New Delhi: Amidst tech layoffs and worries about AI taking over jobs, developers today are writing code faster than ever. Tools powered by artificial intelligence
are helping engineers generate large chunks of code in minutes. That speed is exciting, but it creates a new headache. Someone still has to review all that code before it goes live.
Anthropic says the pressure is real. The company revealed that code output per engineer has grown 200 percent in the past year. That spike has turned code reviews into a serious bottleneck for software teams. On Monday, the AI firm introduced a new system called Code Review inside Claude Code, designed to help catch software bugs before human reviewers even begin reading the code.
Introducing Code Review, a new feature for Claude Code.
When a PR opens, Claude dispatches a team of agents to hunt for bugs. pic.twitter.com/AL2J4efxPw
— Claude (@claudeai) March 9, 2026
How does Claude Code Review work?
The new system works a bit like a digital inspection team. When a developer opens a pull request, the platform sends multiple AI agents to review the code.
Each agent looks for a different type of issue. They scan the code in parallel, verify potential bugs and then rank them by severity. The final result appears as a single summary comment along with detailed inline notes for developers.
Anthropic describes the system clearly in its announcement.
“Code Review dispatches a team of agents on every PR to catch the bugs that skims miss, built for depth, not speed.”
The tool does not approve code changes automatically. Human engineers still make the final call.
Why AI code review is becoming necessary
Anyone who has worked on software teams knows how code review works. A developer submits a pull request. Another developer reads the code and checks for errors. In theory it is simple. In practice people skim.
Anthropic said that before introducing the new system, only 16 percent of pull requests received detailed review comments. After deploying the tool internally, that number rose to 54 percent.
The company says developers are shipping more code partly because of tools like:
- Claude Code
- Codex
- Cursor
More code means more chances for bugs. And humans often miss tiny changes that break systems.
One real example shared by Anthropic involved a one line code change. It looked harmless in the pull request. The AI review agents flagged it as critical. The change would have broken authentication in a production service.
Honestly, that kind of bug is easy to miss. I have seen similar problems slip through reviews before. One small line can bring down a service.
How long Claude Code Reviews take and what they cost
Claude Code Review focuses on depth instead of speed. According to Anthropic, an average review takes about 20 minutes. Complex pull requests trigger more agents and deeper inspection. Smaller changes get a lighter pass.
There is also a cost factor. Reviews run on token usage and typically cost between $15 and $25 per review depending on complexity.
Administrators can manage spending with several controls:
- Monthly spending caps
- Repository level review settings
- Analytics dashboard for tracking review activity
Feature launched in research preview
Anthropic said the feature is now available in research preview for Claude Team and Enterprise users through the Claude Code web interface. Once enabled by administrators, reviews automatically run whenever a pull request is created.
The company has already used the system internally for months. Large pull requests with more than 1,000 lines of changes generated findings in 84 percent of cases, averaging about 7.5 issues per review. Smaller changes under 50 lines triggered issues in 31 percent of cases.
