Claude Code Review is now generally available

Updated: October 17, 2026, 17:04 CEST

What changed for GA

Code Review is out of research preview. Anthropic says most reviews finish about 3× faster: preview runs of ~20 minutes now typically take 6–10. Findings are shorter: problem, trigger and fix in a few sentences (Claude).

Admins choose Quick, Standard (default) or Deep per repository. Quicker levels cost less and finish sooner; Deep is the most thorough and expensive. Docs describe parallel agents reading the diff and surrounding code, a verification step that filters false positives, then severity ranking (Code Review docs).

How to turn it on

An Owner enables Code Review in Claude Code settings, installs the GitHub App and selects repositories. Reviews can then run automatically on new PRs (drafts and forks excepted). Anyone with write access can also comment `@claude review`. A review never approves or blocks a PR; merge stays with the team.

Enterprise credits

Eligible Claude Enterprise organizations can claim at least $10,000 in Code Review credits by January 15, 2027. Credits expire 30 days after claim; terms apply. Code Review is not available with Zero Data Retention, HIPAA configuration or customer-managed encryption keys yet.

How the Claude Code team reviews its own code

In an October 15 blog post, Anthropic says Claude writes about 90% of Claude Code's own code. From June to September, merged pull requests to the Claude Code codebase more than doubled, while the share that had to be reverted fell by more than 40% (claude.dev).

Review didn't go away; most of the reading moved to machines. Most changes are reviewed and fixed by Claude before a PR even exists. Every PR then gets an AI review plus a separate security review, and a PR can't merge while any finding is still open. When the research preview launched, the share of internal PRs getting substantive review comments went from 16% to 54%.

Human approval scales with risk. The change sets the tier, not the author: a classifier assigns it. Low risk (docs, copy, additive changes behind a flag): no human approval required once automated review and security checks are green. Medium risk (ordinary product changes, caching, the auto-updater, system prompts): one person. High risk: two people. Only trusted authors on the repo get the lighter paths, and an agent's approval never counts as a person's.

Review loops needed limits too: one CLI PR went through more than fifty rounds of review and fixes overnight. Now, after three rounds, the reviewer asks the author to step back and reconsider the approach. Why it matters: the goal isn't that nobody reads code, but that human attention goes to the risky decisions. If you're rolling out Code Review, start with one repo and a short REVIEW.md, then write down your own risk tiers.

On the developer-tools front, voice and agent APIs are also moving fast; we recently covered Grok Voice 3.0.