A company decides to get into AI and does the thing that seems most sensible: it sends the sharpest person. The marketing lead, or the one who “gets technology.” Advanced course, three days, the best on offer. The idea behind it is simple and almost always said out loud: he learns, then comes back and infects the others. Months pass. The contagion never comes. The person is better than before — and everything else is identical to before.
The rejected transplant
It didn’t go wrong by chance. It went wrong by how it was set up. You trained an individual hoping to change a system, and an individual isn’t a system. A new person inside an old environment doesn’t transform the environment: they get reabsorbed. He comes back with a different way of working and looks around for where to plug it in. There’s nowhere. The processes are the same as before, the colleagues speak the same language as before, work gets handed off the way it was handed off before. His new method has no place to latch onto.
From here, three endings, identical in substance. He pushes uphill: he tries to drag the others along, runs into “we’ve always done it this way,” and wears out. Or he gets isolated: he becomes “the AI guy,” a quirk tolerated in a corner, while the department closes around him like an organism around a foreign body. Or he leaves — and that day the knowledge you paid for walks out the door inside his head, and you start over from zero. A single upgraded node, on a network still running the old protocol, doesn’t speed up the network. It stays a node the others can’t read.
The fake delegation
The same mistake repeats one floor up, and there it costs more because it’s more invisible. The owner funds the course to get the problem off the desk. “Here’s the budget, modernize yourselves, then show me the results.” It looks like healthy delegation: you give a job to whoever has to do it, and you judge by the outcome. It works for almost everything. It doesn’t work here, for a structural reason: adopting AI isn’t learning a program, it’s changing processes. And processes are changed only by whoever has the authority to change them. If that authority delegates the doing but holds back the understanding — never puts their own hands on the tool, doesn’t see where it touches the workflows, doesn’t decide what gets dropped to make room — the trained people run into a wall it’s not their place to knock down. It’s not laziness on the part of whoever’s in charge. It’s that they delegated the one part that couldn’t be delegated.
Entering as a system
So what does it mean to really bring it in. AI enters a company the way any real change does: as the shared practice of a group, not the superpower of one. That doesn’t mean training everyone at once, blindly — that’s just the other way to get it wrong. It means a core, not a hero: a few people who actually work in contact, so that a new method has someone to bounce off. It means redrawing together the processes the tool touches, because an output produced in a new way has to be handed off, checked, and signed by someone who speaks the same language. And it means whoever’s in charge is in deep enough to clear the obstacles when they show up — because they show up. The unit of measure for adoption is the team. Never the individual.
Put that way it sounds like more work, and it is — at the start. But it’s the only kind that pays. A core that adopts the tool together builds something that stays even when one person leaves, because the method lives in the group and not in a single head. And there the tool stops being a toy for one and becomes what it really is: a multiplier for those who know where they’re going. The speed it gives an isolated person is a number on a sheet. That same speed inside an aligned group is an advantage the competition doesn’t see coming.
Antibody or implant
There’s a question that saves the entire cost of the hero, and it has to be asked before signing off on any course. Who else, besides the person I’m sending, has to change how they work for this to take — and is that group in the room? If you’re training one person to change a system of ten, stop. You’re not adopting AI: you’re preparing an antibody. You’re putting them in a position to be rejected — and then you’ll blame them, or the tool, when the fault is in the implant.
A company doesn’t learn because a person learned. It learns when the way of working changes around that person — and that change is the one thing, in this whole story, you can’t delegate to anyone.