How to Lead a Team That Is Split on AI

Three groups, not two

Almost any team splits into the same three groups. A small number are already using AI daily, often on their own accounts, often without telling you. A small number are firmly against it, and say so. The largest group is in the middle, waiting to see which way this goes and who gets embarrassed.

Managers spend their energy arguing with the second group. That is the wrong target. The middle group decides whether this works, and they are watching how you treat the other two.

The sceptic is free quality control

The reflex is to persuade the sceptic. Do not. You will spend weeks on it, you will lose, and you will convert a useful colleague into an opponent with a grievance.

Look at what the sceptic actually does. They read output and disbelieve it. They notice when a figure looks off. They ask where a claim came from. You spent the last lesson trying to build exactly that instinct into a process, and here is somebody doing it by temperament, for free.

So give them the job. Ask them to own the verification standard: what gets checked, what an error looks like for your work, what should never be sent without a second reader. Then genuinely use what they produce.

Two things follow. The work gets safer, because a determined doubter designed the checks. And the objection changes shape from "this is reckless" to "this is only acceptable if we check properly", which is a position you can both hold.

The enthusiast needs guardrails and a job

The enthusiast is not a problem, but unmanaged they become one. They will try everything, share prompts in the team chat, occasionally paste something they should not, and set an example of speed without checking that the middle group copies.

Give them scope and a boundary. The job is to find and document what works: which tasks, which prompts, where it fails. That uses their energy and converts scattered enthusiasm into something the middle group can pick up. The boundary is lesson two said directly rather than implied. Approved tools, the red list, and nothing customer facing without a named reviewer.

Name the job security fear out loud

Underneath most stated objections is an unstated one. People are not usually worried the output will be inaccurate. They are worried about what happens to them if it is accurate.

Leave it unsaid and it turns into a hundred small frictions: work that never quite gets round to the new process, a tool that keeps having problems, a quiet retreat to the old way. Say it plainly and you at least get an honest conversation. But here is the part managers get wrong.

Say what is true and within your gift instead. What you have been asked to achieve. What you will tell them and when. That decisions about people will be made by people. Those are commitments you can keep, and a small promise kept beats a large one broken.

Shadow AI is a symptom, not a discipline problem

If your official rules are unusable, your team does not stop doing the work. They use a personal account on a personal phone, and you lose every scrap of visibility you had.

Say this clearly to anyone above you who wants a ban: a ban does not remove AI from your team's work, it removes your knowledge of it. The work still happens, now on consumer accounts you have no agreement with, with no anonymising habit and no chance of anyone reporting a mistake, because reporting means admitting a rule was broken.

Which makes the conclusion straightforward. Usable rules are a security measure. Not a nicety, not a culture thing. The most secure policy is one your team can follow while doing their real job, because that is the only kind that stays inside your walls.

Here are our current AI rules for staff:
[paste your rules]
And here is the work my team actually does each week:
[main tasks]
Identify every point where following the rule would make a normal task
impossible or much slower. For each one, tell me:

What a busy person would realistically do instead
Whether that workaround would be visible to me
A version of the rule that is followable without losing the protection
Assume people want to do the right thing and are short of time.

Give the middle group something to copy

The middle group are not waiting for a presentation. They are waiting to see one normal colleague use AI on one real task, safely, without drama, and not get burned for it. That is your rollout. One task, one visible example, checks done properly, mistakes discussed openly. Watching that happen once moves more people than any amount of internal communication.

I need to talk to my team about how we will use AI. Some are enthusiastic,
some are opposed, most are waiting to see.
Help me prepare. Give me:

Three things I can honestly commit to that are within my control as a
[job title] who does not decide headcount
Three things I must NOT promise, with a one line reason each
Two questions I should expect and a straight answer to each
Keep every answer plain and short. No corporate phrasing. Do not suggest
I reassure people about things I cannot guarantee.

A member of my team is sceptical about AI and I want to give them
ownership of our checking standard rather than argue with them.
Our first use case is:
[task, who does it, how often, who it goes to]
Draft a short brief for them, under 200 words, that:

Defines what an error means for this specific task
Lists what must be checked before anything is sent
Gives them explicit authority to say something is not ready
Asks them to report patterns, not individual slips
Write it as a genuine role, not as a token gesture.