The situation
The mobile games industry that employed me for most of two decades got hit hard — contraction, layoffs, and an AI shift that has a lot of brilliant people scared for their craft. I was out of a full-time seat for a while (partly the market, partly a deliberate choice to be home with two young kids). The easy move was to treat AI as the thing that came for my job.
I made the opposite bet: that the tool that disrupts a craft is also the best thing that ever happened to anyone who can direct craft. So I stopped reading about it and started building with it.
What I set out to prove
- That a designer who can’t hand-write a backend can still ship one — by directing, not typing.
- That seventeen years of systems judgment transfers directly to working with AI.
- That “solo” doesn’t have to mean “small” anymore.
The approach — direct it like a team
I’ve spent my career telling engineers and designers what “good” looks like and reviewing until it got there. It turns out directing a model is the same job: set the vision, define done, hand over the problem instead of the steps, then review hard and own the calls it can’t make.
So I built a system for it — a personal operating system with specialist roles (capture & planning, building, research) running over a Markdown knowledge graph that gives the AI persistent memory across sessions. The guardrails are borrowed from good engineering culture: think before coding, build the simplest thing, make surgical changes, work toward a definition of done. The skill that carries the whole thing isn’t syntax. It’s decomposition, taste, and knowing the moment an answer is wrong.
I’d spent a career telling a team what “good” looks like and reviewing until it got there. Directing a model turned out to be the same job.
What I built
A self-governing personal OS
A three-agent system over a structured knowledge base that plans my week, keeps a journal, runs research routines, and remembers context between sessions instead of starting cold each time. It’s the workbench everything else gets built from — and proof that I think in systems and operating models, not one-off prompts.
An 11-game web arcade
A scatter of prototype game repos, consolidated into a single deploy with a gallery front end and one clean home for everyone to play. Midway through, the platform blocked the obvious path; rather than stall, I had the build re-route its own architecture — repurpose an existing home, re-point everything, ship. Turning a hard constraint into a one-move workaround is the same instinct I’ve always brought to design.
A card-game engine with live multiplayer
Multiplayer Deck Gateway — a browser workshop where card games are modules on one shared engine: a real 52-card deck, a poker hand evaluator, and host-authoritative networking. A dozen games and design tools plug in, and any of them can be dealt to friends live over a room code. Where the arcade was the content, this is the platform underneath it — the clearest proof that I build the systems, not just the things that run on them.
A private backend control plane
Auth, databases, and infrastructure across Neon, Supabase, and Cloudflare Workers for a handful of projects — unified behind one private “control plane” with a strict no-secrets-in-git discipline and a registry that maps exactly where every credential really lives. I ran a security pass on my own repos in the process and closed the gaps I found.
What actually made it work
- Judgment over syntax. The muscle that balanced a 200+ character roster — decompose the problem, define done, review for correctness, decide under ambiguity — is the exact muscle that gets good work out of a model.
- Systems thinking as the unlock. The reason I could direct a backend I couldn’t hand-write is that I understood the shape of the system and could tell when the pieces were wrong.
- Honesty about the edges. I’m a director and a systems designer, not a from-scratch software engineer. I know what good looks like, and I can get a model there fast. In this era, that’s the job.
Where this points
The next era rewards the people who can aim these tools with taste — who can turn intent into shipped systems and tell the difference between impressive and correct. I want to do that work, and I especially want to help the people whose industries are being remade (the way mine was) learn to wield it instead of fear it. Whether that’s a studio of my own or a team building this future, it’s the same craft I’ve practiced for seventeen years — pointed at a new kind of collaborator.