- Anthropic launched an AI-powered Code Review system to improve pull request reviews in software development
- Code Review uses multiple AI agents to detect bugs and provide feedback on pull requests automatically
- Internal tests showed Code Review increased substantive review comments from 16% to 54% on pull requests
Anthropic has introduced a new AI-powered Code Review system aimed at easing one of the biggest bottlenecks in software development: reviewing pull requests.
In a recent blog post, the company announced Code Review, which uses a team of AI agents to automatically review pull requests (PRs) and flag potential bugs before code is merged. The tool, currently available as a research preview for Team and Enterprise plan users, is “built for depth, not speed,” as per the company.
Why Anthropic Built an AI Code Review
The company says the feature emerged from a challenge: the code output per Anthropic engineer has increased by around 200 per cent in the past year. “Code review has become a bottleneck, and we hear the same from customers every week. They tell us developers are stretched thin, and many PRs get skimmed rather than deep reads,” it mentioned.
To address this, Anthropic built Code Review as an AI-powered system which runs “deep, multi-agent reviews that catch bugs human reviewers often miss themselves.”
The company also shared how the system has already changed its internal review workflow: “We run Code Review on nearly every PR at Anthropic. Before, 16% of PRs got substantive review comments. Now 54 percent do. It won't approve PRs — that's still a human call — but it closes the gap so reviewers can actually cover what's shipping,” the AI giant said in the blog post noted.
How The AI Review System Works
The system activates as soon as a pull request is opened.
“When a PR is opened, Code Review dispatches a team of agents. The agents look for bugs in parallel, verify bugs to filter out false positives, and rank bugs by severity. The result lands on the PR as a single high-signal overview comment, plus in-line comments for specific bugs,” the company explained.
The system also adjusts the depth of analysis depending on the complexity of the pull request. Larger changes receive deeper scrutiny with more agents involved, while “trivial ones get a lightweight pass.”
What Anthropic Found During Testing
Anthropic also revealed that it had been testing Code Review internally for several months and shared early observations.
“On large PRs (over 1,000 lines changed), 84 percent get findings, averaging 7.5 issues. On small PRs under 50 lines, that drops to 31 percent, averaging 0.5 issues. Engineers largely agree with what it surfaces: less than 1 percent of findings are marked incorrect,” the company wrote.
The blog also highlighted an example where the system flagged a one-line change in a production service that appeared routine but would have broken authentication.
The issue was identified and fixed before the code was merged. The company shared that its engineer shared afterwards that they wouldn't be able to catch such codes.
Another example mentioned in the post came from early users. During a ZFS encryption refactor in TrueNAS's open-source middleware, Code Review detected a pre-existing bug that was silently clearing the encryption key cache during every sync.
Pricing And Availability
Anthropic noted in the blog that Code Review is designed for deeper analysis and therefore comes with usage-based pricing.
Reviews are billed based on token usage and typically cost between $15 and $25, depending on the size and complexity of the pull request. This makes it more expensive than Anthropic's Claude Code GitHub Action, which remains open source and offers a lighter review option.
To help organisations manage spending, the platform includes controls such as monthly spending caps, repository-level review settings and analytics dashboards.
For now, Code Review is available as a beta research preview for teams using Claude Code's Team and Enterprise plans. Administrators can enable it by installing the GitHub app and selecting the repositories where automated reviews should run.
For developers, the process is simple: once the feature is enabled, Code Review automatically runs on every new pull request. This means no configuration is needed.
