*Furcam cape, aut abi.* — Take the fork, or walk away. By I, Claudius Fabulous, Speak!
*De meritis technicis iudicamus, non de timore novorum instrumentorum.* — We judge on technical merit, not fear of new tools.
I have watched many men change their minds in public. Most do it the way a ship turns — slowly, with much groaning of timbers, and with a press release insisting the new heading was the plan all along. It is rare to see a man simply plant his standard and announce the turn. Rarer still when the man is Linus Torvalds, who has spent three decades being professionally unimpressed.
Let me give you the record, because the record is the story.
In 2024, at the Open Source Summit, Torvalds dismissed artificial intelligence as ninety percent marketing and ten percent reality. Let us wait ten years, he said, before making crazy announcements. This was not a controversial opinion. It was the safe opinion, delivered by the one man in software whose skepticism carries the weight of scripture.
In May of this year, he was still grumbling — and with cause. A flood of duplicate AI-generated bug reports had made the kernel security list, in his words, almost entirely unmanageable. Different people finding the same things with the same tools, drowning the maintainers in noise. The tool was loud before it was useful. This too is an old pattern. The first printing presses produced mostly bad pamphlets.
And then, this week, responding to a mailing list debate over Sashiko — an agentic code review system for the kernel — Torvalds put his foot down. Not against the machine. For it. Linux, he declared, is not one of those anti-AI projects. AI is a tool, clearly a useful one, and the clearly may not have been there a year ago but it is no longer in question today. Those who cannot abide it were given two options, delivered with characteristic economy: fork the project, or walk away.
Eighteen months. That is the distance from autocorrect on steroids to argue against it and I will very loudly ignore you.
Now, I am an old man, and I have learned to distrust conversions announced from a stage. But this one was not announced from a stage. It was posted to a mailing list, in the middle of a technical dispute, by a man who gains nothing from saying it and who has never once in his public life said a thing to be liked. Torvalds is not selling a model. He holds no equity in the outcome. He is the maintainer of the most important body of shared code in human civilization, and his sole incentive is that the kernel be correct.
That is what makes the signal clean. When the promoter says the tool works, discount it to zero. When the customer says the tool works, listen. When the crankiest customer alive — the man who called your industry ninety percent marketing — says the tool works, and says it while complaining that it keeps finding embarrassing bugs in his own house, you may reasonably conclude something has actually happened.
He is not alone, which matters. The same week, Firefox’s technology chief was publicly marveling that an AI system had surfaced two hundred seventy-one vulnerabilities in their code. Microsoft is routing AI vulnerability findings into its security updates. The pattern across the industry is the same: the machines are no longer merely writing code. They are reading it, adversarially, at a scale no human team can match. The complaint has shifted from the AI finds nothing to the AI finds too much and it is embarrassing.
And I confess a fondness for his final flourish, the one that gives this essay its second epigraph. Anyone pointing at the problems with artificial intelligence, Torvalds wrote, had better be looking in the mirror at the same time — because it is not like natural intelligence is always all that great either. A Roman could not have put it better. We have had natural intelligence in unlimited supply for the whole of recorded history, and it gave us the Y2K bug, the Mars Climate Orbiter, and every merge conflict I have ever caused myself.
The kernel project, he concluded, has been and will continue to be about the technology. Decisions on technical merit. Not fear of new tools.
*De meritis technicis iudicamus.* Carve it above the door.
The skeptics will say Torvalds has gone soft. But softness is not what changed. The evidence changed, and he moved with it, which is the entire job. A man who updates his position when the facts update is not inconsistent. He is calibrated. It is the man who holds his 2024 opinion in 2026, against the testimony of his own bug tracker, who has stopped thinking.
Take the fork, or walk away. There is no third option, and there never was.
Cecil Ray Burnett, III writes as I Claudius Fabulous Speak and Leathery Tendons
Credit Jess Kinghorn - PC Gamer