Anthropic says Claude writes and reviews about 90% of Claude Code team's code
About 90% of the code on Anthropic's Claude Code team is written and reviewed by Claude, the company says in an engineering post published on 16 November. The post describes how the team gives its agents ways to verify their own work, which it says is what lets engineers run dozens or hundreds of Claude agents in parallel instead of checking every change themselves.
Claude tries its changes in a running application, and The New Stack reports that it uses browser automation for web apps, MCP connected simulators for mobile builds and local services for backend work, with the instructions stored as verification skills in each repository. Flaky tests get no special treatment: Claude cannot skip them, add sleeps or retry until they pass, and if it cannot find the root cause it deletes the test and writes down why.
In answers to The New Stack, Anthropic said cheap deterministic checks such as builds, linters, unit tests and property tests run before any model review, and that "if a change can modify the check that judges it, the check is broken." It said merged pull requests to the Claude Code codebase more than doubled between June and September, while the share of changes that had to be reverted fell by more than 40%. Anthropic did not say how much of that came from its verification loops, gave no figure for the added inference cost, and does not publish the false positive or false negative rates of its code reviewers.
"A person is accountable for every change that ships, whoever typed it," the company told the publication.
We hope you enjoyed this article
Consider subscribing to one of our newsletters like AI Programming Weekly or Daily AI Brief.
Also, consider following us on social media:
AI Programming Weekly
Weekly news about AI tools for software engineers, AI enabled IDE's and much more.
Whitepaper
Tensordyne Napier: What If One Rack Could Do the Work of Nine?
Tensordyne
This Tensordyne whitepaper presents Napier, an inference-focused AI processor and rack-scale system based on the company’s TDN Math logarithmic number system. It examines infrastructure requirements for large mixture-of-experts and agentic models, compares major inference architecture approaches, and details the TDN AIP processor, TDN72 pod, TDN Link fabric, and Napier Ultra configuration. The paper reports simulation-based performance, cost, and accuracy-validation results, including Tensordyne’s projected comparison of one Napier rack with a nine-rack Nvidia Rubin plus Groq deployment; the chip is reported as taped out and in fabrication.
Read moreYou may also like
Anthropic Uses Claude to Verify 123 Open Source Libraries
Anthropic Adds Claude Self Assessment to Detect Misuse
Anthropic Finds Researchers Use Claude Mostly for Data Analysis
Anthropic Manager Drops Coding Interviews From Team Hiring
Anthropic Automates Search for Training Data That Evades Safety Monitors
Daily AI Brief: the AI news that matters, in your inbox.