I built my parents an AI system they don't use

I'm going to tell you about one of my builds that failed. Not because it broke. It worked exactly as designed. It failed because nobody uses it, and the reasons why changed how I build everything now.

The build

My parents own The Cowboy Saloon in downtown Laramie. Bands email them wanting to play, dozens of inquiries piling up, and vetting each one properly took a couple hours: check their Spotify numbers, their YouTube, their online presence, their reviews. Real time, every week, on top of running a bar.

So I built them something I was genuinely proud of: a five-agent AI system. One agent sorted genuine band inquiries from spam. One analyzed Spotify data. One scouted their internet presence. One evaluated YouTube. A final agent scored everything into a ranked dashboard of the most promising acts.

It worked. The research that took hours came back in minutes, ranked and sourced. Technically, it might still be one of the cooler things I've built.

They don't use it.

What actually went wrong

The failure wasn't in the code. It was in everything around the code, and every reason is one I now check for before I build anything for a client.

I automated the interesting part. Picking bands isn't pure drudgery to a bar owner. It's taste, it's identity, it's their stage. I treated it as a research problem when a big piece of it was a judgment call they liked making. Remember the line between AI work and human work? I drew it in the wrong place, on a task that was partly relationship and craft.

The tool lived outside their workflow. The dashboard was its own thing to go check. Their actual workflow was an inbox. Every trip to a separate tool is a toll, and tolls kill habits. If I built it today, the results would land as a reply in the same inbox the inquiry came into.

It solved the problem I found interesting, not the one that hurt most. Five agents talking to each other is a fun engineering problem. My parents' most painful problems were elsewhere, and the systems I built them for those (their end-of-night reporting automation, the head counter app their bouncers run on busy nights) get used constantly. The contrast is the lesson.

What this bought me

That build taught me the questions I now ask before writing a line of code. Does this task hurt enough that you'd change a habit to kill it? Does the solution live inside a tool you already open every day? Is this task actually drudgery, or is part of it work you secretly like?

If the answers are wrong, I say so and we pick a different task. My pricing model says a build should pay for itself within months, and a tool nobody opens pays for itself never, no matter how clever the agents are.

Sometimes the tuition for that kind of lesson is a client relationship. I'm lucky mine only cost me a dinner conversation with my parents.

-Thayne