Anthropic Revolutionizes AI Code Control with Code Review

The Importance of Feedback in Software Development

In the field of programming, peer feedback exchange is an essential element for quickly identifying bugs, ensuring code consistency, and improving the overall quality of software. This practice has become even more crucial with the rise of artificial intelligence tools that generate code from natural language instructions.

The Impact of "Vibe Coding" on Development

"Vibe coding," which refers to the use of AI tools capable of rapidly producing large amounts of code from simple instructions, has profoundly changed the daily lives of developers. While these tools accelerate the development process, they also introduce new challenges, such as unexpected bugs, security vulnerabilities, and sometimes limited understanding of the generated code.

An Innovative Solution: Code Review by Anthropic

To address these challenges, Anthropic has developed an AI-based code review tool called Code Review. This product, integrated into Claude Code, was officially launched on Monday. Its mission is to detect bugs before they are integrated into the software's codebase.

Cat Wu, product manager at Anthropic, explained to TechCrunch that the tool was designed in response to a growing demand from companies using Claude Code. These companies were seeking an effective method to manage the increasing requests for code reviews generated by this tool.

Managing Pull Requests: A Growing Challenge

Pull requests are proposals for code modifications submitted for review before their final integration into the software. With the increase in code production thanks to Claude Code, the number of these requests has exploded, creating a bottleneck in the deployment process.

Wu emphasized that Code Review is Anthropic's response to this issue, allowing for a smoother and more efficient review of pull requests.

A Strategic Launch for Anthropic

The launch of Code Review comes at a pivotal moment for Anthropic. On Monday, the company filed two lawsuits against the Department of Defense following its designation as a potential risk to the supply chain. In this context, Anthropic is heavily betting on expanding its business with enterprises, which has seen its subscriptions quadruple since the beginning of the year.

A Tool Designed for Large Enterprises

The recurring revenue from Claude Code has exceeded $2.5 billion since its launch, according to company statements. Wu specified that Code Review primarily targets large companies like Uber, Salesforce, and Accenture, which are already using Claude Code and looking to optimize their pull request management.

Integration and Customization of Code Review

Development leads can enable Code Review to function by default for each team member. Once activated, the tool integrates with GitHub and automatically analyzes pull requests, leaving comments on the code to highlight potential issues and suggest corrections.

The focus is on detecting logical errors rather than coding style. Wu explained that this is crucial because developers are often frustrated by automated feedback that is not immediately actionable.

A Methodical Approach to Code Review

The AI in Code Review explains its reasoning step by step, identifying issues, their potential impact, and possible solutions. Problems are categorized by severity using a color code: red for critical issues, yellow for issues to be reviewed, and purple for issues related to pre-existing code.

Wu clarified that the tool operates effectively thanks to several agents working in parallel, each analyzing the codebase from a different angle. A final agent aggregates and ranks the results, eliminating duplicates and prioritizing the most important corrections.

Security and Customization

While Code Review offers lightweight security analysis, engineering leads can customize additional checks based on internal best practices. Anthropic also offers Claude Code Security for more in-depth security analysis.

A Premium Product for a Growing Market

The multi-agent architecture of Code Review makes it a potentially resource-intensive product. Like other AI services, pricing is based on tokens, and the cost varies depending on the complexity of the code. Wu estimates that each review will cost between $15 and $25 on average.

Wu concluded by emphasizing that Code Review meets a growing market demand. Engineers using Claude Code are experiencing reduced friction in creating new features and an increased demand for code review. Anthropic hopes to enable companies to develop faster and with fewer bugs.