An Agent Running Now Beats a Perfect One

A builder in one of the cohorts came to a 1 on 1 in week two still redesigning the architecture. The spec had grown since day one, the data model had two more tables than it started with, and nothing was running yet. I asked what the smallest version was that someone else could actually try and tell him where it broke. He didn't have an answer. He had a diagram.

If you're still in planning mode, or still asking Claude to redesign the architecture, or extend the specs, the thing you're missing usually isn't more design. It's feedback from someone who isn't you. This is a good moment to stop and think if your idea is concrete enough and how a potential user can push you towards clarity.

A working demo that does one thing clearly will teach you more than a half-built system that does five things poorly, because only the working one can be handed to someone and get you an honest reaction back. A diagram can't surprise you. A person looking at your rough demo can.

The scope question worth asking yourself is: what is the smallest version of this that I could hand to someone right now, and would using it tell me anything I don't already know? If the honest answer is still weeks away, the scope needs to shrink.