The Reproduction of Skill

In the AI Story

The senior software architect in The Orange Pill who felt like a master calligrapher watching the printing press arrive was mourning something specific. Not his job, not his income — a relationship. The particular intimacy between a practitioner and the thing practiced, built over twenty-five years of embodied engagement with systems whose behavior he could feel before he could articulate it. He could read a codebase the way a doctor reads a pulse: not through analysis alone but through somatic knowledge deposited by thousands of hours of direct contact. Berger would have recognized this mourning immediately, because it is the same mourning he documented among peasant farmers in the Haute-Savoie whose knowledge lived in their hands.

Berger's analysis of drawing illuminates the dynamic with particular force. In his essay 'Drawing on Paper,' he argued that the value of drawing lies not in the drawing produced but in the seeing that the act of drawing enforces. When you draw a face, you look at that face with an intensity and duration no other activity demands. The drawing that results may be crude. It does not matter. The seeing was precise, and the precision was produced by the friction of the hand's attempt to record what the eye observed. Remove the friction — give the artist a camera, or an AI image generator — and the artifact may be technically superior. But the seeing is gone, because the seeing was produced by the friction, and the friction was the point.

The Orange Pill describes an engineer in Trivandrum who spent eight years on backend systems and had never written a line of frontend code. Using Claude, she built a complete user-facing feature in two days. The book presents this as liberation — and it is liberation, in the same way the printing press liberated knowledge from the scriptoria. But liberation always liberates from something, and the question is whether the thing liberated from was purely a constraint or also a source of knowledge that the liberation destroys. The eight years of backend work included drudgery. It also included the moments when something unexpected happened — a configuration failure that forced her to understand a connection between systems she had not previously seen. These moments were rare, perhaps ten minutes in a four-hour block. But they were the moments that deposited the thin layers of understanding that, accumulated over years, produced the somatic knowledge that made her an engineer.

The most dangerous failure mode of working with AI, as Segal acknowledges, is confident wrongness dressed in good prose. The model produces a reference to Gilles Deleuze that sounds authoritative, reads beautifully, and is philosophically wrong. The error is caught because the author has read enough Deleuze to recognize the misuse. But the prose is so smooth that the error nearly passes. This is the reproduction of skill at its most treacherous: the surface is perfect, the artifact is polished, and the wrongness lives beneath the surface where only embodied knowledge — the knowledge built through years of friction — can detect it. And that knowledge is precisely what the tool's facility erodes.

Origin

The concept is developed in Chapter 2 of this volume, extending Benjamin's 1935 framework through Berger's writing on drawing, peasant knowledge, and embodied practice. It draws also on Harry Collins's work on tacit knowledge and on the extensive literature in cognitive science on the role of friction in skill acquisition.

Key Ideas

The artifact and the practitioner are both outputs of skilled work. Reproducing one does not reproduce the other.

The friction of making produces embodied knowledge. Remove the friction and the artifact survives; the knowledge does not.

The loss is not always visible at the individual level. A single person who uses Claude to write a frontend feature does not obviously lose anything. The loss accumulates across generations as fewer practitioners develop the embodied knowledge that friction once produced.

The detection of fluent error requires the knowledge the tool replaces. This is the circular danger: the users most qualified to catch AI errors are the ones whose qualification comes from not having used AI.

The question is not whether to mourn. Mourning is appropriate. The question is what new forms of seeing will be demanded, and what embodied knowledge will replace the embodied knowledge the tool has rendered unnecessary.

Debates & Critiques

The economic and technical arguments for AI-assisted development are serious and well-documented: productivity multiplies, barriers to entry collapse, the global distribution of capability broadens. Berger's framework does not dispute these gains. It insists they be counted alongside the losses — which are harder to quantify because they unfold across generations and do not show up on productivity dashboards. The debate, on the framework's terms, is not whether AI is useful but whether its use can be structured in ways that preserve the friction that produces embodied knowledge without sacrificing the gains that reducing friction makes possible. This is an institutional design question, not a technical one.