Anthropic has launched a Code Review feature for Claude Code, now in research preview for Team and Enterprise users. The system dispatches multiple AI agents in parallel when a pull request is opened, searching for bugs, verifying findings to reduce false positives, and ranking issues by severity before posting inline comments. The number of agents scales with PR size and complexity, with an average review time of around 20 minutes and a cost of $15–25 per PR. Internally, Anthropic reports substantive review comments rose from 16% to 54% of PRs after adoption, with fewer than 1% of findings marked incorrect by engineers. Community reactions are mixed: developers praise the multi-agent depth, but some question the cost and practicality for high-volume workflows, and others raise concerns about Claude both writing and reviewing code. The feature competes with GitHub Copilot code review and CodeRabbit, differentiating itself through deeper, slower multi-agent analysis.
Sort: