The Terminal

I never wanted to go into the software development business.

I was in the parking business. Garages, ramps, night shifts, damage claims, guests who want their car faster than physics allows. That was my world, and I liked it there. The software came into my life the way a tow truck does — because something was broken and nobody else was coming.

So for 12 years, I got software built. Not by writing it — by carrying it. I described every screen, every flow, every detail to a human developer. I tested what came back. I broke it on purpose, explained how it should actually work, and tested again. When the hardware didn't exist, I made it exist — my garage access control was assembled in my backyard with my own toolbox, because the scanner wouldn't mount on anything in the catalog. I walked the garages. I took the support calls at 3am. That was AVPMi. Airports, hotels, hospitals.

In 2016, I sold the business.

And I went sailing.

Almost ten years. I was done — genuinely, happily done. A beach chair, a cooler, and not one single thought about software. I had built the thing, sold the thing, and owed the industry nothing.

Then AI arrived. And from that beach chair, I watched the AI companies make the same mistake I had spent 12 years paying a middleman to work around: they built for the people who already speak their language, and waited for the rest of the world to learn it.

I've written about that mistake before — the 95% problem, and what it looked like from inside a garage in 2004. That series asked a question I couldn't answer yet: could an operator, alone, take AI all the way to enterprise-grade systems? When I wrote "the real test starts now," I meant it. I just didn't tell you I was running the test on myself.

Ground 0 is the result.

Four months ago, I opened a terminal for the first time in my life. A black screen and a blinking cursor. I'm not exaggerating — first time ever. Twelve years in the software business and I had never once seen the machine without a person in front of it.

Before you decide what that says about me, be clear about what a terminal is: the one part of building software that was never the qualification. Whether the builder is a human or an AI, software gets used and projects get finished on everything else — knowing exactly what needs to happen before anyone builds anything; designing the flow the way the night shift actually runs it, not the way a diagram wishes it ran; catching the edge case that only shows up at 1am in the rain with a full ramp; breaking the thing on purpose before a customer breaks it by accident; testing until it stops surprising you; and staying on it until it is finished — because a system that's 90% done is 0% usable. That was my job for 12 years. Each of those is a skill, and the coming episodes take them apart one at a time.

So nothing about my process changed. Describe what needs to happen. Test what was built. Break it. Explain how it should work. Test again. The same loop I ran for 12 years — except the answers came back before my coffee cooled, and nobody billed me for the conversation.

Four months of that loop, and here is the scoreboard: ValetParking AI, as an on-site system and as SaaS. Belldesk for hotels. Locker — for anything a guest hands you and expects back: cars, coats, bags, skis, repairs; the phone number is the claim ticket, and a tap takes the payment. And my own CRM, built the way I run sales instead of the way a vendor decided I should. Five systems. In production. Tested the way human-built software gets tested — bugs found, fixed, tested again, every single field. PCI compliance. Security standards. Live credit card payments. Validation. Fully automated deployment. That list gets its own episode, because a claim without receipts is just marketing.

Now understand what the five systems actually are. They're the receipts — not the product. Parking is where I proved the method, because parking is the industry I know down to the 1am shift change. But the method itself — one operator with decades of domain knowledge, AI as the builder, no middleman — doesn't care which industry it's pointed at. Restaurants. Clinics. Marinas. Warehouses. Every operation in the 95% has the same shape: someone who knows the job cold, and software that never matched it. What I did for parking in four months, any deep operator can do for their industry with this method. The systems are worth what systems are worth. The method that produces them, repeatably, across every industry AI can't reach on its own — that's the part with a B in the price tag.

I came back because everything AI needed to reach the parking business was sitting in my head, not in anyone's code. The terminal just finally let me deliver it without permission.

This series is the record: how it started, what broke, what I'd tell an operator starting tomorrow, and one warning the software industry will not enjoy.

Is five systems enough? No. We're building the unthinkable. That's the finale.