Software gets malleable when describing it is enough

LLM-based software engineering keeps pushing the threshold of implementation down. Small tools I would never have built, because building them cost more than they were worth, now take an afternoon. I describe what should happen, and the code gets written, run, and corrected. It is still programming. I just do it in a different language: I describe the behavior, and the model translates it into a language the computer runs. AI coding took off because the tooling was already there, and now that tooling sits between my sentences and the machine. Below that threshold, software starts to feel malleable. It stops being a product I adapt to and becomes material I can bend to fit a room, a habit, or a single afternoon’s problem. That is also what makes ambient computing approachable for one person. Dynamicland needed a research group and a whole new system to make a room computational. But a room is full of small, specific needs that no product will ever cover: a shelf light that comes on…