The system that paid for itself

In 1993 I helped build a machine that mailed you a ticket for running a red light in New York City.

Not the whole thing, and not by myself — I want to be careful about that, and I’ll come back to it. But I was an engineer on the EDS team that built New York’s first automated red-light camera enforcement system. It’s one of the earliest things I ever shipped that a lot of strangers felt in their actual lives. Some of them, I’d guess, are still a little annoyed about it.

Here’s how it worked, roughly. A camera watched an intersection. Somebody ran the light, and it caught the moment. Then a person — an actual human being — reviewed the image to make sure it was a real violation and not a glare, or a bad angle, or a car that had every right to be there. If it held up, the system generated the notice and put it in the mail. Capture, review, notice. That was the loop.

We built it in PowerBuilder, with a stack of imaging code underneath — a library I’d found the way you found things then, from an ad in the back of a Dr. Dobb’s. And in 1993, “imaging” wasn’t a checkbox you ticked — it was the whole ballgame. Get a picture off a camera, into a system, and in front of a reviewer, clear enough and fast enough to stand up as evidence. There was no Stack Overflow. There was no Internet to speak of. I remember going into a dusty lab to find the technical manuals — eight inches thick, the kind you carried with both hands — and somehow pulling together the integration that got photos into PowerBuilder. You read, you tried it, you read again. That was the loop underneath the loop.

It was also where I was learning to think like an architect. I carried James Rumbaugh’s Object Modeling Technique — that beige book — back and forth to work, reading it on the way in and the way home, teaching myself object-oriented design one diagram at a time while the real thing was being built around me. And the nights were long. I did a lot of that late, with Pearl Jam’s Ten turned up loud — an album that flat-out blew me away and still takes me straight back to that desk.

But the part I actually think about, 33 years on, isn’t the code. It’s the people.

We were a crew of twenty-somethings from all over, thrown at one hard problem, and somewhere in the middle of it we stopped being a roster and became a team. That was the first time I ever watched that happen up close — and I’ve been chasing the feeling in every team I’ve built since. Everybody had their thing. One guy would wander into your cube looking like he had something to discuss, and the actual purpose of the visit turned out to be to fart and leave. Another had a victory move — an inimitable run-in-place — that came out whenever anyone shipped something good. When the project was done, they all came to my house and we played water volleyball in the pool. I can still see it.

And I haven’t even told you about the hotel. We stayed somewhere near Times Square that was, let’s say, economical, and a few of us were fairly sure some kind of business was being conducted a little too close to our rooms for comfort. This was pre-cleanup Times Square. You slept with one eye open and swapped notes at breakfast.

The part I still think is a little wild: in 1995, ABC’s PrimeTime Live ran a segment on it. Something I’d worked on, on a network newsmagazine, on a Thursday night, in front of millions of people. I was not on camera. I was the guy who knew why the review queue behaved the way it did.

Now the honest part, because the numbers from that program are good — and they’re better precisely because they aren’t mine to make up. New York’s DOT reported that over the three-year contract, violations at the enforced intersections dropped about twenty percent. Sixty-five percent of the people who got a notice paid it on the first letter. And the one that stuck with me: the program collected $18.5 million against a total cost of $15.5 million. A public-safety system that more than paid for itself. They extended the law and added twelve more cameras.

Let me be precise about what’s mine in that, though, because it’s easy to round up. Those are the program’s outcomes, not my personal box score. My boss, Bill Purcell, was the team leader. I was an engineer on his team. I didn’t cut anyone’s violations by twenty percent — the city did that, with the law and the cameras and the people who ran the whole operation. What I did was help build the machine underneath, the one that had to actually work, every day, at the scale of New York City traffic. That’s the part I’m proud of, and it’s plenty.

Because here’s what that job taught me early, and I’ve never unlearned it. Software that touches the real world doesn’t get to be almost right. A fantasy football league can survive a weird score. A red-light notice cannot go to the wrong person — or the whole thing collapses, the trust and the law and the money all at once. There’s a human reviewing every single image for a reason. The stakes are a letter in a stranger’s mailbox, with their name on it.

It’s been 33 years. The stack is unrecognizable — I build with AI and Rails now, not PowerBuilder and imaging libraries and eight-inch manuals. But the job hasn’t changed a bit: build the thing that has to work when you point it at the real world, and then let it work.

And after all this time, all the projects, all the tools that came and went — that one is still practically impossible to beat. A machine that paid for itself, a Thursday night on network TV, a sketchy hotel off Times Square, and a pool full of twenty-somethings who’d just built something real together. Of all the projects, all the teammates, all the teams, it’s the one I think about the most.