Somebody waits behind after the meeting. They let the room empty first, which tells you they have been working up to this, and then they ask whether AI is going to take their job.
You have about four seconds. Most leaders use them to say some version of "no, nothing is going to change for you." It is kindly meant. It is also the wrong answer, and the person walking away knows it, because three of the things they did last year they no longer do, and nobody ever told them that was the plan.
Should you tell staff their jobs are safe?
No. Not because their jobs are in danger, but because you cannot know, and because a promise about the role answers a question they did not ask. When someone asks whether AI will take their job, they almost never mean "will this organization employ someone with my title next September." They mean something narrower and much more answerable: parts of my work have gone somewhere, I was not consulted, and I do not know who is deciding what happens to the rest.
That is a question about tasks, and about who is making decisions over them. A leader can answer it. A leader who answers about the role instead has changed the subject in front of somebody who came to them frightened, which is a poor trade for both of you.
It is worth checking which conversation you are in before you open your mouth. A skeptic doubts the machine works; a frightened person has already concluded that it does. The two need opposite things from you, and bringing AI skeptics with you is a different job from the one below.
What the fear is, and what the change actually is
Two bodies of evidence sit oddly beside each other, and the gap between them is where this conversation goes wrong.
The fear is the majority position almost everywhere. Pew Research Center's 37-country survey, published September 17, 2026 from 42,151 respondents interviewed between February 8 and May 13, 2026, found that in 34 of those 37 countries people tend to believe AI will lead to fewer jobs rather than more. In high-income countries the numbers are starker: in the United States, Australia and South Korea, roughly seven in ten or more expect job losses over the next twenty years. A separate Pew survey of 3,488 US adults, conducted June 22 to 28 and published August 18, 2026, put the American figure at 71 percent, up from 64 percent in 2024, and found the sharpest movement among adults under thirty, where it reached 73 percent, up from 61 percent two years earlier. Among the youngest people you employ the worry has a particular shape, because it is about the first rung rather than the current one, which is what happens to entry level work. The people you lead are not unusual for worrying. They are typical.
What is measurably happening, though, is not the thing they are worried about. IBM's Institute for Business Value surveyed 1,500 chief human resources officers and senior workforce strategy executives alongside 8,800 employees with Oxford Economics between April and June 2026, and published the findings in September as the 2026 CHRO study. Fifty-two percent of employees said their assigned tasks had changed over the previous year as AI reshaped what they were responsible for. Roles are not vanishing at that rate. Task lists are being rewritten at that rate, and mostly without an announcement.
The same study contains the finding that explains why nobody has been told. Only 26 percent of organizations clearly define the split between work that is human-led, work that is AI-assisted and work that is AI-executed. Three quarters of organizations have not made that distinction at all. So when a member of staff asks what is happening to their work, the honest institutional answer in most places is that nobody has decided, and the reason the leader reaches for reassurance is that reassurance is the only thing available when the decision has not been made. That is the same gap, seen from one desk, as the one that appears when your team knows more about AI than you do.
In schools the same gap shows up with different numbers. A study IBM commissioned from Morning Consult in July 2026, surveying 2,048 US adults comprising 1,019 educators and 1,029 parents and published on September 2, 2026, found 76 percent of middle school and 73 percent of high school classroom educators using AI weekly, while only 20 percent reported extensive training in it and only 24 percent thought the education system was adapting well to AI. A teacher who asks you what is happening to their job is not being paranoid. They are reading the same gap you are, from underneath.
There is a third finding worth putting beside those two, because it is the part staff feel first. Sixty percent of employees in the IBM study said skill erosion affects them directly, and among those who were worried, three in four said AI had already begun to erode at least some of their skills, with critical thinking named most often. That is not a fear about next year. It is a report about this year. The person in front of you may be less worried about being laid off than about becoming someone they would not hire.
The Replacement Question
The Replacement Question is three questions I ask before answering anyone who wants to know whether AI is going to take their job: Which parts of this person's work have already moved to a machine, and who decided that they should? What is this person now accountable for that they were not accountable for a year ago? And if their role were to change, would they hear it from me first, or find out? A leader who can answer all three is having a conversation about work. A leader who cannot is offering comfort, and comfort has a shelf life.
This is my suggested way of hearing the question rather than a tested method. It is useful because it moves the conversation from the future, where you have nothing to offer but sincerity, to the recent past and the near present, where you have facts.
The four answers you can give that are true
There are four, and they work in order, because each one earns the right to the next.
1. Name what has already moved
Not "AI is changing how we work." The three specific things this person used to do that a machine now does or drafts, and roughly when each one changed. If you cannot name them, say so and go and find out before you continue the conversation, which is a better answer than confident vagueness.
This is the single most calming thing a leader can say, and it surprises people that it works, because it sounds like bad news. It works because it is checkable. The person already knows their own week. Hearing their own week described accurately by their leader tells them somebody is paying attention, and that is most of what they came for.
2. Separate what has been decided from what has not, and date the rest
"We have decided that all first drafts of the end-of-term report can be machine-drafted. We have not decided what happens to the moderation meeting. That decision sits with the leadership team and we will have made it by the end of November."
Half of what a frightened person imagines is being decided in secret is not being decided at all. Saying which is which converts a conspiracy into a schedule. A date is the load-bearing part. Without one, "we have not decided" is indistinguishable from "we are not telling you."
3. Say what they are now accountable for
This is where most organizations have quietly moved the ground without telling anyone. The IBM study measured the gap precisely: 71 percent of executives named the ability to supervise, validate or override AI output as a priority capability, against 38 percent of employees. A 33-point gap. Executives have already decided that checking the machine is the job. A third of the people expected to do it have not been told.
So tell them. "The drafting is no longer yours. The judgment about whether the draft is right is yours, it always was, and it is now the larger part of what I am asking you for." That is a real answer to "am I still needed," and it is the one most staff have never been given.
4. Promise the sequence, not the outcome
You cannot promise the role survives. You can promise this: if it changes, they hear it from you, before anyone else, and not in a meeting where thirty other people learn the same thing. That promise is entirely within your power, it is the one people actually want, and breaking it is the only part of this a leader can be justly blamed for.
The mistake I see most often
Leaders answer the question they were asked rather than the question underneath it, and then they are puzzled when the reassurance does not hold.
Across the leadership teams I work with, the pattern is consistent. Somebody senior says, in good faith, that no jobs are going. Four weeks later the same anxiety is back in the same person, because in the intervening month a new tool appeared, a task quietly moved, and nothing the leader said in that first conversation could be checked against any of it. Reassurance that cannot be checked is indistinguishable from a leader who does not know. Worse, once it fails, it takes the next thing you say down with it.
I spent years as a teacher and then as an assistant headteacher standing in front of staff to explain changes I had not chosen and could not fully explain, and the lesson from every one of those was the same: people forgive uncertainty and do not forgive being handled. Since then I have trained more than 150,000 educators in over thirty countries, and this question arrives in the room more reliably than any other. It is almost never asked by someone looking for a promise. It is asked by someone looking for evidence that an adult is watching the work.
What you cannot promise, and what to say instead
Honesty has edges, and pretending otherwise is how leaders get caught.
You cannot promise there will be no restructuring. If your organization is under pressure and you say otherwise, you are not reassuring anyone, you are borrowing against a conversation you will have to have later on worse terms. Say instead: "I am not going to tell you that nothing will ever change, because I cannot see that far. What I can tell you is what we have decided so far, and that you will hear any change from me first."
You cannot promise that the skills they are proud of will stay valuable. Some will not. What you can do is name which parts of their expertise the organization is now relying on more heavily than before, because there is nearly always something, and a person who knows what they are relied on for behaves very differently from a person who is guessing.
And you cannot answer this once. Task lists moved for half of all employees in a single year. A conversation held in September is stale by February. The leaders who handle this well do not have one good conversation. They have a standing item.
When the honest answer is that the role is changing
Sometimes it is. Then the four answers still hold, and only the content changes.
Consider a hypothetical example: a school business manager whose week was largely purchase orders, supplier correspondence and first-draft policy documents, all three of which are now machine-drafted and checked. Telling that person their job is safe is not kindness, it is a fiction they can disprove by looking at their own calendar. The honest version names the three tasks that have gone, states that the role as written no longer describes the week, and says what happens next and when: the role is being rewritten, they are being consulted rather than informed, and there is a date.
That conversation is harder and goes better. It goes better because it is true, and because it treats the person as a participant in a change rather than a bystander to one.
What good looks like three months on
Four observable things, none of which needs any fluency with a tool.
Every person on the team can name the parts of their work a machine now does. The leadership team has written down, for at least the largest processes, which work is human-led, which is AI-assisted and which the machine does on its own, which puts you in the 26 percent that had done it by mid-2026. Nobody in your organization has learned about a change to their own work from a group email. And the question has stopped being asked nervously after meetings, because there is a place it gets asked properly.
None of that is a communications exercise, and if you are early enough that the whole change still needs sequencing, the first ninety days of leading AI change sets out the order to run it in.
Where this leaves you
The reason this conversation is so hard is not that the answer is upsetting. It is that most organizations have not yet made the decision the answer depends on. A leader who cannot say which work is the machine's and which is the person's has nothing true to offer, and so offers comfort, and the comfort expires.
Make the decision first, for one process, this month. The conversation gets easier afterwards.
If your leadership team is working out how to have these conversations without either frightening people or misleading them, that is the kind of work I support through coaching for school and organizational leaders. I write about it most weeks in my newsletter.
Sources and further reading
- Pew Research Center, Globally, More People Expect AI to Cause Job Loss Than Growth, September 17, 2026. 42,151 respondents across 37 countries, surveyed February 8 to May 13, 2026.
- Pew Research Center, Young adults in the US are increasingly wary of AI, concerned it will take jobs, August 18, 2026. 3,488 US adults, surveyed June 22 to 28, 2026.
- IBM Institute for Business Value, 2026 CHRO study: Designing the Thinking Organization, published September 2026. Surveys conducted with Oxford Economics between April and June 2026 of 1,500 CHROs and senior workforce strategy executives and 8,800 employees.
- IBM and Morning Consult, New IBM study finds AI adoption is outpacing K-12 readiness, September 2, 2026. 2,048 US adults, comprising 1,019 educators and 1,029 parents, margin of error plus or minus 3 percentage points.
Dan Fitzpatrick is the founder of The AI Educator and works with schools, trusts and organizations on AI strategy, readiness and leadership. More about Dan.
