The Last Rung
Work, AI, and the Absence of Opposition
In January 1912, Frederick Winslow Taylor arrived at the House of Representatives for what became several days of withering cross-examination by William B. Wilson, chairman of the House Committee on Labor. Taylor had published The Principles of Scientific Management the year before his manifesto for time-and-motion studies, the stopwatch as instrument of production, the systematic extraction of craft knowledge from workers’ bodies into management protocols and the labor movement regarded it, with some accuracy, as a blueprint for their replacement. Wilson wanted to know about the workers bearing the cost. Wasn’t it true, he pressed, that when any system is introduced requiring fewer men to produce the same output, the workmen “have to bear the entire burden” of the readjustment? Taylor repeated his talking points. Workers got their fair share. Scientific management had always netted workers “an increase of 30 per cent to 100 per cent in wages.” Wilson pushed back: what guarantee existed that employers would actually share the gains? Taylor said the workers could always slow down if they felt cheated. Wilson found this less reassuring than Taylor appeared to.
Three days. Public record. Congress subsequently banned time-and-motion studies from federal facilities — a partial and temporary response, easily eroded, ultimately insufficient, but the hearings existed. Taylor had to sit there and defend his claims to someone whose job it was to be skeptical of them, in a forum with consequences for people who refused to engage seriously.
Sam Altman, whose company is overseeing a transformation of the labor market at least as significant as anything Taylor wrought, goes on Lex Fridman’s podcast. Dario Amodei goes on Dwarkesh Patel’s. These are not unintelligent conversations. Dwarkesh will press on capabilities timelines with genuine specificity. Nobody is calling it propaganda. The problem is something both subtler and more damning: the format is constitutively incapable of producing the thing the moment requires. A podcast is not a hearing. The guest explains their worldview, the host asks clarifying questions, the guest elaborates. The audience self-selects. The accountability is optional. Nobody is going to make Sam Altman spend three days defending who bears the cost of the readjustment, because nobody has the institutional power to put him in that room ,in part because the people who might have held that power have been systematically cultivated by the companies whose deployment decisions most need scrutinizing. When OpenAI’s people are drafting executive orders and Anthropic’s people are in the White House and the regulatory framework for the technology is being written by the organizations with the most to gain from its lightness, the Wilson/Taylor hearing is not a precedent being honored. It is a precedent whose recurrence is being structurally prevented.
This is not a complaint about individual bad faith. It is a description of a political economy. And it is the starting point for understanding why the intellectual response to the present transformation is so drastically, almost insultingly, thin : why the discourse about what AI is doing to work is so impoverished relative to the magnitude of what is being done.
Erik Baker’s Make Your Own Job is about a specific ideological operation that has been performed, in nearly identical form, every time the American economy has displaced workers faster than it has provided alternatives for them. The operation goes: structural problem gets translated into individual opportunity. The specific mechanism Baker traces is the entrepreneurial work ethic.The set of ideas distributed through self-help literature, management theory, and eventually positive psychology, which hold that the secret to navigating displacement is to look inward, locate your unique aptitude, and build an economic identity around it. The argument is not that this was always cynical. Sometimes it was sincere, and the economy was sometimes responsive to what it prescribed. The con, as Baker precisely identifies it, lay in generalization, in converting something that worked for some people in some circumstances into a structural account, in presenting individual strategy as systemic solution, in making the displaced worker’s psychology the relevant unit of analysis when the relevant unit was the labor market.
The original crisis that produced it looks, from 2026, like a rough draft of the present. In the first decades of the twentieth century, manufacturing output was expanding while manufacturing employment was not. Baker documents 100,000 jobs shed between 1919 and 1929 against 65 percent output growth and the concept of “structural unemployment” entered the language specifically to describe a new kind of joblessness that was not the economy failing but, in Baker’s phrase, “the economy of abundance doing well, in a perverse kind of way.” Output expanding, capitalists growing rich, science marching onward, workers being left behind. The concept was politically explosive, and the explosion produced genuine competition between frameworks, not a single coherent response but several, each with institutional backing, real constituencies, and serious intellectual foundations.
This is the part of the story that the present moment most conspicuously lacks, and it is worth being specific about what it contained, because the contrast is not flattering to us.
Eugene Debs ran for president in 1912 on a platform that named the structural problem as structural, not an invitation to individual self-invention but evidence that the wage system required collective dismantling and received six percent of the popular vote, which is not an asterisk in a history of American radicalism. The IWW was printing newspapers, organizing miners in Bisbee and harvest workers in the Palouse, building the theory and the practice of industrial unionism on a foundation that was the explicit negation of Wilmans’s individualism: what the displaced worker needed was not self-discovery but solidarity, not inward development but horizontal organization, and anyone telling you otherwise was doing the boss’s work. The Social Gospel movement was arguing that the Sermon on the Mount had opinions about the labor question and that the church had obligations to the structurally unemployed that charity alone could not discharge. Du Bois was writing in The Crisis founded 1910, reaching a hundred thousand readers within five years — that the “unique indwelling aptitudes” argument, applied to Black workers facing systematic exclusion from the labor market, was precisely the wrong prescription for a structural problem: that Booker Washington’s accommodationist version of the entrepreneurial work ethic, his counsel to prove economic utility within the existing racial order, was a trap, because individual excellence could not solve a problem that was collective and political in its origins. The argument between Du Bois and Washington is usually rendered as a debate about strategy, but it was more fundamentally a debate about the right unit of analysis, whether the self or the system was the correct scale for understanding displacement. Du Bois understood that the choice of scale was itself political: to insist on the individual as the relevant unit was already to have conceded something about where the intervention would happen.
Veblen was dissecting the leisure class with surgical contempt. Rosa Luxemburg was writing The Accumulation of Capital. The young Reinhold Niebuhr was in Detroit watching Ford’s welfare capitalism perform its contradictions in real time, watching the company that professed Christian concern for workers break the early UAW with methodical brutality, and drawing conclusions about the relationship between institutional self-interest and moral rhetoric that would become Moral Man and Immoral Society and then the intellectual foundation of mid-century social democracy.
All of this was happening simultaneously, in genuine tension, competing for the allegiance of workers who were trying to understand what was happening to them. New Thought won commercially — better retail, faster feedback loop, no waiting for the revolution, but it did not win by default. It had to contend with opponents who understood the structural problem as structural and had built organizations around that understanding. The labor movement had better politics for the scale of the problem. What they built together in the thirty years of labor’s patient, contested, frequently violent accumulation of institutional power between the Taylor hearings and the New Deal, was the architecture that managed the transition. Not the entrepreneurial work ethic. The labor movement. The Social Gospel. The political left in its various and competing forms. The recognition, fought for and defended against enormous resistance, that displacement at this scale was a collective problem requiring collective response.
Karl Polanyi’s The Great Transformation, published in 1944 as he watched the consequences of the 1930s from exile, theorized this dynamic with a precision that the current moment makes newly urgent. Polanyi’s argument was that the self-regulating market is not a natural phenomenon — it is a political construction, built through deliberate decisions to commodify labor, land, and money, to subject domains of human life to market logic that were previously organized around different principles. The market’s advance is always dislocating: it destroys the communities, the craft traditions, the social relationships that people depend on. But Polanyi observed that this penetration reliably provokes a counter-movement, not ideologically coherent, not organized by a single class, but generated by the sheer fact of dislocation. Conservatives protecting traditional communities, socialists organizing workers, churches claiming obligations to the displaced, farmers defending subsistence; all of them, with nothing in common except that the market was destroying something they needed, generating the institutional resistance that eventually embedded the market in social constraints. The New Deal was the counter-movement. Labor law was the counter-movement. The welfare state was the counter-movement. None of it was inevitable. It was the product of organized people who refused to let market penetration be the final word.
The critical implication and Polanyi was writing this while the consequences of a failed counter-movement were visible across Europe is that when the double movement doesn’t activate, the outcome is not equilibrium. Displaced populations without adequate institutional protection don’t quietly accept their displacement. They become available for a different kind of politics. Polanyi did not think this was inevitable; it required specific political failures, specific betrayals by institutions that should have held. But it was the predictable consequence of markets penetrating faster than counter-institutions could form.
Now consider, by contrast, what the present transformation has produced intellectually and politically. Consider it honestly, which means sitting with the embarrassment of the comparison.
The optimistic version of the argument the one Altman’s “The Intelligence Age” gestures toward and Amodei’s essay spends fifteen thousand words not quite making, is that productivity gains from AI will generate wealth significant enough that the distribution question is really about mechanisms of sharing, about getting the institutions right so that the abundance flows outward rather than concentrating. This is what Keynes believed in 1930 and what Silicon Valley believes now. But the distribution argument is worse than Keynes’s version, because Keynes’s failure was about redistribution, he assumed productivity gains would flow through an economy that still generated broad employment and consumer income, and was simply wrong about the political conditions required to make that happen. The current failure is happening at a prior step. Keynes at least imagined a pie growing large enough that the distribution question was about sharing it. What AI is doing to the knowledge economy is not obviously growing the pie. It is, in significant and measurable ways, eating existing industries and concentrating the proceeds.
The SaaS economy Salesforce, Workday, the hundreds of productivity and analytics and workflow tools that generated an enormous expansion of knowledge work over the last two decades is being disrupted not because AI created something new that complements it, but because AI agents are becoming capable enough that firms can build internal alternatives rather than pay subscription fees. SaaS firms drop prices to compete, compressing margins, forcing layoffs, workers replaced by the AI tools their former employers are now paying for companies funding the infrastructure of their own demise. The AI firms capture the licensing revenue while the incumbents that employed the knowledge workers get hollowed out. No new market is being created. An existing market is being absorbed, and the value embedded in it built by workers and engineers and companies over decades is being transferred to a handful of private firms that were not there when the value was created.
This is what makes the distribution argument not just insufficient but structurally incoherent at its root. The implicit premise of “things will probably work out” is that productivity generates wealth that can be distributed, given the right mechanisms. But the mechanism of value formation in the AI transition is itself the problem. The firms most positioned to capture the gains from displacing the knowledge economy are private, built on training data extracted without compensation from the workers whose output made the models possible, structured to ensure the people who generated the underlying value have no ownership claim on what was built from it, designed to remain private long enough that value is fully captured before any public participation mechanism becomes relevant. Previous waves of capital concentration at least had IPOs. You could own Standard Oil and then AT&T and then Microsoft and then Salesforce. The wealth was concentrated, but it was at least nominally accessible, and the industries that formed around these companies expanded employment even as they concentrated ownership. The specific firms extracting value from the AI transition are not accessible in this way, were built on inputs, the collective cognitive output of the professional class, the creative labor of writers and coders and analysts — for which those inputs’ owners received nothing, and operate in a regulatory environment largely staffed by their own alumni. Gorz’s question about whether productivity gains would flow outward as broadly shared leisure or upward as concentrated profit was always a political question. Here it has been substantially pre-answered by the ownership structure before the political process is even able to form.
Sam Altman’s “The Intelligence Age” is three pages long. It describes AI as the next thing that happens, as natural and inevitable as fire or electricity, the distributional question not answered but structurally absent, not a problem Altman has considered and resolved, but one that has not registered as a problem requiring consideration. David Noble documented in Forces of Production how numerically controlled machine tools were chosen over record-playback systems in postwar American manufacturing: both worked, both were technically viable, but one transferred control to management while the other left it with the machinists, and that was the choice that was made, not because it was more efficient but because of who held the decision and the lesson is that when a technology is described as inevitable, the interesting question is always who is doing the describing and what the alternatives were. There is no theory of who captures the gains from the productivity increase. There is no engagement with the empirical record, with Acemoglu and Restrepo’s documentation that automation already accounts for 50 to 70 percent of the growth in wage inequality between 1980 and 2016, more than unions, more than trade, more than market concentration and none with Acemoglu and Johnson’s argument in Power and Progress that technology produces shared prosperity only when counter-institutions force it to: that the productivity bandwagon has historically included workers not because markets are self-correcting but because unions, regulation, and democratic pressure made inclusion the path of least resistance, and that absent those institutions, the default is what they call the “so-so technology” trap, automation that replaces workers without generating sufficient new demand to compensate, producing permanent harm rather than transitional pain. There is no Debs, no Wilson, no adversarial encounter with someone whose job is to be skeptical. There is a vibe. The vibe is that it will probably be fine. Altman will give you his personal number for questions.
Dario Amodei’s “Machines of Loving Grace” is more serious — fifteen thousand words, a genuine attempt to think through what powerful AI might produce, written with evident care and some humility about the limits of prediction. He even acknowledges distribution as a concern. The acknowledgment occupies perhaps three paragraphs in fifteen thousand words and consists mainly of the assertion that things will probably work out, which is what John Maynard Keynes said in 1930 when he predicted fifteen-hour workweeks by 2030 and called the primary challenge of the coming century the management of abundance he was right about the production trajectory, catastrophically wrong about distribution, not because he was a bad economist but because he had trained himself out of the capacity to think politically about who made the choices and captured the gains. Amodei named his essay after a Richard Brautigan poem about a “cybernetic meadow where mammals and computers live together in mutually programming harmony”. Brautigan, Beat-adjacent, 1967, the counterculture at its most utopian and whatever else you want to say about Amodei, the choice of register is precise. The aesthetic vocabulary of liberation is borrowed to announce the dissolution of what workers depend on, by the CEO of one of the companies doing the dissolving. Thomas Frank documented in The Conquest of Cool how Madison Avenue in the 1960s discovered that the counterculture could sell cigarettes. Silicon Valley has discovered it can sell the future of work.
What they share Altman and Amodei and the broader AI discourse that orbits them —is a specific and telling absence: any engagement with the workers. Not as an abstraction, not as a category to be retrained and managed, but as people whose planning, whose educational investments, whose sense of self-worth and social legibility were built on a structural fact that cognitive judgment was the kind of work the machine could not do, that is now, suddenly and without their consent, contested. The discourse talks about workers the way you talk about a UX problem. “Human in the loop” is the phrase of art, and it is a phrase that reveals everything: the human has been grammatically demoted, from agent to component, from the person doing the work to the residual function the system still requires before full automation is technically feasible. The project is not secret. The disdain for having to include the human at all, the impatience with the friction they introduce, surfaces constantly in how the deployment is discussed internally and occasionally in how it is discussed publicly, in the confident predictions about which professions will be “disrupted,” in the eagerness to demonstrate that the model can do the work better and faster, in the barely suppressed irritation when someone points out that the people whose work is being replicated have not been compensated for the training data their labor produced. There is a specific Silicon Valley quality to the moral reasoning here, which is that the technology is so obviously and enormously good that concerns about the people affected by it are essentially a form of selfishness and a failure to appreciate the magnitude of the improvement, a parochialism of the currently employed.
Ford had Bennett’s enforcers, but he also had Niebuhr watching. He had the UAW. He had the National Labor Relations Act. He had an adversarial political environment that he could not simply opt out of by doing podcasts. The accountability was structural, not personal. It did not depend on his goodwill or his guests’ willingness to ask hard questions.
To understand what is being lost, precisely & not abstractly, not as economic category but as the specific thing that workers actually had and are having taken, you need Harry Braverman’s Labor and Monopoly Capital, published in 1974, widely read, absorbed into the academic left, and now substantially forgotten in exactly the circles where its argument would be most useful.
Braverman’s insight was about what Taylor’s real innovation had actually been. Not efficiency .Taylor’s detractors and admirers both misread him on this, but the systematic separation of conception from execution as a deliberate strategy of labor control. The craftsman held in their body and tradition the knowledge of how to do the work. Management could not easily replace or fully surveil craft workers because management did not know what they knew: the feel of the material, the judgment about tolerances, the accumulated experience of what went wrong and how to correct it. The craft knowledge was, in a real and not merely sentimental sense, owned by the worker. Taylor understood this as a problem — not because he was uniquely villainous, but because he was looking at the same production process as Wilson and seeing something different. Time-and-motion studies were about extraction: codifying the knowledge that lived in the worker’s body, transferring it into management protocols, rendering the individual worker interchangeable. Once the knowledge of how to do the work belonged to management, it could be used to surveil, discipline, and ultimately replace the person who previously was irreplaceable because they held it.
This pattern repeats across every subsequent wave of automation, which is why Braverman remains analytically essential despite his specific predictions aging. The word processor extracted from secretarial work the judgment about correspondence and formatting that had made skilled secretaries necessary, encoded it in software, distributed it to anyone who could navigate a menu. The spreadsheet extracted the contextual knowledge of the bookkeeper and put it in formulae. Each time, the extraction preceded and enabled the replacement. Each time, what workers held in their bodies or their heads became a protocol, and protocols don’t require the person who generated them.
But all of these extractions operated within a categorical boundary. Mechanization could extract physical work but not cognitive judgment. Early computing could extract routine cognitive work but not contextual reasoning, not the integration of ambiguous evidence, not the tacit knowledge of what an expert recognizes before they can say why. The entire history of the work ethic’s adaptation to displacement is built on this boundary holding, because the escape from each previous wave was always upward: manual to clerical, clerical to professional, routine professional to advisory and creative. The ladder went somewhere because at every transition there was a tier of work that was, genuinely, beyond what the machine could do. Braverman’s critics in the 1980s were not wrong that computerization created new skilled roles even as it destroyed old ones, it did but their argument depended on this boundary persisting.
The training of large language models is, at the level of Braverman’s analysis, the same extraction operation performed at a depth that previous iterations could not reach. RLHF is reinforcement learning from human feedback it works by having human annotators evaluate outputs, producing preference data on which the model is fine-tuned. The annotators are structurally in the position of Taylor’s workers during the time-and-motion studies: their judgments are being observed, codified, used to train a system that will replicate those judgments without needing to employ the people who made them. What Taylor extracted with a stopwatch — the optimal physical movement, the efficient action sequence.RLHF extracts with a preference dataset: what James Scott would call metis: the practical, embodied, locally specific knowledge that resists full codification, that exists in the worker’s judgment precisely because it accumulated through doing rather than through instruction.The sense of appropriate register, the recognition of what good output looks like before you can say why. The model learns to replicate not just the execution of tasks but the conception of how tasks should be executed, which is exactly the tier Braverman identified as the last refuge. And unlike previous waves of automation, this one is not targeting a single industry or a single tier of work. It is hitting cognitive labor across domains simultaneously law, medicine, finance, software, writing, analysis at a speed that makes the sequential absorption of previous transitions structurally impossible. There is nowhere to be absorbed into. The lateral move is not available because everything is moving at once.
There is a scene in Industry — the HBO series about junior analysts at a London investment bank , where Harper, having made a catastrophic error on a trade, is told by her senior that the bank will cover it this time, but she should understand what this means: she has used whatever protection she had, and she is now exposed. The scene is bleak in a specific way, not the bleakness of exploitation Harper is well-paid and knows it, but the bleakness of a person who has completely absorbed her own disposability as background fact. She does not appeal to principle. She does not argue. She has been formed by an environment in which the question of what her labor is worth is answered entirely by whether it is cheaper than the alternative. The show is set in the present tense, before the full deployment of AI in finance, and it depicts with unusual precision the workers who will be first in line when that deployment arrives: people who have internalized the conservative work ethic so completely, you owe relentless performance under whatever conditions are imposed, your continued employment is entirely contingent on output..that they have no framework for collective response, no language for structural grievance, no union, no Debs, nothing but the individual and the deal and the terror of being dropped.
Simone Weil understood this before anyone theorized it. In 1934 she took leave from her philosophy post and went to work on the factory floor at Renault not to observe but to understand from inside what industrial labor was doing to the people performing it. Her factory journals describe something that Braverman’s structural analysis, for all its precision, does not quite reach: what the extraction of conception from execution does to the inner life of the worker. Not just their economic position, not just their bargaining power, but their capacity for attention, the sustained, directed engagement with a problem that is the precondition for thought itself. The factory, organized to minimize the time between motions, to reduce each action to its most mechanical and legible form, produced in her something she described as a kind of affliction: not unhappiness exactly, but the progressive destruction of the ability to think, to plan, to maintain the thread of an intention across time. She was a trained philosopher and it still did it to her. What she was experiencing was the phenomenological consequence of Braverman’s structural argument: when conception is separated from execution, what remains is not just worse-paid or less secure. It is less. Anderson’s Hijacked is trying to recover this claim from political philosophy. Weil made it from a factory floor in Paris in 1934, and the journals have a weight that philosophical reconstruction cannot replicate. The thing being lost in the current extraction is not only economic. It is the thing Weil was trying to name.
What the Cunningham thread shows Scott Cunningham, Baylor economics professor, methodologist, someone who knows Martin Weitzman’s prose well enough to commission an imitation of it, is what intellectual unsettlement looks like when the competing frameworks have been removed, when the only available scale is the self.
In March 2026 he posted about having spent 3.5 hours watching Netflix while Claude designed a research project, collected data, chose an estimator, and produced a paper he thinks could clear the desk at a top field journal. His response to this is genuine and not stupid and contains real vertigo. “Do I look at that event study he drew from his CS estimate and refuse to believe it bc he wrote it?” he posted. “Is this picture ‘true’? Like when am I obligated to believe it? Never?”⁴ He wonders whether the quality assessment of AI-written work suffers from repugnance bias in Al Roth’s sense, people rejecting the output not because it is worse but because they know what made it. He proposes an RCT: give people a hundred AI-written papers, tell them the authorship randomly, measure quality ratings. He is, throughout, a thoughtful person trying to find his epistemic footing.
What he does not wonder about: the graduate students. The people who spent years developing the craft now being replicated, who trained on the literature and accumulated its tacit knowledge and learned the difference between a well-specified instrument and a poorly specified one through the expensive, time-consuming process of being wrong in seminars and whose career timeline just compressed, not as a philosophical puzzle about Cunningham’s epistemic obligations but as a material fact about the job market they are entering. The productivity gain from the Netflix afternoon is real; Cunningham is honest about it. The question of where that gain goes, who captures it, what institutional structure would need to exist to distribute it rather than concentrate it does not appear in the thread. Not because Cunningham is indifferent but because the conversational form available to him routes genuine unsettlement into questions about the self. His bias. His obligations. His epistemology. “When am I obligated to believe it?” rather than “who is being asked to bear the cost?”
This is the entrepreneurial work ethic’s deepest and most durable achievement and not that it convinced people displacement is opportunity, though it did that too, but that it made the individual the only available scale for measuring anything at all. Baker treats this as an ideological operation, recurring and powerful but in principle contestable, a story people are told, which means in principle they could be told a different story. Wendy Brown’s argument in Undoing the Demos is harder and more troubling: neoliberalism has not merely circulated an ideology but reconstituted the subject. Every domain of life: education, healthcare, civic participation, professional development, self-understanding has been made legible only through market rationality, restructured at the institutional level so that homo oeconomicus is not what people are told to be but what the institutions around them can process. The university is an enterprise. The discipline of economics is organized around individual optimization. The grant system, the journal, the tenure case, the seminar: all of it runs on the same operating system. Cunningham’s vertigo expressing itself as self-interrogation is not a failure of imagination or political will. It is the correct response to an institutional environment that has no surface on which a different grammar could be practiced. The competing frameworks that existed in 1912 required institutions to sustain them: newspapers, unions, churches, political parties with genuine mass membership. Those institutions have been systematically defunded, deunionized, and in many cases actively dismantled over the same forty years whose intellectual history Baker is tracing. The absence of the alternative grammar is not accidental. It was produced. Du Bois did not ask whether his assessment of Jim Crow labor markets might suffer from repugnance bias. The IWW did not wonder whether its opposition to Taylorism was partly a failure of openness to new management techniques. They had the structural analysis already and were arguing about what to do with it. The loss is not only the institutional response that never materialized. It is the competing frameworks, the vocabulary for talking about collective problems at collective scale, the theory that the self is the wrong unit of analysis, the organizations that trained people to think structurally about their own displacement that made the institutional response conceivable in the first place. Without Debs, no New Deal. Without the IWW’s model of industrial unionism, no Wagner Act. Without Du Bois’s insistence that individual excellence could not solve a structural problem, no... well, the structural problem remains unsolved, which is its own argument.
And we are sitting here, in 2026, with the discourse about the most significant transformation of work since industrialization being conducted through podcast networks controlled by the companies doing the transforming, through regulatory processes staffed by the alumni of those companies, through the genre of the long-form essay where a CEO can spend fifteen thousand words on the production side of the equation and three paragraphs on the distribution question and receive applause for the seriousness of the engagement. Taylor at least had to go to Washington. He at least had to sit across from someone whose job was adversarial. He had Wilson, who was frustrated and partially wrong and ultimately insufficient but Wilson named the problem in public, with consequences for people who refused to engage with it seriously, in a form that created a record.
The workers who are being displaced right now are not stupid and they are not passive and they are not without rage. What they are without is the thing the 1910s had: an intellectual and political framework adequate to the structural scale of what is happening, built by people who understood as Braverman understood in 1974, as Gorz understood in 1980, as Du Bois understood in 1903 is that the individual is the wrong unit of analysis for a problem that is collective in its origins and must be collective in its response.
Elizabeth Anderson’s Hijacked is a rescue operation: recover the progressive work ethic from its conservative captors. It is a serious and careful book and it arrives at exactly the wrong moment, arguing for the preservation and democratization of meaningful work in an economy where the object of the argument is in the process of being dissolved, not reformed, not recaptured, but made structurally irrelevant by systems trained on the collective output of the people whose work is being replaced. Anderson’s deskilling critique that the capitalist firm “aims to minimize labor costs by reducing labor to a commodity... replacing nuanced expert judgment with uniform protocols, reducing workers to interchangeable parts” is accurate as a description of what Taylorism did to craft labor and what corporate restructuring did to professional labor. It presupposes, though, that the deskilling is unnecessary from a production standpoint, that the professional judgment being destroyed is genuinely valuable in ways the price system cannot measure, such that restoring it would be both just and efficient. The AI case complicates this normative claim in ways that are uncomfortable to state plainly: the model doing document review in seconds is not, in a significant range of cases, producing worse output than the junior associate. The argument for preserving and rewarding human judgment cannot rest on the claim that it always produces better outcomes. It has to rest on something else, on the value of the practice to the practitioner, on the claim that the development of human capacities through meaningful work is a social good the price system cannot measure, on Gorz’s “autonomous activity” and that argument, defensible as it is, is an extremely hard one to make in an economy whose institutions are organized entirely around the price system’s authority, especially when the political actors who might impose alternative metrics have been systematically defunded, deunionized, and in many cases actively recruited into the industry whose deployment decisions most need scrutinizing.
The moral vocabulary of work is not disappearing. It is being transferred. The founder creates. The investor deploys capital with vision and discipline. The model generates. The “human in the loop” curates, which is a job description for someone whose function is to approve the machine’s output until the machine no longer requires approval. The craft, the professional judgment, the accumulated tacit knowledge, these are being extracted, codified, and replicated; the people who held them are being told, in the same language Wilmans used in 1899 and Csikszentmihalyi used in 2004, that this is their invitation to discover what is irreplaceably human about themselves. The vocabulary is of liberation but the function is demotion.
Weitzman’s fat-tails argument developed for climate risk, where standard expected-value calculations fail because the worst outcomes are genuinely irreversible applies here as structure rather than metaphor. You do not need to believe AI automates all cognitive work. You need only accept that the uncertainty about the escape hatch is substantially greater than it was in previous transitions, and that the cost of treating this as a normal adjustment — building individual reskilling programs, writing books about recovering the progressive work ethic, publishing careful essays about what Keynes got wrong about abundance, while the political window for structural response closes is not measured in years of transitional pain. It is measured in what Gorz watched happen to the deindustrialized communities, in the political geography of the last decade, in the populations for whom the promise of the work ethic failed and no alternative framework was provided and who were therefore available for mobilization by whoever could most convincingly promise to name and restore what was lost.
Baker ends with three taxi drivers across a century.Harold Lloyd’s Speedy, entrepreneurial and resourceful in 1928; Travis Bickle, his work ethic curdled into ideology and violence in 1976; Damani, queued in the gig economy with her anthropology degree and her rideshare app in 2023. The progression is not triumphant. It is a history of a promise that kept failing to be fulfilled and kept being reissued, in updated vocabulary, to people who had fewer and fewer alternatives to accepting it.
The version of this essay that ends with a rescue project, with a call to recover the progressive work ethic, to reform the regulatory architecture, to distribute the gains through better institutions would be a more comfortable essay to write and to read. The honest version has to acknowledge that the rescue project requires the rescuers to exist: requires an organized political constituency with the will and the institutional capacity to impose collective solutions on a structural problem, requires the intellectual infrastructure to name what is happening at the right scale, requires something that looks less like a podcast and more like Wilson’s hearing room. The Taylor hearings were not sufficient. Wilson was frustrated and partially wrong and ultimately insufficient: but the hearings were a precondition for what came after, for the thirty years of labor’s patient and contested accumulation of institutional power that eventually produced the architecture capable of managing the transition.
What names the problem publicly, right now, in an adversarial and accountable form, with consequences for people who refuse to engage with it seriously? The podcast? The Substack? The long-form essay by a CEO who names it after a Brautigan poem and distributes it through his own newsletter?
The charlatanism of the tech discourse is not even interesting charlatanism. It is charlatanism by subtraction: not the active deception of selling something you know to be false, but the passive deception of declining to ask the questions whose answers would be inconvenient, of mistaking the podcast for the hearing room, of writing fifteen thousand words about production and three paragraphs about distribution and calling it a serious engagement with the future of work. Taylor was not worse than Altman. He was more honest about the adversarial nature of what he was doing, which is why Wilson could subpoena him. The current version performs benevolence “we love humans, we just find them expensive” is the honest translation of the “human in the loop” literature and the performance is so thorough that the performers have, in many cases, convinced themselves. Belson was not lying. That was the joke. That is still the joke. It has just stopped being funny.
In the 1910s there was something to fight with and something to fight against. The entrepreneurial work ethic had to contend with Debs and the IWW and Du Bois and Niebuhr and the Social Gospel and the labor movement building its institutions brick by contested brick, and out of that contention — partial, violent, frequently betrayed, came the architecture that managed the transition. The architecture was not the inevitable product of good intentions. It was the product of organized people who understood the structural nature of the problem and refused to let it be translated back into individual terms.
What we have now is the translation without the refusal. The vocabulary of liberation borrowed to announce the dissolution. The podcast where the hearing room should be. The market rewarding the company that cuts 40 percent of its workforce on a profitable quarter, and the next CEO watching the stock price and drawing the obvious conclusion. The workers whose career timelines are compressing not as philosophical puzzles about epistemic obligation but as material facts about the job market, asking themselves as Cunningham asked himself, as the only scale available, what they are obligated to believe, whether their resistance is just bias, what is irreplaceably human about themselves. Wilmans asked that question in 1899. The answer then, as now, depended entirely on whether there was anyone building the thing that the individual answer could not provide.
There isn’t. Not yet. Possibly not in time.
⁴ Scott Cunningham (@causalinf), Twitter/X, March 5, 2026.
Baker, Erik. Make Your Own Job: How the Entrepreneurial Work Ethic Exhausted America. Harvard University Press, 2025.
Anderson, Elizabeth. Hijacked: How Neoliberalism Turned the Work Ethic against Workers and How Workers Can Take It Back. Cambridge University Press, 2023.
Braverman, Harry. Labor and Monopoly Capital: The Degradation of Work in the Twentieth Century. Monthly Review Press, 1974.
Brown, Wendy. Undoing the Demos: Neoliberalism’s Stealth Revolution. Zone Books, 2015.
Gorz, André. Farewell to the Working Class. Trans. Michael Sonenscher. Pluto Press, 1980.
Acemoglu, Daron, and Pascual Restrepo. “Tasks, Automation, and the Rise in U.S. Wage Inequality.” Econometrica 90, no. 5 (2022): 1973–2016.
Acemoglu, Daron, and Simon Johnson. Power and Progress: Our Thousand-Year Struggle Over Technology and Prosperity. PublicAffairs, 2023.
Frank, Thomas. The Conquest of Cool. University of Chicago Press, 1997.
Keynes, John Maynard. “Economic Possibilities for Our Grandchildren.” In Essays in Persuasion. Macmillan, 1931 [1930].
Noble, David F. Forces of Production: A Social History of Industrial Automation. Knopf, 1984.
Polanyi, Karl. The Great Transformation: The Political and Economic Origins of Our Time. Farrar & Rinehart, 1944.
Scott, James C. Seeing Like a State: How Certain Schemes to Improve the Human Condition Have Failed. Yale University Press, 1998.
Weil, Simone. “Factory Journal.” In Formative Writings, 1929–1941. Trans. Dorothy Tuck McFarland and Wilhelmina van Ness. University of Massachusetts Press, 1987.
Weitzman, Martin L. “On Modeling and Interpreting the Economics of Catastrophic Climate Change.” Review of Economics and Statistics 91, no. 1 (2009): 1–19.
Amodei, Dario. “Machines of Loving Grace.” darioamodei.com, October 2024.