The Day an AI Rollout Ends, a Job Appears That Nobody Owns

The question employees ask me most is asked privately: “Once this is finished, do you still need me?”

It almost never comes up in the meeting room. It comes in the hallway afterward, or when someone follows me to the door and lowers their voice.

I used to answer with reassurance. I don't anymore. Now I ask a question back: after this goes live, who looks after it?

Most people stop short. They can answer — they just never considered that looking after it counts as a job. They assumed the rollout ends the way installing an air conditioner ends.

The Owner Is Doing Subtraction. So Is the Employee.

When an owner brings in a consultant, the arithmetic in his head is usually clear. Hours saved, steps removed, people freed up for something else.

The employee hears exactly the same arithmetic. The difference is that he is the thing being subtracted.

Both are missing the same fact. A rollout does not reduce the total amount of work. It changes the shape of the work. A machine takes over one segment, and the machine itself becomes a new segment — someone has to watch it, feed it rules, and notice when it is wrong. That job did not exist before. It is created on the day the system goes live.

A fastener trading company in central Taiwan, around twenty-five staff, automated its quoting process. The first month went smoothly. Then they called: the system was quoting the wrong prices. Nothing was broken. A discount rule for one product code had changed, and nobody had updated the setting.

That task was never assigned to anyone. It was not on a job description, and it was not in my contract either.

A rollout does not remove a person from the company. It adds a job that nobody owns.

Our Own Website Changed How I Explain This

We built our company website with AI. It went up fast and it works well, but it has no traditional admin panel. Publishing an article or rewriting a paragraph doesn't mean logging into a dashboard and clicking. It means telling the AI what to do.

So the site permanently needs someone to do the telling. That person doesn't have to write code. They do have to know what needs changing, what it should become, and what to check afterward.

An AI can absolutely build a site with a full admin panel. That isn't the point. Whichever version you build, someone has to stand in that position.

The more the tool handles, the more visible that position becomes. When five people shared a task, any of them could cover for the others. When the machine does ninety percent, nobody covers the last ten. One person knows how to instruct it, and that is the whole bench.

The smarter the tool gets, the more important the person standing next to it becomes — because eventually only they know what to ask it for.

No Company Can Afford to Outsource Maintenance Forever

An owner once asked me what maintenance would cost. I quoted him. He didn't answer right away.

He was right to pause. No company can pay an outside firm every month to adjust one discount rule or rewrite one line of web copy. The real problem isn't the money. It's the delay. You want the change today, and by the time it reaches the top of someone else's queue, the customer has moved on.

Tools also grow with the company. What is sufficient this year stops being sufficient when volume doubles next year, and it will have to be rebuilt. When that happens, the person who best understands how the company actually operates is inside the company. It isn't me.

So there was only ever one reasonable ending: the work returns to the company. Sometimes to one person, sometimes split across several.

The system doesn't really start living until the consultant leaves. Whether it survives depends on whether anyone picks it up.

Now I Ask in the First Meeting: Who Will Own This?

I used to save that question for the final meeting. Now I ask it in the first one, before any tool has been chosen.

The room usually goes cold for a moment, because the question puts an unwanted responsibility on the table. That cold moment is the point. It moves the conversation from “who gets replaced” to “who takes this on.” The two questions sound similar. They put you in opposite positions.

I also stopped assuming the person who takes it on will be the youngest or the most comfortable with computers. I've watched a thirty-year-old administrator dodge it completely, and I've watched someone close to retirement volunteer. Age isn't the variable. Willingness to admit you don't know yet is.

The people who get left behind are not the ones who can't use AI. They're the ones who decide not to learn it. Between those two things sits a choice, and the choice belongs to them.

At the fastener company, when the meeting turned to who would own it, nobody raised a hand. Ten seconds of silence.

Then the warehouse manager, nineteen years there, spoke up: “I'll give it a try.”

When the meeting broke up he took my handout with him. Nobody else touched a copy.