Linus Torvalds Loves Vibe Coding — What the Linux Creator Gets Right About AI Tools

Introduction: The Skeptic Who Came Around

If anyone earned the right to be cynical about AI coding tools, it's Linus Torvalds. The creator of Linux — the operating system kernel that runs most of the internet — has spent 2026 publicly complaining that AI bug hunters have made the Linux security mailing list "almost entirely unmanageable." Meanwhile, much of the open-source world is actively hostile: just this week, System76 banned AI-generated content from its COSMIC desktop outright.

So it turned heads when, at Open Source Summit Europe in Prague this week, Torvalds told his longtime friend Dirk Hohndel of Ericsson: "I actually really like using AI." In a conversation reported by ZDNet on October 9, the man who invented Git and Linux explained a philosophy of AI-assisted coding that is more nuanced — and frankly more useful — than most of the hype or the backlash. It boils down to one line: "I use AI to do the things that I'm bad at."

1. 'I Use AI to Do the Things I'm Bad At'

Torvalds is quick to clarify what he doesn't use AI for: kernel code. "I don't do a lot of kernel programming. I'm a maintainer," he said. "The Linux kernel is a project where other people do the real work, and I act as the collection point."

Instead, his canonical example is a personal toy project: an effects pedal for guitars he's building, with a user interface he'd written himself in C running on a small microcontroller screen. "It looks like something from the '80s; I wanted it to look like something from this century. But I don't do Java. I do kernels."

So he did what millions of hobbyists do — prompted an AI model to build the modern UI. "It did it, and it was horrendous," he laughed. But it worked, and it let an expert in one domain ship something in a domain he'd never invested in. That's the pattern worth noticing: AI as a bridge across your own skill gaps, not a replacement for your expertise. He already knew exactly what he wanted and how the underlying code should behave — the AI just handled the parts outside his wheelhouse.

2. Vibe Coding and the Joy Problem

The most quoted line of the talk is his defense of the most mocked term in software: "A lot of people make fun of vibe coding, but I think it's a wonderful way to find joy in programming. It makes you feel like you're doing something relevant. It allows you, as a new programmer, to do things that you would otherwise have a really hard time with."

His argument is historical. When Torvalds started programming around 1981, computers were simple enough to fully understand, and the programs you compared yourself against were simple too. "I made computer games which sucked, but it was fine because everything else also sucked back then," he said at the summit. Today, a beginner's first attempts are measured against decades of polished, professional software — and their "pitiful, small efforts" feel worthless by comparison. That discouragement kills careers before they start.

AI coding assistants flatten that curve. A newcomer with an idea can build something that actually works in an afternoon, get the dopamine hit that keeps them learning, and gradually absorb the underlying concepts. Torvalds calls AI "a gateway drug" into programming — half-joking, but the point is sincere: the entry bar has to be low enough that people can start, enjoy it, and stay long enough to get good.

3. The Warning: Toy Projects Before Serious Work

The endorsement comes with a sharp asterisk. "You need to be very careful using AI when you're doing something real and important," Torvalds cautioned. To use AI well, you have to know what you want and what the correct outcome looks like — which is precisely the knowledge beginners lack. His advice: start with toy and personal projects, and only bring AI into serious work once you can evaluate what it produces.

That maps exactly onto the failure modes making news this year. AI-generated code vulnerabilities are piling up in enterprise codebases; the same week as his talk, Walmart was reportedly reining in unrestricted AI coding after token costs and quality concerns ballooned. The tool amplifies whatever judgment you bring to it. Torvalds could AI-generate a UI for his guitar pedal because he'd already built the hard part correctly in C. A novice shipping AI-written payment code has no such safety net.

4. The Other Side: Kernel Maintainers Are Drowning

Torvalds didn't soften his other message: AI has broken the human side of open source. He revealed that three-quarters of the preceding day's Linux Kernel Maintainer Summit discussion was consumed by how to handle AI-generated patches and reviews. "The bottleneck isn't getting patches," he said. "It's getting useful work through review without exhausting the people responsible for it."

Since AI bug hunters began flooding the kernel's security lists in May, maintainers have faced a deluge of machine-generated reports that are individually plausible but collectively unreviewable. It's the same asymmetry Torvalds flagged with joy: AI makes producing code nearly free, which is wonderful when you're the producer and corrosive when you're the reviewer. Expect tooling that filters, labels, and rate-limits AI contributions to become a whole product category in open source.

5. AI Coding Tools That Fit the Torvalds Philosophy

Torvalds didn't name products, but his criteria are easy to apply: use AI where you lack expertise, keep verification in your own hands, and let it lower the barrier — not the standards. A few tools from our directory that fit:

• GitHub Copilot — the default entry point for in-editor completions; good for the "I already know what I want, speed me up" pattern.
• Claude Code — an agentic terminal assistant that shows diffs and lets you audit every change before it lands.
• Cursor — an AI-first editor popular with vibe coders, with enough context handling to keep beginners unblocked.
• Replit — zero-setup, browser-based building; the closest thing today to the "start with a toy project" on-ramp he describes.

The common thread: pick tools where you stay in the review loop. That's the whole Torvalds doctrine in one sentence — treat AI as a tool, keep the judgment human.

Frequently Asked Questions