Anthropic on March 9 introduced Code Review, a feature for Claude Code designed to catch bugs and other problems in pull requests that human reviewers frequently miss. The company positions it as a response to the growing volume of AI-generated code that engineering teams now have to check.
Rather than a single pass, Code Review runs multiple AI agents in parallel over the same pull request. According to Anthropic, each agent examines the change from a different angle, looking for potential bugs, security weaknesses and logic errors, and the results are combined into a single review.
Filtering the noise
Each finding is scored on a confidence scale, and the tool surfaces only issues it judges to be serious or high confidence. Anthropic says this is intended to cut down on the low-value alerts that have made many automated review tools easy to ignore, so that developers can focus on the problems that matter.
The company said a full review typically takes about 20 minutes and costs in the range of $15 to $25 per pull request, a price that reflects the amount of model computation involved in running several agents against a codebase. Several commentators noted the cost is high relative to existing review tools, although Anthropic argues it should be weighed against the cost of a bug reaching production.
Availability and competition
Code Review is available as a research preview to customers on Claude Team and Enterprise plans. It integrates with GitHub pull requests and runs in Anthropic’s cloud rather than on a developer’s machine.
The launch intensifies competition in developer tooling, where Microsoft’s GitHub Copilot already offers automated review and OpenAI has pushed its Codex agent into the same workflow. Automated review is emerging as the next battleground after code generation itself, as teams discover that producing code faster than it can be checked creates problems of its own.
