With the rise of "Vibe Coding" (generating large amounts of code quickly through natural language instructions), programmers' development speed has significantly increased, but it has also led to logical vulnerabilities and security risks. To address this, Anthropic has officially launched an AI code audit tool called Code Review, aimed at providing enterprises with automated "expert-level" feedback.
The product is now available to Claude for Teams and Enterprise users as part of Claude Code. Cat Wu, a product leader at Anthropic, pointed out that AI tools have caused a surge in pull requests, and manual review has become a bottleneck in software delivery. The introduction of Code Review is designed to address this bottleneck.
The core features of Code Review include:
Deep Logic Detection: Unlike simple style checks, this tool focuses on identifying logical errors in the code. Wu said that developers often get tired of non-actionable, trivial feedback, so Code Review prioritizes high-level functional fixes.
Severity Level Identification: The system uses color coding to classify risks—red indicates extremely high severity, yellow represents potential issues, and purple highlights risks related to historical bugs or legacy code.
Multi-Agent Collaboration Architecture: Internally, it consists of multiple parallel AI agents, each analyzing the code from different dimensions (such as security, logical integrity, etc.). A final aggregation agent then compiles and removes duplicates to ensure high-quality recommendations.
Deep Integration with GitHub: The tool can leave comments directly within GitHub pull requests, explaining its reasoning process, potential impact, and specific repair suggestions.
Currently, large companies such as Uber, Salesforce, and Accenture have started using this tool. Although the cost of each review is expected to be between $15 and $25, in the context of AI-generated code becoming prevalent, this high-efficiency, low-bug-rate automated audit has become a critical need for enterprises.
