Twenty years that look like scatter — and weren't.

Half a dozen businesses, starting with a lawnmower

Landscaping first, then a string of others, up through photography and the agency. Some worked, some taught me what doesn't. All of them meant the same loop: work out what somebody actually wants, build the thing, charge for it, and find out immediately whether you were right.

Running your own business is the fastest available education in being wrong. There's no department to absorb it. The market tells you, in money, within weeks.

A boutique ad agency, and the layer I never got tired of

I built clientele, ran the work, and did every part of it — but the layer I loved was always conceptual direction. Not the execution: the idea that makes the execution obvious. Get that right and everything downstream stops being guesswork.

The clearest case was an air show pilot in his first two years, up against the marketing departments of Red Bull, Oracle and the major oil companies — every one of them with a sponsored pilot and a real budget. I was the entire department: identity, photography, video, print, booth, collateral, web. We placed 2nd and 3rd in consecutive years in the International Council of Air Shows (ICAS) Marketing Competition.

One hand across every surface produces a coherence a committee structurally can't. Their identity went through brand guardians and sign-off chains. Ours went through one person at every touchpoint, so nothing drifted. The shoestring didn't win despite being one person — it won partly because of it.

The first ceremony was a dinner in a ballroom on the last day of ICAS Vegas. I stepped out to the bathroom, and that's when they announced it. I came back to the table and Jon was sitting there with an enormous grin, so I asked him what was funny. He got to be the one who told me.

• 
  Jon Melby Airshows — the full case study Identity, air-to-air photography, the campaign film, the booth and the website — every surface, alone. It lives on the Creative door, where there is room to read it.

Photography — where all of it actually started

My first creative role was shooting weddings in 1997. Then portraiture, editorial, commercial, studio. Eventually a self-published coffee-table book, which meant the photography, the marketing, the print production and the writing were all mine — my first taste of owning every surface of a thing rather than one slice of it.

Weddings in particular teach something no studio can: one take, no reshoot, and a client who will look at those images for fifty years. You learn to see the shot before it happens, and to know instantly whether you got it.

Video — story with a clock running

The eye carried over, but moving image adds the thing a photograph never has to solve: time. Order, pace, what you withhold and for how long. I write the story and cut the production, and I hold a Part 107 for the aerial work.

It's also the discipline that maps most directly onto directing AI video now — the models have made generation cheap, and left every hard problem exactly where it was. Consistency, continuity, whether a cut earns its place.

Design and brand — the systems underneath

Identity, type, color, layout — and the rules that hold them together across surfaces that never sit side by side. A logo is the easy part. The hard part is a system coherent enough that someone else could extend it without asking you.

That's the same skill as writing an operating contract for an AI build: you're not making the artifact, you're making the thing that decides what the artifact should be.

The one that doesn't fit

Six chapters in, this is the one that looks like it wandered in from someone else's résumé. It didn't. It was here first.

From 1998 to 2000 I designed, built and maintained Intel's internal web sites — back when the way you got that job was being the person who had already worked out how. Then two decades of everything above: the agency, the cameras, the businesses, the brand systems.

I came back in 2019, this time validating silicon and server hardware — processors, motherboards, the occasional PCIe card. No cameras, no clients, no concepts. Something has been designed and built, and before it goes near a customer, somebody determines whether it actually does what it claims. Before the org restructured I led that team: seven reports, and I did the hiring.

Here's the part that belongs in this timeline. In 2021 the lab had a problem nobody owned: over a thousand pieces of pre-production silicon across several labs and hundreds of trays, with no tracking and no tray identity. Finding a part meant looking by hand — thirty minutes on a normal day, an hour for anything that hadn't moved in a while.

Nobody asked me to fix it. I got tired of spending thirty minutes a pop looking for silicon. So I designed a coding system that gave every tray and product a real identity, built a web database on top of it, and the hand-search became a lookup. At peak, one program alone ran two pieces of silicon per board across a couple of hundred test systems. It's still what the lab uses, five years on.

Read that with the rest of this page in mind, because it's the same sentence every time. The renovation suite: nothing tied design to cost to permits, so I built one. The finance app: I wanted my money private and mine, and nobody makes that. The lab: I was tired of looking for silicon. See the missing tool, build the missing tool.

The other half of this job is what it taught me about being unimpressed. “It ran without errors” is not “it's correct.” The defect that kills you is never the loud one — it's the plausible one, the confident wrong number on a document that looks finished.

It converges

So when I say I build products with AI, understand what I mean by build. I set the architecture, the design language, and an operating contract the work has to hold to — then I direct the AI against it and reject what doesn't clear the bar. That last part is most of the job, and it's the part the tools don't do for you.

Screenshotting my own apps for this portfolio should have been a capture session. Twice it became a review pass instead. Both times an automated audit had just run clean — 344 assertions green on the second one — and a person still found seven defects no test could have caught, because not one of them was a calculation. A permit document counting a garage as living space. Every expense in a ledger signed +. Setup deleting records silently and then reporting success.

AI made producing things nearly free, so the whole game moved to knowing which output to keep. Every chapter above was practice for that, and none of them knew it at the time.

And it closes where it opened. In 1998 I was building web software at Intel. Twenty years of cameras, clients and concepts in between — the part that supposedly doesn't belong on a technical résumé. I'm back at the same company, directing software again, and the twenty years are the reason I'm any good at it.

Nobody changed careers. The two halves just took a while to arrive at the same place.