Anthropic merges 30,000 code changes a day: Claude reviews first, humans check evidence
dotey · x · 2026-11-05
Addy Osmani, who leads developer experience for Claude Code at Anthropic, revealed at an 18-minute AI Engineer fireside chat that daily merged code changes grew from 2,500 in May to over 30,000 by September.
- Claude reviews first: each change is used and audited by Claude like a real user, with issues auto-fixed
- Humans verify evidence, not code: engineers watch browser session recordings and cross-device screenshots instead of reading lines
- Verification is the bottleneck: giving AI a way to check itself improves output quality 2-3x, he argues
- The claude.ai redesign (3x faster) merged 3,000 PRs in a short window, involving designers and non-engineers
Osmani spent 14 years at Google on Chrome DevX before joining Anthropic in September; questioner Manish Kapur is at Sonar.
More from coding & agent
- Grok asked for the best coding harness recommends rival Claude Code — mhdfaran · 2026-11-05
- Agent benchmark scores hinge on container configs: 6-point swing from resource limits alone — krishnan · 2026-11-05
- Revolut cut AI coding inference costs ~90% by self-hosting GLM on 56 B200s instead of Claude — sebkrier · 2026-11-05
- VibeCoding a 3D fishing game in 4 days with Kimi K3.1 Pro and DeepSeek: full workflow breakdown — lxfater · 2026-11-05
- Agent eval design: testing partial updates vs full snapshots with a truck dock case — jylusdev · 2026-11-05
- Sentry founder says AI agent 'junior' handles hardest coding tasks well — zeeg · 2026-11-05