One argument keeps surfacing in discussions about AI and software development, repeated often enough that it has started to feel like common sense: AI will take over the routine work, but judgment will remain human. Developers will shift from writing code to evaluating it. From production to oversight. That sounds reasonable and in fact, it is already happening from yesterday. But it contains a hidden assumption that rarely gets examined. Judgment is outcome of learning. The ability to recognize a…

Yes, this is not a new thought today, I just wanted to return to this analogy and do another comparison for me. The history of technology is full of examples when a tool does more than make work faster. It changes what the work is, who can do it, and where value is. The loom was one of those tools. And so is AI in the sense of LLM. That is why the comparison between weavers in the first industrial…

Tom has never really used a hand saw. He has seen it working. Clean cuts, straight lines, wood behaving as if it wanted to cooperate. But he had never felt what it takes to get there – the resistance, the small corrections, the way a cut slowly drifts if you stop paying attention for even a second. Still, when he walked into the workshop and saw the chainsaw on the wall, he reached for that. Why wouldn’t he? That same…

TL;DR ChatGPT/Codex converted old Java applet code to JavaScript surprisingly well. But when I asked it to do what sounded like a simpler task – reformat and unify a set of old HTML tutorial pages – it struggled badly. The main lesson was simple: LLMs are excellent at well-defined transformations, but much less reliable when the task depends on many unstated assumptions. Why This Project Existed in the First Place This project started in 1998, when I was a computer…