# Linus Torvalds uses Claude Code to develop new Linux kernel feature

> **Key takeaway:** Linus Torvalds used Anthropic's Claude Code LLM assistant to write code, tests, and documentation for a new kernel memory management feature, going beyond his previous use of AI limited to debugging.

**Published:** 2026-10-15T17:27:19+00:00 
**Source:** Linux Magazine News 
**Category:** distro-news 
**Canonical URL:** https://fosswire.org/news/linus-torvalds-uses-claude-code-to-develop-new-linux-kernel-.html 

## Summary
Linus Torvalds has expanded his use of AI tools in Linux kernel development from debugging to active feature creation. In a patch message for the 'noref-filemap-read-with-llm' work, Torvalds stated that Claude Code generated the code, test programs, review, and first draft of his explanation under his direction. The feature replaces reference-counted reader tracking with RCU-based lockless reads to reduce CPU locking overhead. Phoronix's Michael Larabel noted the unusual appearance of a new branch in Torvalds' repository, where he typically only hosts the master branch. There is no confirmation whether the patches will merge into kernel 7.4.

## What Torvalds built with Claude Code
- **Feature**: `noref-filemap-read-with-llm` — a folio page-cache removal optimization
- **Core idea**: A folio removed from page cache returns to the page allocator only after an RCU grace period
- **Performance change**: Replaces reference-counted reader tracking with **lockless RCU Read** via deferred memory freeing
- **Problem solved**: Previous approach suffered heavy CPU locking under load
- **Torvalds' disclosure**: Explicitly credited Claude Code for code, tests, review, and first draft of his patch message
- **Branch location**: New kernel branch appeared in Torvalds' main repository — unusual per Phoronix observation

## Why it matters
Torvalds' public adoption of LLM-assisted coding for core kernel features, with explicit attribution, may accelerate acceptance of AI tools among maintainers of critical infrastructure projects who have been skeptical of generated code quality.