Anthropic has announced a new feature for Claude Code called Code Review. The company has said that whenever a pull request (PR) is opened, Claude dispatches a team of agents to hunt for bugs.
The feature is now in research preview for the Team and the Enterprise plans. Anthropic added that this new feature is more thorough than the existing Claude Code GitHub Actions feature, which can automate review and fix bugs on a GitHub workflow.
The agents ‘dispatched’ by the Code Review feature 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,” Anthropic said in the blog post.
The company also stated that the depth of reviews is scaled based on the PR, with large or complex changes getting more agents at disposal, whereas trivial ones get a ‘lightweight pass’.
“Based on our testing, the average review takes around 20 minutes,” Anthropic added.
Boris Cherny, the co-creator of Claude Code, said the code review feature was built first for Anthropic itself, which internally uses Claude Code at a large scale. “Code output per Anthropic engineer is up 200% this year, and reviews were the bottleneck,” he wrote in a post on X.
“We run Code Review on nearly every PR at Anthropic. Before, 16% of PRs got substantive review comments. Now 54% do,” the company further said. “It won't approve PRs—that's still a human call—but it closes the gap so reviewers can actually cover what's shipping.”
In its internal testing, Anthropic said that if a one-line change to a production service looked routine, a diff like that would normally be approved quickly. “But Code Review flagged it as critical. The change would have broken authentication for the service, a failure mode that’s easy to read past in the diff but obvious once pointed out,” the company said.
“It was fixed before merge, and the engineer shared afterwards that they wouldn't have caught it on their own.”
Jarred Sumner, founder of the Bun open source framework (now part of Anthropic), shared his experience in testing the feature inside Bun’s repository for a week. “It regularly catches extremely subtle bugs and rarely makes mistakes,” he wrote in a post on X.
Your reaction
Discussion
What's your take on this story?
No comments yet — be the first to share your view.