This post was created with AI assistance and reviewed for accuracy before publishing.
Anthropic launched Code Review for Claude Code on March 9, 2025. It is a multi-agent system that runs on every pull request, finds bugs humans often skim over, and posts findings directly on the PR. If your team ships more code than reviewers can keep up with, Claude Code Review is worth knowing about. Here is how it works, what the numbers say, and how to get started.
Why code review became a bottleneck
AI coding tools increased output sharply. At Anthropic, code written per engineer grew about 200% in the last year. Human review capacity did not. The result: many PRs get a quick skim instead of a deep pass. Before Code Review, only 16% of Anthropic's PRs received substantive feedback. The rest were rubber-stamped or lightly glanced at.
The gap is real for most teams. Developers are stretched thin, PRs pile up, and reviewers focus on the biggest or riskiest changes. Small changes and "obvious" fixes often go through with minimal scrutiny. That is where bugs hide.
Anthropic's announcement frames Code Review as the reviewer you can run on every PR: built for depth, not speed.
How Claude Code Review works
When a PR is opened, Code Review starts a team of agents. They run in parallel, each looking for different kinds of issues. The agents:
- Search independently for different error types (logic bugs, type mismatches, security issues, etc.)
- Adjust depth based on PR size and complexity (small PRs get a lighter pass, large ones a deeper analysis)
- Cross-check findings to reduce false positives
- Rank issues by severity
- Consider the full codebase, not just the diff
Results show up as a single overview comment on the PR plus inline comments on specific lines. The whole run takes around 20 minutes on average, scaling with PR size.
Code Review does not approve PRs. That stays with humans. It surfaces issues so reviewers can focus on the real problems and approve with more confidence.
What the numbers show
Anthropic has been running Code Review internally for months. The official blog reports:
- 54% of PRs now get substantive feedback, up from 16%
- Large PRs (1,000+ lines): 84% get findings, averaging 7.5 issues
- Small PRs (under 50 lines): 31% get findings, averaging 0.5 issues
- Accuracy: Less than 1% of findings are marked incorrect by engineers
Two concrete examples stood out. In one case, a one-line change to a production service looked routine and would normally get a fast approval. Code Review flagged it as critical: the change would have broken authentication. The failure mode was easy to read past in the diff but obvious once pointed out. It was fixed before merge.
In another, on a ZFS encryption refactor in TrueNAS middleware, Code Review found a pre-existing bug in adjacent code: a type mismatch that was silently wiping the encryption key cache on every sync. A human reviewer scanning the changeset would not typically go looking for that.
Cost and admin controls
Code Review is built for depth, so it costs more than lighter options like the Claude Code GitHub Action, which remains free and open source. Reviews are billed on token usage. Anthropic says typical cost is $15–25 per PR, scaling with size and complexity.
Admins get controls to manage spend:
- Analytics dashboard: PRs reviewed, acceptance rate, total review costs
- Repository-level control: Enable reviews only on selected repos
- Monthly org caps: Set total monthly spend across all reviews
If you are on Team or Enterprise, you can tune where and how much Code Review runs.
Who can use it and how to turn it on
Code Review is in research preview for Claude Code Team and Enterprise plans. It is not available for individual or Pro tiers.
To enable it:
- Admins: Go to Claude Code settings, enable Code Review, and install the GitHub App.
- Select repositories: Choose which repos should run Code Review.
- Developers: Once enabled, reviews run automatically on new PRs. No extra config.
Full setup details are in the Code Review docs.
When Code Review makes sense
Code Review fits teams that:
- Ship a lot of PRs and struggle to review them all
- Have sensitive or critical code paths where missed bugs are costly
- Are on Claude Code Team or Enterprise and can afford the per-PR cost
It is overkill for tiny side projects or when you already have strong review coverage. For teams drowning in PRs, it is a way to get a consistent second pass on every change without scaling headcount.
Claude Code Review is still new, but the internal numbers and early customer stories show it can catch bugs that slip through quick skims. If your team relies on AI to write more code, it is worth considering how you will review it.