Claude launches multi-agent PR Review tool: AI deep code review Anthropic (Claude Developer) officially released Claude Code's new feature - Code Review (Research Preview) on March 9th. Core Highlights -As soon as PR is opened, a team of Claude agents will be automatically dispatched to work in parallel: each hunting bugs → cross validation filtering false positives → sorting by severity. -Output: A high signal summary comment on PR+inline annotation. -On average, it takes 20 minutes per session, and the scale is adaptive (large PR requires multiple agents to dig deep, small changes require light weight). -Internal testing (Anthropic uses it on most PRs): The actual feedback PR ratio has skyrocketed from 16% to 54%, with an error rate of<1%, which can catch fatal problems that humans miss at a glance. Who can use it? Only for Claude Team&Enterprise users. Enable GitHub App with just one click in Claude Code settings, select the repository and it will automatically run in the cloud without the need for additional developer actions. Will not automatically approve PR, purely assisting human reviewers. Price comparison: Based on token billing, the average cost is $15-25 per transaction (the higher the complexity, the more expensive). Administrators can set monthly limits, warehouse level switches, and fee dashboards. The official statement clearly states that it is more thorough (and expensive) than the existing open-source Claude Code GitHub Action, which is still completely free and lightweight, suitable for simple automatic review. In the era of AI coding, the number of PRs is exploding, and this tool is enterprise level "code insurance" - helping you block bugs that are easily overlooked by humans in advance, allowing reviewers to focus on architecture rather than trivial matters. Want to experience it first? Team/Enterprise users can simply go to Claude Code settings to enable it!