Linus Torvalds, the creator of Linux, has been spotted using Anthropic's Claude Code to develop a significant improvement for the Linux kernel's memory management subsystem. While refreshing his torvalds/linux.git repository in anticipation of the Linux 7.3-rc7 release, observers discovered a new branch titled noref-filemap-read-with-llm. This branch targets a specific performance bottleneck: contention on folio reference counts during small, concurrent file reads.

A Technical Deep Dive into Folio References

The core issue addressed by this patch series is the contention that occurs when many threads attempt small reads from the same file. Under the current implementation, these operations contend on folio reference counting, creating a scalability bottleneck. The new code aims to adapt the memory management code to read from the page cache without relying on these folio references, potentially offering substantial performance gains for multi-threaded workloads.

Torvalds Admits AI Involvement

In a rare and transparent admission, Torvalds confirmed the role of artificial intelligence in the development process directly within the main patch message. He stated, "The code, the test programs, the review behind the list above and the first draft of this text were done with an LLM coding assistant (Claude Code) under my direction." He further clarified that while the AI handled the drafting and coding, "The measurements were run on my machine," maintaining human oversight over the validation process.

The Evolution of Kernel Development Tools

This development marks a shift in how even the most traditional figures in open source view AI tools. Torvalds has previously expressed that he views AI as a useful tool and has experimented with Google's Antigravity for vibe coding outside the kernel tree. However, using Claude Code for core kernel features suggests that LLMs are moving beyond simple debugging or peripheral tasks into the heart of complex systems programming.

Key Takeaways

  • The new branch is named noref-filemap-read-with-llm, explicitly signaling LLM involvement.
  • The patch series aims to eliminate folio reference contention for small, concurrent file reads.
  • Torvalds used Claude Code for code generation, test programs, and the initial cover letter draft.
  • Human validation remains critical, with measurements performed on Torvalds' own hardware.

The Bottom Line

If Linus Torvalds is using Claude Code to draft kernel patches, the debate over AI's utility in serious systems programming is effectively over. We are entering an era where the most scrutinized codebase in the world is being assisted by LLMs, and the results are significant enough to warrant a dedicated branch.