We review irreversible decisions first (e.g. schema changes). Once there's agreement on that, tools like CodeRabbit or Claude review the code, and humans focus mostly on whether the behavior is correct.

We also try to build guardrails so bad code is hard to write in the first place: CI checks, custom lint rules, repo guidelines (e.g. CLAUDE.md), etc.