Anthropic Launches Claude Code Review to Solve the Industry's Massive Engineering Bottleneck

Anthropic announced Code Review for Claude Code today, a new system modeled after the company's internal development process. The tool addresses a growing industry crisis where AI-assisted coding has increased output so rapidly that human reviewers can no longer keep pace. It is currently available as a research preview for Team and Enterprise users.

A Multi-Agent Approach to Quality

When a developer opens a pull request, Code Review dispatches a fleet of specialized AI agents to scan for logic errors and security flaws. These agents work in parallel to verify bugs, a process that significantly reduces the noise of false positives. Findings are delivered as a high-signal overview comment directly within GitHub to help developers iterate faster.

Proven Results in Production

Internal testing at Anthropic revealed that the percentage of pull requests receiving substantive feedback jumped from 16% to 54% after implementation. Engineers found that less than 1% of the automated findings were incorrect. The system effectively scales its depth based on the complexity of the change, taking roughly 20 minutes for an average pass.

Real-World Bug Detection

In one instance, the agents flagged a routine one-line change that would have broken authentication for a production service. Another review on the TrueNAS open-source middleware discovered a pre-existing type mismatch that was silently wiping encryption keys. These are critical failures that human eyes often miss during quick scans of large diffs.

Cost and Integration

Reviews are billed based on token usage, typically averaging between $15 and $25 per session. Administrators can maintain control through monthly spend caps and repository-level toggles. The feature is designed to bridge the gap between high-speed code generation and secure, stable deployment.