Anthropic Introduces AI-Powered Code Review to Ease Developer Workloads

As AI coding assistants become commonplace in modern software development, a new challenge has quietly emerged: reviewing the flood of machine-generated code. Developers may be producing more than ever, but ensuring quality and reliability still demands careful oversight. Anthropic believes it has an answer.

The AI research company has unveiled Code Review, a new capability inside its Claude Code platform designed to automatically analyze pull requests and detect potential issues before they reach production. The announcement comes as engineering teams increasingly rely on coding agents to accelerate development cycles, sometimes generating more code than humans can realistically examine line by line.

Anthropic says the feature was built in response to customer feedback about growing review bottlenecks. In its announcement, the company stated, “They tell us developers are stretched thin, and many PRs [pull requests] get skims rather than deep reads.”

The tool is aimed primarily at large enterprise teams that already depend on Claude Code for development workflows. Organizations handling massive codebases and high submission volumes are expected to benefit most, particularly companies managing hundreds of pull requests daily.

How the system works

Code Review evaluates pull requests — collections of code changes submitted for approval — using multiple AI agents that operate simultaneously. Once connected to GitHub, the system can be configured to run automatically for every engineer, embedding itself directly into the development pipeline.

Each AI agent examines the codebase from a different angle, allowing for broader and more nuanced scrutiny. The system then posts comments directly within the pull request, flagging potential bugs and design concerns using color-coded severity levels:

• Red indicates high-severity issues requiring urgent attention
• Yellow marks areas that warrant further review
• Purple highlights historical or pre-existing bugs

A coordinating agent consolidates the feedback, removes duplicate findings, and prioritizes issues based on importance. The result is a structured review that helps developers focus on the most critical problems first.

Despite the automation, Anthropic emphasizes that humans remain the final decision-makers. Engineers still approve or reject pull requests after reviewing the AI’s recommendations.

Measured effectiveness

Anthropic reports that internal development metrics show rapid growth in code output, with engineers now producing 200 percent more code than they did a year ago. Within the company, Code Review significantly improved review quality: the share of meaningful review comments increased from 16 percent to 54 percent.

Testing also showed the system performs best on larger pull requests. Submissions exceeding 1,000 modified lines surfaced issues 84 percent of the time, while smaller changes under 50 lines triggered findings 31 percent of the time.

Engineers largely trusted the system’s assessments. According to Anthropic, developers agreed with the findings in nearly all cases, with less than 1 percent considered incorrect.

Availability and pricing

Code Review is currently offered in research preview for Claude Code Teams and Enterprise users.

Anthropic has adopted a usage-based pricing model tied to review size and complexity. Costs typically range between $15 and $25 per pull request. To help organizations manage spending, the company includes administrative tools such as monthly budget caps, repository-level permissions, and analytics dashboards.

As AI continues reshaping software engineering, tools like Code Review signal a shift from writing code to supervising intelligent systems that write — and now review — it.