I chair a group that sells AI and automation to regulated businesses. I also build software most weeks: products, internal tools, hobby projects, this website. People sometimes ask why someone in my seat bothers.
The honest answer is that building is how I stay honest about what I'm selling.
Selling without building drifts
I spent around twenty years selling technology. The best salespeople I worked with understood the product well enough to say no. The worst sold the brochure. They weren't dishonest. They just had no feel for where the product stopped and the slide began.
AI makes that gap wider than it has ever been. The demos are extraordinary. The distance between a demo and something that runs every day on real data, with real exceptions and a real bill at the end of the month, is where most of the work is. You only really feel that distance by walking it yourself.
What building teaches
Three things I'd never have learned from a deck.
What it costs. Not the list price per token. The real cost: the retries, the context you didn't need to send, the model that's three times the price for a result you can't tell apart from the cheaper one. Building is how I learned to ask about the bill before I ask about the capability. It's also where Analect, ETT's platform for forecasting AI cost, started.
Where it breaks. Every system I've built has broken somewhere I didn't expect. Usually at the edges: the odd input, the slow system on the other end, the case nobody wrote down. That's exactly where client projects break too. Having felt it on my own work, I go looking for it first on theirs.
How fast things move. The tools I use to build today are different from the ones I used six months ago. If I only read about that, I'd be guessing. Because I use them, I can tell a client what's genuinely ready and what's still a demo.
What I build
Most of it is on the build log. A platform for trading card collectors and the shops that sell to them. A live map for amateur radio operators. Cost and accuracy tools that became part of Analect. This site, which I built in a day with an AI coding assistant doing a lot of the typing.
None of it is client work, and none of it needs to be impressive. It just needs to be real.
Why it matters to the people I work with
When a board asks me whether an AI project will work, they're not asking for my opinion on AI. They're asking whether I've seen this kind of thing hold up in the real world. Building gives me a small, constant supply of the real world.
It also makes me a better buyer's advocate. I know which questions make a weak vendor uncomfortable, because I've had to answer them about my own work.
So I'll keep building. It's the cheapest way I know to keep my judgement sharp, and the most fun.