Anthropic took Claude Code Review out of research preview on 15 October 2026. It is available to Team and Enterprise organisations, and once an admin enables it, reviews run automatically on every new pull request.
The headline numbers from Anthropic’s release notes: most reviews now finish in 6 to 10 minutes, down from around 20, so about three times faster. Findings are written to state the problem, what triggers it and the fix. Admins can set each repository to Quick, Standard (the default) or Deep. Developers do not have to configure anything. Anthropic has not published per-review pricing, only a credits offer for eligible Enterprise customers who claim it by 15 January 2027.
Why this keeps appearing in the news
This is the second major AI code review product to go general availability in a month, after GitHub Copilot’s in September (see our earlier signal). The pattern is clear. AI tools now write a large share of the code, and the bottleneck has moved to checking it. Vendors are answering that with more AI.
What it does well
Automated review is good at consistent, tireless first-pass checks: obvious bugs, missed edge cases, risky patterns, inconsistencies. It never gets bored on pull request number forty.
What it does not do
It does not know your business. It cannot tell whether the feature solves the right problem, whether a pricing rule matches your contract, or whether a data flow meets your compliance duties. A “Deep” setting is still a model reading code, not an accountable engineer signing off.
So what
When a development partner says their code is “AI reviewed”, treat that as a good sign and then ask three follow-ups: which parts get a human reviewer, who is accountable for architecture and security decisions, and what happens when the tool and the human disagree. Those answers matter more than the tool name.
That is how we structure AI-assisted development at BuildApps: automated tooling for volume, named people for judgement. If you want a project scoped with that balance built in, start a project.