Notes on the builds I take on, the products I’m making on the side, and what I’ve worked out along the way. Mostly about shipping software end to end as one person.
A Melbourne manufacturer had tracked its work on paper cards and a spreadsheet that was always a day behind. Here's what it took to replace all of it with tablets on the workshop wall and a live dashboard in the office.
I said I'd take on four fixed-price builds in twelve months and write down whatever I learned. The year is up, so here it is, including the one that nearly didn't ship.
I get asked about this often enough that I wanted to write the answer down properly once, including the part that isn't especially flattering to me.
Most senior design roles now are mostly coordination. The AI tools got good enough that I could go back to building things myself, so I did.
The AI features I build take roughly a week of model work and three weeks on everything around it. That ratio surprises people, and it probably shouldn't.
I say no to more projects than I take. Four builds a year means I'd have to. Working out how to say it well took longer than working out what to say it to.
An AI itinerary tool for travel agents. Sixteen months of building it, what stuck, and the mistake I made early on that I now bring to every AI build I take.
A small practice on purpose. Four fixed-price AI builds a year, no retainers, and no consulting work unless something gets built at the end of it.
I won't run the build-versus-buy decision for you, since that's strategy work I don't sell. But I've seen enough of these go both ways to have a view worth writing down.
AI doesn't take over a whole process. It changes one or two specific moments inside one. Here's where I keep finding the useful ones, and where I'd be careful.
The question I get most from operations leads is where to start. The honest answer is smaller than you'd think, and sooner than you'd think.
The most common SaaS architecture mistake I see is leaving scale until later. Later usually turns out to mean a rewrite you didn't budget for.
A roadmap assumes you already know what's worth building. With AI you usually don't, and you find out much faster by building one small thing.
Twenty years of building software and the last twelve months are the biggest shift I've been through. Here's the honest account of what's different day to day.