Anthropic introduces a code review tool to examine the flood of AI-generated code

When it comes to coding, peer feedback is essential to catch bugs early, maintain consistency across the codebase, and improve overall software quality.

The rise of “vibe coding” — using AI tools that take instructions given in simple language and rapidly generate large amounts of code — has changed the way developers work. While these tools speed up development, they also introduce new bugs, security vulnerabilities, and poorly understood code.

Anthropic’s solution is an AI reviewer designed to catch bugs before they enter the software codebase. A new product, called Code Review, was launched on Monday at Claude Code.

“We’ve seen a lot of growth in Claude Code, especially in business, and one of the questions we keep getting from business leaders is: As Claude Code puts in so many pull requests, how do I make sure those are reviewed efficiently?” Cat Wu, Anthropic’s head of product, told TechCrunch.

Pull requests are a method used by developers to submit code changes for review before those changes make it into software. Wu said that Claude Code has significantly increased the output of code, which has increased the number of pull request revisions that have caused the bottleneck of the shipping code.

“Code Review is our answer to that,” Wu said.

Anthropic’s launch of Code Review – a first for Claude Wampahet and Claude for Enterprise customers in a research preview – comes at an important time for the company.

On Monday, Anthropic filed two lawsuits against the Department of Defense in response to the organization’s designation of Anthropic as a procurement risk. The dispute will likely see Anthropic lean heavily on its booming business, which has seen subscriptions quadruple since the start of the year. Claude Code’s run-rate revenue has exceeded $2.5 billion since launch, according to the company.

“This product is mainly aimed at large business users, so companies like Uber, Salesforce, Accenture, who are already using Claude Code and now want help with a small amount [pull requests] that it helps to produce,” said Wu.

He added that developer leaders can turn on Code Review to run automatically for all developers on the team. Once enabled, it connects to GitHub and automatically analyzes pull requests, leaving comments directly on the code describing potential issues and suggested fixes.

The focus is on correcting logical errors over style, Wu said.

“This is very important because many developers have seen automatic AI feedback before, and they get upset when it doesn’t happen quickly,” Wu said. “We have decided that we will only focus on logical errors. This way we catch the important things that we need to fix.”

The AI ​​explains its thinking step by step, explaining what it thinks the problem is, why it might be a problem, and how it can be fixed. The system will label the severity of issues using colors: red for high severity, yellow for potential issues that need to be reviewed, and purple for issues related to pre-existing code or historical bugs.

Wu said it does this quickly and efficiently by relying on multiple agents working in parallel, each agent examining the codebase from a different perspective or dimension. The final agent aggregates and ranks the findings, removing duplicates and prioritizing the most important.

The tool provides simple security analysis, and engineering leads can perform additional customized testing based on internal best practices. Wu said Anthropic’s recently introduced Claude Code Security provides in-depth security analysis.

The multi-agent design means this can be a resource-intensive product, Wu said. Similar to other AI services, pricing is based on tokens, and costs vary depending on the complexity of the code — though Wu estimates that each update will cost $15 to $25 on average. He added that it’s a premium experience, and it’s needed as AI tools make more code.

“[Code Review] it’s something that comes from a crazy market pull,” said Wu. “As developers develop with Claude Code, they see a conflict in creating a new feature. [decrease]and they see a very high need for code reviews. So we’re hoping that with this, we’ll enable businesses to build faster than ever before, and with far fewer bugs than ever before.”