Anthropic launches tool to check errors in AI-generated codes

Anthropic has launched ‘Code Review’, a tool that uses agents to review errors in codes generated by Claude’s artificial intelligence (AI), which will improve efficiency during the process.

The new ‘Code Review’ system arises from the need to balance workload that Anthropic developers face, since, as the company has indicated, code production per engineer has grown by 200 percent in the last year, which has led to a “bottleneck.”

To put it in context, Anthropic has revealed that, before implementing ‘Code Review’, 16 percent of review requests received “substantial comments”, but that has now been moved to 54 percent thanks to this new tool. However, the company has clarified that review requests must be approved by humans.

Anthropic has made this tool available to more developers through Claude Code, where it is available as a preview on the beta version of Team and Enterprise plans.

This code review tool allows you to perform more comprehensive analysis thanks to your multiple agents that “detect errors that humans often miss,” Anthropic has indicated on its blog.

In this way, when a review is processed, ‘Code Review’ deploys a group of agents that work in parallel to look for errors, verify them and classify them depending on its severity. Finally, the result of the analysis is included as a general comment in the review, in addition to a series of specific comments.

Anthropic has clarified that reviews vary depending on the complexity of the code being analyzed, but has explained that a review takes around 20 minutes on average.

‘Code Review’ coexistirá con ‘Claude Code GitHub Actions’which will continue to be available, maintaining its open source modality. However, this new tool entails higher costs for Anthropic, which has set a price per revision between 15 and 25 dollars (12.87 to 21.46 euros), depending on the size and complexity of the code.