TechCrunch reported on the 10th (local time) that Anthropic has launched "code review," a feature that automatically reviews large amounts of code generated by artificial intelligence (AI). Code review was newly added that day to Anthropic's developer AI coding tool Claude Code.
TechCrunch said, "Workflows are changing as vibe coding—where developers give instructions in natural language and AI generates code—gains traction," adding, "AI tools that support vibe coding have greatly accelerated development, but they have also increased issues such as new bugs (errors), security risks, and code that in-house developers do not fully understand."
Anthropic explained that code review is a feature for catching such bugs. The company said that as AI generates code at scale, pull requests have surged. A pull request is a process in which a developer asks teammates to review code that has been modified or newly written before it is merged into existing software.
Cat Wu, Anthropic's head of product, said, "As Claude Code generates a large number of pull requests, related bottlenecks are emerging in corporations' development teams," adding, "Code review is Anthropic's solution to this problem."
According to Anthropic, over the past year the amount of code produced per company engineer increased by about 200%. While coding speed has increased, review time has not kept pace. In response, Anthropic introduced code review with a multi‑agent architecture in which multiple AI agents analyze code simultaneously. Multiple AI agents analyze the codebase in parallel and detect bugs, after which other agents verify the results.
Code review focuses on finding logical issues that are likely to be actual errors rather than code style. It indicates severity levels by color. The most severe issues are marked in red, potential issues worth reviewing in yellow, and issues related to existing code or past bugs in purple.
Wu said, "We hope code review will help corporations develop software faster than before while also reducing bugs."