Three days after I built a place to keep what I knew, I got on a plane, and by the time I landed I had written down the bet that everything since has been an execution of.
The bet, in the words I used at the time: most software frameworks require code changes to extend. Ours requires configuration. That single architectural fact is the entire opening. If building a new capability is a matter of describing it rather than coding it, an agent that understands the description can do the actual building, not merely assist a human who does it. The unit of production stops being "an app" and becomes "an agent that can build a class of apps." Output capacity, from that point forward, is bounded by agent capability rather than by how many engineers you can hire.
I wrote a flywheel underneath the bet, and it is worth stating plainly because it is easy to read as a slogan and it was not meant as one. More apps digitise more local knowledge. Richer digitised knowledge makes the ecosystem more attractive to whoever wants to buy access to that data. More demand brings more funding and more clients. More clients fund more apps. Every stage of that loop, before this bet, scaled with headcount: more engineers, more apps, slower and slower as the codebase grew under them. The bet says the loop can scale a different way. Ideas can fail cheaply and often, because the thing that survives every failed idea is the data it captured on the way down.
I want to be honest about what this was and was not, at the time. It was not a plan. The document I wrote that day is explicitly a brain dump, functions numbered and half-formed, several of them marked "to define" rather than defined. What it was, was a constitutional bet made early enough to shape everything that came after it, before there was a name for the suite that would eventually carry two of its functions out from under the original company entirely. The framing sharpened over the weeks that followed, into the exact sentence this post borrows for its title. But the sentence describes a bet already placed on day three, on a plane, with nothing built yet to prove it.
Humans design and supervise the agents. Agents do the construction. I did not know yet what the agents would be called, or that there would eventually be more than one kind of them, building apps on one substrate and conforming data on another and reading the public record on a third. I knew that the alternative, more engineers writing more code by hand, could not get us where the ecosystem needed to go, and that betting the whole strategy on agent capability instead of headcount was the only version of the plan that scaled the way the problem demanded.
Everything with an agent's name on it today, every product this suite has built, is that bet, kept.