The Pitchforks Are Coming for A.I.

In a recent episode of The Good Fight on Americans’ loss of faith in universities, Yascha Mounk, speaking with Yale professor David Bromwich, identified a contradiction at the heart of American meritocracy: The stakes are enormous, but the criteria are often obscure. America, he said, combines “a highly meritocratic culture, with outsized returns for getting into Yale or into the investment bank you want to work for,” with “a culture that is very reluctant to be explicit about criteria.”

They were talking about college admissions, but the point reaches further. American life is organized around competition, while the rules of that competition are often concealed.

That may be one reason American inequality has remained politically tolerable. The country has never promised equality of outcome. It has promised mobility. Work hard in school, get into the right college, take the entry-level job, endure the humiliations of apprenticeship, and eventually you may reach the class of people who make decisions rather than merely absorb them.

The path was never as open as the mythology suggested. Race, class, geography, inheritance and luck have always shaped who rises and how far. But the promise of mobility gave inequality a moral defense. The rich were not supposed to be a permanent caste. The poor were not supposed to be trapped forever. The ladder might be steep, and the rules opaque, but it was said to exist.

A.I. threatens that story at its most vulnerable point: the first rung.

Even if artificial intelligence does not eliminate all work, it may damage the ladder by which people become valuable. Entry-level tasks are often boring, repetitive, and badly paid. They are also how people learn. The junior associate reviews documents before arguing the case. The assistant edits copy before running the campaign. The young worker performs tedious tasks not because the tasks are noble, but because they are the way judgment is acquired. They are the apprenticeship hidden inside the modern office.

Automate those tasks, and one does not simply liberate the young from drudgery. One may strand them below the point of entry. The danger is not only that people will lose jobs, but that they will lose the path by which they become the kind of people who can get better jobs.

This is where the comparison to earlier moments of political rupture becomes useful. Not because America is eighteenth-century France, or because today’s engineers are aristocrats in powdered wigs, or because every new machine is a guillotine in embryo. History rarely repeats so neatly. It returns instead as pressure. It returns as a question: What happens when a society organized around inequality can no longer plausibly promise escape?

The ancien régime collapsed not simply because people were poor, but because poverty came to look permanent, artificial, and politically arranged. The peasant did not merely lack bread. He lacked a plausible path out of dependence. The burdens of the old order fell downward, while privilege remained protected by birth, law, and custom. To be born into the wrong estate was not merely to begin life at a disadvantage, but to discover that disadvantage had been made hereditary.

Before it became blood and empire, the French Revolution was a crisis of arithmetic. The monarchy was bankrupt. The nobility held privileges it would not surrender. The peasantry paid in taxes, feudal dues, and labor owed to men who had been born already exempt from much of the burden. Bread grew scarce. The price of a loaf rose faster than patience. In the countryside, rumors spread: of hoarded grain, aristocratic plots, deliberate starvation. Rumor did what policy had failed to do. It organized fear.

In 1789, Louis XVI summoned the Estates-General, the old representative body that divided France into three orders: the clergy, the nobility, and the Third Estate. The Third Estate meant everyone who belonged to neither privileged order; it was, by far, most of France: lawyers, merchants, artisans, laborers, and peasants. These were the people who had begun to suspect that a society could not endure forever when its burdens and rewards were so extravagantly misaligned.

When the Third Estate declared itself the National Assembly, it announced that legitimacy had migrated. Power, once understood as descending from crown and altar, now appeared to rise from the crowd.

Then came the familiar sequence, always simplified in retrospect: the Bastille, the abolition of feudal privileges, the Declaration of the Rights of Man, the march on Versailles, the king’s execution, the Terror, Thermidor, Bonaparte. We remember the Revolution as violence because violence supplied its images: heads on pikes, the guillotine in the square, silk stockings giving way to the righteousness of the street.

But violence was not the first fact of the Revolution. The first fact was a collapse of belief. People stopped believing that suffering was natural when it was arranged for the comfort of others. They stopped believing that hunger was merely an unfortunate weather system while abundance remained protected by law. They stopped believing that the powerful were guardians of order rather than its chief beneficiaries.

Most important, they stopped believing that patience, labor, or obedience would deliver them from dependence. Poverty becomes explosive when it ceases to look temporary, and inequality becomes intolerable when those at the bottom no longer believe there is a route upward. The French Revolution began, in part, when millions of people stopped seeing misery as fate and started seeing it as a structure.

That is the history worth remembering now, as the artificial intelligence economy takes shape.

Inside the world of A.I., a grim view has become increasingly ordinary, and it is no longer confined to outside critics. In May 2025, Dario Amodei told Axios that A.I. could eliminate half of all entry-level white-collar jobs and push unemployment to 10 to 20 percent within one to five years. He described one possible future this way: “Cancer is cured, the economy grows at 10% a year, the budget is balanced — and 20% of people don’t have jobs.” Axios reported that Amodei was saying publicly what other leading A.I. executives had expressed privately.

OpenAI’s own 2026 policy blueprint points in the same direction, warning that A.I. may expand corporate profits and capital gains while reducing reliance on labor income and payroll taxes. The industry is not merely admitting that disruption is possible. It is beginning to plan around it.

More than that, the industry is optimizing for it. A.I. companies do not simply announce that models are improving at abstract tasks, or that they can write poems, summarize documents, and pass exams. Increasingly, they measure progress by economic utility: how well models can perform real jobs, complete real workflows, and substitute for real employees. Benchmarks that evaluate models against economically valuable tasks convert job replacement into a technical milestone. A model that can perform the work of an analyst, paralegal, customer service agent, junior software engineer, or marketing associate is not just “capable.” It is valuable because it threatens a wage.

Once that metric becomes respectable, everyone downstream adjusts. Executives describe layoffs as modernization. Investors reward head-count restraint. Managers freeze hiring before they fully understand the tool they are invoking. A.I. becomes an alibi for decisions that may have been desired anyway: fewer workers, weaker obligations, more leverage for capital. The machine need not actually replace every person it is said to replace. The rumor of replacement can do some of the work.

The fear, then, is not merely unemployment. It is structural demotion: a world in which human labor is no longer the central means by which ordinary people claim income, status, discipline, purpose, and political leverage.

This is why the entry-level job matters so much. It tells the young worker, “You are not yet powerful, but you are inside the system; you do not yet know enough, but you can learn; you are not yet trusted, but trust can be earned.” Remove that first point of entry, and the promise changes. The worker is no longer waiting to ascend. He is waiting outside a locked door.

OpenAI occupies a revealing position in this drama because it has long spoken in the grammar of public benefit while increasingly operating through the structures of private capital. The company was founded as a nonprofit with a mission to ensure that artificial general intelligence “benefits all of humanity,” but in 2019 it created a capped-profit subsidiary so it could raise the billions of dollars it said would be needed to build advanced A.I. Sam Altman has also acknowledged the basic distributive problem. In “Moore’s Law for Everything,” he wrote that software would do more and more of the work people now do, and that “even more power will shift from labor to capital.” His proposed answer was an American Equity Fund, financed by taxes on large companies and privately held land.

More recently, OpenAI has floated a broader industrial-policy agenda: higher capital-based taxes, possible taxes related to automated labor, a public wealth fund, and 32-hour workweek pilots with no loss in pay.

These ideas at least recognize that abundance without distribution is only another name for oligarchy. But the posture remains uneasy, as though the industry knows the negative case and would prefer not to lead with it. The darker research sits in tension with the brighter press release. The egalitarian rhetoric sits in tension with corporate valuation. The language of shared destiny sits in tension with the legal and financial structures through which the gains will actually be allocated.

Jack Clark, through the Anthropic Institute, has framed the matter as a societal choice: A.I. could destroy worker leverage, or society could use the wealth it creates to fund human-centered work — teaching, nursing, elder care, child care, all the intimate labor markets have traditionally underpaid because it cannot be scaled cleanly.

While there is moral seriousness in that framing, there is also an unresolved contradiction. Anthropic has not committed itself to a redistributive program commensurate with the risks it describes. Its business depends on selling enterprise agents to firms that want efficiency, and efficiency in this context often means fewer people doing the work. The company can warn of the fire while selling accelerants to the warehouse. This does not make its warnings false. It makes them more disturbing.

The psychological contradiction inside the industry is by now almost gothic. Men and women in hoodies and conference badges, with stock options vesting and calendars full of alignment meetings, believe they may be summoning either paradise or ruin. They speak of curing disease, accelerating science, ending drudgery, democratizing expertise. They also speak, after the panel is over, of economic chaos, political instability and jobs evaporating faster than institutions can respond. The creature is still on the table, still being assembled from lightning and ambition, and already its makers are afraid of its shadow on the wall.

Some justify their participation by saying that someone else would build it anyway. This is the oldest absolution in technological history. It has the advantage of being partly true and morally insufficient. Others try to offset the damage through philanthropy, policy research, safety institutes, universal basic income pilots, or private guilt metabolized into public concern.

The political system is beginning to notice what the industry already fears. Voters do not need a technical paper to understand vulnerability. They know what it means when a manager says productivity. They know what happens when a company discovers it can do more with less. They know, even if they cannot name the benchmark, that every new claim of “economic utility” may be translated into someone else’s lost bargaining power.

That shared vulnerability may become politically clarifying. A.I. could produce something like a modern Third Estate: not a class unified by occupation or education, but by the suspicion that the promised route upward has been closed. It would include the warehouse worker, the call-center employee, the adjunct instructor, the junior coder, the graphic designer, the bookkeeper, and the recent graduate sending résumés into a vacancy. What unites them is not identical work, but a common perception: that the American promise — that education would move you upward, that diligence would stabilize you, that the future needed your labor — has been withdrawn.

That perception matters. A society can tolerate inequality when it believes inequality is temporary, or at least permeable. It can tolerate hierarchy when it believes the people at the bottom have some chance, however slim, of rising. But when people conclude that the ladder has been pulled up, inequality changes character. It stops looking like a race, however unfair, and starts looking like a wall. It stops producing ambition and starts producing rage.

This could create a new and unstable solidarity. The truck driver and the junior programmer may not have shared much political language before. The customer support worker and the law graduate may not have imagined themselves as members of the same economic class. A.I. could change that. It could reveal that vulnerability is no longer confined to those without degrees. It could make the central divide less cultural than structural: those who own the machines, and those whose labor is disciplined by them.

The ancien régime fell because poverty became politically intelligible as captivity. The privileges of the few could no longer be justified as part of a natural order. The people below could no longer imagine that patience, work, or obedience would deliver them from dependence. Once that belief collapsed, the whole moral architecture of the regime collapsed with it.

A.I. may bring America to a similar crisis of belief. The country has tolerated obscene inequality because it has preserved, or claimed to preserve, the possibility of movement. But if artificial intelligence hollows out the entry-level jobs through which people become skilled, employable, and socially recognized, then the old promise begins to fail. The young graduate sending résumés into a vacancy, the junior coder replaced before he has learned enough to become senior, the paralegal whose first tasks are now done by software, the designer told that taste itself can be automated — these people are not merely losing jobs. They are losing the story that made the system bearable.

When people believe they have a chance, they may accept inequality. When they believe the chance has been withdrawn, they begin to look at wealth differently. They stop seeing it as achievement and start seeing it as enclosure. They stop seeing the successful as winners of a difficult contest and start seeing them as owners of a rigged machine.

This is the political danger A.I. companies consistently underrate. The danger is not only regulation, lawsuits, bad headlines, or the technical difficulty of building what they promise. The deeper danger is legitimacy. If the public comes to believe that A.I.’s gains are privatized while its losses are socialized, then the technology itself may become a target — not merely a product to be rejected, but an order to be resisted.

Resistance need not begin as revolution. It usually does not. It can begin as municipal fights over data centers, anger over water and energy use, bans on automated systems, labor actions, lawsuits, refusals, boycotts, sabotage, political campaigns, and a general withdrawal of consent. None of this should be romanticized. Violence is not justice. It is what appears when legitimacy has failed and politics has arrived too late.

This may be the ultimate check on A.I. wealth growth: not technical limitation, not market saturation, not even formal regulation, but unrest. A society can tolerate great wealth when it believes, however dimly, that the wealth serves a common future. It can tolerate disruption when it believes there is a bridge to the other side. It can tolerate automation when the displaced are offered more than slogans about reskilling. But it cannot indefinitely tolerate a future in which human beings are asked to applaud their own obsolescence while a narrow class compounds its claims on everything valuable.

The builders of A.I. often speak as though the future is a problem of alignment between machine and human intention. That is one problem. A bigger problem is the alignment between wealth and legitimacy. France discovered what happens when that alignment breaks. The lesson was written first in pamphlets, then in bread lines, then in law, then in blood.

The machine may make fortunes. It may discover medicines, write code, tutor children, run offices, design factories, and compress into minutes the labor of entire departments. It may also produce a class of people who look upon those fortunes and see not progress but dispossession. At that point the question will no longer be what A.I. can do. It will be whether the people left outside the future will consent to stay there.