Linus Torvalds has started vibe coding, just not on Linux

Skye Jacobs

Posts: 1,987   +58
Staff
Ripple effect: Linus Torvalds, the creator of Linux and Git, has quietly joined the ranks of developers using generative AI to write code – though only as part of a small personal project. His recent experiment with vibe coding shows how even the most traditional programmers are beginning to incorporate AI tools into their workflows, at least for low-stakes development.

The project, called AudioNoise, appeared on Torvalds' GitHub earlier this month. Written primarily in C, the program explores digital audio effects and signal processing. It grew out of a hardware experiment he conducted last year called GuitarPedal, a series of homemade guitar effects pedals he built to learn more about analog circuits. Torvalds has since given the devices away as gifts to kernel developers and even to Bill Gates.

For AudioNoise, Torvalds used Google's Antigravity AI, a coding assistant built to generate and refine programs via natural-language input. In this case, the tool generated a Python-based audio sample visualizer that complements Torvalds' C routines.

In the project documentation, Torvalds wrote that "the Python visualizer tool has been basically written by vibe-coding," adding, "I cut out the middleman – me – and just used Google Antigravity to do the audio sample visualizer."

His comments suggest he was satisfied with the results and felt no need to manually rewrite the AI-generated code. He also noted that when working in unfamiliar languages, he often relies on online resources or code snippets, a habit most programmers can relate to.

While vibe coding has been mocked online, it has gained traction as AI tools like Copilot, ChatGPT, and Claude AI replace traditional platforms such as Stack Overflow for quick code assistance. That Torvalds – long known for his skepticism toward hype-driven technologies – would take up vibe coding at all has surprised many in the open-source community.

The experiment also reminds us how AI is steadily permeating the Linux ecosystem itself. Developers are beginning to use AI models for tasks like debugging, patch triage, and code maintenance, work that was once handled entirely by human contributors.

Torvalds has continued to warn against the marketing excess surrounding artificial intelligence, stating, "I hate the whole subject of AI, not because I hate AI, but because it's being such a hype word." Yet, he has also emphasized that he is "a huge believer in AI as a tool."

Torvalds' use of Antigravity for AudioNoise is less about automation and more about creative exploration. By applying AI to a project far removed from the Linux kernel, he frames the technology as an extension of craft rather than a challenge to it.

For developers still questioning whether AI-assisted code generation has a place in real programming, Torvalds' hobbyist approach may mark a subtle turning point, one where even the most committed traditionalists are willing to let the algorithms jam along.

Permalink to story:

 
Recently I've spent hours studying the Vulkan API – read all the docs, checked out Intel's guide, used AI for 'explain it like I'm five style'. But the sad truth? I still can't 100% manually(no docs, no AI) initialize Vulkan to show a single triangle without external help. It's because there are hundreds of things that need defining just during initialization.
 
I’ve recently taken to having different AI double check each others work. They catch a lot of things, then I look over it. Still faster than me typing everything myself.
 
Coding a little hobby tool? Sure.
Coding an operating system? No.

Why? You know how many humans are insanely bad at optimizing code? Many are programmers but only a handful are actually good at it and very few are extremely good

Bad software optimization and code is the main reason for bugs, slowdowns and crashes

AI would improve code in most cases and AI will work 24/7 without breaks, food or sleep - It is a nobrainer going forward

I trust AI code more than most human code - Just review the code.
I have seen my fair share of terrible terrible human code, even from big companies - Most coders are really bad and do too many workarounds to get stuff to work - lost ressources

Most coders will be replaced by AI in the coming years, especially the bad and mediocre ones. It is already happening.
 
I’ve recently taken to having different AI double check each others work. They catch a lot of things, then I look over it. Still faster than me typing everything myself.

Using AI to generate code that you don't understand what is happening? Bad.
Using AI on human generated code to determined simple mistakes? That's fine.
Using AI on human generated code to determined what this other f#*(@$* person was thinking because they didn't leave comments? Awesome.
 
Using AI to generate code that you don't understand what is happening? Bad.
Using AI on human generated code to determined simple mistakes? That's fine.
Using AI on human generated code to determined what this other f#*(@$* person was thinking because they didn't leave comments? Awesome.
I hadn't thought about using to on noncommented code (I'm usually on solo projects). Genius!
 
I just wonder how many obscene words did he use in the prompt before reaching the desired effect. It appears, with ChatGPT itis a common practice, driving people to learning new languages for extending their swear vocabulary.
 
Back