Anthropic: 80% of code is written by Claude, and CI jobs exploded 25x in six months
新智元 · wechat · 2026-09-18
In an official blog post, Anthropic disclosed that 80% of its code is now written by Claude, with engineers shipping 8x the 2021-2025 average per quarter. But AI coding patterns broke the pipeline: tiny fragmented PRs, 24/7 submissions, and 10x more tests drove CI runs up 25x in six months, crushing continuous integration.
Three rapid patches all failed: doubling cores lasted 70 days; per-package sharding lasted 29; daily restarts didn't survive a single day due to data loss. The root cause was a singleton write path in the test impact analysis service — a 20-minute Listener lag desynced tens of thousands of test states, causing bad merges and regressions. The team finally took Claude's own earlier advice and rebuilt it as a distributed stateless service with an in-memory store, async aggregation, and read-only decoupled selection, flattening a queue that had grown to hundreds of thousands of events.
The lesson: AI coding's strain has moved from "will devs lose jobs" to "can engineering infrastructure survive" — the bottleneck is now the system, not the code.
Related event: Anthropic Discloses Internal Metrics: Claude Now Drives 26% of Its R&D(41 posts)→
More from coding & agent
- Google Rolls Out Browser Use to All Antigravity Users, Raising Doubts About Gemini Desktop App — testingcatalog · 2026-11-12
- OpenRouter launches Extractions API: typed JSON field extraction from text, images and PDFs — michellechen · 2026-11-12
- AgenticRL-Bench debuts: best agents pass only 14% of real-world agentic RL tasks — cuijiaxun · 2026-11-12
- SemiAnalysis: Persistent agents now eat 45% of our AI spend, and Perplexity's harness beats Claude Tags — dylan522p · 2026-11-12
- Gemini Agent Spans Workspace, Mobile, Desktop, and Slack via @-Mention — samcharrington · 2026-11-12
- Carbon + /swarm in Antigravity: Big Model Thinks, Lighter Models Work — samcharrington · 2026-11-12