Anthropic Launches Multi-Agent Code Review for Claude Code

A Multi-Agent System for the AI Code Flood

Anthropic launched Code Review in Claude Code on Monday, a new product that deploys multiple AI agents in parallel to automatically analyze pull requests for logic errors, bugs, and security vulnerabilities. The tool arrives as enterprise development teams grapple with an explosion of AI-generated code that has overwhelmed traditional review processes.

The system assigns multiple AI agents to each pull request, with each agent examining the codebase from a different perspective. A final aggregation agent ranks the findings by severity and filters out false positives before posting inline comments on the specific lines of code where issues were detected. Larger pull requests automatically receive more agents for deeper analysis, while smaller changes get a quicker review.

Internal Results: 200% More Code, 3x More Reviews

Anthropic shared internal data that underscores both the problem and the opportunity. Code output per developer at the company has jumped 200% over the past year, driven by AI coding assistants. Before the internal AI review system launched, only 16% of code changes received substantive comments. After deployment, that figure rose to 54%, suggesting the tool catches issues that human reviewers were consistently missing under time pressure.

The product is explicitly targeted at large enterprise customers. Anthropic cited companies like Uber, Salesforce, and Accenture as the type of organizations that already use Claude Code and now need help managing the sheer volume of pull requests it helps produce. Code Review won't approve PRs on its own — that decision remains with human reviewers — but it aims to surface the most critical issues so developers can focus their review time effectively.

Pricing and Availability

Code Review is available now in research preview for Claude Code Team and Enterprise plans. Pricing is token-based, with Anthropic estimating each review will cost $15 to $25 on average depending on code complexity. Administrators can set monthly spending caps to keep costs predictable.

The feature integrates directly with GitHub, analyzing pull requests and posting findings as inline comments. This GitHub-first approach reflects the reality that most enterprise development workflows are centered on pull request reviews, making it a natural integration point for automated analysis.

Addressing the "Vibe Coding" Problem

The launch addresses a growing industry concern sometimes called "vibe coding" — the practice of using AI tools to generate large volumes of code quickly without thorough review. As AI coding assistants become more powerful, the gap between code generation speed and review capacity has widened, creating potential blind spots for bugs and security vulnerabilities.

Anthropic's approach of using multiple agents with different review perspectives mirrors how experienced development teams conduct reviews, where different reviewers focus on different aspects like security, performance, and correctness. The aggregation step that removes duplicates and ranks by severity aims to reduce the noise that often makes automated code analysis tools frustrating to use.

What This Means for Software Engineers

For development teams, Code Review represents a new category of tooling that sits between code generation and human approval. Engineers can expect faster feedback loops on pull requests, but the $15-$25 per review cost means teams will need to evaluate the tool's return on investment carefully. The launch also signals that AI companies increasingly see developer tools as a major revenue stream, with Anthropic now competing directly with GitHub Copilot and other AI-assisted development platforms for enterprise budgets.