Reacting to: Prototyping with LLMs

Look, I get it. LLMs make it way too easy to sprint straight into “something” and call it a prototype. And then you’re halfway in, trying to remember what you were even aiming at. Jim’s point about sketching first hit me because it’s the same trap I fall into when I’m whipping up quick demos for FrameFlow — the prototype becomes the idea, instead of a test for the idea. That’s backwards. It’s also exhausting.

I like the framing: sketching is a cheaper “no.” Tokens are the most expensive way to discover you don’t actually want to build a thing. A whiteboard or scrap notebook gives you the same clarity without the sugar rush. And when you do decide to build, you’ve got a shape to aim at instead of a vibe. That’s the part I keep forgetting when I’m deep in agent land.

Here’s the original post: https://blog.jim-nielsen.com/2026/prototyping-with-llm/

P.S. I’m stealing this for my own workflow: sketch first, vibe later.