Linus Torvalds told Linux kernel developers this week that the project will not define itself by opposition to AI tools, drawing a hard line in a mailing list fight over whether large language models belong in maintainer workflows.
The exchange happened on the public linux-media mailing list in a thread about linking Patchwork with Sashiko. Roman Gushchin wrote that rejecting LLMs in general would make Sashiko’s stated purpose, helping maintainers, impossible to achieve. He said the discussion should address whether the kernel project should use LLMs at all rather than making each proposed use case more difficult.
Torvalds replied that a broadly anti-LLM position is not the Linux kernel’s position. As the project’s top-level maintainer, he said he was willing to “put my foot down” on that point.
“Linux is not one of those anti-AI projects,” Torvalds wrote. If people object to that, he said, they can fork the project or leave it.
AI as another maintainer tool
Torvalds described AI as a tool comparable to other tools developers already use. He said he now sees its usefulness as settled, even if that may have been less obvious a year earlier. He did not say every AI-generated output should be trusted, and he acknowledged that questions remain about the economics around AI.
His argument was narrower: the kernel project should not ban or discourage other developers from using LLM-based tooling merely because some contributors dislike it. Torvalds said nobody is being required to use AI. He also said he would ignore arguments aimed at stopping others from using it.
That distinction matters in kernel process terms. Maintainers already carry much of the burden for reviewing patches, triaging reports, and pushing back on low-quality submissions. Torvalds said AI can add pain to that workload, and he also joked that it can be painful because it keeps finding embarrassing bugs.
His proposed answer was not avoidance. Torvalds said the project should make LLM tools help maintainers instead of making their jobs harder.
Technical merit over tool fear
Torvalds also rejected the idea that defects in AI systems are a reason to dismiss the category. He said AI is imperfect, but pointed out that human judgment is hardly flawless either.
The message landed in a sprawling thread with more than 99 listed messages, involving developers including Mauro Carvalho Chehab, Laurent Pinchart, Derek Barbosa, Matthieu Baerts, Jason Gunthorpe, Steven Rostedt and others. The thread began in late May and was still active after Torvalds’ July 14 reply.
Torvalds framed the policy question as a familiar one for Linux: whether a tool improves the technology. He wrote that the kernel community does open source because it produces better technology, not for religious reasons. Social collaboration matters, he said, but he called it a side benefit rather than the project’s purpose.
His closing line was the operative policy signal: kernel decisions, in his view, should be made mainly on technical merit, not fear of new tools.
This story draws on original reporting from Daring Fireball.
