The pilot is the easy part
Most companies get as far as a promising AI pilot. Few get a business that runs on it. This is what we think sits in between.
In this essay
A demo proves the model. It proves nothing about the company.
A pilot answers one question: can the technology do this task once, under supervision, with a motivated team watching. It usually can. What it does not tell you is who owns the result on a Tuesday afternoon when nobody is watching, or what happens to the one case in twenty that goes wrong.
The gap is made of ordinary things.
Unclear ownership. No agreed definition of a correct result. Exceptions with nowhere to go. A team that was shown the tool but never asked to change how they work. None of this is glamorous and none of it is solved by a better model.
A company that runs itself is still run by people.
The goal is not to remove people from the work. It is to move them to the part only they can do: deciding what good looks like, handling the cases that matter, and improving the system that handles the rest. That is a change in how a company operates, and it has to be built with the people who will live in it.
So we do slow, specific work.
One process at a time, in production, measured, owned by your team. It is less exciting than a strategy offsite and it is the only thing we have seen work.