--- title: Linus Torvalds Used AI in Linux Kernel Development. The... description: Linus Torvalds used Claude Code and other LLM tools in two separate Linux kernel efforts. In one, he explored an experimental branch addressing folio reference... canonical: https://daily.dev/posts/linus-torvalds-used-ai-in-linux-kernel-development-the-interesting-part-wasn-t-just-the-code--mri7b3yk9 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Linus Torvalds Used AI in Linux Kernel Development. The Interesting Part Wasn’t Just the Code. | daily.dev og:description: Linus Torvalds used Claude Code and other LLM tools in two separate Linux kernel efforts. In one, he explored an experimental branch addressing folio reference... og:url: https://daily.dev/posts/linus-torvalds-used-ai-in-linux-kernel-development-the-interesting-part-wasn-t-just-the-code--mri7b3yk9 og:image: https://api.daily.dev/og/posts/mri7b3yk9.png og:image:alt: Linus Torvalds Used AI in Linux Kernel Development. The Interesting Part Wasn’t Just the Code. og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Linus Torvalds Used AI in Linux Kernel Development. The Interesting Part Wasn’t Just the Code. **[Julius AQ](https://daily.dev/sources/zkzgtvnwpfu9mycqdvz6r)** · [@iaparatudia](https://daily.dev/iaparatudia) · 4 min read · 76 upvotes · 14 comments ## Summary Linus Torvalds used Claude Code and other LLM tools in two separate Linux kernel efforts. In one, he explored an experimental branch addressing folio reference count contention during page-cache reads, crediting Claude Code with code, test programs, and draft documentation, while running performance measurements himself; the change remains experimental and unmerged. In the other, he used LLM tools to make the hard-to-read assoc_array library more understandable, splitting functions and running a user-space test harness that confirmed a reported bug and uncovered three additional bugs dating back to 2013. The piece argues the real value of AI coding tools lies less in generating code and more in reducing uncertainty: helping developers understand, test, and trust existing code. ## Content _Two recent experiments offer a useful perspective on AI-assisted engineering: generating code matters, but understanding and validating it matter just as much._ Two reports published on October 11 and 12 describe Linus Torvalds’ recent use of AI-assisted development in the Linux kernel. The headline is interesting enough: the creator of Linux is using AI coding tools. But after looking at both examples, I think the more useful story is how those tools fit into the engineering process. These were two separate efforts, and they illustrate different ways AI can help. 1. Claude Code and an experimental performance improvement In an experimental branch named noref-filemap-read-with-llm, Torvalds explored a change to how Linux reads data from the page cache. The technical issue involves contention over folio reference counts when multiple threads perform small reads from the same file. The proposed approach aims to avoid taking those references during page-cache reads. According to Phoronix’s October 11 report, Torvalds explicitly credited Claude Code with helping produce the code, test programs, review work, and an initial draft of the patch explanation. He also stated that he ran the measurements on his own machine. That detail matters. AI assisted with implementation and supporting work, but the reported performance measurements were conducted by Torvalds. The initial results showed substantial improvements in the specific workload being tested. They do not establish a general performance improvement across Linux systems. And as reported on October 11, the changes remained experimental, with no confirmed merge into the mainline kernel. A promising benchmark is a starting point, not the end of the engineering process. 1. Using LLMs to understand code and uncover bugs The second example involves the Linux kernel’s assoc_array library. As Phoronix reported on October 12, developer Nils Ollrogge had identified a bug and submitted a patch. Torvalds found the existing code difficult to follow, so he used LLM tools to help make it more understandable. The work involved splitting some large, hard-to-read functions into smaller pieces. Then the revised code went through a test harness running in user space. According to Torvalds’ account, reproduced by Phoronix, the tests confirmed the originally reported bug and exposed three additional bugs. He said those additional problems dated back to the library’s introduction in 2013. The reporting does not establish that Claude Code was the specific tool used in this second effort, so it’s important not to conflate the two cases. Nor should we describe this as an AI independently discovering and fixing four bugs without human involvement. What makes the example valuable is the combination of AI-assisted code analysis, improved readability, and testing. The AI wasn’t just being used to produce more code. It was helping make existing code easier to investigate. A better question for AI-assisted development These examples made me think about how we measure the value of AI coding tools. We often focus on how quickly an agent can generate a patch, modify a repository, or complete a task. Those are useful measurements. But I’m increasingly interested in another question: How much uncertainty did the AI help us eliminate? Did it help us understand unfamiliar code? Did it make a proposed change easier to review? Did it generate tests that challenged our assumptions? Did it help uncover a problem we might otherwise have missed? And can we explain why the resulting code should be trusted? I don’t mean that these two kernel experiments prove AI reduces uncertainty by some measurable amount. They don’t. I’m proposing uncertainty reduction as an additional way to evaluate AI-assisted engineering. In one case, AI helped explore a performance change that still requires broader validation. In the other, AI-assisted analysis and testing helped expose problems in longstanding code. Both examples remind me that generating an implementation is only part of the work. The takeaway I build with AI tools, and I don’t think the goal should be to maximize how much code an agent writes. The goal should be to build software we can understand, test, maintain, and trust. Sometimes AI helps by writing code. Sometimes its most valuable contribution may be helping us understand the code that’s already there. For developers using AI coding agents: are you spending more time generating code, or improving the tests and review processes needed to trust it? ⸻ Julius Founder, IA para tu Día Sources - Phoronix — Linus Torvalds Working On A Hefty Linux Kernel Improvement With Claude Code (October 11, 2026) - Phoronix — Linus Torvalds “Did What I Preach”, Used AI To Make Code Decipherable & Uncovered More Bugs (October 12, 2026) - Linux kernel — Experimental noref-filemap-read-with-llm branch ## Community discussion Top comments from developers on daily.dev. **@orkeilius** · 32 upvotes > So once again, IA is just a tool, result depend on who,when,where,how and for what your using it ? **@ezpie** · 10 upvotes > he knows what he's doing but I'm pretty sure majority of my bones don't **@nosfair** · 5 upvotes > AI coding productivity shouldn't be measured in lines of code generated, but in uncertainty eliminated. **@eliasdevv** · 4 upvotes > ai‑generated patches have to survive the kernel's static analysis and the full kselftest suite, otherwise they can easily break lock ordering and introduce hidden races. did you also run a targeted fuzzer or use lockdep traces to catch subtle regressions? **@dignity31** · 3 upvotes > i agree ## Similar posts on daily.dev - [Linus Torvalds "Did What I Preach", Used AI To Make Code Decipherable & Uncovered More Bugs](https://daily.dev/posts/linus-torvalds-did-what-i-preach-used-ai-to-make-code-decipherable-uncovered-more-bugs-tsqqcmos3) · Phoronix · 1 upvotes · 0 comments --- Tags: [#linux](https://daily.dev/tags/linux), [#ai-coding](https://daily.dev/tags/ai-coding), [#claude-code](https://daily.dev/tags/claude-code), [#code-review](https://daily.dev/tags/code-review) [View this post on daily.dev](https://daily.dev/posts/linus-torvalds-used-ai-in-linux-kernel-development-the-interesting-part-wasn-t-just-the-code--mri7b3yk9) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"DiscussionForumPosting","mainEntityOfPage":"https://daily.dev/posts/linus-torvalds-used-ai-in-linux-kernel-development-the-interesting-part-wasn-t-just-the-code--mri7b3yk9","headline":"Linus Torvalds Used AI in Linux Kernel Development. The Interesting Part Wasn’t Just the Code.","text":"Linus Torvalds used Claude Code and other LLM tools in two separate Linux kernel efforts. In one, he explored an experimental branch addressing folio reference count contention during page-cache reads, crediting Claude Code with code, test programs, and draft documentation, while running performance measurements himself; the change remains experimental and unmerged. In the other, he used LLM tools to make the hard-to-read assoc_array library more understandable, splitting functions and running a user-space test harness that confirmed a reported bug and uncovered three additional bugs dating back to 2013. The piece argues the real value of AI coding tools lies less in generating code and more in reducing uncertainty: helping developers understand, test, and trust existing code.","url":"https://daily.dev/posts/linus-torvalds-used-ai-in-linux-kernel-development-the-interesting-part-wasn-t-just-the-code--mri7b3yk9","datePublished":"2026-10-13T13:49:05.562Z","dateModified":"2026-10-13T13:49:36.149Z","author":{"@type":"Person","name":"Julius AQ","url":"https://daily.dev/iaparatudia","image":"https://media.daily.dev/image/upload/s--fyjHBSHd--/f_auto/v1790748730/avatars/avatar_zkzgTvNwPFu9mYCQDVZ6R?_a=BAMAMicg0","description":"Creator of IA para tu día · I explain AI, test it and build with it. 🇲🇽","interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"EndorseAction"},"userInteractionCount":3200}},"image":"https://media.daily.dev/image/upload/s--_C25T4Zz--/f_auto,pg_1/v1791899355/posts/mri7b3yk9?_a=BAMAMicg0","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":76},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":14}],"comment":[{"@type":"Comment","text":"So once again, IA is just a tool, result depend on who,when,where,how and for what your using it ?","datePublished":"2026-10-13T18:52:57.201Z","dateModified":"2026-10-13T18:53:13.382Z","url":"https://daily.dev/posts/mri7b3yk9#c-Zd2mNSzzH","author":{"@type":"Person","name":"orkeilius","url":"https://daily.dev/orkeilius","image":"https://avatars.githubusercontent.com/u/63045634?v=4"},"interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":32}},{"@type":"Comment","text":"he knows what he’s doing but I’m pretty sure majority of my bones don’t","datePublished":"2026-10-14T01:26:46.456Z","url":"https://daily.dev/posts/mri7b3yk9#c-QDWd6kanT","author":{"@type":"Person","name":"Ezpie","url":"https://daily.dev/ezpie","image":"https://avatars.githubusercontent.com/u/104765117?v=4"},"interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":10}},{"@type":"Comment","text":"AI coding productivity shouldn’t be measured in lines of code generated, but in uncertainty eliminated.","datePublished":"2026-10-14T06:10:43.926Z","url":"https://daily.dev/posts/mri7b3yk9#c-qgiJ27GV5","author":{"@type":"Person","name":"Thialon","url":"https://daily.dev/nosfair","image":"https://avatars.githubusercontent.com/u/70216808?v=4"},"interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":5}},{"@type":"Comment","text":"ai‑generated patches have to survive the kernel’s static analysis and the full kselftest suite, otherwise they can easily break lock ordering and introduce hidden races. did you also run a targeted fuzzer or use lockdep traces to catch subtle regressions?","datePublished":"2026-10-14T01:59:05.724Z","url":"https://daily.dev/posts/mri7b3yk9#c-bJ7BthlHw","author":{"@type":"Person","name":"elias-devv","url":"https://daily.dev/eliasdevv","image":"https://avatars.githubusercontent.com/u/338986921?v=4"},"interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":4}},{"@type":"Comment","text":"i agree","datePublished":"2026-10-14T00:43:10.997Z","dateModified":"2026-10-14T00:43:32.942Z","url":"https://daily.dev/posts/mri7b3yk9#c-CzItqqzFk","author":{"@type":"Person","name":"Dignity","url":"https://daily.dev/dignity31","image":"https://media.daily.dev/image/upload/s--xbFCRIaC--/f_auto/v1789511341/avatars/avatar_27xFR2iYDQ2YHngeEMexT?_a=BAMAMicg0"},"interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":3}}],"isPartOf":{"@type":"WebPage","url":"https://daily.dev/sources/zkzgtvnwpfu9mycqdvz6r","name":"Julius AQ"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Julius AQ","item":"https://daily.dev/sources/zkzgtvnwpfu9mycqdvz6r"},{"@type":"ListItem","position":3,"name":"Linus Torvalds Used AI in Linux Kernel Development. The Interesting Part Wasn’t Just the Code."}]} ```