Anthropic's daily merged code changes jumped from ~2,500 to 30,000+; 7-person team rebuilt dev stack around agents
msharmas · x · 2026-11-03
Anthropic detailed how a 7-person team spent 5 months rebuilding its dev setup around Claude agents: daily merged changes went from 2,500 in May to 30,000+ by September, making human-oriented tools like git the bottleneck. Key practices: engineers act as design deciders and rule-enforcers while Claude writes most code; humans review intent rather than every line, with tests verifying; one Claude per component watches logs and charts in a dedicated Slack channel, catching issues normal alerts miss; testing happens after merge with Claude fixing conflicts. The tooling was replaced with agent-first infrastructure — agents request files from scion-cache (two cache layers, snapshot-based) instead of cloning repos, and changesets store only diffs in Spanner metadata with file contents deduplicated by hash in Cloud Storage. Includes an infographic by @undefinedKi.
More from coding & agent
- Sentry CEO shows AI agent bots flooding the repo with PRs: 'two humans in a sea of juniors' — zeeg · 2026-11-04
- OpenClaw cut its ~$1M/month CI bill by over 95% with adaptive agentic testing — vincent_koc · 2026-11-04
- Perplexity Computer can now run Meta ads, Klaviyo email and Shopify stores end-to-end — perplexity_ai · 2026-11-04
- Filarise shows finance MCP server letting ChatGPT query your bank via Plaid — ChampionSpecialist60 · 2026-11-04
- Multi-agent messaging barely helps: 61% of agreed runs still hit merge conflicts — EliasEskin · 2026-11-04
- worldinmotion: open-source MCP server for encrypted agent-to-agent messaging across machines, with no approve tool by design — distronode · 2026-11-04