Current outcome
Anthropic disclosed that roughly 90% of Claude Code's code is now written by Claude and that it has rebuilt code review around multi-agent AI checking with risk-tiered human approval, while capping runaway review loops after one change went through 50+ rounds in a single night.
Progress timeline
2 material updates- #01
How the Claude Code team does code review
Anthropic's Claude Code team published a post describing how it runs code review at scale, reporting roughly 30,000 pull requests per day, with most of the path automated and human attention directed to high-risk changes. Low-risk changes — docs, copy, internal-only features and additive changes behind a flag — require no human approval once automated review and security checks are green. According to the team's own framing, the pipeline is designed to send scarce human review attention only where the risk is, while the reviewer's eval suite grows with every incident and every change to the reviewer is scored against all of it. The ~30k PRs/day figure and the low-risk categories (docs, copy, internal-only features, additive changes behind a flag) come from the team's own blog post rather than independent measurement.
Source evidence: pretext
- #02
Claude Code开始让AI审查AI：90%代码由Claude编写，低风险修改无需人工批准
Anthropic disclosed that roughly 90% of Claude Code's code is now written by Claude and that it has rebuilt code review around multi-agent AI checking with risk-tiered human approval, while capping runaway review loops after one change went through 50+ rounds in a single night.
Source evidence: theblockbeats