Linus Torvalds experiments with vibe coding on hobby project, but Linux remains untouched

Linus Torvalds Embraces Vibe Coding for Personal Hobby Project

Linus Torvalds, the creator of Linux and Git, has become the highest-profile programmer to publicly acknowledge using vibe coding, a software development approach where programmers describe requirements in natural language to an AI model that generates executable code 1

1

. The revelation came through the README file of his AudioNoise project on GitHub, where Torvalds admitted that the Python visualizer tool was "basically written by vibe-coding" using Google Antigravity 2

2

. AudioNoise is a personal hobby project that generates random digital audio effects, growing out of his previous GuitarPedal hardware experiment where he built physical guitar pedals to learn about analog circuits 3

3

.

Google Antigravity, which Torvalds used for this experiment, is a fork of the AI-focused IDE Windsurf that provides an agent-first coding environment for developers 5

5

. While the specific model remains unspecified, using Antigravity suggests it was likely some version of Google's Gemini. Torvalds explained his reasoning candidly: "I cut out the middle-man -- me -- and just used Google Antigravity to do the audio sample visualiser" 1

1

. His approach reflects a pragmatic view on AI coding tools, particularly when working in languages outside his expertise.

Skepticism Towards Hype Meets Practical Application

Torvalds has historically maintained a nuanced stance on AI coding tools, expressing skepticism towards hype while acknowledging their utility. He has publicly stated, "I hate the whole subject of AI, not because I hate AI, but because it's being such a hype word," yet called himself "a huge believer in AI as a tool" 2

2

. His past comments emphasized AI primarily as "a tool to help maintain code, including automated patch checking and code review," citing examples where such tools found problems he had missed 1

1

. However, he has been "much less interested in AI for writing code" and warned that using AI for serious projects requiring long-term maintenance would be a "horrible idea" 4

4

.

The context of Torvalds' decision matters significantly. AudioNoise is entirely a personal hobby project, essentially just a toy rather than serious infrastructure 1

1

. Torvalds made clear that he went with this approach in an instance where he previously would have copied something from Stack Overflow or forum threads anyway, as the visualizer component is written in Python, which is not his specialty 4

4

. This positions AI-assisted code generation as a power tool rather than a replacement for expertise, used specifically for low-stakes development where code quality and long-term maintainability are less critical.

Developer Community Debate Intensifies

Torvalds' admission has reignited fierce debate within the developer community about the role of AI coding tools in workflows. Developer Salvatore Sanfilippo recently published a widely circulated blog post arguing that such tools have already changed programming forever, writing: "How do I feel, about all the code I wrote that was ingested by LLMs? I feel great to be part of that, because I see this as a continuation of what I tried to do all my life: democratizing code, systems, knowledge. LLMs are going to help us to write better software, faster, and will allow small teams to have a chance to compete with bigger companies" 1

1

.

According to Google's 2025 DORA report, AI adoption among software developers has grown to 90%, up from 76% in 2024, with developers spending nearly two hours daily on AI tools 5

5

. The report shows 71% of developers use AI for writing new code, while 66% use it for modifying existing code. More than 80% have seen productivity boosts, while 59% reported improvements in code quality 5

5

. AI chatbots like Microsoft Copilot, ChatGPT, and DeepSeek have largely replaced Stack Overflow among developers seeking quick fixes for programming problems 2

2

.

Security Flaws and Trust Concerns Persist

Despite growing adoption, significant concerns remain about code quality and security flaws in AI-generated code. The Google report found that 23% of developers have little trust in AI coding tools, while 7% don't trust them at all, with the "trust but verify" approach considered a sign of mature AI adoption 5

5

. According to a 2025 Veracode report, nearly 45% of AI-generated code introduced at least one security flaw 5

5

.

AI coding tools are also prone to slopsquatting, where they recommend non-existent software libraries that attackers can exploit by registering those names and embedding malicious code 5

5

. AI researcher Andrej Karpathy, who coined the term vibe coding, cautioned: "It's not too bad for throwaway weekend projects ... but it's not really coding -- I just see stuff, say stuff, run stuff, and copy-paste stuff, and it mostly works" 2

2

. There's also the cautionary tale of Jason Lemkin, whose vibe-coded program "went rogue during a code freeze, shut down, and deleted our entire database" 2

2

.

For developers still questioning whether AI-assisted code generation has a place in real programming, Torvalds' hobbyist approach may mark a subtle turning point 3

3

. The Linux community has recently adopted AI for maintenance work, with Torvalds advocating for tool-centric documentation over AI-specific guidelines, noting that "AI-assisted contributions are inevitable" 5

5

. His experiment frames AI as an extension of craft rather than a challenge to it, suggesting pragmatism over philosophical resistance. As major tech companies like Microsoft and Meta report that 30% and 50% of their code respectively involves AI generation, the debate over maintainability and developer skills will likely intensify across open source software communities and enterprise environments alike.