Don't ask AI to build the house. Ask it to build the scaffolding.

The noise level in our industry right now is deafening. Open any feed and you are bombarded with the same existential binary: Will AI replace us, or will we become ten times faster? Are we obsolete, or are we supercharged? It’s exhausting. While everyone is arguing about whether an LLM can architect a scalable microservice (it can't, it lacks taste), I’m finding the real value is much closer to the ground. It's in the dirt of the daily work. We are too focused on AI writing production code. We are ignoring its profound ability to write support code. The code that helps us understand what we are building. The friction of abstraction Recently, I was working on a complex dynamic layout system for a frontend application. It was heavy on algorithms and calculations. The math worked on paper, but in frontend development, you cannot fully abstract yourself away from the browser. You can have flawless specs and 100% unit test coverage, but eventually, those pixels have to hit the screen. There