Add as a preferred source on GoogleClaude Code Merged Pull Requests More Than Double as Reverted Share Falls
From June through September, the share of merged pull requests that had to be reverted fell by more than 40%, as the team used automated checks and risk-based human review.
Anthropic’s Claude Code repository saw merged pull requests more than double from June through September as its review process adapted to higher code volume. Over the same period, the share of merged pull requests that had to be reverted fell by more than 40%.
Claude produces about 90% of Claude Code’s code. Before human review, the team uses Claude to check changes alongside automated security checks, tests, builds, linting and type checks.
Trusted authors can move low-risk changes forward without human approval when automated review and security checks are clear. Sensitive paths always require a person, and other authors need another person’s approval. Medium-risk changes require one human approval; high-risk changes require two. The engineer directing an AI agent remains accountable for its change.
Engineers do not maintain a complete mental model of Claude Code. When they need to understand a particular area, they ask Claude to explain it.
Across Anthropic’s broader monorepo, changes landing per day rose from about 2,500 in May to more than 30,000 in September. Claude Code is in a separate repository, and its team also faced pressure from the faster pace of changes. Anthropic’s Code Review product is available on Team and Enterprise plans, with details in the earlier rollout story.
Simon Yoon
ReporterSimon Yoon reports on blockchain technology for TokenPost. Send corrections or tips to info@tokenpost.com.