The Part That Holds

Nobody takes a picture of a foundation.

The ribbon-cutting photo is always the building. The glass, the lobby, the name over the door. The thing that decided how tall it could stand was poured months earlier, by people who were gone before the cameras arrived, into a hole nobody will ever look at again.

Enterprise AI has the same problem. The model is the building. It demos well, it photographs well, it gets the headline. But the model is not the part that holds. I have spent five weeks writing around the word for the part that does. Foundations.

None of it looks like AI work.

A curriculum that compounds. We run an AI literacy program with more than 2,000 monthly learners, and the number I care about is not the first thousand who got fluent. It is that the next thousand arrive faster, because the first cohort now teaches.

A governance gate built to reshape. Sixty-five percent of our first-year use cases left that gate looking different than they walked in, and shipped because of it.

Infrastructure a non-engineer can build on. A sandbox where the person who used to file a ticket now builds the agent themselves, and has it running by Wednesday. We don’t let localhost become Production.

A community of those builders. Close to a thousand of them in six months, their agents logging over a million workflow executions, most of them people no org chart would have called technical.

And enablement as an actual job. Someone whose full-time work is to sit next to a non-engineer at 4pm on a Thursday and unblock them. We pay people to do this, and it is on the org chart.

Models change. Vendors get bought. The infrastructure decision you made eighteen months ago already looks quaint. What compounds is the human capability you built around all of it. It might also be the hardest part.

Stop admiring the building. Make sure the foundation can hold it up.