Linus Torvalds posts his first LLM-assisted Linux patch series

Linus Torvalds used an LLM coding assistant to tackle a long-standing Linux scalability problem: when many threads read the same file data, they fight over a shared page reference count. Several earlier attempts, including his own rough draft, had problems. His new series of ten patches lets frequently read pages be read safely without taking that reference. In a deliberately artificial test with 64 threads, reads per second went from 13.3 million to 304 million, about 23 times faster; he says real workloads will see much less. He noted the commit messages are wordier than he would write himself, and review has already found a race condition that needs fixing. On Hacker News (about 90 comments), there is no consensus. Some see it as proof that AI help is now normal even in critical software, and one commenter who read the diffs said they are moderate in size and readable, not the huge dumps people often send with LLMs. Others worry beginners will take the wrong lesson, pointed out that a human reviewer still caught the race, disliked the long AI-written commit messages, or were surprised he used a closed commercial tool. Sorted take: this is the pattern worth copying. An expert picks a hard, well-understood problem, uses the AI for the grind, measures the result honestly, and still sends it through normal review.