We're in an Age of Build

Show Notes

Dan opened with a confession that ended up being the spine of the whole episode: he's been resistant to most of the recent tech waves, quietly filing them under fad, and with this one he's decided he's not missing it. "Nope, I'm not missing this one." The capitalizing on it would be a bonus — mostly he just wants to plug in, because if he doesn't, he's straight up left behind. I told him I don't see how this one doesn't take hold, and we were off. We took the long way there, though. Most of the first half is me coming back from Indiana with complicated feelings about midsize American cities — Indianapolis, Lafayette, the two-thirds-empty downtown mall they're trying to atomize into something smaller and mixed-use — and Dan's "capital flight" theory about why downtowns hollow out every couple of generations. I'll save you the urban-planning detour, except to say it set the table: we're in an age of flux, things are getting rebuilt, and the question is always who gets to do the building.

Then Dan dropped what he's actually been doing, which is living in AI land. He's been playing with Cursor, Lovable, and Bolt — three AI-assisted ways to build software — and the thing that stopped him cold is how fast you go from not-a-developer to way past hello world. "The future of code is using English, which is nuts." I agreed it's nuts and also that it makes perfect sense: the layers of abstraction keep climbing, the languages keep getting friendlier, so of course we eventually arrive at plain English. His honest take after a couple weeks of poking around: you still have to know some software development. The tools nail your first prompt, then the second prompt wanders off and uses a different library, so you have to walk in and set your boundary conditions — your architecture, your auth, your packages — up front. Cursor even has a config file for your preferences, though Dan and I both wondered out loud whether resubmitting all that every turn is a quiet token problem.

The part I keep chewing on is the strategy conversation. Dan's pitch was clean: build a highly niched piece of software, prove it works, land paying clients, get your ARR to a number, and sell on a healthy multiple to a Constellation or a Vista or a strategic buyer. He's right that this has worked for years and is still working — those serial acquirers really do run "buy and build" roll-ups of small vertical-market software firms, and the multiples really are nuts. The line that stuck was niche is where you can be a minnow and play. Dan doesn't need to be much bigger than a minnow; a couple of those payouts would be enormous for him and barely register for the buyer. He also flagged the two-speed reality: being an automation agency right now is "like printing money," but he doesn't think that lasts — the durable move is building the actual product, the engine that sits on top and doesn't change even when you swap the data underneath.

Naturally I tried to find the oil-and-gas angle, because Dan asked why we aren't already running a vertically integrated AI software studio. My honest answer was that at the scale I live in, the big players are already partnered with Microsoft and Google to build use-case-specific models on top of foundation models — SLB really is turning its petabytes of subsurface and well-logging data into an AI-ready platform, exactly the kind of correlation work I used to do by hand as a reliability engineer, slicing a million well-run logs by longitude and latitude and depth. So Dan moved me down-market: forget the 50,000-person company, go to the driller or the small operator who's "still just a little bit cowboy." That reframed it. The minnow play isn't competing with the majors — it's serving everyone the majors won't bother with. We even sketched a real one: a front-end tool that helps suppliers get their invoices through SAP Ariba clean the first time so they're not stuck waiting out net-70. Dan pointed me at CrewAI, where you can stand up little single-purpose agents — turn the unstructured data into structured data, then handle it programmatically — and sure enough their homepage had an SAP logo flying by.

We closed on the window itself, which is the real thesis. Get on it or get left behind, Dan said, and probably within two years — I'll stretch it to two-to-five, but I won't argue the direction. His prediction from last time still stands: there's now an army of people who don't need to code well, just technical enough, and they're going to go transform every crappy little app with some MRR until the whole app store turns over. As long as there's money flowing, there's opportunity, and people will eke it out — "they're just gonna eat up the little fish food until it's gone." That's the age we're in. Not the age of having the idea, and not yet the age of everything being built. The narrow, strange, beautiful middle — the age of build, where the only real qualification is that you actually start.

Niche is enough. Be the minnow. Go eat the fish food before someone else does.

— Sean