Anthropic has made Code Review for Claude Code generally available, moving the multi-agent pull request reviewing system out of the research preview it entered in March 2026. The general availability release comes with faster reviews, more concise findings, and — according to a report by The Next Web — $50,000 in credits attached for adopting organizations.
The system, first introduced as a research preview on March 9, 2026 for Team and Enterprise plans, dispatches a team of AI agents to examine every pull request. Anthropic said at the time that its own code output per engineer had grown 200 percent over the prior year, turning code review into a bottleneck: before deploying Code Review internally, only 16 percent of the company's pull requests received substantive review comments. After deployment, that figure rose to 54 percent. For more context on this story, see our ongoing AI trends.
How the Multi-Agent Review Pipeline Works
According to Anthropic's Code Review documentation, each review runs on Anthropic infrastructure in four stages. First, several agents examine the diff and the surrounding code in parallel to find potential issues. Second, a verification step checks each candidate finding against the actual code and filters out false positives. Third, the remaining findings are deduplicated and ranked by severity. Finally, the results are posted to the pull request.
Each review leaves inline comments on the specific lines concerned, a summary, and a check run named "Claude Code Review" listing the findings. Notably, the check run never concludes as a failure, and a review never approves or blocks a pull request — the merge decision stays with the human team. When a review finds nothing worth flagging, the check run reports that no issues were found.
By default, Code Review focuses on correctness: bugs that would break production, rather than formatting preferences or missing test coverage. Repositories can tune what gets flagged using REVIEW.md and CLAUDE.md guidance files, and Anthropic's documentation states that REVIEW.md is read from the base or default branch — never from the pull request's own changes, a design choice that prevents a pull request from rewriting its own review criteria.
Faster Reviews and Quieter Findings
Anthropic said it used the research preview period to make reviews faster, better at finding the bugs that matter, and easier to tune to each repository. According to the company, most reviews now finish roughly three times faster than during the preview, with typical review time moving from about 20 minutes to 6–10 minutes.
The findings themselves have also been reformatted. Anthropic said findings now state the problem, what triggers it, and the fix in a few sentences. Administrators can assign each repository a Quick, Standard, or Deep review level, with more thorough levels costing more. Anthropic noted that preview users wanted reviews that were "quieter" and sized to each change — feedback that shaped the general availability release.
What Anthropic's Internal Numbers Show
The company has published statistics from running Code Review on nearly every internal pull request. On internal pull requests with more than 1,000 lines changed, Anthropic said 84 percent received findings, averaging 7.5 issues each. Pull requests under 50 lines dropped to 31 percent receiving findings, averaging 0.5 issues — a sign the system scales its scrutiny with the size of the change. Engineers marked less than 1 percent of findings as incorrect.
Anthropic also pointed to a concrete catch from early-access usage: a latent type-mismatch bug in TrueNAS's open-source middleware that had been silently wiping the ZFS encryption key cache on every sync, a pre-existing issue in adjacent code that a pull request had touched.
How Teams Turn It On
To enable Code Review, an administrator turns it on in Claude Code settings, installs the Claude GitHub App, and selects repositories. Once enabled, reviews run automatically on new pull requests with no configuration required from individual developers. Teams can choose whether a repository runs reviews once when a pull request opens, after every push, or only on request, and anyone with write access can trigger one by commenting @claude review. Drafts and forks are excluded from automatic reviews.
Reviews run in a managed environment with read-only repository access, working from the diff and surrounding code without building or running it, and do not run in self-hosted environments. Code Review selects which Claude models each review runs on — admin settings offer no model choice.
Requirements and Limitations
The service requires a Team or Enterprise plan and repositories hosted on GitHub, including GitHub Enterprise Server and GitHub Enterprise Cloud with data residency. It is not available to organizations with Zero Data Retention enabled or the HIPAA configuration applied, is not covered under Anthropic's Business Associate Agreement, and is not yet available to organizations using customer-managed encryption keys. A Claude organization is required even when developers access Claude Code through Amazon Bedrock.
A Crowded Market Gets a New Incumbent
Anthropic is not alone in betting that AI-generated code needs AI-generated review. CodeRabbit, a dedicated AI code review startup, raised $143 million at a $1.5 billion valuation in August, and platform vendors have been folding review agents into their developer toolchains. What distinguishes Anthropic's entry is distribution: Code Review ships inside Claude Code, which many of the same teams already use to write the code in the first place.
The general availability release effectively makes automated pre-human review a default part of the Claude Code workflow for enterprise customers. Whether teams embrace a reviewer that never blocks a merge — but never gets tired of reading a 1,000-line diff — will show up in merge queues over the coming months.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
