There is a difference between owning a piece of something and owning the whole thing. I have mostly done the second one.
When you found a business, there is no one above you to escalate to. The product, the brand, the money, and the customer who is unhappy at nine on a Tuesday night all end with you. I ran that kind of business more than once, end to end. I formulated a men’s grooming line from raw ingredients, manufactured it in small batches, built the brand and the packaging, ran the marketing, shipped the orders, answered the complaints, and carried the profit and loss. Before that I ran an investigations practice the same way: my name on the work, my judgment on the line.
Alongside the businesses, I led inside operations where being wrong was not a bug ticket. Federally regulated work, under DOT and FMCSA oversight, with zero-error standards and real liability. I was certified to train new people and sign them off — to say, on the record, that someone was ready to be trusted with equipment that could hurt people if it were handled badly.
That work teaches a specific discipline. You document the process so it can be repeated. You make your judgment defensible, not only correct. And you assume being wrong has a cost, because you have seen what it costs. None of it is glamorous. All of it carries straight into building software.
These days I build production AI, end to end: the assistant, the tools it calls, the retrieval that feeds it, the evals that keep it honest, and the observability that tells me when it has started to drift.
Here is the part the demos skip. AI acts. It writes the message, routes the request, flags the case, drafts the decision. The moment a system does something in the world instead of only answering a question, someone has to own what it does — the whole outcome, including the parts that go wrong when no one is watching.
That judgment does not come from a framework or a best-practices doc. It comes from having been the one accountable before: from having carried the P&L, answered to the regulator, signed off the person, and shipped the thing and then lived with what happened next.
So I build the whole system. I test it as if being wrong costs something, because being wrong has cost me before. I keep the observability running because I am the one who gets the call. I would rather understand a system completely and ship it a day later than ship something I could not stand behind.
I build with AI, too. It is my daily working partner — I pair with it the way I would with anyone whose output lands under my name: lean on it, check it, own the result. Using it well is part of the craft now. So is knowing when not to trust it.
The industry is about to hand a great deal of consequential work to systems that act on their own. Many of the people building them have never been personally on the hook for an outcome that landed on a real customer, a real regulator, or a person who got hurt. I have. That is the whole of what I bring: I build the software, and I have been the business it serves.
I build the whole thing because I am the one who has to live with it. It is the only way I know how to work, and it is the only kind of work I trust.