Michael Crawford

In the [YOU] on AI Field Guide

[YOU] on AI documents the collapse of what it calls the imagination-to-artifact ratio—the distance between a human idea and its realization, which AI has compressed toward zero. When anyone with an idea and natural language can produce a working prototype in hours, the expansion of who gets to build is genuinely revolutionary. Crawford would not deny this. But his framework insists on a question the celebration tends to skip: What happens to the builder when the building requires no friction? The prototype arrives through a conversational interface. The builder's body has not engaged with the material. The builder's perceptual systems have not been calibrated by the specific feedback that only material engagement provides. The artifact exists. The understanding that would have been deposited by the struggle to produce it does not.

His concept of the circular vulnerability names the structural danger that most AI discourse misses entirely. The effective use of AI tools depends on the practitioner's capacity to evaluate the tool's output. The evaluative capacity depends on sustained engagement with the material of the work. The tool eliminates the engagement. Therefore the tool, over time, erodes the conditions for its own effective use. The senior engineers who currently evaluate AI-generated code developed their evaluative capacity through decades of writing code by hand. The junior engineers trained entirely through AI-mediated production will not develop equivalent evaluative capacity, because the mechanism that produces it—the friction, the failure, the specific struggle of understanding why something does not work—has been designed out of their workflow.

Crawford's ascending-friction thesis—his name for the Orange Pill's claim that AI relocates difficulty to a higher cognitive level rather than eliminating it—he accepts as a description and challenges as an evaluation. The relocation is real. But it assumes that the higher-level work is the work that matters, that judgment is more valuable than implementation. This assumption is the knowledge economy's fishbowl. From inside it, the ascent looks like progress. Crawford's claim is that the twenty percent—the architectural instinct, the taste, the judgment about what will break—was built through the eighty percent. If the eighty percent is eliminated for the next generation, the twenty percent is not transmitted, because it was not produced through instruction but through engagement, and the engagement is what the abstraction has designed away.

Where Merritt Roe Smith provides the historical framework for understanding how institutional contexts shape technology and Meredith Whittaker names the power structure that owns it, Crawford provides the phenomenological anatomy of what is being lost at the level of the individual practitioner—not productivity, not output, but the specific human experience of standing behind your work and being answerable to a world that does not grade on a curve.

Origin

Crawford completed a PhD in political philosophy from the University of Chicago in 2000 and took a position as executive director of a Washington think tank. He lasted five months. The work required summarizing other people's arguments without engaging them seriously—a form of intellectual production that felt, to him, profoundly dishonest. He opened Shockoe Moto, a motorcycle repair shop in Richmond, Virginia, where he still works. The shop was not a retreat from ideas. It was a discovery that certain ideas could only be tested in a particular way: against a machine that could not be bullied, charmed, or talked into starting.

His first book, Shop Class as Soulcraft: An Inquiry into the Value of Work (2009), began as an essay in The New Atlantis and became an unexpected bestseller, generating a public conversation about the cognitive and existential value of manual work that Crawford had not anticipated. The World Beyond Your Head: On Becoming an Individual in an Age of Distraction (2015) extended the argument into embodied cognition, attention, and the ecology of the skilled practitioner's perceptual relationship with her environment. His Senate testimony in 2021 identified AI governance systems as a new form of unaccountable power analogous to the administrative state. His 2023 lecture 'The Rise of Antihumanism' identified the tacit premises that legitimize replacing human judgment with automated systems: that human beings are stupid, obsolete, fragile, and hateful. And his 2025 essay 'Ownership of the Means of Thinking' framed the AI revolution not as a technological event but as a political-economic one—the extension of oligopoly logic into cognition itself.

He served as chair of the AI Ethics Council at the American Enterprise Institute beginning in 2026, where he brought his unusual combination of phenomenological precision and political economy to bear on the governance of AI deployment in professional domains.

Key Ideas

Genuine knowledge requires friction. Crawford identifies three characteristics that distinguish genuine knowledge from functional equivalents. First, it is grounded in experience—not in the thin sense of having processed information, but in the thick sense of having engaged with the subject through the body's full perceptual apparatus, under conditions of genuine uncertainty, where the outcome carried real consequences. Second, it is tested against an incorruptible standard—the motorcycle that starts or does not, regardless of the mechanic's confidence, credentials, or eloquence. Third, it is earned through friction—through the specific resistance that material reality offers to human intentions. AI-generated output lacks all three characteristics, and the absence is invisible in the commodity and catastrophic in the practitioner.

Replacism. Crawford names the metaphysical assumption underlying the cultural embrace of AI: replacism—the premise that every particular thing can be replaced by its standardized double, and thus made more amenable to machine logic. The assumption extends, in Crawford's analysis, from manufactured goods to human cognition itself: the substitution of artificial intelligence for human intelligence is simply a matter of swapping out carbon for silicon. Crawford argues this is false—that the mechanic's situated intelligence is categorically different from the pattern-matching intelligence of a language model, not merely slower or less efficient but different in kind.

The circular vulnerability. The effective use of AI tools depends on practitioner judgment. Practitioner judgment depends on sustained material engagement. AI tools eliminate the engagement. Therefore AI tools progressively erode the conditions under which the evaluative capacity they depend on is produced. This is not a paradox but a structural vulnerability: the system functions, outputs are produced, metrics improve, and the human capacity that separates functional output from genuinely good output thins with each generation trained through AI-mediated production rather than direct engagement with the material.

Authorship versus directorship. Crawford identifies the central psychological transformation of AI-mediated work as the shift from authorship to directorship. The author engages with material that resists, produces something through the specific exercise of embodied skill, and bears genuine responsibility for the quality of the outcome because the outcome depended on her specific judgment. The director specifies what she wants, evaluates whether she received it, and makes important decisions—but does not engage with the process through which the commodity was produced. Both are genuine cognitive acts. But they produce fundamentally different relationships to the work, different depths of understanding, different experiences of agency.

Ownership of the means of thinking. Crawford's most recent political framing: the AI revolution is an extension of oligopoly logic into cognition itself. The concentration of the training data, the computational infrastructure, and the interfaces through which AI-mediated cognition occurs in a small number of firms represents a structural transformation of who gets to think and on what terms. Deskilling, in this analysis, is not a side effect of AI but its business rationale—the hope that AI will substitute for human judgment and discretion, transferring the means of thinking from the practitioner to the corporation.