Using AI to speed up work erodes the skills that made the speedup possible

A new model shows that when workers use AI tools to boost productivity, they lose expertise over time — and that loss can eventually make them less productive than before they started using AI. Managers chasing short-term gains will rationally push workers into this trap, especially if the worker's skill doesn't matter much to the AI's output.

This is the first time someone has modeled the actual tradeoff: immediate productivity gains versus long-run skill decay. The model shows three concrete failure modes. One: a rational manager adopts AI, watches productivity spike, then watches it crater as the worker deskills — ending up worse off than day one. Two: if the manager cares only about this quarter's numbers, they'll deliberately push the worker into permanent deskilling. Three: workers with less experience can deskill to zero while experienced workers keep their edge, depending entirely on how much the AI depends on human judgment. The implication is brutal: the more a tool works without needing human expertise, the more it sorts workers into two classes — those who stay skilled and those who become replaceable.