Revue de code en couches : l'humain ne relit que le risque
Method described by [the Claude Code team](https://claude.dev/blog/solving-code-review-at-anthropic/) (Addy Osmani, Morgan Westlee Lunt, Qing Wang, October 15, 2026): stack automated checks, each in charge of one kind of error, so human review is limited to direction, risky code and architecture. The team states that merged pull requests more than doubled from June to September while the reverted share fell by over 40%, and that the mean number of reviews per internal PR went from 2.08 to 1.32. Internal figures reported by the team, not a benchmark, not independently verified.