{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreigqpm3sq5imn4svia5yzxn73mz7vngseweoucwp4gplhp7hcjfdti",
    "uri": "at://did:plc:bqma3dxvtfkv542aaek7xf6c/app.bsky.feed.post/3mhy64q4hkn52"
  },
  "path": "/blog/2026/03/26/open-source-and-ai/",
  "publishedAt": "2026-03-26T17:50:32.659Z",
  "site": "https://kusma.xyz",
  "tags": [
    "It sometimes can be",
    "critical issue to open source",
    "license-incompatible",
    "output Quake 3 source code",
    "insisting not to worry",
    "made it clear",
    "AI scrapers",
    "taking advantage",
    "robots.txt",
    "residential IP addresses",
    "Anubis",
    "Mesa 3D",
    "fewer competent senior developers",
    "how bad",
    "prevent us from knowing",
    "told that",
    "we should",
    "it for all problems",
    "electricity",
    "memory",
    "storage",
    "new GPU",
    "new computer",
    "any new gadgets",
    "circular economy",
    "actually profitable",
    "currently in a bubble",
    "dot-com bubble",
    "Keeping the brain sharp requires work, not assistance",
    "data",
    "yelling"
  ],
  "textContent": "More and more frequently, I get asked about my stance on AI in the context of programming. This is my attempt to summarize my stance for those who wonder.\n\nThis is a blog post that I don’t want to write, but some recent developments have more or less forced my hand here. I would have preferred to keep pretending that I’m neutral in the issue, and just hoping that the problem goes away. But that doesn’t seem to be happening.\n\nI’m probably not the most qualified person to write about this, I’m sure you can find better informed articles out there. These are my personal opinions, and not those of my employer, their customers or any other of my affiliates. Take them with a pinch of salt, and feel free to disagree.\n\nOn a general note, I’m very reluctant to tell people how they should behave. But in this case, I’ve decided to do just that. I hope it’s clear why from the context.\n\nSimilarly, I would caution every reader to be skeptical of anyone who claims to know what the future holds, me included. People often predict the future that benefits themselves the most. There’s a few times I make some predictions in this post. Those are just predictions, and I might very well be wrong.\n\nA final warning; this is a long post, so set aside some time. I’ve tried to limit the scope somewhat to mostly cover topics concerning open source development, but I sometimes end up discussing wider issues. This is simply because I don’t feel like I can ignore these.\n\nBut yeah, let’s start close to home here…\n\n## Licensing issues - The “plagiarism machine” __\n\nCurrently, the legal status of AI generated code is still far from clear. Is it derivative work of _all of the training data_ or not? It sometimes can be, but it depends on a lot of factors.\n\nThis is a critical issue to open source; I can’t submit code somewhere that I don’t know the origin of and potentially is license-incompatible with the upstream project. This isn’t just theoretical, Copilot has been known to output Quake 3 source code with the wrong license. It doesn’t matter if the people with _the most to lose_ from AI output counting as derivative works keeps insisting not to worry.\n\nThe US supreme court recently made it clear that AI generated code isn’t even copyright-able. A human needs to write the code for copyright to be granted. But with using AI tools, we’re blurring the line, making it hard or even impossible to tell what is written by a human and what isn’t.\n\nI also doubt that a single or a handful of lawsuits is going to be enough to settle this. We’re working in a global ecosystem, and there’s potentially hundreds of jurisdictions that might have to rule, and just as many subtleties to take into account before we have a good understanding around this. It’s going to take a long time to find out.\n\nBut even if this wasn’t an issue, does that _really_ mean we _should_ use AI? Open source software is inherently political, _especially_ when it comes to licensing. I tend to find something _not being illegal_ to be a terribly low standard to have. It should IMO also be the _right thing to do_. This brings us to the other issues…\n\n## The cost of AI scraping __\n\nAs open source developers, it’s important for our infrastructure to be publicly available to everyone. In recent times, AI scrapers have started taking advantage of this, and are now aggressively scraping all content on open source support infrastructure so they can train their models. These scrapers often ignore robots.txt directives, and sometimes even use randomized, residential IP addresses, making it hard or impossible to effectively block them,\n\nAll of this has a major financial impact on the open source projects. It’s not unusual to see over 90% of traffic _provably_ coming from AI scrapers.\n\nAs a result, the open source community has had to introduce barriers, like Anubis, which slows down initial page-loads. Since Anubis is based on proof-of-work, it means people with slow computers can no longer reach our infrastructure. And I can’t reliably browse our GitLab instance from my phone on the bus to work.\n\nWhile the latter is a minor annoyance, the former is a real problem for inclusivity.\n\nAnd because our infrastructure is so heavily affected by this, it feels deeply problematic to me if we use (and pay for) the tools that are built on this behavior. That would be rewarding the behavior. We should vote with our wallets here, and in this case this means to not pay them.\n\n## Maintainability issues __\n\nAnother issue is that code needs to be understood and maintained in the long term. For this to work well, we need to be able to reach out to the people that wrote the code and get input on what led to a decision. Obviously, that’s not always possible, but with AI this is almost _never_ possible. The context is lost, and so is all the insight. Asking an AI again about the same code might lead to completely different reasoning, and miss crucial details.\n\nThe project I’m mostly working on, Mesa 3D, is also arguably critical infrastructure for a lot of computer systems around the world. We need to be lean towards being conservative rather than experimental when building these kinds of systems.\n\n### The junior problem __\n\nAnother related issue is that AI technology tends to be used to take over more “junior” tasks, but the result of this is likely to be that we end up hiring and mentoring fewer junior developers. This will lead us to having fewer competent senior developers in the future.\n\nInteracting with an AI isn’t going to gradually make the AI learn and become more senior, unlike with a junior developer. AIs learn from training, not queries. Mentoring junior developers builds _trust_ , which makes the interactions worthwhile also for the senior. In my experience, interactions with AIs are little other than frustration that never improves. And because working with the AI doesn’t build any meaningful _trust_ , the AIs will always need guard-rails to prevent disasters.\n\nA future where we develop software with few to no human developers (junior _or_ senior) sounds scary to me, but that’s where this path leads.\n\n## Environmental issues __\n\nBuilding and running these huge data-centers is extremely resource heavy. Some of these resources are resources we all have to share on this planet, like water, electricity and rare earth minerals. This is taking a toll on our planet and everything living on it.\n\nI feel like this point got a lot less attention recently than it used to, but it hasn’t really been solved. Instead, the AI giants have just doubled down on wanting to consume all the resources they feel they need, without regard for the planet or people living on it. They are far from truthful about how bad this is, and try to prevent us from knowing just how bad it is.\n\nThe truth is that AI-type solutions are almost always one of the _most resource intensive_ solutions to problems possible. And right now we’re being told that we should use it for all problems. This is a recipe for disaster, nothing less.\n\nAnd it seems like there’s nothing being done on this front. The big AI companies are just slowly boiling the ocean, hoping that we don’t notice or that we forget. I haven’t forgotten.\n\n## Economical issues __\n\nA secondary effect of building all these data-centers is that demand for a bunch of resources goes up, and so does the price. This affects everyone.\n\nWe’re not just seeing electricity and water being more scarce, we’re also seeing memory and storage prices spiking hard as well. Forget about buying a new GPU, and just generally wait a couple of years with buying a new computer, or really any new gadgets.\n\nHow can this _possibly_ not lead to a recession if things are allowed to continue?\n\nAnd then we have the blatant circular economy that the big AI players are doing to try to convince the market this is actually profitable. In reality, very little actual money is changing hands, they’re mostly just making promises to buy tech from each other in the future… Which brings us to the big one…\n\n### The bubble __\n\nYeah, so it seems very likely we’re currently in a bubble. We’ve been for a while, and this bubble _is_ going to pop. The question is when and how.\n\nDon’t get me wrong; not all bubbles pop and erase everything with it. The dot-com bubble took years to pop, and we still have computers and the internet and all that jazz.\n\nBut we’re currently overspending on infrastructure, and the companies selling that are currently raking in, and they are trying hard to make us all dependent on their technology.\n\nFor the last few years, the AI industry has slurped up most of the traditional technology investment capital available. The investors seem less and less interested in investing more money into the AI industry, and want return on their investments instead. So they have started turning to things like pension funds. If they get away with this, _everyone_ is going to pay for this, regardless of their involvement in AI.\n\nWe’ve already been seeing the idea of “too big to fail” being thrown out there, mirroring what happened in the subprime mortgage crisis. We should, as a society refrain from letting them do this. These problems are caused by the AI industry, not by us consumers. We shouldn’t be the ones to bail them out when the time comes.\n\nOpenAI’s CFO has already suggested that the U.S. government should provide a $1.4 trillion “safety net” for AI investments, and while Sam Altman since has walked that back after public outcry, this shows that these companies are already thinking along those lines.\n\n## Keeping the brain active __\n\nOn a more personal note, it’s kinda undeniable; I’m getting older, and part of getting old means that I need to spend more time actively thinking about things to keep up. Letting an AI take over the wheels, even just for the boring bits doesn’t help me, it only makes this worse. Keeping the brain sharp requires work, not assistance.\n\nIn fact, I often feel like I learn something useful, even when I do mundane tasks. Asking an LLM to write up a python script for me to do something robs me of learning in the process of doing it myself.\n\nAdd that to the data that suggests that we actually get _less_ productive by using AI (while thinking we’re more productive), makes this all very unappealing to me. My brain is my most important tool, and I’m not going to risk it because tech CEOs are yelling at the world that they need to use AI to prevent a recession.\n\n## Conclusion __\n\nYou might have noticed that I don’t really address the technical abilities of current AI technologies in this post. The reason is that I don’t feel like I need to; it’s kinda irrelevant.\n\nI think the moral arguments against using AI for open source development are just too large to ignore. In fact, just the licensing and environmental issues _alone_ would probably have been enough for me to draw a hard line in the sand here:\n\nUsing AI for open source projects is in my opinion immoral, and I will _not_ be using it. I _do not_ condone others using AI for anything in the open source ecosystem either. Using it is simply detrimental to our values and directly harms our community.\n\nIf you’re currently playing around with AI out of curiosity for open source projects, I would like to ask you to reconsider. If you’re working in a company that’s encouraging AI usage, I would like to ask you to speak up against it. If you are involved in policy decisions for open source projects, I would like to encourage you to try your best to discourage AI adoption within those projects.\n\nOur entire ecosystem is on the line here. Not just the open source ecosystem, but the entire, global ecosystem. And I feel there’s not enough voices speaking up about it.\n\nMake your voice heard! Allow yourself to be angry; there’s enough nonsense going out there! We need to stop this madness.",
  "title": "Open Source and AI",
  "updatedAt": "2026-03-26T16:30:00.000Z"
}