Claude Code’s team makes review continuous and scales human checks by risk

Anthropic reports that Claude Code’s merged pull requests more than doubled from June to September while the reverted share fell over 40%. Its process combines executable checks, agent review, risk-based human approval and gradual rollout. The team also caps review loops after agents spent over fifty rounds on one change. These are internal results, not proof that other teams should remove human review.

Anthropic · Oct 15, 2026

Why it matters

Agent throughput makes review quality, convergence and recovery as important as code generation.

More context

Try stronger verification and incident-based reviewer evaluations before loosening approvals. The public review product is generally available on Team and Enterprise plans; internal low-risk merge policies go beyond that product.

Keep in mind

Human ownership remains, sensitive paths need people, and the team lacks automatic approval-tightening circuit breakers.

Original read