How the Claude Code team does code review

The post is dated Oct 15, 2026 and written by Addy Osmani, Morgan Westlee Lunt and Qing Wang. Most review now happens before a pull request exists: Claude runs tests, build and linter, then a second agent reviews the diff in a fresh context. A classifier then sets a risk tier. Low-risk changes need no human approval, medium-risk changes need one, and high-risk code or sensitive paths need two. One PR hit 50+ review rounds without converging. The public product came first. Code Review launched on March 9, 2026, and the share of internal PRs getting substantive comments rose from 16% to 54%. It is now generally available on Team and Enterprise plans, with Quick, Standard and Deep settings per repository. In Anthropic's broader monorepo, daily changes grew from about 2,500 in May to more than 30,000 in September. For AI founders, the takeaway is that review capacity, not code generation, becomes the constraint once agents write most of the code. The team's stated target is the least human review time that does not raise reverts or incidents. The numbers are self-reported, and the post does not publish underlying revert counts or a baseline. The team also says it lacks circuit breakers that tighten approval when metrics move.