Doctorow: Chatbots Can't Capture Workers' Process Knowledge

Cory Doctorow's latest Pluralistic post turns a familiar critique of management into a direct challenge for anyone selling AI replacement of human labor. His argument is that 'process knowledge', the hard-won lessons workers carry in their heads across firms and divisions, is systematically undervalued because it can't be listed on a balance sheet, and that the same blind spot is now driving confident bets on AI.

The line is pointed. The people who value process knowledge least, Doctorow writes, are 'the AI bros who think you can replace skilled workers with a chatbot trained on the things they say and write down, as though that somehow captured everything they know.' He puts them alongside 'business geniuses like Elon Musk' firing 'skilled lifers from key government agencies' and Trump assuming new factories can be stood up in response to tariffs, with 'training' backfilling a generation of missing onshore production knowhow.

The examples are deliberately unglamorous. Claudio, a career dishwasher, teaches a novice that the machine 'takes two full minutes to run a cycle' and that mismanaging idle time between loads is why shifts drag into the night. Marisol, a receptionist, warned architects that charts got pulled '20-30 times/day' and that patients couldn't be watched from her desk; her concerns were dismissed. Apple gets the counterexample slot: engineers who 'seized upon a lot of "binned" microprocessors that were headed to the landfill' and designed the MacBook Neo around them by running the salvaged chips slower.

The honest caveat is that this is opinion writing, not a study. Doctorow doesn't quantify how much tacit knowledge has been lost in the current wave of AI-driven layoffs, doesn't distinguish which job categories are most exposed, and doesn't engage with whether agentic or observation-based systems narrow the documentation gap. The takeaway for anyone selling or buying AI labor replacement is the plain one: if your model of a job is what workers write down, you're modeling a thin slice of it, and the parts you're missing tend to be the parts that decide whether the shop actually runs on time.