{"slug": "linus-torvalds-finally-admits-ai-saved-his-ass-on-a-nightmare", "title": "Linus Torvalds finally admits AI saved his ass on a nightmare", "summary": "Linus Torvalds credited Anthropic's Claude AI with helping him debug a Linux kernel locking issue, saying the model spotted a missing `d_lock` acquisition in a code path that human reviewers had missed twice. The fix was five lines, found in forty seconds, and Torvalds noted the AI 'connected dots I didn't have time to connect,' though he remains skeptical about AI writing net-new kernel code. He still ran the stress test suite for twelve hours to verify the fix.", "body_md": "# Linus Torvalds finally admits AI saved his ass on a nightmare\n\n[Claude](/en/tags/claude/)), asked it to reason through the lock ordering, and the model spotted a missing\n\n`d_lock`\n\nacquisition in a code path that human reviewers had missed twice.The fix was five lines. Finding it took the model forty seconds.\n\nWhat's interesting isn't that AI found a bug — static analyzers do that. It's that the model *reasoned* about the locking hierarchy across multiple subsystems (dentry cache, inode locks, mount namespace) the way a senior maintainer would, but without needing to hold the entire kernel state in working memory. Linus's exact phrasing: \"it connected dots I didn't have time to connect.\"\n\nHe's still skeptical about AI writing net-new kernel code — \"I don't want hallucinated locking primitives in my RCU paths\" — but this debugging use case has shifted his stance from \"overhyped autocomplete\" to \"genuinely useful for the boring forensic work.\"\n\nPractical takeaway if you're doing systems work: feed the model *context*, not just the crash site. Give it the header files, the lockdep annotations, the relevant `.c`\n\nfiles within three hops of the fault. The more constraints you provide, the less it hallucinates. I've started keeping a `debug-context.md`\n\nin each subsystem directory — lock hierarchy diagrams, common race windows, recent fixes — specifically to paste into these sessions.\n\nAlso worth noting: the model didn't replace the verification step. Linus still ran the stress test suite for twelve hours. AI proposed the hypothesis; silicon confirmed it. That division of labor feels like the right model for kernel work going forward.\n\n[Linus Torvalds says AI 'enormously helped' a debug session from 7h ago](/en/news/7197/)\n\n[Captain makes travel planning actually usable via Telegram 7d ago](/en/news/6331/)\n\n[Claude Code actually writes a decent novel if you stop treating 8d ago](/en/news/6125/)\n\n[Linus Torvalds thinks AI is fundamentally changing how the Linux 10d ago](/en/news/5905/)\n\n[Linus Torvalds thinks AI is actually helping the Linux kernel 11d ago](/en/news/5778/)\n\n[Building a custom AI code review agent is way cheaper than the 11d ago](/en/news/5745/)\n\n[Next The Leiden Declaration on AI and Math just dropped — has anyone →](/en/news/7246/)", "url": "https://wpnews.pro/news/linus-torvalds-finally-admits-ai-saved-his-ass-on-a-nightmare", "canonical_source": "https://promptcube3.com/en/news/7251/", "published_at": "2026-08-22 01:50:22+00:00", "updated_at": "2026-08-22 02:12:40.933862+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-tools", "ai-research"], "entities": ["Linus Torvalds", "Anthropic", "Claude", "Linux kernel"], "alternates": {"html": "https://wpnews.pro/news/linus-torvalds-finally-admits-ai-saved-his-ass-on-a-nightmare", "markdown": "https://wpnews.pro/news/linus-torvalds-finally-admits-ai-saved-his-ass-on-a-nightmare.md", "text": "https://wpnews.pro/news/linus-torvalds-finally-admits-ai-saved-his-ass-on-a-nightmare.txt", "jsonld": "https://wpnews.pro/news/linus-torvalds-finally-admits-ai-saved-his-ass-on-a-nightmare.jsonld"}}