way to share the gain with the workforce that produced it.

Most organizations will choose the first destination by doing nothing. The hours will flow to margin because no governance mechanism exists to route them anywhere else. No policy. No negotiation. No explicit commitment to the worker’s share of the dividend.

The worker who automated 120 hours of monthly work will not see those hours returned as creative capacity or reclaimed time. They will see a larger job description and a performance review that expects more.

The True Story

The full commercial release of Copilot is not the story. The story is the 120 hours and where they go.

Right now they are being captured silently and systematically, without a ledger entry that names what was taken or from whom, without a governance structure that forces the question of who should benefit, and without a public conversation about whether the default destination is the right one.

The Buick staffer figured this out thirty years ago and kept it to himself. He was rational. Every worker hiding automation today is equally rational. Every supervisor adding friction to protect headcount is equally rational.

The system is working exactly as designed. That is precisely the problem.

A future piece in this series will take up the governance question directly: what a fair accounting of the AI productivity dividend would look like, who has standing to claim a share, and why the current silence is not neutral. It is a choice, made by default, in favor of one party.

James Allen Ramsey writes J Allen Insights on the intersection of technology, geopolitics, and human development. He is a retired tax attorney and consultant whose career included GM, PwC, Shaw Pittman, EY and Bionic System Analytics.*

When a Commerce Department letter can pull a model deployed to hundreds of millions of people, the constraint on American AI is no longer technical. It is political.

Late Friday, the United States government directed Anthropic to immediately suspend all access to its two newest and most powerful AI models, Fable 5 and Mythos 5, for any foreign national anywhere in the world, including foreign nationals inside the United States, including Anthropic’s own employees. According to Anthropic’s public statement, the company received the export control directive from the Commerce Department at 5:21 PM Eastern. Commerce Secretary Howard Lutnick addressed the letter personally to CEO Dario Amodei.

The models had been publicly available for three days.

Anthropic had no technical mechanism to verify user nationality at scale. So it did the only thing it could do to ensure compliance. It shut both models off for everyone.

That is what happened on the surface. What happened underneath it is a more complicated story, and it matters for every institution, every enterprise customer, and every foreign-born researcher working in American AI today.

What They Said and What They Meant

The government’s stated rationale was a jailbreak. A third party had allegedly found a method to bypass Fable 5’s safety systems and access capabilities posing a national security threat. The Commerce Department cited this as grounds for an export control directive under national security authorities, the same legal infrastructure used to restrict semiconductor exports to adversarial nations.

Anthropic reviewed the demonstration and disputed the characterization directly in its blog post. The company said the vulnerability was narrow, non-universal, and limited to a specific cybersecurity use case. More pointedly, Anthropic stated that the identical capability could be elicited from OpenAI’s GPT-5.5, which faces no comparable export control. The company’s conclusion was unambiguous: if this standard were applied consistently across the industry, it would essentially halt all new model deployments for every frontier AI provider.

The government did not apply it consistently. Only Anthropic’s models were recalled.

That asymmetry is not a technical finding. It is a policy choice. The distinction between those two things is what the rest of this piece is about.

The Review That Did Not Happen

Ten days before the kill switch was pulled, President Trump signed an executive order titled “Promoting Advanced Artificial Intelligence Innovation and Security.” Signed June 2, the order established a voluntary framework under which AI developers would submit frontier models to federal review for up to 30 days before public release. The reviewing agencies named in the order include Treasury, the NSA, CISA, NIST, and the National Cyber Director. The stated purpose was to identify and remediate cybersecurity vulnerabilities before commercial deployment.

Fable 5 and Mythos 5 launched publicly on June 9. The executive order framework had been in place for seven days.

There was no 30-day review. There was no classified benchmarking assessment. There was no structured remediation process. There was a Thursday night phone call, a Friday afternoon letter with no specific technical disclosure, and a global model recall by Friday evening. The entire sequence from alleged jailbreak report to Commerce Department directive consumed less than 24 hours.

The executive order framework the administration had just created, the one designed to ensure orderly national security review of exactly this kind of frontier capability, was bypassed entirely. What replaced it was a reactive process triggered by a single company’s report, routed through political channels rather than technical ones, producing an outcome that the administration’s own newly appointed review apparatus had no meaningful role in reaching.

This matters for a reason beyond process. The executive order’s 30-day review exists precisely to prevent the kind of rushed, politically charged determination that produced the Friday night letter. The administration signed that order, then ignored it ten days later. What that tells every other frontier AI developer is that the formal framework is decorative. The actual decision-making runs through different channels entirely.

The Amazon Question

Multiple reports, including a detailed account from Axios, have identified Amazon as the third party that brought the jailbreak report to the White House. According to that reporting, Amazon called administration officials Thursday night. Not Anthropic. Not the company’s security team. Not CISA. The White House. The administrative scramble that followed produced the Friday night letter.

This warrants examination, because Amazon is not a disinterested party. Amazon is one of Anthropic’s largest investors, having committed billions to the company across multiple funding rounds. Amazon Web Services is Anthropic’s primary cloud partner. The commercial relationship is deep and the financial exposure is substantial.

And yet when Amazon believed it had found a security vulnerability in Anthropic’s flagship model, its first call was to the administration rather than to the company whose product it was examining and whose success it had financed. A security expert whom Anthropic subsequently shared the Amazon report with told Axios the government’s response appeared “way out of line with what’s actually in the research report.”

Attributing motive is always hazardous. But it is fair to note that this choice is consistent with several structural incentives operating simultaneously. Jeff Bezos attended Donald Trump’s inauguration. The Washington Post’s editorial posture shifted after his direct intervention earlier in the administration’s term. Blue Origin has deepened its government contracting relationships across this period. The pattern describes a company that has made a deliberate choice to operate inside the administration’s political perimeter.

That positioning carries competitive implications in the AI market. Amazon competes with Anthropic in enterprise AI. Amazon benefits if Anthropic’s IPO is complicated. Amazon benefits if the administration views Amazon as a cooperative intelligence partner. Whether any of those incentives shaped Thursday night’s phone call is not something the public record can establish. What the public record does establish is that the call went to the White House rather than to Anthropic, and that the result was a global model recall bypassing the formal review process the administration had established ten days earlier.

The inference is available. The public record supports it.

The Litigation That Was Already There

This did not arrive in a vacuum. Anthropic and the Trump administration have been in open conflict since early this year, and the dispute is already in federal court.

In March, Defense Secretary Pete Hegseth labeled Anthropic a supply chain risk to national security and prohibited the use of its Claude models across Pentagon procurement chains. According to reporting at the time, the designation appeared to be the first time such a label, historically applied to foreign companies including Huawei, had been used against a domestic American firm.

The trigger was Dario Amodei’s refusal to allow the military unrestricted use of Claude for fully autonomous lethal weapons systems and mass domestic surveillance of American citizens. Amodei sought written guarantees that those uses would be excluded. The Pentagon declined. The supply chain designation followed.

Anthropic filed suit, arguing in federal court that the Constitution does not permit the government to use its regulatory power to punish a company for protected speech, and that no federal statute authorized the actions taken. Amodei said publicly the company saw no choice but to litigate.

That case is active.

The Fable 5 and Mythos 5 export control directive now lands directly on top of that litigation. Anthropic is already in court arguing that the administration has been using regulatory instruments to retaliate for the company’s refusal of military deployment terms. The sequence of events now available to its lawyers includes a Thursday night call from a major investor to the White House, a Friday afternoon letter with no specific technical disclosure, a global model recall, and a formal review framework that the administration itself established and then bypassed entirely. Whether that sequence survives judicial deference to executive national security determinations is a different question. But it is a sequence Anthropic’s legal team will put in front of a court.

The Standard That Was Not Applied Uniformly

The administration’s selective application of the jailbreak standard is the clearest signal that this is not primarily a security decision.

GPT-5.5 is not under export controls. Gemini is not under export controls. Both models operate at frontier capability levels. OpenAI signed a deal to deploy its models on classified Department of War networks. Google has maintained cooperative relationships with defense and intelligence procurement. Neither company refused military deployment requests on ethical grounds.

Anthropic did.

The companies inside the administration’s political perimeter are operating without recall risk this week. The company outside that perimeter had its flagship models pulled three days after launch. That is not a national security policy. It is an industrial policy enforced through national security law, applied selectively based on political alignment rather than technical criteria.

Every other frontier AI company is reading that lesson right now, and drawing conclusions about what the perimeter costs and what staying outside it costs.

Three Firms. Nearly One Million Professionals. No Fallback.

The collateral damage from Friday’s directive extends well beyond Anthropic. Three of the four largest professional services firms in the world have built their AI delivery infrastructure on Claude, and they did so in the last sixty days.

Deloitte rolled Claude out to its entire global workforce of more than 470,000 people, making it Anthropic’s largest enterprise deployment at the time of the agreement. The rollout targets regulated industries including financial services, healthcare, and the public sector. KPMG followed in May with a global strategic alliance deploying Claude to all 276,000 of its employees, embedding it directly into Digital Gateway, the client delivery platform hosted on Microsoft Azure, with full implementation targeted for September 2026. The rollout begins with Tax and Legal and expands into advisory services. PwC announced its own expanded Claude alliance on May 14, committing to train and certify 30,000 US professionals on Claude and positioning itself as the first professional services firm to build an at-scale finance function around a single foundation model.

EY is the only holdout among the Big Four, having announced a separate Microsoft 365 Copilot initiative with a commitment exceeding one billion dollars covering more than 400,000 staff.

The combined Anthropic exposure across Deloitte, KPMG, and PwC represents nearly one million professionals whose client-delivery workflows are now dependent on a vendor whose flagship models were pulled from global access by executive letter with no advance notice and no transition period. The KPMG deployment has not yet reached full implementation. It is scheduled for September. The directive arrived while the rollout is in progress.

Consider what that means in practice. KPMG’s Tax and Legal practice, already mid-deployment on Claude, is now operating on a platform whose most advanced models have been recalled. PwC’s Office of the CFO, built explicitly around Claude as its foundation model, has the ground shifted beneath it. Deloitte’s 470,000 professionals in financial services and healthcare are working with a vendor who received a government directive at 5:21 PM on a Friday with no prior notice.

These firms did not build on Anthropic recklessly. They built on the market leader after extensive due diligence, with full awareness of the enterprise risk frameworks their regulated industry clients require. What no due diligence process could have identified is that the vendor’s regulatory relationship with a single government agency would become a binary availability risk activatable by executive letter. That risk category did not exist in professional services AI procurement a week ago. It exists now.

The downstream exposure extends further still. Every mid-market firm, every PE portfolio company, every regulated enterprise that hired KPMG or Deloitte or PwC in 2026 was also hiring Claude, whether that was in their procurement specification or not. Their audit workpapers, their tax engagements, their regulatory filings are running through workflows that as of Friday night lost access to the models those workflows were built around.

The Engineers Are Reading It Too

The deemed export control that locked out Anthropic’s own foreign-national employees carries a consequence that extends well beyond Anthropic.

Every non-citizen researcher at every American AI lab, at OpenAI, Google DeepMind, Meta AI, and the next generation of labs, read Friday night’s announcement and understood that their passport is now a variable in their employer’s regulatory risk profile. That variable can be activated by executive letter. No notice. No process. No appeal on a useful timeline.

Frontier AI research is among the most globally concentrated skilled workforces in any sector. The top researchers hold passports from India, China, South Korea, Canada, the UK, France, and Israel. The H-1B relationship that brought them to the United States created a conditional attachment to American jurisdiction. Friday’s directive made that condition significantly less attractive.

Canada already has the infrastructure to absorb displaced talent. The Vector Institute in Toronto, the Mila lab in Montreal, and the Waterloo AI cluster have spent a decade building exactly the research environment that competes for this cohort. Canadian Prime Minister Mark Carney said Sunday morning that the Anthropic restrictions demonstrate the danger of over-reliance on American providers. That was a policy statement. It was also a recruitment signal.

The administration appears to be treating AI capability as residing in the artifact, in the model weights subject to export control. It does not. The capability resides in the researchers who know how to build the next version. If that cohort relocates to Toronto or London or Zurich, the export control on Fable 5 is irrelevant within eighteen months. The next model gets built outside the jurisdiction, beyond the reach of any Commerce Department letter.

What This Means for Institutions

The immediate lesson for every enterprise buyer is vendor concentration risk at the regulatory layer. Not the technical layer. Not the product layer. The political layer. A model available today can be pulled by executive letter before your integration is complete, before your September go-live, before your workforce certification program finishes. Multi-vendor architecture is no longer optional discipline. It is the minimum viable posture, and three of the four largest professional services firms in the world learned that this week while their deployments were in progress.

For institutional capital evaluating Anthropic’s IPO, the question is no longer whether the models are technically competitive. The question is whether a company in active federal litigation with the administration, carrying a Pentagon blacklist, a flagship model recall, and the demonstrated willingness of a major investor to route security findings to the White House rather than to the company, can sustain frontier development and global market access at a near-trillion dollar valuation. That is a sovereign risk analysis. It is not a standard technology sector growth story.

For every other frontier AI company, the question is more uncomfortable. The political perimeter that protects OpenAI and Google today was built by accepting military deployment terms that Anthropic refused on ethical grounds. Every lab must now decide whether that price is acceptable, knowing what the alternative looks like.

The constraint is no longer technical. It is political alignment under conditions that can change without notice, enforced by instruments with no statutory standard, no transparent process, and no appeal timeline that fits the pace of commercial AI deployment.

The administration signed an executive order ten days ago to prevent exactly this kind of rushed determination. Then it bypassed that order to issue exactly this kind of rushed determination. That sequence is the clearest signal available about where the actual decision authority resides, and what framework, if any, governs its exercise.

The kill switch exists. Friday was its first use.

James Allen Ramsey writes J Allen Insights on the intersection of technology, geopolitics, and human development. He is a retired tax attorney and consultant whose career included GM, PwC, Shaw Pittman, and EY

Sometime in August 2025, Ukrainian officials pitched the White House on a drone partnership. The pitch came right after President Trump privately praised Operation Spiderweb, the operation in which Ukrainian pilots had guided explosive drones, smuggled into Russia aboard ordinary trucks, to destroy dozens of Russian warplanes parked on tarmacs hundreds of miles from any front line. It was the kind of operation that rewrites what military planners think is possible, and for a moment, Washington noticed.

What followed was not a deal. It was a year of talks. By March 2026, Zelensky was telling reporters Ukraine had proposed a five-year, fifty-billion-dollar framework: ten million drone units a year, the workload split evenly between Ukrainian and American engineers. Still nothing. By May, the State Department had drafted a memorandum with Ukraine’s ambassador in Washington. Still nothing. By the end of May, Zelensky posted on social media that the US had agreed to test Ukrainian systems across air, land, and sea, but “we still don’t have a bilateral Drone Deal, a big framework document.” As of early June, senior Pentagon officials, including the Secretary of Defense himself, were on record praising Ukraine’s drone capabilities. The agreement remained unsigned.

A professor of strategic studies put it plainly to The Hill: the US is putting its own troops in danger by not working as closely as possible with the Ukrainians on drone development. The delay, he said, is potentially kneecapping the United States in an area where it is already trying to catch up.

That phrase, *trying to catch up*, is worth sitting with, because it is doing a lot of work. The United States is not short on capability here. It has the most advanced AI sector in the world, a defense budget larger than the next several countries combined, and, as of mid-2025, an explicit Pentagon mandate to fix exactly this gap. A July 2025 memo from the Secretary of Defense to senior Pentagon leadership did not mince words: drones are the biggest battlefield innovation in a generation, they account for most of this year’s casualties in Ukraine, adversaries produce millions of them annually, and “U.S. units are not outfitted with the lethal small drones the modern battlefield requires.” An executive order followed. A Drone Dominance Program followed, with a billion-dollar budget signal and a goal of fielding hundreds of thousands of low-cost weaponized drones by 2027. Congress, in the FY2026 defense authorization, rewrote acquisition rules specifically to let non-traditional manufacturers in faster.

So this is not a story about a country that lacks the tools. It is a story about a country that had, sitting on the table, an offer to shortcut years of that buildout, and chose instead to build it from scratch. Anduril’s flagship drone factory in Ohio, the centerpiece of America’s homegrown answer to this problem, is not expected to be complete until 2035. Ukraine, meanwhile, produced an estimated four million unmanned systems in 2025 alone, and is on pace to exceed that in 2026. One Ukrainian manufacturer alone plans to build more than three million first-person-view drones this year. The United States built roughly three hundred thousand in 2025.

The cost of that choice is not abstract. The same two former Pentagon officials who built the Defense Innovation Unit’s drone framework warned, in plain terms, that a drone attack on US soil is coming in 2026, and that no American military installation could currently repel an assault like the one Ukraine carried out against Russian airfields. The methodology that made Spiderweb possible, ordinary trucks, ordinary roads, a payload that looks like cargo until it isn’t, does not require Russian geography to work. A container ship makes the same point at a different scale, and at American ports, where inspection rates cover a fraction of total volume, the math is not reassuring.

While Washington deliberated, everyone else moved.

Europe Builds the Institution

The trigger, for Europe, was humiliating and specific. In September 2025, Russian drones entered Polish airspace. NATO’s response was to scramble multimillion-dollar fighter jets against drones that cost a few thousand dollars each. Some of the drones crashed into the Polish countryside anyway. The asymmetry was the whole lesson.

By February 2026, five of Europe’s largest military powers, France, Poland, Germany, the UK, and Italy, had signed onto a program called LEAP: Low-Cost Effectors and Autonomous Platforms. The commitment was explicit: joint development of drone-based strike capabilities, low-cost joint production, and joint procurement of AI-enabled drone payloads, all built on Ukrainian battlefield expertise. The EU’s foreign policy chief framed it as a hedge against what she called a “redefined” relationship with the United States, and as a response to a security environment more uncertain than at any point in decades.

By May, the European Commission had gone further, launching the EU-Ukraine Drone Alliance: an industry-led structure connecting manufacturers, startups, and end-users across the EU, the EEA, and Ukraine, backed by a thirty-five-million-euro combat-testing platform and a separate one-and-a-half-billion-euro EU-Ukraine defense-industrial work program. The first tranche of the EU’s ninety-billion-euro loan to Ukraine is earmarked for buying Ukrainian-made drones. By 2027, the roadmap calls for joint ventures, shared technology standards, and mutual recognition of qualifications between European and Ukrainian defense firms. Six new joint ventures with Germany alone were announced in a single month. Norway signed on to mass-produce Ukraine’s mid-range strike drones.

None of this required Washington’s involvement, and none of it waited for it.

The Gulf Moves Fastest

If Europe institutionalized, the Gulf operationalized, and it did so faster than anyone.

Through the spring of 2026, Iranian-designed Shahed drones, the same family Russia has used against Ukrainian cities since 2022, were striking targets across the Gulf in large numbers. The UAE alone absorbed nearly half of all recorded incidents, facing something like 1,825 drones against just under 400 missiles, by far the most drone-saturated environment of the conflict. Saudi Arabia and Bahrain faced similar patterns.

Zelensky toured the region in late March, visiting Saudi Arabia, the UAE, and Qatar. Within days, Ukraine had signed ten-year defense cooperation frameworks with all three, covering drone warfare, electronic warfare, and air-defense interception, the first time a country actively fighting a war had simultaneously begun exporting the technology it developed under fire. By the time the ink was dry, more than two hundred Ukrainian military specialists were already on the ground, embedded inside Saudi air-defense networks, assessing vulnerabilities and training local forces against the very drones now hitting Gulf infrastructure. Britain announced it would send its own Ukrainian-specialist teams to the region as well.

The economic logic Zelensky offered was the same one Poland had just learned the hard way: it no longer makes sense to burn through multi-million-dollar interceptor missiles against ten-thousand-dollar problems. Ukraine’s newer interceptors run closer to ten thousand dollars each, against Shahed drones that can cost up to a hundred and thirty thousand. For Gulf states that have spent decades importing exquisite, expensive air defense from the US, UK, and France, and that are separately trying to build domestic knowledge-economy sectors as part of broader national diversification plans, this was not just a procurement decision. It was a template.

By May, Zelensky had added Bahrain to the list, and Ukrainian officials said roughly twenty more agreements across the Middle East and partner countries were in the pipeline.

Japan Removes the Wall

Japan’s shift required no new urgency from Ukraine at all, just a legal door swinging open at the right moment.

On April 21, 2026, the cabinet of Prime Minister Sanae Takaichi formally revised the Three Principles on Transfer of Defence Equipment and Technology, a framework that had banned Japanese lethal arms exports since 1976. The change came amid a broader rearmament push: Japan’s 2026 defense budget hit a record fifty-eight billion dollars, reaching two percent of GDP two years ahead of schedule, driven by what Takaichi described as the most severe security environment Japan has faced since World War Two, citing China’s military expansion and closer Russia-China-North Korea coordination.

Ukraine’s ambassador to Tokyo didn’t wait to see how the dust settled. Kyiv was ready, he said, to share four years of battle-tested drone and counter-drone expertise, “especially drone and counter-drone systems,” right as the export ban came down. Zelensky separately offered Japan something more specific: marine drone technology. Ukraine, without a meaningful navy of its own, has used naval drones to push the Russian Black Sea Fleet away from its own coastline. Japan, he noted, already holds licenses for missile and air-defense production. The fit was obvious.

On the ground, it moved immediately. Japanese drone maker Terra Drone, which had previously routed its Ukraine work through a Dutch subsidiary to get around the export restrictions, announced direct partnerships with Ukrainian firms WinnyLab and Amazing Drones on interceptor systems designed to take down Shahed-type drones. Russia summoned Japan’s ambassador in Moscow to protest. Tokyo did not seem to feel the pressure.

The Pattern Around the Edges

Step back further and the same shape repeats. An Australian delegation, including unmanned-systems specialists, visited Ukraine in August 2025 specifically to study its drone industry, its deployment doctrine, and its training pipelines. South Korea, separately, signed a letter of intent with the United States in May 2026 on joint drone and counter-drone supply chains, proof that Washington can move quickly on this kind of cooperation when it wants to, just apparently not with Ukraine. Even without a formal government-to-government channel, Korean defense innovators are already absorbing Ukrainian lessons informally, and Seoul has imported Polish-made loitering munitions that Ukrainian forces use at a fraction of the cost of conventional guided missiles. Europe and Korea are now co-developing autonomy and command-and-control technology directly, a connection that runs laterally between two blocs and bypasses Washington entirely.

What the Choice Costs

Add it up. In the space of about six months, Europe stood up a funded multilateral institution. The Gulf signed decade-long frameworks and put boots on the ground inside weeks. Japan dismantled a fifty-year-old legal wall and watched its companies move within days. Smaller threads, Australia, Korea, Norway, Lithuania, all point the same direction.

The United States, fighting alongside several of these same partners in the same theater, against the same category of threat, spent the same six months with a memo.

This isn’t a story about American weakness. The capability is there, the policy apparatus is there, the money is there. What’s missing is the thing that costs nothing to give and everything to withhold: time. Ukraine’s knowledge is not a secret formula that will retain its value indefinitely while Washington builds its own version on its own schedule. It is a perishable asset, forged under conditions nobody else can replicate, and every month it sits un-shared is a month in which the gap between “what the US could field” and “what the US has fielded” stays exactly as wide as it was the day the first pitch landed in August 2025.

Everyone else read that clock correctly. Whether Washington has is, as of this writing, still an open question, one that troops stationed across a drone-saturated Middle East, and ports across a drone-naive homeland, may not have the luxury of waiting to find out.

James Allen Ramsey writes J Allen Insights on the intersection of technology, geopolitics, and human development. He is a retired tax attorney and consultant whose career included GM, PwC, Shaw Pittman, and EY

Roughly 100,000 people once punched in every day at the Ford River Rouge Complex in Dearborn, Michigan. The Rouge was not a factory in any ordinary sense. It was an industrial civilization compressed onto 2,000 acres. Iron ore arrived by Great Lakes freighter at one end. Finished automobiles rolled out the other. In between: blast furnaces, open-hearth steel mills, coke ovens, rolling mills, a glass plant, a rubber plant, a paper mill, foundries, stamping facilities, machine shops, and its own power plant generating electricity at a scale comparable to a municipal utility. At its peak in the 1940s, the Rouge produced up to 4,000 vehicles a day and employed a workforce larger than the population of most American cities.

I walked through the same complex recently. Not the Rouge at peak. The Dearborn Truck Plant that now occupies roughly 600 acres of what the Rouge once covered. A truck rolls off that line every 53 seconds. The F-150 is the best-selling vehicle in America. The plant is a marvel of managed complexity, running hybrid and internal combustion variants, multiple trim levels, and configurations that would have been unimaginable in Henry Ford’s era.

Roughly 5,000 to 6,000 people work there.

That number requires explanation. Not an excuse. An honest accounting of what actually happened between then and now, and why the political stories told about it, from both parties, have been wrong in ways that matter enormously for what is coming next.

What the Rouge Was

Henry Ford’s vision for the Rouge was total self-sufficiency. He called it an “ore to assembly” complex. His Ford Motor Company once owned 700,000 acres of forest, iron mines, and limestone quarries in northern Michigan, Minnesota, and Wisconsin. The Rouge was the terminus of that supply chain. Every input arrived by Ford’s own railroad, Ford’s own fleet, Ford’s own docks.

The vertical integration was not just an engineering achievement. It was an employment model. The Rouge needed ironworkers, steelworkers, glassworkers, rubber workers, electricians, machinists, carpenters, painters, janitors, and assembly workers. At its peak, more than 5,000 workers were employed just to keep the grounds and buildings clean, painted, and maintained. If the Rouge had been a city, it would have ranked fourth largest in Michigan.

It also became, almost accidentally, the foundation of the American middle class in metropolitan Detroit. UAW wages at the Rouge were not charity. They were the price of labor in a complex that could not function without tens of thousands of skilled and semi-skilled workers performing tasks that, in the 1940s, no machine could replicate.

That is the world that is now gone. But it did not disappear the way the political narrative describes.

The First Subtraction: Vertical Integration Abandoned

By the 1960s, Henry Ford’s grandson had already concluded that vertical integration was more liability than asset. The Rouge’s sprawling internal supply chain was expensive to maintain, inflexible to change, and increasingly uncompetitive compared with specialized outside suppliers operating at scale across multiple customers.

The dispersal happened gradually and then all at once. The tire plant went. The foundries. The power generation. By the 1980s, Ford’s steel operations were losing money as foreign competition and recession compressed margins. In 1989, Ford sold the steel mill, giving up ownership of a large portion of the Rouge River frontage, the boat docks, and 45 percent of the original 2,000 acres. The glass plant closed in 1998.

What Ford retained was the assembly function. The one operation where the product still required human hands at sufficient density to justify keeping it in-house.

The steel jobs did not transfer intact to the new Rouge Steel Company. They entered a separate corporate structure that faced its own automation cycle, its own competitive pressures, and eventually its own bankruptcy. Around 2004, Russian steelmaker Severstal purchased the bankrupt Rouge Steel. The jobs dispersed, compressed, and in many cases disappeared. Just more slowly, and out of frame.

The vertical integration that anchored roughly 100,000 jobs was not dismantled by a political decision. It was dismantled by a capital allocation logic that determined each non-assembly function could be performed more cheaply by someone else, somewhere else, at lower overhead. Every individual decision was defensible. The aggregate was the slow demolition of an employment model.

The Second Subtraction: Two Models of Offshoring

I spent seventeen years in GM as a tax professional and clearly saw the results of the offshoring decisions play out financially.

What most people do not understand is that China and Mexico represented entirely different strategies. They are not the same story.

When GM and Ford moved production to China, the Chinese government extracted automation capability as the price of market access. Beginning with the 1994 Automobile Industry Policy, China codified a requirement that foreign automakers form joint ventures with Chinese partners, capped foreign ownership at 50 percent, and mandated technology transfer as a condition of doing business. The policy was explicit: trading market access for knowledge.

China did not primarily want the jobs. It wanted the process knowledge embedded in the equipment, the quality systems, the production logic, and the engineering expertise. China absorbed employment in those early decades, but capability accumulation was the strategic objective, not employment maximization. Research has since documented that affiliated Chinese domestic firms systematically improved quality in the specific dimensions where their joint venture partners were strongest. The mechanism was worker flows and shared supplier relationships. The student was in class, taking notes, for thirty years.

Mexico was a different transaction entirely.

I remember watching ten workers manually marry a transmission to a vehicle body at a GM facility in Mexico. That same operation was automated in US plants. The automation stayed in Michigan. Mexico got the bodies and the repetitive task.

This was not an oversight. It was the structure of the deal. Mexico had no mandatory technology transfer requirement. The Mexican maquiladora program was built on one premise: labor cost arbitrage. US companies imported components duty-free, assembled them with Mexican labor at a fraction of US wages, and exported finished goods back across the border. Mexico’s government accepted the jobs and the foreign exchange. It did not demand the knowledge. Mexico did develop some supplier integration over time, particularly in Tier 1 automotive components. But it did so without the state-directed capability extraction that China pursued as explicit national policy.

The consequence, playing out now, is that the two partners exit the offshoring era in entirely different positions. China spent thirty years extracting process knowledge, building automation capability, and is now iterating humanoid robots in its own gigafactories with data accumulated during the joint venture era. Mexico built a manufacturing employment base without the technological leverage that would allow it to move up the value chain. When the labor arbitrage closes, through automation or reshoring incentives, Mexico faces the loss of jobs without a capability platform to replace them.

Detroit gave China the automation alongside the jobs. Detroit gave Mexico the jobs and kept the automation. China used what it received to build a competitor. Mexico was not given the tools to do the same. Washington now calls it a trade policy problem.

The Third Subtraction: Spinning Out What Could Not Yet Be Automated

When GM could not automate a job and could not offshore it, it found a third instrument.

In May 1999, GM completed the spinoff of Delphi, its components division, into an independent company with 200,000 employees. Ford followed a year later with the Visteon spinoff, carrying 77,000 workers. In two years, the two largest American automakers moved roughly 277,000 workers off their books.

The UAW had always opposed Delphi’s separation from GM. Union officials understood what the structure implied. An independent Delphi was required by its spinoff agreement to pay GM-level wages, approximately $27 an hour, to workers performing tasks that competing suppliers were paying $12 to $14 an hour to accomplish. The company was structurally constrained to compete at a cost disadvantage from the day it opened as an independent entity. The only question was how long it would take to reach the point of rupture.

The answer was six years. Delphi filed for bankruptcy in October 2005, one of the largest bankruptcy filings in US history. In the restructuring that followed, the company sought to cut 24,000 hourly jobs and reduce wages by 60 percent. GM, rather than rescuing its former division, issued a statement welcoming the opportunity to reduce the premium it paid for Delphi parts. The bankruptcy court did what the collective bargaining agreement had prevented: it voided the union contracts, restructured the wages, and eliminated the jobs that could not survive at market rates.

Visteon followed the same arc, emerging from its own bankruptcy in 2009 at a fraction of its former size.

The component work that remained labor-intensive and could not yet be automated was not eliminated directly. It was placed in companies structurally constrained to fail under Big Three wage structures, then restructured through bankruptcy to wages competitive with offshore labor. Policy shaped the pace and the distribution of pain. But it did not originate the direction.

The Rouge’s workers did not all lose their jobs to robots. Some lost them to a federal judge voiding a union contract in a Manhattan courtroom in 2005.

What I Saw on the Floor

The Dearborn Truck Plant today is, by any objective measure, extraordinary. I watched a truck box marry to its platform without a human hand involved. That is a high-mass, high-precision, repeatable operation that was one of the last to be automated because of the tooling investment and the consequence of error. It now runs without intervention.

What remained visible on the floor was light manual work. Dexterous tasks requiring positional awareness. Two-handed operations with judgment components. Fit and finish adjustments. The kind of work that fixed automation, with its dedicated tooling and rigid repeatability, could never economically capture across the full range of vehicle configurations an F-150 requires.

The plant manager framed the new technologies carefully during a recent tour. “A lot of people think it’s a reduction or replacement,” he said. “It’s not. It’s more of how to use technology to enhance the job or make it easier.”

That is the same sentence, in different words, that has accompanied every automation wave since the 1950s.

The work I watched on that floor has a name in the robotics industry. It is called the last-constraint problem. Seven of eight operations automated, one remaining because the engineering cost of the final solution exceeded the tolerance of the program cycle. You assign a person with a power tool to the eighth station. They read a book between cycles. They are not a worker in any full sense. They are a placeholder until someone solves the last constraint.

That is precisely what is now being solved, offshore, at scale, in preparation for arrival.

The Fourth Wave: What Is Being Built in Shanghai

This is where the prior three subtractions converge into a single completion.

Tesla has reported deploying humanoid robots in its Shanghai Gigafactory, with early pilots handling parts, installing seats, and conducting quality inspections at claimed efficiency levels approaching human performance. These figures are industry-reported rather than independently verified, and the systems remain in early-stage development. But the strategic logic does not depend on the precision of the current numbers. It depends on the trajectory.

BYD has confirmed it is building humanoid robots, deploying them first in its own EV and battery factories before an eventual consumer rollout through its dealer network. Their executive vice president stated that automotive AI and robotics share common technological foundations. China’s UBTECH is targeting 5,000 humanoid units annually in 2026 and 10,000 by 2027, transitioning from selling standalone units to delivering turnkey operational systems covering deployment, training, and factory-floor scenario setup.

The strategic frame is the same across all of them: use high-density domestic production lines as the proving ground. Accumulate real production data in an environment with no institutional friction, no legacy wage structure, and no blue light politics (see section below for Blue Light explanation). Then scale.

The offshore iteration loop is not randomized experimentation. It is the fourth wave of deskilling applied to the physical layer, using the same logic that moved component manufacturing to Delphi, process knowledge to China, and assembly jobs to wherever the labor cost penciled out. The last-constraint problems visible on the Dearborn floor today are the training curriculum for systems being built in Shenzhen tonight.

When those systems arrive in US plants, they will not arrive as experiments requiring human cooperation. They will arrive as validated systems with production records, needing no introduction, carrying no blue light, and offering no visible accusation.

The Blue Light

I want to tell a story from my years in this industry that explains something important about why US automation deployments have repeatedly stalled, and why the iteration happening in Shanghai will not stall the same way.

When I owned part of a robotics firm, we installed a robot in a Ford plant. The implementation team put a blue light on the machine. The light was meant to signal when the robot was being held up waiting for a human operator to feed the next part. It was a diagnostic tool. A productivity indicator.

What it became was a public humiliation device. Every time the blue light activated, it announced to everyone on the floor that a specific human being was the reason the line was not moving. The person at that station was not slower than before. They were doing exactly what they had always done. But now their performance was visible, quantified, and broadcast in real time to their peers and supervisors.

The workers did not hate the robot because it was going to take their job. They hated it because it shamed them. The sabotage that followed was not irrational. It was a rational response to an intolerable social architecture. Nobody had costed the human environment into the deployment plan.

The humanoid systems training in Shanghai carry no blue lights. They accumulate failure data silently. They improve in the next training cycle. They do not require the human worker to be reframed as the bottleneck. And critically, they are not being deployed into the social architecture of a UAW plant where that reframing carries thirty years of institutional memory. They are being deployed into environments purpose-built for their iteration, with no legacy of the blue light and no one who remembers what it meant.

That asymmetry, not the hardware specifications, is why the deployment timelines now look real.

The Political Misdirection

US manufacturing employment peaked in June 1979 at 19.6 million workers. By 2019, that number had fallen to 12.8 million. Six million jobs gone over forty years, across ten presidencies, under six Democrats and four Republicans.

The political argument assigns blame by administration. Republicans blame Democrats for NAFTA. Democrats blame Republicans for the China trade normalization of 2000. Each party points to a specific agreement, a specific signature, a specific betrayal.

The data does not support the framing.

Roughly half of the decline in manufacturing’s share of employment occurred during a period when the US trade balance was essentially flat, before the deficits that define the political narrative. The downward trend in manufacturing employment began in 1980, preceding every agreement used to explain it. The decline predates NAFTA, predates China’s WTO accession, predates every villain either party has nominated.

What the data shows is a continuous, directional substitution of capital for labor operating beneath the surface of politics across the entire postwar era. Automation drove the displacement from roughly 1950 through 1990. Offshoring and trade became the dominant mechanism from 2000 through 2015, when investment in domestic automation actually slowed while trade deficits expanded. The capital logic operated with the grain of consumer behavior as well: Americans buying cheaper goods reinforced the same direction that capital allocation was already moving. Now automation is returning as the primary instrument, completing the work the prior waves left unfinished.

Each mechanism served the same underlying logic. Each wave was described in the language of the problem it solved: quality, safety, efficiency, competitiveness, shareholder value. None was described honestly as a workforce reduction policy. None required a vote. The politicians arrived after each wave, named a villain, and asked for your support. The capital had already moved.

General Motors now employs roughly a third of the 600,000 workers it had in the 1970s. It produces more vehicles. The steel industry shed approximately 265,000 jobs over roughly two decades while domestic production increased substantially. More output. Far fewer workers. That is not a trade story. It is not a Democratic or Republican story. It is the capital allocation story told across a century, in its cleanest form.

The Reshoring Illusion

The political promise of reshoring runs like this: China took your jobs, we will bring the factories back, the jobs return with them. Every ribbon cutting at a new semiconductor fab or battery plant carries that implicit promise.

The Dearborn Truck Plant quietly refutes it.

A reshored assembly plant in 2026 arrives with the automation frontier already embedded. The operation that once required ten workers in a Mexican facility runs without hands on day one. The seventh screw is already automated before the ribbon is cut. The eighth screw position that once required a placeholder human now has a learning system training in Shanghai to fill it before the US facility opens.

Lear Corporation’s CEO recently told investors that the company’s new Rochester, Michigan facility demonstrates “how a plant can run without human intervention,” calling the elimination of labor “key to winning onshoring.” He was direct: “Reduced labor overhead can offset the cost of moving production from Mexico.”

The factory is coming home. The workforce is not.

The worker who is told the factory is returning is being told a true thing about geography and a false thing about employment. The building returns. The relationship between that building and the community surrounding it does not. That relationship was already gone before the first job left for China. It left quietly, one automation decision at a time, across the careers of people who watched it happen from the inside and understood what each decision meant.

What Comes Next

The light manual work I watched on the Dearborn Truck Plant floor is the final frontier of the displacement wave that began when the Rouge’s first steel furnace replaced a gang of ironworkers in the 1920s.

It is not being approached from one direction. It is being surrounded.

Offshore iteration loops are solving the last-constraint problems that US plants still assign to humans. Humanoid systems purpose-built for factory floor morphology, designed to fit into buildings built for human ergonomics without a facility rebuild, are accumulating production hours in environments with no institutional friction. The data being generated in those facilities today will arrive in Dearborn, in Louisville, in Warren, as a validated system with a performance record. Not as an experiment requiring cooperation.

This is not an argument against technology. Automation creates wealth. The Rouge created wealth too. The question the last century has not answered honestly is who captures that wealth, and through what mechanism.

The Rouge employed roughly 100,000 people in a direct relationship between their labor and their income. Each subsequent wave of displacement broke one link in that chain. The components went to Delphi. Delphi went bankrupt. The steel went to Rouge Steel. Rouge Steel went to a Russian conglomerate. The assembly work went to a plant with 5,000 workers producing what once required 100,000.

The productivity gains were real. The distribution of those gains was a choice that was never put to a vote.

That is what is being completed in Shanghai. Not a technological revolution. The final chapter of a century-long capital allocation story that has arrived, one locally rational decision at a time, at the last task the machine could not yet do.

It can do it now.

James Allen Ramsey writes J Allen Insights on Substack, examining AI economics, labor displacement, and political economy through the lens of forty years in tax law and enterprise management. He is a former Director at General Motors and founder of Bionic System Analytics LLC.

The Trap That Closed

The wheat did not ask permission.

That is Yuval Noah Harari’s most unsettling observation in Sapiens. The agricultural revolution was not a human decision so much as a biological negotiation we lost. Wheat needed us to clear land, pull weeds, and carry its seeds across continents. We needed calories. The arrangement looked mutual until the fields were planted, the granaries were full, and the population had doubled past the point of return.

By then the trap had closed.

We had traded leisure, variety, and the deep social bonds of the small band for backbreaking repetitive labor, epidemic disease, and the chronic anxiety of a harvest that might fail. Harari is honest about what this cost. The individual suffered while the species succeeded. The hunter-gatherer woman who bent her back over the first grain field lost something real. More varied food. More sleep. More time. A life scaled to human senses rather than to the logic of surplus and storage.

Harari does not sentimentalize the loss. He simply refuses to look away from it.

This is admirable. And also, I think, incomplete.

The Threshold Nobody Noticed

Somewhere in that same sweep of time, something else happened. Something Harari records but does not fully weight.

Roughly seventy thousand years ago, before the fields, before the granaries, before any of the traps were set, the human animal crossed a threshold that no other species has approached. We began to think in terms of past and future simultaneously. We began to hold in our minds things that do not exist. Gods. Laws. Obligations that outlast the people who made them. Stories that carry knowledge across centuries without a single gene changing.

Harari calls this fictional thinking and means it as a neutral description.

I want to call it something else. I want to call it breaking the code.

Every other species lives inside present-tense experience. The gazelle cannot grieve yesterday or plan tomorrow with deliberate foresight. It responds. It adapts. It does not contemplate. We stepped outside biological time entirely at that threshold and we have never stepped back. We can model the past and simulate the future simultaneously. We can accumulate knowledge across generations without waiting for genetic encoding.

We can ask why.

That is not a wrong turn. That is the whole point of the enterprise.

What Teilhard Sees That Harari Cannot

Pierre Teilhard de Chardin, the Jesuit paleontologist who spent his life holding science and faith inside the same frame, would look at Harari’s data and arrive at a different conclusion.

For Teilhard the agricultural trap, the industrial trap, the current AI trap, all of it is the Noosphere thickening. The layer of human thought and connection that wraps the planet is getting denser, more interconnected, more capable of what he called complexification. Each wave of disruption, however painful at the individual level, is the mechanism by which consciousness arrives at greater coherence.

The bee does not understand the hive. The neuron does not understand the brain. But the hive and the brain are real. Their reality does not depend on the consent or comprehension of their constituent parts.

This is where Teilhard becomes either profound or unbearable, depending on what you bring to him. He does not deny the suffering. He contextualizes it within a trajectory the sufferer cannot see from inside the moment. The woman bent over the grain field is real. Her loss is real. And she is also, without knowing it, a neuron firing in a nervous system that will eventually produce Augustine, Aquinas, and the capacity to ask whether any of this means anything at all.

Where Materialism Runs Out of Road

Harari’s frame has no place to put that question.

Meaning keeps showing up in his account as illusion or cognitive bias. A story we tell ourselves to manage the anxiety of being the only animal that knows it will die. He is too honest to pretend otherwise and too committed to materialism to follow the question where it leads.

But the question does not go away.

I look at the complexity and beauty of the world and find Harari’s wrong turn tough luck account genuinely disheartening. Not because it is too dark but because it is too small. A universe that spent thirteen billion years building a nervous system complex enough to contemplate its own existence, and then produced that contemplation as a mere accident of geography and wheat reproductive strategy, is a universe that has dramatically underexplained itself.

The Direction Is Real

Teilhard’s Omega Point is the alternative. Not a comfortable one.

It does not promise that the pain was worth it in any accounting we can run from here. It promises something stranger and more demanding. That the pain is the mechanism. That complexity does not emerge without pressure. That the direction is real even when the destination is not yet visible.

We broke the code at the cognitive threshold. We won the evolutionary contest in the only way that matters, by becoming the species that can ask whether winning means anything. Every trap we have walked into since, the fields, the factories, the algorithms, has been the price of carrying that question forward.

The wheat did not ask permission. Neither did the Omega Point.

But one of them is pulling toward something. And I do not think it is the wheat.

James Allen Ramsey writes J Allen Insights on the intersection of technology, geopolitics, and human development. He is a retired tax attorney and consultant whose career included GM, PwC, Shaw Pittman, and EY.

In June 2026, the National Science Foundation began directing research vessels to do something unusual. Not to deploy instruments. Not to recover them for service. To physically remove them. The target is the Ocean Observatories Initiative: more than 900 deep-sea sensors anchored across the Pacific Northwest, Alaska, North Carolina, and the Greenland-Iceland corridor. A $368 million network. Operational for over a decade. Engineered to continue delivering data for another fifteen years. Being pulled out by hand. [Maritime Executive, June 2026]

This is what makes DOGE different from any budget cut in living memory. It is not deferral. It is not reorganization. It is destruction in the present tense, executed faster than the surrounding institutions can react. Whether the motive is ideological (a regime decision to end climate observation as a federal function) or fiscal (a DOGE ploy to remove a recurring O&M line and book the savings) is a question that has not been answered. The structural result is the same either way.

If you only count dollars and positions, the OOI removal looks roughly proportional to its budget line. It is not. The buoys were the easy part. What is gone, and gone in a way no policy reversal can recover, is the layered tacit knowledge that kept the array calibrated and operational. Site-specific drift profiles. Biofouling rates by region. The judgment to distinguish a sensor that has wandered out of spec from one that is correctly reading a genuine anomaly. None of this is in a manual. As Jim Edson, head scientist of the OOI, put it on the way out, the array was the product of “extraordinary efforts of the scientists, engineers, operators, educators, students, and partners who made this facility possible.” Those efforts cannot be retrieved. They are scattered into private sector positions, retirement, or out of the field entirely.

The OOI is one case. The pattern is general. A [Science magazine analysis] found that 10,109 doctoral-trained experts left 14 federal research agencies in 2025. Eleven exits for every one hire. The departures concentrated in exactly the positions where tacit knowledge runs deepest: long-tenured operators, senior site-specific scientists, inspectors and examiners whose judgment is the system.

What Was Lost Is Not What The Org Chart Says Was Lost

Michael Polanyi made the distinction in the 1950s. There is the codifiable knowledge that can be written down. And there is the tacit knowledge that resides in the practitioner’s body, judgment, and accumulated site-specific experience. The latter is what makes a craft a craft. It cannot be transferred by handoff document. It can only be transferred by working alongside the person who has it, for years.

This was always the structural weakness of the federal scientific and supervisory workforce. The codified outputs were the visible deliverable: the regulation, the dataset, the inspection report. The institutional capacity that produced them was the iceberg below the waterline. DOGE made the bet, implicitly or explicitly, that the visible outputs were the operation and the iceberg was overhead. The bet was wrong, and the cases below show how wrong.

The Cases

The National Weather Service entered hurricane season 2025 with 30 of its 122 forecast offices missing their Meteorologist-in-Charge. [Global Extremism Project] Five former NWS directors, spanning Republican and Democratic administrations, jointly wrote Congress warning of “needless loss of life.” On July 4, 2025, a flash flood struck Texas’s Guadalupe River valley and killed at least 89 people. The deadliest thunderstorm-driven flash flood in the United States in nearly 50 years. The New Braunfels office covering that stretch of river had lost 22 percent of its staff. The Warning Coordination Meteorologist position, the role specifically designed to bridge forecasters and county emergency managers, sat empty that night.

The USGS stream gauging network, 11,800 stations in all 50 states, had its field maintenance hollowed out by a different mechanism. DOGE-imposed travel restrictions prohibited overnight field travel, which physically severs technicians from the remote stations they calibrate. The Western Land Owners Association documented the specificity of what is at stake: “Most USGS gage techs are experienced professionals with detailed knowledge about the intricate and complex gaging systems.” Each station requires calibration against the behavior of that particular streambed and watershed. Not streams in general. ([Western Land Owners Association]

The CDC Vessel Sanitation Program was harder to write into a memo. In April 2025, all full-time VSP employees were terminated, including the lead epidemiologist directing cruise ship outbreak response. The May 2026 hantavirus outbreak on the MV Hondius exposed the gap directly. The CDC did not issue a Health Alert Network notice to clinicians until more than a week after confirmation, a delay STAT News called “a critical and previously uncharacteristic lag.” ([STAT News, May 2026] The CDC’s training cycle for a vessel inspector is six months. The training cycle for a lead outbreak epidemiologist with operational judgment is closer to a decade.

The Department of Energy lost mission-critical staff at Hanford, the most contaminated nuclear waste site in the United States. The GAO reported in May 2026 that federal project directors at Hanford require “an extensive qualification process that can take up to 10 years.” Even employees who have been on site for two to five years, the GAO noted, can still be considered “new hires” because of the technicality of the role and the amount of learning required. The 46 percent vacancy rate in mission-critical positions at Hanford cannot be closed before the mid-2030s, and only then if hiring resumes at scale, which the federal hiring freeze has prevented. ([GAO via Tri-City Herald, May 2026]

The Nuclear Regulatory Commission was restructured by personnel substitution rather than only by headcount reduction. A DOE lawyer who entered government through DOGE, Seth Cohen, is reported to have stated in an internal meeting: “Assume the NRC is going to do whatever we tell the NRC to do.” ([New Republic] Senior inspectors with reactor-specific anomaly memory have retired. Their judgment-based verification, what former NRC second-in-command Scott Morris described as “I’m a prove-it-to-me guy. I’m not just going to take your freaking word for it,” is precisely the function being phased out.

The bank examination force is the case with the cleanest historical analogue. The FDIC lost 20 percent of its total workforce between the end of 2024 and early 2026. Its resolutions division, the unit that handles bank failures, lost 22 percent of staff with another 28 percent eligible to retire this year. The Federal Reserve’s Washington supervision division was cut by 30 percent. ([Banking Exchange/Kiah Lau Haslett, May 2026]

The 1983 FHLB Dallas Natural Experiment

The historical parallel is exact and the result is documented. In 1983, a Federal Home Loan Bank district office relocated, and most examiners did not follow. Within two years, unsupervised institutions took on significantly more risk, had higher rates of failure, and required slower, costlier resolutions. Not a regulatory change. Not a capital requirement change. Just examiner departure. The same institutions, the same rulebook, materially worse outcomes. The variable that moved was the embedded knowledge of examiners about their specific institutions: management candor history, off-balance-sheet concentrations, the gap between what gets reported and what actually exists.

Kirsten Muetzel, CPA and former bank supervisor, named the substitution being proposed in 2026: “The agencies are migrating toward offsite supervision that is data driven rather than the ongoing interactions with board members and executive management. There are some people who believe that bank supervision can be efficiently and effectively managed solely through data and analysis rather than the ‘boots on the ground’ approach. I tend to get more out of the interactions.” That is the codified-versus-tacit distinction in supervisory clothing. The 1983 experiment already ran this trial. It did not work.

The Irreversibility Math

The qualification timelines establish that this is not a cycle but a step function. Bank examiner: 3 years minimum training, supervised on real institutions. NRC reactor inspector: 3 to 5 years. OSHA Compliance Safety and Health Officer: 3 years minimum. CDC Epidemic Intelligence Service field investigator: 5 to 7 year class cycle. Hanford federal project director: up to 10 years. EPA Office of Research and Development integrated cross-disciplinary research capacity, in Bryan Hubbell’s own words after leaving, will be “difficult to replicate this type of large-scale, integrated research program outside of the federal government.” ([Resources for the Future, May 2026] Add the apprenticeship channel collapse. CDC Epidemic Intelligence Service applications dropped to roughly 20 percent of typical levels. The senior leaves, the apprentice never enters, and the institutional ladder loses both rungs in the same year. The next senior cohort will not be ready in the mid-2030s. It will be ready, if at all, in the 2040s.

The National Law Review observed the structural implication in plain terms: “The disbandment of internal organizations, and the abandonment of key policies (with accompanying enforcement and support staff) will render it considerably more difficult to resurrect similar climate-focused policies in a subsequent administration, due to a lack of appropriate personnel and institutional knowledge. Indeed, this may be an intended consequence of such a re-shaping of government structures.” ([National Law Review, May 2026] The intent matters. If irreversibility is a side effect, the next administration may attempt to rebuild. If irreversibility is the goal, the rebuild will not be authorized to start in time.

What This Is

In the private sector deskilling waves I have been tracking, each round arrived with a sales pitch. ERP would standardize the back office. Offshoring would lower labor cost. Shared services would consolidate the routine. AI agents would absorb the rest. The pitch was always the same: efficiency now, capacity preserved, the senior judgment will be there when it is needed. It was not. The bilingual seniors who carried the institutional memory walked out one round at a time, and the compensating controls they had silently provided walked out with them. Same pitch four times.

DOGE is that same operation executed without the pitch. There is no claim that an AI agent will absorb the Warning Coordination Meteorologist function. There is no claim that the OOI sensors are being replaced with a more modern array. There is no claim that bank examination is moving to a higher-quality method. The capacity is being deleted, and the buildings are being closed behind it. The GSA terminated leases on USGS regional offices in Boulder, Bozeman, Klamath Falls, Moab, Spokane, and Cheyenne. The EPA’s Office of Research and Development was eliminated as a unit. The OOI buoys are being pulled out of the water.

Nathan Anderson of the GAO offered the description that captures what was lost: people who were “equipped by both education and experience to perform roles independently.” That phrase is the operational definition of tacit knowledge. The credential is education. The judgment is experience. You can hire for the first. You cannot hire for the second.

The Risk That Replaces Them

The visible outputs of all these agencies will continue, in some form, for some period. Forecasts will be issued. Stream readings will be reported. Examination cycles will be scheduled. The codified machinery is intact. What is missing is the judgment layer that calibrated the machinery against reality.

The Texas flood was the early signal. The hantavirus delay was another. The next signal will arrive on a schedule no agency can now predict, because the people who could predict it are the ones who left.

You feel it not when it leaves. You feel it when you need it and it isn’t there.

James Allen Ramsey writes J Allen Insights on the intersection of technology, geopolitics, and human development. He is a retired tax attorney and consultant whose career included GM, PwC, Shaw Pittman, and EY.

In the strawberry fields of central Florida and the citrus groves of California, a new kind of crew has started to arrive. Harvest CROO’s machines roll down the strawberry rows picking and packing. Tevel’s drones hover through the orange canopy, plucking fruit one piece at a time. Nanovel’s rigs work the citrus. The harvest is real and the fruit is good enough to ship. But the most revealing thing about these machines is not what they pick. It is what they cost.

Harvest CROO does not sell its robots. It rents them to growers on a modified piece-rate basis, similar to a human crew. The grower pays per unit harvested, the same way he has always paid the people who did this work. No capital outlay, no risk, no equipment to own. The robot is priced at exactly the wage it replaces.

The Confession in the Price

For half a century, American agriculture has told one story about labor. It cannot find enough workers at a wage the crop can bear. This is why we need the guest worker, the H-2A visa, the seasonal crew. The shortage was always presented as a fact of arithmetic.

The piece-rate robot ends that story, because it sets the wage and accepts it. When a company prices its machine at the same piece rate a human crew earned, it is telling you the wage was never the problem. The grower is willing to pay that rate. He is paying it right now, to a machine. What he was not willing to do was pay it to a person who came with a person’s requirements: housing, transport, water breaks in the heat, the thin layer of legal protection even a guest worker carries, the possibility that the person might one day organize, complain, or stay.

I do not have to allege the motive. Harvest CROO’s own materials promise the system greatly reduces dependence on guest workers. That is the sales pitch. The product is sold on the explicit promise that it ends the need for the man who used to do the picking.

They Started With the Hardest Job

Here is the part that should unsettle anyone who thinks automation arrives in order of difficulty. Picking a ripe strawberry without bruising it is one of the hardest physical tasks in the entire economy. The fruit is soft, the ripeness window is days, the stem must be taken cleanly, and the quality gate is brutal. A berry that arrives damaged past the threshold is lost. There is no second pass and no correction. By any engineering measure, this should be among the last jobs a machine takes, not the first.

It went first anyway. Not despite the difficulty. Because of who does it.

The strawberry picker has no senator. The orange picker has no union local, no church bulletin that will run his story, no hometown paper that will note his absence. When he is replaced he does not file for unemployment. He is not counted in the jobless rate, because he was never counted in the labor force the way a citizen is. He goes home, and the statistic that measures American employment does not so much as flicker.

So the difficulty was worth paying. A company will pour years and capital into solving the hardest version of a problem when it can do so in a field where no one will fight back. The voiceless worker is not just the cheapest to displace. He is the safest place to prove the machine.

The Model Is the Product

The robot is impressive. The robot is not the point. The point is the rental contract.

Once a company can rent a machine to an employer on a piece-rate basis, priced at the wage, with no capital risk to the employer, it has built something far more portable than a fruit picker. It has built a procurement model. The employer keeps paying what he always paid, the labor cost converts from a workforce into a subscription, and the human being quietly drops out of the arrangement. That model does not care what is being picked, lifted, sorted, or assembled. It cares only that the task can be automated and the contract can be written.

And the model gets exported up the stack. The next worker it reaches is the warehouse picker, who moves totes a machine can already handle more easily than a strawberry. After him, the auto worker on the line, doing repetitive motions a machine performs without fatigue. These are people with unions, with contracts, with senators, with a vote. They will see it coming and they will resist. But by the time it arrives at the loading dock and the assembly line, the rental model is no longer a novelty being tested. It is a proven procurement structure with a price sheet, normalized in fields where no one could object to it.

Up the stack the arithmetic shifts in the employer’s favor. The migrant wage was held down, so the rental merely met it. The union wage is higher, so the rental comes in under it, and the gap is no longer a swap but margin. Once the rental sits clearly below the wage, the next step writes itself. The employer stops renting and buys the machine, and the discount becomes ownership of the thing that replaced the worker.

The migrant field was the demonstration plant. The hardest task, solved first, on the workers least able to push back, precisely so that the financial machinery would be mature and ordinary by the time it reached the workers who can.

An Account That Was Opened Long Ago

This is not a new kind of injustice. It is the closing of an old account.

The Bracero Program brought millions of Mexican workers north to build California agriculture, and when it ended in 1964 those workers were sent home with their backs and nothing else. The growers kept the groves. The arrangement that followed kept the same shape under new legal names. As long as the work needed hands, the system needed a next wave of hands it could obtain at the price it wished to pay, and each wave let it defer the bill to the last group while recruiting the next.

The piece-rate robot is the moment there is no next group. The hands are no longer needed. The deferral has run out of people to defer to.

What Is Owed

The bill for this has a strange property. It becomes undeniable at the moment the person owed it disappears. As long as the picker was in the field, his low wage could be called a market outcome. The instant he is replaced by a machine priced at his own wage, the pretense collapses. The employer has stated, in dollars, exactly what the work was worth to him. He is paying that amount still. He has simply arranged that it no longer reach a human hand.

The picker cannot send the invoice. He has gone home, uncounted. But the price is on the contract, and the same contract is already being written for the warehouse and the line. The question is whether the workers who can still raise their hands will recognize, before their turn comes, that the bill was first presented in a strawberry field to someone who was never allowed to ask for it. of crew has started to arrive. Harvest CROO’s machines roll down the strawberry rows picking and packing. Tevel’s drones hover through the orange canopy, plucking fruit one piece at a time. Nanovel’s rigs work the citrus. The harvest is real and the fruit is good enough to ship. But the most revealing thing about these machines is not what they pick. It is what they cost.

Harvest CROO does not sell its robots. It rents them to growers on a modified piece-rate basis, similar to a human crew. The grower pays per unit harvested, the same way he has always paid the people who did this work. No capital outlay, no risk, no equipment to own. The robot is priced at exactly the wage it replaces.

The Confession in the Price

For half a century, American agriculture has told one story about labor. It cannot find enough workers at a wage the crop can bear. This is why we need the guest worker, the H-2A visa, the seasonal crew. The shortage was always presented as a fact of arithmetic.

The piece-rate robot ends that story, because it sets the wage and accepts it. When a company prices its machine at the same piece rate a human crew earned, it is telling you the wage was never the problem. The grower is willing to pay that rate. He is paying it right now, to a machine. What he was not willing to do was pay it to a person who came with a person’s requirements: housing, transport, water breaks in the heat, the thin layer of legal protection even a guest worker carries, the possibility that the person might one day organize, complain, or stay.

I do not have to allege the motive. Harvest CROO’s own materials promise the system greatly reduces dependence on guest workers. That is the sales pitch. The product is sold on the explicit promise that it ends the need for the man who used to do the picking.

They Started With the Hardest Job

Here is the part that should unsettle anyone who thinks automation arrives in order of difficulty. Picking a ripe strawberry without bruising it is one of the hardest physical tasks in the entire economy. The fruit is soft, the ripeness window is days, the stem must be taken cleanly, and the quality gate is brutal. A berry that arrives damaged past the threshold is lost. There is no second pass and no correction. By any engineering measure, this should be among the last jobs a machine takes, not the first.

It went first anyway. Not despite the difficulty. Because of who does it.

The strawberry picker has no senator. The orange picker has no union local, no church bulletin that will run his story, no hometown paper that will note his absence. When he is replaced he does not file for unemployment. He is not counted in the jobless rate, because he was never counted in the labor force the way a citizen is. He goes home, and the statistic that measures American employment does not so much as flicker.

So the difficulty was worth paying. A company will pour years and capital into solving the hardest version of a problem when it can do so in a field where no one will fight back. The voiceless worker is not just the cheapest to displace. He is the safest place to prove the machine.

The Model Is the Product

The robot is impressive. The robot is not the point. The point is the rental contract.

Once a company can rent a machine to an employer on a piece-rate basis, priced at the wage, with no capital risk to the employer, it has built something far more portable than a fruit picker. It has built a procurement model. The employer keeps paying what he always paid, the labor cost converts from a workforce into a subscription, and the human being quietly drops out of the arrangement. That model does not care what is being picked, lifted, sorted, or assembled. It cares only that the task can be automated and the contract can be written.

And the model gets exported up the stack. The next worker it reaches is the warehouse picker, who moves totes a machine can already handle more easily than a strawberry. After him, the auto worker on the line, doing repetitive motions a machine performs without fatigue. These are people with unions, with contracts, with senators, with a vote. They will see it coming and they will resist. But by the time it arrives at the loading dock and the assembly line, the rental model is no longer a novelty being tested. It is a proven procurement structure with a price sheet, normalized in fields where no one could object to it.

The migrant field was the demonstration plant. The hardest task, solved first, on the workers least able to push back, precisely so that the financial machinery would be mature and ordinary by the time it reached the workers who can.

An Account That Was Opened Long Ago

This is not a new kind of injustice. It is the closing of an old account.

The Bracero Program brought millions of Mexican workers north to build California agriculture, and when it ended in 1964 those workers were sent home with their backs and nothing else. The growers kept the groves. The arrangement that followed kept the same shape under new legal names. As long as the work needed hands, the system needed a next wave of hands it could obtain at the price it wished to pay, and each wave let it defer the bill to the last group while recruiting the next.

The piece-rate robot is the moment there is no next group. The hands are no longer needed. The deferral has run out of people to defer to.

What the Price Revealed

The price on that rental contract did one thing the fifty-year labor-shortage story never could. It put a number on what the work was worth to the employer, and then showed he was willing to pay it. He is paying it still. He has simply arranged that it no longer reach a human hand. The shortage was never about the wage. It was about everything the wage came attached to when a person earned it.

The picker is gone now, uncounted, back across a border. But the price is on the contract, and the same contract is already being drafted for the warehouse and the line. It was proven in a field where no one could object, on the hardest task anyone could find, so that it would be ordinary by the time it reached the workers who can.

The question is whether those workers, the ones with unions and contracts and a vote, will recognize what was done in that strawberry field before the same thing is done to them.

James Allen Ramsey writes J Allen Insights on the intersection of technology, geopolitics, and human development. He is a retired tax attorney and consultant whose career included GM, PwC, Shaw Pittman, and EY.

On April 30, in an all-hands meeting at Meta, Mark Zuckerberg explained to his workforce why the company is now logging their mouse movements, keystrokes, and periodic screenshots. The program is called the Model Capability Initiative. It runs through approved applications including Gmail, GChat, VSCode, and Meta’s internal AI assistant Metamate. The CTO, Andrew Bosworth, has reportedly told staff Meta is “not accepting refusal” to participate.

The audio leaked to More Perfect Union and was made public on May 19. The next day, Meta began laying off roughly 8,000 employees.

In the recording, Zuckerberg told his engineers why their behavior in particular is being harvested. “The average intelligence of the people who are at this company is significantly higher than the average set of people that you can get to do tasks if you’re working through these contractors.” And then the line that does the actual work: “The AI models learn from watching really smart people do things.”

He also acknowledged the deliberate opacity: “It is not strategically in your interest for us to communicate everything in all the detail that we normally would on this.” He said it knowing it would leak. He said it anyway.

The deal he was offering was this. Your intelligence is the most valuable input we have. Your captured behavior is what makes our model better than our competitors’ models. We will not let you decline. And we are laying off eight thousand of you tomorrow.

The Same Company Running Both Ends

Meta is also one of the largest harvesters of child and adolescent behavioral data in human history. Instagram and Facebook have spent two decades cataloging which colors hold a thirteen-year-old’s eye for how long, which kinds of bodies make her tap, which captions make her share, which combinations of music and motion will keep her thumb moving for the next forty minutes. That harvest built the recommendation engines that built the company that now employs the engineers whose VSCode sessions are training the next system.

The engineers being surveilled at Meta this spring grew up on Meta’s products. The data Meta took from them as adolescents trained the platform they came to work on. The data Meta is taking from them now will train the AI that, by their CEO’s own account in other public remarks, is meant to replace software engineers within this calendar year. It is the same harvest at two ages, by the same company, of the same people.

There is a piece I have been carrying that I call The Factory Moved Into the Playroom. Its argument is that when a six-year-old taps a tablet, the company on the other end is not entertaining her. It is taking her labor. Her attention, her preferences, her small choices about which animation made her laugh and which one she swiped past, all of it is captured and recoded as training data.

We do not call this labor because she is a child and because the activity is called play. But the captured behavior is what makes the product valuable. The relationship is not different in kind from the Model Capability Initiative. It is only different in candor.

Zuckerberg told the engineers what was happening. Nobody tells the children.

The Petition the Children Cannot Sign

Meta workers, after the announcement, circulated a petition with the following language: “It should not be the norm that companies of any size are permitted to exploit their employees by non-consensually extracting their data for the purposes of AI training.” That is a precise and useful sentence. It identifies the act, the parties, the absence of consent, and the purpose. With the substitution of two words, it also describes what happens to a child on a tablet.

The child has no petition. She has parents who handed her the tablet because the meeting ran long and the dishes were not done. She has a school that requires a tablet for fourth-grade math. She has a state that has not yet decided whether her behavioral data is hers, her parents’, the company’s, or no one’s in particular.

She has nothing comparable to the legal structure of employment. The adult engineer can at least quit. The child cannot quit being six.

This produces a situation that, when stated plainly, is genuinely strange. The most articulate and well-paid workers in the country are filing a petition saying their behavioral data should not be taken without consent. The most defenseless humans in the country, our children, are having precisely the same data taken without consent and there is no petition, because nobody under twelve is in a position to draft one.

The two harvests run on the same servers, in the same data centers, by the same company, for related purposes. Only one of them generates a press cycle.

The Ledger That Was Never Kept

Watching really smart people do things is half the harvest. Watching really small people do things is the other half. They are the same harvest. They have always been the same harvest. The only new thing is that someone in the room admitted it.

But the harvest of the children does not have value on its own. The recommendation engine that knows a thirteen-year-old wants the newest sneakers monetizes only because someone in her house can buy the sneakers. The advertiser pays for the impression because the impression reaches a household with disposable income.

The child’s attention is the input. The parent’s paycheck is the dividend. When Meta lays off eight thousand engineers, those are eight thousand households where the dividend has collapsed and the children’s attention is suddenly worth less.

The same company is harvesting both ends of the lifecycle, and it is also dissolving the link between them. The recommendation engine eats the wage that pays for what it recommends.

When the labor side is closed out, what remains is a pure profit play, and it accrues entirely to whoever owns the capital. There is no countervailing claim, because there is no remaining counterparty to assert one.

This is worth pausing on, because the accounting convention that produced this outcome is the central artifact of modern commerce. In a properly structured corporate formation, the founders who contribute labor and know-how receive shares for that contribution. The sweat equity is booked on the capital ledger as paid-in contribution.

It becomes a permanent claim on the entity. The founder owns part of what the company will become because the work that built the company was treated as capital input, not as expense.

For more than five hundred years, a large portion of the labor and the know-how that built the modern economy was harvested without ever hitting the capital ledger. The contributors were real. The contributions were real. The shares were never issued.

The Africans whose labor cleared the cotton fields and refined the sugar were a capital input booked as zero. The indentured Europeans whose seven-year contracts built the colonial estates were a capital input booked as a debt to be retired against passage. The children who tended the spinning frames in Manchester and the textile mills of Lowell were a capital input booked as a wage held below subsistence. The miners who hauled the anthracite that fed the steel that built the railroads were a capital input booked as a wage and a cemetery plot. The women in the maquiladoras stitching seams for the global apparel chains, the warehouse pickers walking twelve miles a shift, the meatpackers losing fingers in the disassembly lines, all of them were capital inputs booked as operating expense.

Every one of them put in what a founder puts in. None of them got what a founder gets. The shares were issued only to whoever held the title.

The harvest at Meta this spring is the same accounting maneuver, conducted on more articulate subjects and with cleaner instrumentation. The engineers whose keystrokes train the next system are contributing the labor and the know-how that the system will embody and sell.

They are doing what a founder does. The system will keep what they built. They will be issued no shares. They will be issued severance.

What is different now is that the labor portion of capitalism is being closed out entirely. After this round of engineers, the harvest does not require new contributors. It only requires the accumulated capture of every contributor who came before.

The book is being closed on the labor account. Every share that was never issued is becoming visible at once, because there is no more new contribution on the other side to keep the absence hidden.

The petition Meta workers filed is the first sentence of a much longer bill. It is in the same handwriting as the petition the slaves did not get to file, the indentured did not get to file, the children in the mills did not get to file, the miners with destroyed lungs did not get to file. They never closed the labor account before. They are about to.

James Allen Ramsey writes J Allen Insights on the intersection of technology, geopolitics, and human development. He is a retired tax attorney and consultant whose career included GM, PwC, Shaw Pittman, and EY.

William Shatner is ninety-five years old and still working. The work is commercials. Travel sites, supplements, breakfast cereal. He is good at it. The voice still carries that mid-Atlantic certainty that made him a star, the slight pause before the punch line, the willingness to commit absolutely to the bit. Younger actors study him for it.

Meanwhile the Enterprise is still flying.

What Kirk Actually Was

The standard reading of Star Trek puts Spock at the center of the ship’s intelligence. He is the logic engine. He runs the calculations, identifies the anomalies, surfaces the threats. The episodes turn on his computations. By the metric of raw cognitive output, Spock is the smartest entity on the bridge by a wide margin.

Kirk is not smarter than Spock on logic. He is not stronger than Sulu at the helm. He is not faster than Uhura at communications. He cannot rewire a warp coil under fire the way Scotty can. By any single-axis measurement of capability, Kirk is the least specialized officer on the bridge.

What Kirk has is the corpus. He has walked every deck of that ship during a crisis. He has watched Spock’s literal answer be the right answer, and he has watched it be the wrong one. He has seen Scotty say “she canna take it” and meant it, and he has seen Scotty say it as an opening bid in a negotiation. He has thousands of missions metabolized into instinct. The bridge works because Kirk is the integrator. He carries the model of the whole ship in his body.

That is the part the show never quite explains, because the show takes it for granted. The captain’s chair is not a job description. It is a corpus seat.

The C-Suite Parallel

For forty years inside large corporations, the equivalent of Kirk was the senior partner, the long-tenured division head, the engineering fellow with thirty years on the floor. They were not the smartest people in the room on any given technical question. The associates ran circles around them on tax code minutiae. The new MBAs had better spreadsheets. The young engineers wrote cleaner code.

What the senior had was the corpus. They had watched the same kind of deal go sideways in 1997 and again in 2008. They knew which auditor was bluffing. They knew when the model was missing a number that did not yet have a column. They knew which client request meant what the client said and which meant something else entirely.

The technology waves of the last forty years have all been sold the same way. The senior person is expensive. The senior person is the bottleneck. The technology absorbs what the senior person used to do, and the firm runs leaner, faster, more profitably without them. ERP. Offshoring. Shared services. Now agents.

In each case the hidden assumption is that the senior was redundant. The empirical record is that the senior was carrying the apophatic load, the tacit pattern recognition that catches the wrong answer before it ships. Strip that, and the literal-execution layer does not fail noisily. It fails plausibly.

The Bridge Without Kirk

Picture the bridge of the Enterprise with the captain’s chair empty.

Spock is still there. He still gives perfect literal answers to whatever question is put to him. The helm is still staffed. Engineering still reports up. The ship still flies, and flies beautifully, at warp speed in whatever direction the last literal answer pointed.

The moment the situation requires someone to say “Spock, I hear you, and we are doing the opposite,” there is nobody on the bridge with the standing or the corpus to say it. The first officer logically promoted into the chair is brilliant on logic and empty on integrated judgment. The second officer is sharper still on his particular console and has never set foot in engineering during a crisis. The chair is filled. The seat is empty.

This is the structural fact the AI sales literature is asking you not to notice. The agent does not replace Kirk. It replaces Spock, and it replaces him at scale, with confidence, around the clock. What it cannot do is sit in the captain’s chair, because the captain’s chair was never a logic problem. It was a corpus problem.

The Training Program Will Not Save You

The standard response from the consulting class is that firms will simply train their younger people to be the new integrators. There will be AI literacy programs, prompt engineering certifications, executive education seminars on human in the loop oversight.

This will not work. You cannot teach a thirty-year-old associate to recognize the wrong answer in a multi-state nexus memo by sending them to a two-day workshop. That recognition is built from a thousand small moments of being wrong and being corrected by someone senior who had themselves been wrong and corrected. The corpus is not content. It is metabolized experience.

Cutting the senior tier does not just lose the current integrator. It closes the only road that produces the next one. The associates who would have become the next generation of senior partners learned by doing the work the senior partner reviewed. Take the senior partner out and the associates do not develop. Take the work itself out, give it to the agent, and the associates do not even start.

Starfleet Academy can produce a four-stripe captain in twenty years if it has four-stripe captains to produce them. It cannot produce a four-stripe captain from a fleet that retired all of them in 2026 because the technology made them redundant.

The Civilizational Register

This is the same structural move three times. ERP hollowed out the middle layer of the firm, the people who used to integrate departments. Offshoring hollowed out the floor, the people who used to learn the work by doing it. The agent wave is hollowing out the bridge, the people who used to integrate everything into a coherent mission.

Each wave was sold as a productivity gain. Each wave was a transfer of compensating controls from the firm’s balance sheet to nowhere, because the controls were never priced. The firm took the labor savings into earnings. The risk went into a category nobody costed.

The senior integrator was the last unpriced control. The agent wave is removing him.

Kirk in Burbank

Shatner is ninety-five and doing commercials. The image is funny because we know who he was. The captain of the Enterprise is reading cue cards about fiber content in a green room in Burbank.

The deeper joke, the one the C-suite is not yet in on, is that Shatner is the lucky one. He retired voluntarily. He had a long career, a recognizable face, an annuity stream of residuals. He chose the commercials.

Most of the senior integrators in American firms did not choose. They were packaged out, retired early, told their function was being absorbed. They are tending gardens. They are volunteering at soup kitchens. They are writing on Substack.

The ship is still flying. Spock is still on the bridge. The captain’s chair is empty. The training school is closed.

The question is not whether anyone notices. The question is what the ship hits first.

James Allen Ramsey writes J Allen Insights on the intersection of technology, geopolitics, and human development. He is a retired tax attorney and consultant whose career included GM, PwC, Shaw Pittman, and EY.