Claude Code team doubles merged pull requests from June to September while the revert rate falls by more than 40%, by moving most review to agents
Claude writes about 90% of Claude Code's own code, so the team rebuilt code review around human time. Claude verifies its own work and a fresh-context subagent reviews the diff before a PR exists; every PR then gets an AI review whose findings are cross-checked by other agents, plus a blocking security review and CI. Human approvals scale by risk tier set by a classifier: none for docs and flagged additive changes, one person for ordinary changes, two for the riskiest paths. A merge queue, feature flags, staged releases and fast rollback catch the rest, and past incidents form a daily eval for the reviewer. After one CLI PR went through more than fifty overnight review rounds, they capped rounds at three.
The numbers (as reported)
| Merged PRs | More than doubled from June to September |
|---|---|
| Revert rate | Share of merged PRs reverted fell by more than 40% |
| Substantive review comments | Share of internal PRs getting them went from 16% to 54% at the research-preview launch |
| Agent-written code | About 90% of Claude Code's own code |
Facts
- Company / subject
- Anthropic (Claude Code team)
- Tools
- Claude Code
- Codebase
- brownfield
- Work type
- Company rollout
- Stack
- —
- Published
- 2026-10-15
- Last checked
- 2026-10-26
What the source claims Vendor-published
Claimed by the vendor that sells the tool. Not checked independently.
- Merged PRs more than doubled from June to September while the share reverted fell by more than 40% 2x+ PRs; -40% reverts
“From June to September, merged pull requests to Claude Code's codebase more than doubled while the share that had to be reverted fell by more than 40 percent.”
- Share of internal PRs getting substantive review comments went from 16% to 54% 16% → 54%
“the share of internal PRs getting substantive review comments went from 16% to 54%”
Caveats
Vendor post about its own product, which also promotes Code Review for Claude Code. No absolute PR counts, incident numbers or cost. About review and merging rather than one build.
More stories by tool: Claude Code
Related topics: Company rollouts and costBrownfield codebasesVendor case studies
Source
claude.dev blog (Addy Osmani, Morgan Westlee Lunt, Qing Wang)