{"slug": "rocket-to-nowhere", "title": "Rocket to nowhere", "summary": "Developer Mario Zechner expresses exhaustion with the AI landscape, calling it pointless. The author argues that AI-generated software feels dead because it lacks a step-wise creation process with feedback, unlike natural or living structures. Speed without direction creates 'rockets to nowhere.'", "body_md": "“The past months have been mentally exhausting. I’m kind of sick of the entire AI landscape. It all feels pointless.” —\n\n[Mario Zechner]\n\nI share this feeling. Here I’ll try to understand why.\n\nLately, I’ve been finding a lot of solace and guidance in Christopher Alexander’s book, The Process of Creating Life (which is volume 2 of his magnum opus, The Nature of Order). Brutally simplifying, his core argument is: natural processes have a high degree of life. Inanimate objects can also have a high degree of life. This degree of life is ultimately an arrangement in space. This degree of life is objective, but is *felt* by individuals. An objective feeling of being represented by something is what triggers our sense of that something being alive.\n\nMy intuition is that most of what we’re doing with AI feels pointless because it has an extremely low level of life. In general, software has been quite dead, even when we handcrafted it. But most AI generated software, I wager, has an even lower level of life. It’s deadening to use and to experience. That’s why it feels pointless.\n\nWhy most AI generated software is more dead than traditional, dead-ish handcrafted software? In book 2, Alexander makes the crucial point: you can only generate objects with life if you follow a certain step-wise process where each step gives you feedback to correct each change.\n\nWhen you fabricate something all at once, assembling different parts, without a clear step-wise sequence that allows for adjustments, you end up with dead structure. This is already the case most of the time in everyday modern architecture or software. When we use LLMs to quickly create structure, doing hundreds of steps at once without feedback, we create structure that’s even more dead than before. This is why it feels pointless, no matter how superficially impressive it is. Perhaps the term “slop” simply means “dead”.\n\nDead structure feels disconnected: disconnected from yourself, from its surroundings, and from itself. It has a distinct “nobody home” feeling that cannot be pinpointed to any specific flaw. Interacting with dead structure induces a subtle deadening by a thousand paper cuts: little mistakes that compound into a disconnected whole.\n\nNote: if you quite disagree with Alexander’s view above, you might want to stop reading. The rest of the article will be pointless unless you already are in alignment with the above, since the rest builds upon it.\n\nI believe it is possible to create living software using LLMs, if we employ Alexander’s step-wise approach. LLMs are fantastic information sources, sparring partners and checkers. But the human must drive the process, step-wise, using their own feeling to perceive (not judge) the rightness of each step. There is a very interesting open question (and perhaps a fertile field): will AI be ever able to perceive feeling, and thus direct the creation of structure with a high degree of life? I have no idea. In the meantime, we can use AI to create living software, but we need to drive this process ourselves.\n\nIn the mainstream there is today an incredible emphasis on the speed of AI. This is understandable, because indeed it is mindboggling what’s now possible to do in a short amount of time. But this speed, without a proper feedback mechanism of generation and verification (through feeling, no less), we’re just creating rockets to nowhere. We’ll get there superfast, but there’s nothing there awaiting us that makes sense. Speed only has value if we are working on a right direction.\n\n(I wonder if an emphasis on speed is the telltale of every collective mania, financial or otherwise. Time will tell about this one. I do find that those most focused on the speed of this revolution are the ones least aware of its shortcomings.)\n\nHow to go about creating living structure? I recommend Christopher Alexander’s last books. I’m making some notes about them in [these development notes](https://github.com/altocodenl/vibey/blob/main/notes.md) but my quotes don’t do the books justice.\n\nIronically, Mario Zechner, who I quoted above, has built a tool (the [pi harness](https://pi.dev)) that is by far the most living piece of software I’ve seen emerge since the onset of AI. Nothing else that’s new comes even close to the sheer level of “rightness” that pi has. It’s not about features (Lord knows everything’s packing tons of features nowadays). You have to experience that rightness by yourself.", "url": "https://wpnews.pro/news/rocket-to-nowhere", "canonical_source": "https://federicopereiro.com/rocket-to-nowhere/", "published_at": "2026-06-18 14:58:25+00:00", "updated_at": "2026-06-18 15:23:31.665561+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-ethics", "ai-research", "generative-ai"], "entities": ["Mario Zechner", "Christopher Alexander", "The Process of Creating Life", "The Nature of Order"], "alternates": {"html": "https://wpnews.pro/news/rocket-to-nowhere", "markdown": "https://wpnews.pro/news/rocket-to-nowhere.md", "text": "https://wpnews.pro/news/rocket-to-nowhere.txt", "jsonld": "https://wpnews.pro/news/rocket-to-nowhere.jsonld"}}