Linus Torvalds uses Claude Code to write Linux kernel patches

By Splitfeed EditorialLast updated Oct 16, 2026, 13:51 UTC

Between October 13 and 16, 2026, Linus Torvalds confirmed he used Anthropic's Claude Code to write an eight-patch series targeting the Linux page cache, along with the test programs used to validate that work. Synthetic benchmarks produced performance figures ranging from 3x to 23x depending on the specific test, with the lower end widely cited and the higher figure appearing in select reports. Torvalds publicly acknowledged the tool's role in generating both the patches and the supporting test code. Reviewers subsequently identified a race condition in the submitted series, raising questions about the code's readiness for merging into the mainline kernel.

The significance of this event sits squarely in who did it and what he had previously said. Torvalds is not a peripheral figure in open-source software; he created Linux and has maintained final authority over its kernel for decades. His earlier characterization of AI as "90% marketing" made his adoption of Claude Code a direct reversal of a publicly held position. The patch series also targets the page cache, a foundational component of kernel memory management, meaning any defect there carries real consequences. The race condition found by reviewers is not a minor cosmetic issue; race conditions in low-level kernel code can produce intermittent data corruption or system instability, which is why the bug's discovery anchors the central dispute over whether the patches are net positive or a cautionary example.

No published reactions from the Pro-AI, Anti-AI, or Middle Ground camps had appeared by the time this story was filed, so what follows reflects what each camp would typically argue. The Pro-AI camp would likely treat Torvalds's adoption as a landmark endorsement, pointing to the performance gains as evidence that AI tooling can contribute meaningfully even to the most demanding software in existence. The Anti-AI camp would be expected to focus on the race condition as validation of longstanding concerns that AI-generated code introduces subtle, hard-to-detect bugs that can slip past initial review in safety-critical systems. The Middle Ground camp would probably argue that the episode illustrates both the promise and the limits of current AI coding tools, framing the race condition not as a disqualifying failure but as a reason why human expert review remains indispensable.

The immediate question to watch is whether the patch series is revised to fix the race condition and ultimately merged into the Linux kernel mainline. If Torvalds submits a corrected series and it passes review, that outcome would strengthen the case that AI-assisted kernel development is viable. If the patches are withdrawn or sit unmerged, it would give critics concrete grounds to argue the tooling is not yet ready for work at this level.