Twenty Years of Range

People keep asking how one person ships this much: a memory platform, a 32-worker runtime, a nightly chain, this site and the case studies under it. The assumption baked into the question is that the answer is the AI. It isn't. The AI is real leverage — but leverage needs something to push against. What it's pushing against is twenty years of range.

Here's the range, and it does not read like a plan.

Film school. Visual effects. Super Bowl spots. Three years with the Sherman Bears. Theatrical runs. A stretch at B&H, learning gear at the counter where working professionals actually buy it. iOS since version 4 — back when "there's an app for that" was still a novelty. Then a decade of enterprise product design: Wegmans, Walt Disney World, AB InBev — shipping native software to millions of people who'd never know my name. Program management. Project management. The connective tissue where you learn that the hardest part of building something is deciding what to build, and in what order.

That's not a résumé. A résumé is a list of jobs. This is a trained eye and a pair of builder's hands, assembled out of crafts that don't obviously belong together — and that lack of obvious belonging is the whole point.

The bottleneck was never the eye

Nobody tells you the real consequence of spending twenty years across that many disciplines: the taste arrives long before the ability to act on it.

By year ten I could look at almost anything — an interface, a cut, a system, a pitch — and know, fast, whether it was right. What was off. What it wanted to be. That judgment is the expensive thing, and it compounds; every discipline you add sharpens it against the others. Film taught me pacing. VFX taught me that the last ten percent is the job. The retail counter taught me how professionals really decide. Enterprise design taught me that the user is never you. Program management taught me that a great idea shipped late is a worse idea.

But for two decades, the eye was always faster than the hands. I could see the finished thing with total clarity and then spend three weeks — or three months, or a whole budget cycle — dragging reality up to meet it. The bottleneck was never taste. It was throughput. The tools made me wait.

The tooling finally stopped making me wait

This is the part the "AI makes anyone a builder" framing gets exactly backwards.

AI didn't hand me a speed I never had. It externalized a speed I'd had internally for years and could never get out. The eye was always fast. Now the hands can keep up with it. When I stand up a worker, or restructure a dashboard, or build a memory graph in an afternoon, I'm not discovering the answer as I go — I mostly already know what right looks like, and for the first time the distance between seeing it and having it is short enough to cross in one sitting.

That's a genuinely new feeling, and it is not the feeling of getting smarter. It's the feeling of a throttle coming off something that was always straining against it.

Anyone can prompt. Very few can aim.

This is where the "it levels the playing field" story falls apart, and it's worth being honest about.

Prompting is typing. Everyone can do it, and the models are good enough now that everyone gets something back. Aiming is different. Aiming is knowing what to ask for, recognizing when the answer is subtly wrong, killing the plausible-but-bad direction before it costs a week, and holding the whole shape of the thing in your head so the pieces actually compose. That judgment isn't in the model. It's the twenty years. A benchmark will never surface it, because a benchmark measures the tool — and the tool was never the variable.

The uncomfortable corollary: this cuts both ways. AI is a multiplier, and multiplication is brutal on small numbers. It amplifies whatever judgment you bring — including none. It isn't the great equalizer it's sold as. It's the great amplifier, which is a different and much less comfortable thing. The people it makes dramatically more capable are the ones who already had the eye and were starving for the hands.

The rare thing is the combination

I can build Coquina alone — design it, decide what it's for, and actually write the infrastructure — because I spent twenty years collecting the positions. I can play designer, product manager, and engineer in the same afternoon, not because I'm the best at any one of them, but because I've done enough of each to know what good looks like and what to skip. The AI is what lets me play all three at speed instead of one at a time.

Neither half is the story alone. Range without throughput is a person full of things they can see and can't ship — I was that person for a long time. Throughput without range is a very fast way to build the wrong thing, confidently. The rare, valuable thing is the seam between them, and almost nobody is standing on it, because it takes two decades of unrelated-looking work to earn the range and a very recent tool to unlock the throughput.

The speed was never new to me. It was just, finally, external.