Anthropic takes its AI code reviewer out of preview, with repository-level controls

Anthropic has made Code Review generally available to Claude Team and Enterprise organisations, bringing automated, multi-agent pull-request reviews to GitHub repositories. The practical change is not just that AI can comment on code: administrators can choose how deeply each repository is reviewed, while people retain the merge decision.

Anthropic Watch analysis

What happened

Administrators enable the Claude GitHub App and select repositories. Reviews then run automatically on new pull requests, with settings for Quick, Standard or Deep review. Anthropic says most reviews now take six to ten minutes, down from about 20 during the research preview. That speed comparison is the company’s own figure.

The system’s agents inspect the code changes in parallel, then a verification stage checks candidate findings before the results are ranked and posted as comments. Reviews focus by default on correctness issues, rather than formatting preferences. Teams can adjust what gets flagged with repository guidance files.

Why it matters

This puts AI code review into a routine team workflow, where it may surface bugs before a human reviewer gets to the pull request. The controls matter: a team can use a lighter pass on ordinary repositories and a deeper one where the code warrants it, rather than asking every change to endure the same computationally enthusiastic inspection.

Reviews do not approve or block a pull request, and the check run does not fail it. A clean review is not a safety certificate either: Anthropic’s documentation warns the system can miss bugs and vulnerabilities. Teams still need human review and their usual security checks.

Our read

The useful proposition is assistance, not an automated sign-off. Repository-level settings and a verification step make this more considered than simply sending every change to a model and hoping for the best. But the company’s speed and accuracy claims are not independent results, and the product’s limits make the human decision-maker more than a ceremonial final click.

For eligible Enterprise organisations, Anthropic is also offering between $10,000 and $50,000 in review credits, according to Unite.AI and The Next Web. The offer has eligibility restrictions, and the credits run for 30 days after they are claimed, so organisations should check the terms before treating the headline amount as a free lunch.

What to watch

• Whether teams report useful bugs caught, rather than simply more comments generated.
• How review cost changes with repository size and the Quick, Standard and Deep settings.
• Whether Anthropic publishes evidence on missed bugs and false positives beyond its own use of the tool. Activity teaser Anthropic’s AI code reviewer is now generally available to Claude Team and Enterprise organisations. Teams can set each repository to Quick, Standard or Deep review, while the tool comments on pull requests without approving or blocking them. The key question is whether this catches worthwhile bugs without turning code review into a second inbox. Should AI reviewers be a routine first pass, or does adding another reviewer risk creating more noise than useful scrutiny?

Discussion spark: Should AI code review become a routine first pass on pull requests, or will the extra comments create more noise than useful scrutiny?

Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.