Cat Wu had a simple answer to an increasingly common question from enterprise clients. “Now that Claude Code is putting up a bunch of pull requests,” she said, “how do I make sure that those get reviewed in an efficient manner?”
The question itself tells the story. Anthropic‘s coding assistant has generated so much code, so fast, that the human review process meant to catch errors before they ship has become a bottleneck. The company’s answer, launched Monday, is a tool called Code Review — an automated reviewer built directly into Claude Code.
The timing is notable. On the same day the product launched, Anthropic filed two lawsuits against the Department of Defense, which had designated the company a supply chain risk. That legal dispute puts added pressure on its enterprise business, which the company says has seen subscriptions quadruple since the start of the year. Claude Code‘s run-rate revenue has surpassed $2.5 billion since launch, according to the company.
Logic Over Style
Code Review launches first to Claude for Teams and Claude for Enterprise customers in research preview. Named clients already using Claude Code include Uber, Salesforce, and Accenture. Developer leads can switch the tool on by default for every engineer on a team.
Once active, it connects with GitHub and automatically analyzes pull requests — the formal submissions developers make before code enters a shared codebase. It leaves comments directly on the code, identifying potential issues and suggesting fixes.
Wu, Anthropic‘s head of product, was specific about what the tool will and will not flag. “We decided we’re going to focus purely on logic errors,” she said. “This way we’re catching the highest priority things to fix.” The decision to avoid style feedback is deliberate: developers, she said, grow frustrated with automated tools that flag things that don’t require immediate action.
The tool labels issue severity using a color system. Red marks the highest severity problems. Yellow flags potential issues worth a second look. Purple identifies problems tied to existing code or historical bugs in the repository.
Agents Working in Parallel
Behind that color-coded output is a multi-agent architecture. Multiple agents examine the codebase simultaneously, each approaching it from a different angle. A final agent then aggregates the findings, removes duplicates, and ranks what matters most.
Wu acknowledged that running parallel agents makes the product resource-intensive. The system also includes a light security analysis, and engineering leads can add custom checks based on internal standards. For deeper security work, the company points to a separate product: Claude Code Security.
The rise of “vibe coding” — using plain-language instructions to generate large volumes of code quickly — has accelerated development timelines while introducing bugs, security gaps, and code that developers themselves may not fully understand. Code Review is positioned as a check on that output, designed to catch what speed creates before it becomes someone else’s problem to fix.
Photo by Boitumelo on Unsplash
This article is a curated summary based on third-party sources. Source: Read the original article
