A leadership team sat down last quarter to stress-test an AI email-response system they were already skeptical of.
They picked a live thread off one leader’s inbox. Ran the AI. The reply came back in under a minute. Ready to review. Ready to send.
Then a highly experienced person on the team wrote the same reply from scratch. It took about twelve minutes. The reply was off-brand and lower-margin than the tool’s version.
Nobody in the room called the meeting a win for the tool.
The conversation snapped back to whether the tool was safe. Whether the team would adopt it. Whether people would trust it. Whether they should schedule another training session.
Here’s the thing. The tool was the smallest question in the room.
The bigger question, the one nobody asked out loud, was what the twelve-minute rewrite meant about the person who did it. Whether their instinct was safer than the machine’s. Whether the room valued their judgment or was about to route around it.
Nobody had scheduled that conversation. So the room aimed everything at the tool instead.
The mixed pile
Here’s the pattern I keep watching. A leader runs an AI rollout meeting. A team member raises an objection. The leader hears the objection as low buy-in, poor training, or stubbornness. So the leader books another demo. Assigns a champion. Adds a training session. Sends around a one-pager.
Right? Every one of those moves says the same thing.
The objection is ignorance. If we can just get them to understand the tool, they will accept it.
The objection is not ignorance.
Function is what the tool will do. Fear is what the person thinks that change says about their value. They need two different conversations.
Most leaders run them as one. A single meeting, a single deck, one thing labeled “AI training,” everyone in the room. Function and fear go into the same pile. The pile gets called a training problem. Another demo gets scheduled. The three team members quietly working around the tool by Wednesday do not go away. They just get quieter.
A better demo can make the fear worse if the human role is still unclear. Every time the tool answers faster, cleaner, closer to the brand than the person could, the person hears the same thing. Not “the tool is good.” “I might not be needed.”
You cannot beat that fear with another slide about the API.
The Chef has to move to DECIDING
This is where our own architecture does the diagnostic work. We have a piece called the Four D’s, the four possible stances a Chef can take relative to a given dish. Doing it themselves. Dispatching the recipe to a station. Dialing at the pass while the station cooks. Deciding what the dish has to be.
Most leaders running an AI rollout stall at the wrong stance.
They stay in DOING. Running the rollout themselves, standing at the demo, answering every question live, keeping the tool close because it feels safer that way. Or they slide sideways into defending. Arguing the objection down. Explaining why the fear is wrong. Trying to talk the person into a stance the person did not walk in with.
Neither one asks the question that actually needs to move.
The stance that has to move is DECIDING. Who is standing at the head of the kitchen and holding what the dish has to be, before the line cook is even shown the recipe. What is the tool for. What standard must it meet. What does the person still own, no matter what. What does the tool not touch, ever.
That last one is where fear gets separated from function. The person doing the work needs to know, in plain language, what the tool is not allowed to do. Not because you promised it. Because you decided it. There is a difference.
If you read the earlier piece on writing your lens down, this is that same discipline aimed at the team instead of the buyer. You wrote the lens so the buyer could recognize you. Now write the lens so the person on the line can see where their judgment still lives.
What the person is actually asking
Take a dispatcher who has run the phones for eleven years. Ask her what she loves. She will not say “assigning trucks.” She will say handling the customer whose furnace died the night the baby came home from the hospital. That is the story she tells at Christmas.
If you drop an AI dispatch tool in front of her and say “this will save you an hour a day,” she hears something different. She hears that the story might get flattened. The nervous read of urgency, the memory of who this customer was last year, the split-second call to send the newer tech with the older tech riding along. All of that gets treated as overhead somebody would like to remove.
You are not removing overhead. You are removing the story she is proud of. Until you name that out loud, and decide which parts of it the tool never touches, the training session will not land. It cannot.
A CSR who is genuinely happy to lose the fifty repetitive follow-ups a week is still nervous about the fifty-first one. The one where the tone was slightly off. The one where a customer wrote back angry. She is asking a real question. Whose name is on that reply. If the answer is “yours, but the tool wrote it,” she is going to quietly stop trusting the tool and rewrite every reply herself. You are going to say the tool did not stick. She is going to know exactly why it did not.
Office manager wants to lose the status chasing. Great. Ask her which exceptions she is proud of catching. She has a list. She has watched the same three edge cases blow up rollouts at other firms. She has literally lost sleep over one of them. If the tool touches those cases without her sign-off, she is not going to raise her hand in the demo and say “I have a concern.” She is going to keep chasing the status the old way in a shadow spreadsheet, and the rollout will look like it succeeded from the leadership dashboard while nothing has actually changed on her desk.
Three different roles. Same underlying question. Where does the value I bring still live once this thing is on the floor.
Yeah. That is a fear conversation.
It does not belong in the same meeting as the function conversation. It goes badly there every time.
The guardrails belong on paper before the demo
Once the two conversations are separated, the function side needs guardrails the person doing the work can actually see.
Not aspirational guardrails. “We will monitor this.” “We will iterate based on feedback.” Those are promises to review the rollout, not commitments about the tool. The person doing the work reads them as marketing.
Named guardrails. “The tool cannot quote a price.” “The tool cannot close a complaint thread.” “The tool cannot send anything to a customer who has flagged an injury.” “The tool cannot answer without the CSR reviewing it if the customer has been with us more than three years.” The person needs to know what the tool physically will not do, so the fear has somewhere to land besides everything.
Then a feedback loop the person actually uses. Not a Friday survey. A one-line note attached to any reply the tool wrote that the person had to rewrite substantially. Two weeks of those notes tells you where the recipe is weak, without asking anyone to be brave in a meeting.
Guardrails and feedback are two ingredients of a longer discipline for how a station gets built. This piece is not the discipline. This piece is the conversation that has to happen before the discipline is even worth writing down.
The Monday Move
Before the next AI rollout meeting with a team, sit with one team member. Ask three questions. Write the answers on one page.
- What part of this work do you hate.
- What part are you proud of.
- What would make you nervous if the tool handled it.
Guardrail. Do not defend the tool during the answers. Do not correct. Do not translate. Write the answers first. Then read them.
Owner. The manager or ops person responsible for the workflow. Not HR. Not a consultant. Not the person who bought the tool.
Metric. One page. Three answers. In the team member’s own language. Before any tool demo or training runs.
Next step. Split the rollout plan into two columns on paper. Column one is function the tool may handle. Column two is human value that must stay visible.
If both columns are empty for a role, the rollout is not ready. Write the columns first. Book the demo second.
Does that make sense?
The close
You cannot train your way out of a role conversation. Function is what the tool will do. Fear is what the person thinks that change says about their value. Run the two conversations separately. Then decide what the tool is actually for.
Original framework. Distilled from client work.
~ source material · Original framework. Distilled from client work.
