Anthropic's Claude Gets Multi-Agent Code Review
Anthropic unveiled a multi-agent code review system for "Claude Code," available in research preview for Team/Enterprise that dispatches teams of AI agents to scrutinize every pull request, uncovering bugs humans often miss. Claude Code integrates as a GitHub Action, adding AI review to CI/CD pipelines without manual setup. Anthropic is also targeting enterprise AI procurement bottlenecks with a unified, contract-based model for deploying Claude tools across organizations.
Anthropic's new Claude Code Review feature aims to alleviate code review bottlenecks created by the rise of AI-generated code. The tool dispatches a team of AI agents to analyze pull requests, identifying logic errors and suggesting fixes within GitHub. This multi-agent system is currently available as a research preview for Claude for Teams and Claude for Enterprise customers. The Code Review system integrates with GitHub, automatically scanning pull requests upon opening. Multiple AI agents examine the code from different angles, aggregate their findings, and prioritize issues using a color-coded severity system. Red signifies critical issues, yellow indicates review-worthy items, and purple denotes historical problems. The tool focuses on logic errors rather than style issues and includes light security analysis with customizable checks. Anthropic's internal testing showed a significant increase in thorough pull request reviews, rising from 16% to 54% with the tool's implementation. For large pull requests exceeding 1,000 lines of code, 84% received findings, averaging 7.5 issues. Smaller pull requests under 50 lines saw findings 31% of the time, averaging 0.5 issues. Engineers agreed with over 99% of the findings. Code Review's pricing is token-based, estimated to cost $15 to $25 per review, depending on code complexity. Organizations can configure monthly spending limits, enable/disable settings at the repository level, and access an analytics dashboard for review statistics and costs. Admins can enable the feature by installing the GitHub app in Claude Code settings and selecting repositories for review. Once enabled, reviews run automatically on new pull requests. Alternatives to Claude Code include GitHub Copilot, OpenAI Codex, and OpenCode. GitHub Copilot focuses on individual coding speed through autocompletion, while Claude Code targets team review work. OpenAI Codex offers cloud sandbox isolation and a Rust CLI. OpenCode supports multiple LLM providers, including local models. Anthropic also provides an official GitHub Action called `anthropics/claude-code-action` that allows users to integrate Claude directly into GitHub issues and pull requests. By mentioning `@claude` in a comment, developers can have Claude respond with access to the codebase. This action requires a Claude Pro/Max subscription and the installation of the Claude GitHub App. Claude Code is available on Claude for Pro, Max, Team, and Enterprise users. Pricing for Claude Opus 4.6 starts at $5 per million input tokens and $25 per million output tokens. The Claude Pro plan costs $20/month or $17/month when billed annually. The Max plan starts at $100 per person per month. Anthropic's Claude Code faces constraints related to usage limits and model lock-in. The $20 Pro tier may only provide 10-20 meaningful coding sessions per week for most developers. Claude Code runs Anthropic models exclusively, preventing users from swapping in other models. The tool's verbosity can also lead to higher token consumption compared to alternatives.
