Claude Code's team lets low-risk PRs merge with no human

From June to September, merged pull requests to Claude Code more than doubled, and the share that had to be reverted fell by more than 40%. Anthropic's team kept review but handed most of the reading to machines.

A classifier reads each change and sets a risk tier. Docs and additive changes behind a flag merge with no human approval once automated and security checks are green. Ordinary changes and sensitive areas like the auto-updater need one person, the riskiest paths two.

Agents also had to learn to stop: one PR went through more than fifty review rounds overnight. After three rounds the reviewer asks the author to rethink the approach.

Circuit breakers that tighten approval when revert numbers move don't exist yet, the team admits. Would you let low-risk code merge on automated checks alone?