Claude Code: 78 percent pass in 1 review
Anthropic's Claude Code team now lets AI agents check most of the code its AI writes, InfoQ reported on the 15th of November 2026. The share of changes that pass after one round of review and fixes rose from 65 to 78 percent.

How Anthropic reviews Claude CodeAnthropic's Claude Code team now lets AI agents check most of the code its AI writes, InfoQ reported on the 15th of November 2026. The share of changes that pass after one round of review and fixes rose from 65 to 78 percent.
Claude Code is Anthropic's AI coding helper. Its team said the amount of finished code changes more than doubled between June and September.
How the checks work
Before a change goes up, Claude runs tests and a second AI agent reviews it with fresh eyes, InfoQ reported. Several AI reviewers then look for bugs, and others double check what they find.
A separate security review can block a change if problems remain.
When people step in
Low risk changes, like notes in documents, may need no human sign off. Medium risk changes need one person, and high risk changes need two.
Changes to the systems that build and ship the code always need a person, the team said. The person who directs an AI agent stays responsible for its code.
A limit on rounds
AI reviews stop after three rounds, InfoQ reported. After that, the author is asked to rethink the change.
Below we break down the results
Changes that needed five or more rounds fell from 8.9 percent to 1.5 percent. Changes that had to be undone fell by more than 40 percent, the team said.
The one round pass rate went up 13 points, by our arithmetic.


Key facts
Anthropic's Claude Code team uses AI agents to review AI written code.
78 percent of changes now pass after one round.
High risk changes still need two human approvals.
Undone changes fell by more than 40 percent.
Source: InfoQ (Robert Krzaczynski): Anthropic Outlines its AI-Powered Code Review Process for Claude Code