How the Claude Code team does code review / claude.dev Blog

Why it matters

Anthropic's engineering team explains how it reviews code written by Claude, which the company says now produces most of Claude Code's own code. As agents generate changes far faster than humans can read them, the team rethought review as a continuous process rather than a gate at the end: Claude verifies its own work by running tests, builds, and screenshots, and a subagent examines each diff with fresh context before a pull request even exists. Every pull request then receives a Claude review against the whole repository and the change's stated intent, while other agents vote on findings so only verified issues reach humans. Instructional files guide what the reviewer checks. The team tracks human approvals, comments, and wait times, with revert and incident rates as guardrails, aiming for pull requests that merge with at most one human involved. The public tool, Code Review for Claude Code, is generally available on Team and Enterprise plans, with a lighter local version.