Claude Code Team Automates Most Code Review

This is less about eliminating humans than redefining review as an automated control system with humans focused on judgment, accountability, and risk. The approach is compelling, but only teams with strong evals, telemetry, and rollback discipline can safely copy it.

• –Every pull request receives AI review against the full repository, with separate agents verifying findings to reduce false positives.
• –Risk tiers determine whether changes need zero, one, or two human approvals; agent approval never substitutes for required human sign-off.
• –Incidents and reverts become daily evaluation cases, creating a feedback loop for improving the reviewer.
• –Anthropic had to cap review rounds and switch to diff-only re-review after agents began producing endless fix-and-review loops.
• –The practical takeaway is to automate routine correctness checks while keeping humans responsible for scope, architecture, sensitive code, and outcomes. [Docs](https://support.claude.com/en/articles/14233555-set-up-code-review-for-claude-code)