Anthropic's monorepo went from 2,500 to 30,000 merges/day with Claude writing ~90% of code
ccerrato147 · x · 2026-10-29
Addy Osmani (Claude Code team) published a thread on how Anthropic reviews code when Claude writes about 90% of it. Headline stat: Anthropic's monorepo went from 2,500 merges a day in May to over 30,000 in September — nobody reads 30,000 diffs a day, so per-PR human review is dead.
The full system:
- Review moved before the PR: Claude must prove completion — run tests and builds, match screens to designs — then a fresh context-free agent reads the diff. Boris Cherny (Claude Code creator) puts the quality gain at 2-3x
- Machines arguing with machines: several agents hunt bugs in parallel, a vote throws out false alarms, style nits filtered; PRs getting a real review comment rose from 16% to 54%
- Ownership: you land it, you own it; bot approvals never count as human ones; risk, not author, sets review intensity
- Undo is the last line of defense: feature flags, staged rollouts, one-step reverts; June–September merged PRs more than doubled while the revert share fell over 40%; login/billing/build-release code always gets a human approver
The thread's cover art renders UNDO and CHECKS as temple-sized stone mechanisms — very memeable.
Related event: How Anthropic reviews code when Claude writes 90% of it(6 posts)→
More from coding & agent
- Jason Kneen shares slop_eisley 0.6.22 roadmap for his AI-built FPS game — jasonkneen · 2026-10-29
- Free Oct 29 Every livestream: Context Engineering With DSPy, open to all — danshipper · 2026-10-29
- Open-source lead gen agent Jev at $0.0024/lead; blogger ran a job hunt for $0.13 — JaynitMakwana · 2026-10-29
- Hands-on exercises and resources for GitHub's new Copilot features — jennifermarsman · 2026-10-29
- YC open-sources QM agent harness; E2B adds one-command sandbox deploy — badphilosopher · 2026-10-29
- Ploid launches a living index of 2 billion people built for AI agents — ycombinator · 2026-10-29