Anthropic on Monday launched a code review tool inside Claude Code that automatically reviews pull requests before a human ever opens them and the numbers behind it are hard to dismiss.
The tool, simply called Code Review, deploys multiple AI agents in parallel across a pull request. Each agent hunts a different class of problem: logic errors, security vulnerabilities, broken edge cases. A separate verification step then filters out false positives before anything surfaces to the developer. Results land as a single overview comment in the PR thread, with inline annotations for specific issues.
What makes this different from similar tools is the rejection rate. Anthropic says developers push back on fewer than 1% of flagged issues which suggests the verification layer is actually doing its job.
Anthropic Built This for Itself First
The backstory here is worth knowing. Anthropic’s own engineering team saw code output jump roughly 200% over the past year, driven largely by AI-assisted development. Human reviewers couldn’t keep up. Substantive feedback was landing on only 16% of pull requests.
After deploying Code Review internally, that number climbed to 54%.
The company says it now runs the tool on nearly every internal PR which is about as close to a real-world stress test as you can get before shipping something to customers.
What It Actually Costs
Code Review is billed on token usage. An average review runs between $15 and $25, scaling up for larger or more complex PRs. That’s meaningfully more expensive than open-source alternatives like Anthropic’s own Claude Code GitHub Action.
By the numbers:
- PRs under 50 lines: flagged 31% of the time
- PRs over 1,000 lines: 84% get findings, averaging 7.5 issues each
- Average review time: around 20 minutes
- Developer rejection rate on findings: under 1%
Whether that cost makes sense depends entirely on your team’s risk tolerance. One security vulnerability reaching production can cost far more than a year of reviews.
How to Turn It On
It’s not a one-click install. Admins need to enable the tool in Claude Code settings, install the GitHub App, and select which repositories it covers. From there, reviews run automatically on every new PR with no additional setup.
Monthly spend caps and an analytics dashboard are available so teams can track usage and control costs. The tool won’t approve or merge anything that stays with the human reviewer. Code Review is available now for Claude Teams and Enterprise users.
No comments yet. Be the first to share your thoughts!