Practice - November 2025

AI in the daily loop

The useful question about AI on a dev team is not whether it writes code. It does. The question is where it sits in the loop, and who is still holding the pen. Get that wrong and you get plausible nonsense at speed, which is worse than no help at all.

What has worked for us is narrow and boring. AI drafts the tedious middle - the test scaffold, the migration, the third near-identical component - and a person owns the edges: the intent going in, and the review coming out. It is an intern with an enormous library and no judgement, and you would not merge an intern's work unread either.

Keep the taste human

The moment the team stops reading the output is the moment the codebase starts drifting.

So we treat generated code as a proposal, never a decision. It has to pass the same review as anything else, and the reviewer has to be able to explain it. That slows the flashy demo and speeds the actual work, because the bugs that cost you a week are the ones nobody understood well enough to catch.

None of this is exotic. It is the same discipline good teams already had about copy-paste and clever abstractions, pointed at a faster source of both.

All writing - CV