AI coding tools aren’t a new abstraction layer. I think that’s why the productivity gains aren’t showing up

Recent studies from Anthropic and METR show AI coding tools may actually harm developer productivity—lowering code comprehension by 17% and slowing experienced developers by 19%—despite developers believing they were faster. The author argues this is because AI code generation lacks the key properties of true abstraction layers: determinism, automatic verification, composability, and a precise intent language. A genuine abstraction would let developers define behavior through types and formal constraints while AI handles implementation, with automated verification replacing manual code review.