Torvalds allows AI to hunt kernel bugs yet bars it from running Linux development — a carve-out that turns his personal AI rule into a lesson in trust.
Two verdicts on AI-written code, issued by the same engineer, pointing in opposite directions. Prompting a model to generate software for fun — vibe coding — is, in the words of Linux creator Linus Torvalds, "a wonderful way to find joy in programming." Pointing AI at the Linux development process itself is not on his menu.
The logic behind both positions sits in a single line from Torvalds: "I use AI to do the things that I'm bad at." His framing treats models as a compensator for personal weak spots, not as a stand-in for competence at the center of his work. Where he feels strong, the tool stays out. Where he feels weak, it gets invited in. Read that way, his AI policy is less a stance on the technology than a precise inventory of his own abilities.
Inside Linux work, the boundary gets finer still. Using AI to find bugs is acceptable, per his comments; letting it run development is not. That carve-out separates the mechanical pattern-matching of defect hunting from the judgment calls of shepherding a project — and he puts the latter firmly in human hands.
The tone matters as much as the line. Nothing in his comments dismisses AI coding outright; the objection is to scope.
Vibe coding earns its praise, in his telling, precisely because the stakes are low — it is recreation, a way to enjoy the craft. The kernel is where consequence lives, and there the models stay outside the process. For anyone tracking how deeply AI embeds itself in critical software, that distinction between playground and production is the takeaway.
萬安算交所 AIXX · 算力硬件实名会员制交易平台 · 进入货源大厅