AI doesn't replace engineers - it removes the knowledge ceiling

For two years I've been embedding AI tools across engineering teams - not just personally, but at scale. Custom prompts. Agent-driven CI. LLM-powered code review. The results aren't what most people predicted.

The lazy take is that AI replaces engineers. The honest answer, after thousands of merged PRs across @sociable-tech and @pitb-iep, is that AI removes the knowledge ceiling. There was always a gap between knowing what to build and being able to build it precisely. That gap is collapsing.

What changed, concretely

A junior engineer on my team shipped a Postgres migration with row-level security on her third week. Five years ago that would have been a senior task. Not because she suddenly knew RLS - because the tool let her ask the right questions in the right order, and her judgment about whether the answer made sense was the actual skill.

Across a 7-person team, we shipped what a 16-person team shipped in 2022. Same domain, same compliance, same client expectations. The geometry of work changed; the discipline did not.

"AI doesn't replace the engineer. It removes the knowledge ceiling - and that makes craftsmanship more important, not less."

What didn't change

Taste. Architecture. The ability to say no. The willingness to refactor instead of bolting on. None of these get easier when generation is cheap. If anything, the cost of bad calls compounds faster because you can ship them faster.

Good code is still an art form. AI just gives every artist a better brush.