Point of View

AI tools are only as interesting as the person using them.

There’s a version of this moment where builders hand off more and more to models and assume quality will follow. It won’t. A model with no direction produces an approximation. Technically correct, shaped like a good idea, but isn’t one.

I prompt all day. At work, my job has shifted from writing code to directing a system that writes it. In my personal projects, I use LLMs to sharpen my writing, pressure-test my thinking, accelerate the boring parts of building. The tools are in every part of my work now.

One-shotting anything good is nearly impossible. Not because the models aren’t capable. Because I’m not capable of externalizing my vision well enough on the first try. The gap isn’t the prompt. The gap is between what I think I want and what I actually want. The model makes that gap visible faster through conversation. That’s useful. But closing it still requires knowing what I’m after.

I can’t delegate the interesting part. I can’t prompt my way out of not having a point of view. The goal, the taste, the judgment about when it’s done is not in the model. It’s mine. If I don’t have a compelling point of view going in, I won’t end up with anything worthwhile.

The tools are good now. They’ll take me most of the way. But most of the way only matters if I know where I’m going.