TL;DR: Anthropic reports Claude now authors roughly 90% of Claude Code's codebase, with merged PRs doubling from June to September as reverted-share dropped over 40%.
Summary: Anthropic's Claude Devs account shared that Claude writes about 90% of the code for Claude Code, its terminal-based coding agent. Between June and September, merged pull requests more than doubled while the share of changes later reverted fell by more than 40%. The team published a writeup on how it restructured code review to support this level of AI-generated contribution and how other teams can adopt the same process.
Why it matters: It is a concrete data point that high AI-authored code ratios can coexist with improving merge quality, not just rising review burden. Worth reading the code-review changes if you are scaling agent-generated PRs in your own repo.
Source: x_com