I build AI products, and the tools teams build them with.

What I do for a team: build the AI product that goes in front of real users, and then build the thing the team builds the next one with — the harness, the typed tool surface, the data access, the guardrails that keep it safe. Those are usually two jobs. Last year I did both at once.

The rest is how I got here — thirteen companies, ten industries, and the same few instincts running through all of them.

I commit early, and rarely

I'm skeptical of a technology's staying power until I find one that looks like an outlier. Then I commit fully. I've done it three times, and until recently I would have said it happens about once a decade.

Ext.js, around 2007. I had built internal systems at the NHL on YUI, and yui-ext filled in everything YUI was missing. Once it split off as Ext and it was clear it had direction and active development behind it, I went all in. It spent the next decade as one of the most dominant web frameworks there was.

Next.js, from 2018 — not just React. Through Jet, Grubhub and Sleeper, and the App Router migration at Sona while the App Router was still in beta.

Eve, now. I built the Clanker World agent on it, and today I wouldn't build an agent without it. Vercel hit a home run, and I don't think the market has a full understanding of that yet.

The decade framing is the part I would hedge. It described the first two bets well, and Eve could hold that long. I just wouldn't claim it in advance. AI has compressed the timeline; software is becoming far more bespoke, and what an AI application needs from its framework could change substantially over the next few years. Whether Eve holds its lead is genuinely unsettled. It is where I am placing my bet for now, and I would make the same call again tomorrow — I just wouldn't ask anyone to hold me to ten years.

I build the tools other people build on

At the NHL I wrote Longana, which generated web front ends straight from Lotus Notes views. Instead of migrating applications one at a time, we built the thing that migrated them; it was still driving internal applications about twenty years later. Then Adaptrex, a REST framework that spoke ExtJS natively so nothing had to be translated at the API layer. After I left, another team grew it into AG-Rest, which lets clients shape the response much the way GraphQL does.

That is the instinct behind most of what I build: find the friction, especially the friction slowing down the whole team, and go after that instead of the symptom. Automated deck generation for M&A proposals at Broadview. Migration tooling and shared libraries for every web team at Grubhub. A design system and a transactional email system at Sona. A Turborepo and shared packages at Sleeper. At Neynar, agent toolkits that gave web and React Native the same runtime, and a generator that turns an onchain contract address into type-safe viem clients and wagmi hooks.

And when the tool is a technique rather than a package, I write it down. The article I published in April 2005 on doing real navigation in Domino web views is the reason a recruiter called me two months later.

Twelve years at the NHL, from a three-month contract

The league needed a Lotus Notes engineer, and the article showed I could do the work. I started on a three-month contract in the final weeks of the 2004–05 lockout, and once the lockout ended I was in Ottawa for Sidney Crosby's draft, because the draft ran on a Notes application.

I stayed seven years, left, and came back a year later for four more: twelve years in total across two stints, and I worked the draft in every one of them. Both stints began with the league coming out of a lockout, and the first ended because of the second — costs were being cut and consultants released while there was no hockey being played. A long run for a three-month contract.

The domain has never been the obstacle

Health insurance, food manufacturing, investment banking, financial education, professional hockey, wealth management, retail, food delivery, NFTs, music, fantasy sports and crypto infrastructure.

I learn the business on the fly, every time. It is the part of the job I like most.