Coding agents like Claude Code or OpenAI’s Codex have become synonymous with software development, as many companies state that their engineers have virtually stopped writing code manually, and are “vibe coding” instead. While this does significantly increase productivity, it also creates the problem of reviewing each and every line of code. Now, Anthropic may have a solution, in the form of Code Review.
Anthropic announced that it is bringing Code Review to Claude Code in a blog post. The AI startup states that the tool is aimed at solving the problem of this “bottleneck,” after listening to user feedback. The company stated, “They tell us developers are stretched thin, and many PRs [pull requests] get skims rather than deep reads.”
The product is focused on large-scale enterprise users, such as Uber, Salesforce, and Accenture, who already use Claude Code and now want help with the volume of pull requests.
How does Anthropic’s Code Review work?
Code Review analyses pull requests or PRs by using a team of AI agents. Pull requests refer to bundles of code changes submitted for review.
The AI integrates with GitHub and can be set to run by default for every engineer. It reviews each pull request, leaves comments directly on the code, and highlights potential issues with severity labels – red for high-severity, yellow for review, and purple for historical or pre-existing bugs.
Each review is carried out by multiple agents working in parallel, each examining the codebase from a distinct perspective. A final agent aggregates the results, removes duplicates, and ranks findings by importance.
This not only reduces the workload for a human, but also makes it easier to recognise the potential bugs in large blocks of code. Though Anthropic clarified that a human will make the final call to approve the pull request.
How effective is Code Review?
Anthropic states that the company has seen a rise of 200 per cent in code output per engineer in the last year. Internally, Code Review increased the rate of substantive review comments from 16 per cent to 54 per cent.
According to Anthropic, testing showed that large pull requests (over 1,000 lines changed) flagged issues 84 per cent of the time, while smaller requests (under 50 lines) showed findings 31 per cent of the time. Engineers agreed with the system’s findings in nearly all cases, with less than 1 per cent marked as incorrect.
Code Review availability and price
The feature is currently available in research preview for Claude Code Teams and Enterprise users.
Pricing for Code Review is based on review size and complexity, averaging $15–$25 per pull request. Anthropic provides administrative controls to help manage costs, including monthly spending caps, repository-level controls, and analytics dashboards.
