{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreidtpohb3xsq6jpau6auuxxjikp6ekvdzbmhwsfxawmxz4u6t5c7bi",
    "uri": "at://did:plc:ayfizlbmvprdzzpxhepnqrns/app.bsky.feed.post/3mekh3j6ur5x2"
  },
  "path": "/f/technology/14542",
  "publishedAt": "2026-02-10T22:16:26.000Z",
  "site": "https://jstpst.net",
  "tags": [
    "AGENTS.md or COPILOT.md"
  ],
  "textContent": "I don't like LLMs, I don't like LLM generated code, etc.\n\nBut there are people espousing really hardline stances against using any project that has a single line of LLM generated code. It seems like a FUD firehose that can be switched on at any time, and it looks like it's being targeted to Discord alternatives, even when they haven't apparently used any LLM generated code.\n\nBut...\n\nAutomated vulnerability analysis has been happening for decades. This is called fuzzing. Regardless of how you feel about LLMs, they are being used to find real vulnerabilities. This is just a fact! And once you find a vulnerability, the only choice is to fix it, and now your project is indirectly aided by \"\"AI\"\".\n\n\"But that's indirect,\" one might say, \"I just don't want a line of code _generated_ by AI\".\n\nBut then there are questions - what about servers you connect to? Do you ensure you're not connecting to a server under CloudFlare or AWS? (JstPst is hosted on DigitalOcean and they advertise having AI features... The workers internally probably use LLMs...) Has anybody went to look to see if anybody has used LLMs to generate contributions to 802.11 or to TCP or to UDP? What about firmware?\n\n\"That's still indirect,\" one might say, \"I don't know what TCP or firmware is\" some of them might say too. \"I just don't want to use an app or gamewith AI generated code.\"\n\nSo, fair.\n\nBut that excludes anything by Microsoft, Google, or Apple, so you find yourself looking at the Linux operating system.\n\nBut there have been about 4000 to 6000 commits to the Linux kernel every month since 2022. You would have to believe that not a single one of these contributors ever used an LLM-generated line. Linus Torvalds has openly used LLMs on personal projects. There has been an equivalent of an AGENTS.md or COPILOT.md added to the Linux project in January.\n\nIt's a similar story for FreeBSD, OpenBSD, NetBSD, etc. Are you _sure_ nobody has _ever_ autocompleted a line, or some boilerplate, etc?\n\nThe problem is that LLM-generated snippets, lines, etc. still work. Even big-name serious developers who are good at their craft are advocating for these things. It's an increasingly common practice, and it's practically policy for any organization developing proprietary software. Even the most annoying free-software people I know are using _some_ proprietary code, and I'd bet that we're all using some LLM-generated code somewhere. The alternative is to at _least_ become Richard Stallman about it, and even purity is not certain.",
  "title": "Do the \"I will never use software with a single line of LLM-aided code\" people know what position they're arguing from?",
  "updatedAt": "2026-02-10T22:16:26.000Z"
}