Building as Research

For a long time, good product development meant validating an idea before you built it. That made sense when building was expensive. If an idea would take months and real investment to test, you wanted confidence before you committed.

AI is changing that equation. When you can go from an idea to a working product in days or weeks, building itself becomes part of the research.

I’ve been testing that myself. I collect trading cards, and for the past several months I’ve been building something called Cellopack.

It started with a problem I had. I’ve bought and sold mostly on eBay, where I’ve built up years of history and feedback. But as I started doing more deals through Facebook and other collector communities, I realized how little of that reputation traveled with me. The best proof I had that I was legitimate was basically: “Go look at my eBay.”

That felt like a problem worth solving. So I built Cellopack: one profile where a collector brings together their identity, their marketplace and social accounts, and vouches from other collectors, then shares it wherever a deal happens.

I didn’t spend months validating the idea first. I built the thing I wanted, started using it, and put it in front of other collectors. Now I’m learning where my problem overlaps with theirs, and where it doesn’t. Some of my assumptions are holding up. Others aren’t. The learning is happening against a real thing, not a slide.

There is a trap in it, though. Building for yourself gives you unusually good context on the first user. It also gives you a sample size of one. So I’m holding what I built loosely, and paying attention to where other collectors want something different than I did.

AI made the building cheap enough to start this way. It didn’t make discovery less important. It just means discovery doesn’t always have to come first. For me, Cellopack is an experiment in exactly that.

If you want the longer version, how the trust model works and which parts I rent, I wrote that up in the Studio.