By now, you must know that Linus Torvalds, the leader of the Penguin army, uh, I mean the Linux kernel project, does not shy away from being blunt, especially when it comes to making a point.When he has one, the Linux kernel mailing list hears about it, and so do we, the ones who use Linux. For the past few years, his opinion on AI was quite simple, being that it was mostly noise, and he was not interested.But something changed; Linus has gone from calling AI "90 percent marketing" to publicly telling critics inside the Linux kernel project to fork it or walk away. Here's how that change happened along the years.AI is hypeAt the Open Source Summit Japan in December 2023, Dirk Hohndel asked Linus a direct question. Would he ever see AI-written code submitted to the kernel?Linus said yes and guessed it might already be happening on a small scale. He wasn't particularly worried about the risk of hallucinated code either, pointing out that plain human error causes plenty of bugs on its own. "I think we're doing just fine at making mistakes on our own," he said.Four months later, at Open Source Summit North America in Seattle, that curiosity had turned into something closer to mockery. Where he joked that he might "be replaced by an AI model," and brushed off most of the surrounding hype as nonsense, suggesting everyone wait a decade before drawing any real conclusions.By October 2024, speaking to TFiR at the Open Source Summit in Vienna, Linus put a number on his skepticism. He called the AI industry "90 percent marketing and 10 percent reality," and said he'd mostly just tune it out until that changed. He left himself an opening though, predicting real use cases might show up in five years, once the hype cycle burned out.The slow burnA year later, in November 2025, something had changed. At Open Source Summit Korea, Linus told Dirk that he was "fairly positive" about vibe coding, though only for people learning to code, not for anything that actually mattered.He compared it to typing programs out of computer magazines as a kid, pitching vibe coding as a low-stakes way to get people excited about doing things on computers "that maybe they couldn't do otherwise."That openness turned into a holiday project a few months later, where he built an audio effects tool called AudioNoise, writing a portion of the code himself (mostly C) and then handing over the Python visualizer tool to Google's Antigravity IDE.Even so, the kernel itself stayed off limits. Just before announcing the project, he had replied to a heated thread about anti-LLM sentiment creeping into kernel documentation, pushing back on turning the issue into a political statement.His argument was that people submitting genuinely careless AI patches were never going to label them as such anyway, so writing rules aimed at catching them was pointless.That tension between personal enthusiasm and institutional caution showed up again in April 2026, when the kernel shipped its first official AI Coding Assistants policy alongside Linux 7.0.The rules were narrow but clear. AI-assisted code still had to meet GPL-2.0 licensing requirements. AI tools were barred from adding their own Signed-off-by tags, and a human always had to take responsibility for what got submitted through a new Assisted-by tag.That same month, stable kernel maintainer Greg Kroah-Hartman was found quietly running an AI-assisted fuzzing tool against the kernel from a branch nicknamed "clanker."Coming aroundBy May 2026, the volume of AI-generated activity around the kernel had become its own problem. Linus complained that a release candidate had ballooned with what he called pointless pull requests, much of it AI-assisted bug hunting that produced trivial fixes late in the development cycle. He said that he would get tougher about it, promising to be "a bit more hardnosed" from here on.At his keynote for Open Source Summit North America in Minneapolis, Linus had pushed back hard against a different kind of AI narrative, one he found more annoying than the code itself.Claims that "99 percent" of a project's code came from AI genuinely angered him, he said, since the same people never credited compilers for writing effectively all of their code either.By July, at Open Source Summit India in Mumbai, that optimism came with a correction. He admitted that his own earlier "10x" productivity estimate for AI tools had been guesswork, not a real measurement.He pointed out a real downside too. Separating a fabricated bug report from an actual one could take hours, since a hallucinated issue often doesn't announce itself until someone digs in and finds nothing there.Then, more recently on July 14, while replying to a mailing list thread discussing perceived anti-LLM sentiment in a recent post from the Software Freedom Conservancy, Linus settled the question for good."Linux is not one of those anti-AI projects," he wrote, telling anyone who disagreed to fork the project or simply walk away. While also saying that AI could still be a painful tool, both for maintainer workloads and for how often it turned up embarrassing bugs. But in his view, the answer wasn't to ignore the problem. It was making sure the tools actually helped the people maintaining the code, not just added to their load.Closing wordsThree years isn't long by kernel standards. Linus has been doing this since 1991, and some subsystems have taken longer to get right than that. But watching him go from "90 percent marketing" to "fork it or walk away" in such a short span says something.Unlike some people, he didn't stay stuck on the idea that AI is bad. He kept updating his opinion, in public no less, admitted his slip-ups, and still called out what's lacking.💬 So, are you still against AI? Let me know in the comments below; it's always insightful to see where our readers stand on this.