Thesis

AI is already intelligent enough to run trillions of dollars of digital enterprise work. Enterprises still can’t run on it.

The models are already good enough to do enormous amounts of real enterprise work. It still isn’t happening. Even when AI is packaged into purpose-built agents from world-class engineering teams, it’s still not running economically valuable work right now. There is a gap between today’s agents and running in production for enterprises.

An agent re-derives the task from scratch on every run. You pay full price for the model’s effort each time it does the same thing, and for repetitive work that fires thousands of times a day, the bill is enormous and the spend rarely shows up as value. A long line of companies has already learned this the hard way, having poured fortunes into AI usage with almost nothing to point to.

An agent improvises. It starts fresh each time, so the same input doesn’t reliably produce the same output. That’s fine for a draft or a prototype, where eighty percent is a win. It’s useless for the work the economy actually runs on, where one wrong underwriting call or misposted payment wipes out the value of a thousand correct ones. Eighty percent isn’t most of the way there. It’s zero.

There is a better shape for this work, and it is an old one: code. Plain deterministic code is cheap, instant, and identical every single time. The mistake of the last two years was treating the model as the thing that does the work. The model’s real job is to build the thing that does the work, not to be it. You use it to turn the work into code once, run that code forever for almost nothing, and bring the model back in only on the few steps that genuinely need judgment, or when the rules themselves change.

We turn a company’s real work into code, and we keep it alive. Not a model that improvises the work forever, but the work itself, captured once as cheap and reliable code. The model is rented and swappable; we use it to write that code and to handle the few steps that genuinely need judgment. What we build and own is everything around it: the rules pulled out of people’s heads, the integration into the systems of record, the upkeep as the world shifts, and the record of everything the work has done.

When AI eventually makes execution of work free, enterprises will reveal their true shape: a small core of identity, its human relationships, judgment, capital, accountability, and trust, wrapped by a vast, cheap layer of automated work. We’re the ones connecting the two.