There’s nobody quite like Linus Torvalds to reframe the whole debate around AI assisted coding.
The creator of the Linux kernel and Git has never been one for diplomatic hedging, and a recent statement from him cuts right through the breathless claims that have been flooding the tech industry. While CEOs and venture capitalists compete to announce that AI is writing ninety, ninety-five, or even a hundred percent of their code, Torvalds has a pointed question: what were compilers doing all this time?
“My opinion has always been that AI is a great tool, but it’s a tool,” Torvalds said. “And when I see people saying, ‘Ninety-nine percent of our code is written by AI,’ I literally get angry. Because those same people — I pretty much guarantee that one hundred percent of their code is written by compilers. But they never say that, right? It’s exactly the same issue.”
The analogy is sharp and, once you sit with it, hard to dismiss. Compilers have been translating human-readable code into machine instructions for decades. Nobody holds a press conference to announce that their C++ is being “written by GCC.” The abstraction is invisible, and so it goes unmentioned. Torvalds is essentially pointing out that AI code generation is another layer of the same abstraction — useful, meaningful, productivity-enhancing, but not the categorical transformation it’s being sold as.
He went further: “I’m personally one hundred percent convinced that AI is changing programming, but it’s not changing the fundamentals. Exactly the same way that you all use compilers to actually generate your code, you will all — well, not maybe all of you, but a lot of people will — use AI to generate the code, that the compilers use to generate the code, that the assemblers then use to generate the machine code.”
That’s a precise description of where we are. AI sits above the compiler in the abstraction stack, not outside of it. The chain — human intent to AI-generated source to compiled binary to assembled machine code — is longer now, but the underlying logic of abstraction layering is the same one that has been building since the earliest days of computing.
“This is revolutionary in the same sense that we’ve seen revolutions before,” Torvalds said. “And AI will increase your productivity by a factor of ten, and I claim that compilers increase your productivity by a factor of a thousand. So AI is great, but AI is not changing programming. It may be changing other areas, don’t get me wrong, but I’m a programmer, so I don’t care.”
That last line has a very Torvalds quality to it — the programmer who is unbothered by what AI is doing to every other field, focused entirely on the question that actually matters to him.
The broader context makes Torvalds’ framing all the more useful. The claims about AI-written code have been escalating rapidly. Anthropic CEO Dario Amodei has put the timeline for AI writing ninety percent of code at three to six months. Microsoft CTO Kevin Scott has said ninety-five percent of code will be AI-generated within five years. Sam Altman has predicted that AI will make coders ten times more productive — a number that lines up exactly with Torvalds’ own estimate, though Altman frames it as a revolution and Torvalds frames it as more of the same. A quarter of Y Combinator startups are already reported to have more than ninety-five percent of their code generated by AI.
What Torvalds is objecting to isn’t the capability claim — it’s the framing. Describing AI as “writing your code” while ignoring the compiler that has always been writing your code is a selective use of abstraction to make a marketing point. And it’s one that obscures a more honest conversation about what’s actually changing and what isn’t. The fundamentals of programming — problem decomposition, logical reasoning, architectural judgment — remain where they have always been. The tools to express those fundamentals keep evolving. That’s been true since assembly language gave way to FORTRAN, and it’s true now.
For working programmers, that’s a more grounded way to think about the moment. AI tools like Cursor, Claude Code, and GitHub Copilot are real and genuinely changing how code gets written day to day. But the programmer who understands what the code is doing — who can frame the right problem, catch the wrong abstraction, and judge whether the output is any good — remains the irreplaceable part of the stack. Torvalds has been making that argument for decades, just about different tools. The argument hasn’t changed because the underlying truth hasn’t.