Nobody is documenting their job to lose it: how to introduce AI to a team

Picture the all-hands. The CEO has brought in someone to help with AI. The someone is friendly and says they will be asking a lot of questions, sending round a survey, and would people mind recording a few minutes of how they do their work?

Now picture the person three rows back. They do the Monday numbers. It takes them most of Sunday evening and they are rather good at it. They hear “record how you do your work” and a small, cold thought arrives: am I documenting my own job so a machine can have it?

Nothing said after that point lands, for that person, unless the thought is answered. And it usually is not, because the people at the front of the room are excited and have not had the thought themselves.

The sentence that has to come first

I have come to think only one thing works, and it has to be said by the leader, not by me. Some version of: we are not doing this to remove people. Roles will change, and they will change for the better. Everyone here is going to end up with an assistant that does the tedious parts. Nobody is documenting their job to lose it.

Said plainly, by the person whose name is on the door, and then, crucially, behaved. Because the team will watch what happens to the first person whose process gets automated, and that is when they decide whether the sentence was true.

The reason it has to be the leader is not ceremony. A consultant saying “your job is safe” is worth nothing; I do not control the headcount. The leader saying it is a commitment they can be held to. And in my experience the leaders who bring in help for AI in a lean business mean it, because the whole point of what they are doing is to grow without hiring at the same rate, not to shrink. Their problem is that the volume keeps going up and lands on the same people. They want those people to have more capacity, not fewer colleagues.

A story I tell every leadership team

Years ago, in a former life doing business intelligence, a consultant friend of mine was called in to counsel an executive team. Their CEO had discovered he could generate a report in an hour with AI, and had used that discovery to tear into his team for the “crap” reports they had been producing for years.

Every fact in his complaint was probably true. And it was the single most destructive thing he could have done with the tool, because from that day on, nobody on that team would ever show him how they worked. The machine had become a weapon before it had a chance to become a help.

I tell that story because it names the fear precisely. It is not only “will I be replaced”. It is “will this be used against me”. So there are two norms I now ask leaders to state and model, alongside the headcount sentence. AI output is checked before it is acted on or forwarded. And using AI to critique a colleague’s work is done with them, not to them. The second one is the culture item. It is the difference between a team that shares its prompts and a team that hides them.

The mechanics of feeling safe

Beyond the framing, a few practical things make the difference between a survey people fill in honestly and one they fill in defensively.

Every ask says who it is for, why, and how long it will take. Ten minutes for the form. Five to ten for an optional recording. An hour for the workshop with the key people. When people can see the size of the request they stop imagining a larger one.

Nothing is measured by name. The baseline I take at the start, how many hours a week go to recurring admin, how confident people feel, whether they know the rules, is re-asked at three and six months, and shown as a team. The first time a survey number is used against an individual, the survey stops being honest forever, so it never is.

Recordings stay in the company’s own drive, transcripts are made privately, raw material is destroyed after the findings, and nothing is used to train anything. That paragraph goes at the top of the survey, not in an appendix, because the safety point is only real if people actually read it.

And the order of operations matters more than anything. Individuals first: everyone gets the same tools connected the same way and learns on their own work, so the first thing anyone experiences is their own week getting easier. Processes second, and only the ones that have visibly outgrown one desk. Roles last, and gradually, as the recurring work is carried by machines and people move up to the exceptions and the judgement. Businesses that start at the end, announcing a restructure around AI, get resistance and nothing else. Businesses that start at the beginning get volunteers.

What you get for doing it this way

Volunteers, mostly. In every rollout there are two or three people who take to the tools natively, and once they feel safe they become the examples that persuade everyone else, far more effectively than I could. The person who does the Monday numbers turns out to be the person who most wants them to assemble themselves, because she has better things to do with a Sunday evening and always did.

You also get honest data. A team that believes the framing tells you where the real recurring work is, which numbers they do not trust, and what they have quietly been doing with a personal account. A team that does not believe it tells you what it thinks you want to hear, and you build the wrong thing.

If you are about to have this all-hands

Write the sentence down first. Say it in your own words, at the start, before the friendly consultant speaks. Then behave as if you meant it when the first process gets automated, because that is the moment everyone is watching.

And if you would like a second pair of eyes on how to frame it for your particular team, send me a description of the room. I would rather help you get the first ten minutes right than spend the following six months undoing them.