Anthropic introduces a code analysis tool to assess the extent of AI-generated code

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

The rise of “vibe coding” – the use of AI tools that take instructions given in simple language and quickly generate large amounts of code – has changed the way developers work. While these tools have accelerated development, they have also introduced new bugs, security risks, and unintelligible code.

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

“We’ve seen a lot of growth in Claude Code, especially in businesses, and one of the questions we get from business leaders is this: Since Claude Code is making so many pull requests, how can I get them reviewed?” Cat Wu, Anthropic’s chief product officer, told TechCrunch.

Pull requests are a method that developers use to submit code changes for review before they are released into the software. Wu said Claude Code There has been an increase in the number of code releases, which has increased the number of request comments that have led to problems with delivery.

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

The launch of Anthropic Code Review – the first arrival of the Claude for Teams and Claude for Enterprise client in the preview review – comes at a very important time for the company.

Monday, Anthropic filed two lawsuits against the Department of Defense in response to the designation of Anthropic as a supply chain threat. The deal sees Anthropic lean heavily on its growing business, which has seen subscriptions quadruple since the start of the year. Claude CodeFunding has exceeded $2.5 billion since its inception, according to the company.

“This product is focused on our large business users, so companies like Uber, Salesforce, Accenture, who use Claude Code and now want to be supported by the number of (requests) they are helping to create,” said Wu.

He added that product managers can turn on Code Review to work seamlessly for every engineer on the team. When enabled, it integrates with GitHub and analyzes pull requests, leaving comments directly on the system describing potential issues and recommended changes.

The goal is to correct logical errors in traffic, Wu said.

“This is very important because many developers have seen AI solutions in the past, and they get angry when they can’t act quickly,” Wu said. “We decided to focus on the mistakes in our thinking. This way we are finding the most important things to fix.”

The AI ​​explains its reasoning step by step, explaining what it thinks the problem is, why it might be difficult, and how to solve it. The system will mark the severity of problems using colors: red for critical problems, yellow for problems that may need to be reviewed, and purple for pre-existing problems or old bugs.

Wu said it does this quickly and efficiently by relying on multiple agents that work together, and each agent evaluates the codebase from different perspectives or metrics. The latter consolidates and organizes the findings, removing duplicates and prioritizing the most important ones.

The device provides light security analysisand engineering leads can modify additional checks based on internal best practices. Wu said Anthropic has recently been launched Claude Code Security provides in-depth security analysis.

A multi-agent architecture means that this can be a multi-product design, Wu said. As with other AI services, pricing is based on benchmarks, and the cost varies depending on the complexity of the code – although Wu says each review will cost $15 to $25 on average. He added that it’s a win-win situation, and an important one because AI tools generate more code.

“(Code Review) is something that is coming because of market pull,” Wu said. “When engineers are developing with Claude Code, they are seeing the friction to create something new (less), and they are seeing a greater need for code. So we are hopeful that with this, we will help businesses create faster than they did before, and with fewer bugs than before.