Linus Torvalds Submits First AI-Assisted Linux Kernel Patches, Reports ~3x Throughput Gain

Linux founder Linus Torvalds pushed a new development branch named noref-filemap-read-with-llm to his personal Git repository on October 13, officially submitting his first series of kernel patches developed with the assistance of a large language model. The patches aim to address scalability bottlenecks in Linux memory management related to page reference counts, specifically targeting CPU race conditions during high-frequency concurrent file reads.

According to reports from tech media outlet LWN, Torvalds stated that the core code, test programs, and initial documentation drafts were collaboratively completed by Anthropic's AI coding tool, Claude Code, under his direct guidance. In specific test loads with certain security protections disabled (8 threads reading the same file simultaneously), Torvalds' self-reported data showed read throughput jumping from 9.9 million to 10.4 million operations per second to 30 million to 33 million operations per second, representing an approximately 3x performance improvement. Kernel clock cycles consumed per single read dropped sharply from about 2,950 to 725.

The patches currently remain in an experimental branch and have not been merged into the official Linux 7.4 mainline. Additionally, Torvalds recently used LLMs to refactor obscure legacy code and uncover three hidden bugs that had persisted for 13 years. This event marks the entry of AI coding tools into the development workflow of core operating system infrastructure, though their practical applicability and safety in production environments still require independent verification.