Stop reassuring your team about AI

Somewhere in your company there is a marketing manager doing the arithmetic on a mortgage. They have read that AI is coming for work like theirs, and the reading is not wrong. They have also sat through the all-hands where someone said AI will augment rather than replace and nobody is going anywhere. They did not believe it, and they were right not to.

They have two places to go for an answer. One is a feed saying the technology may end the species, with a probability attached and no instruction of any kind. The other is their own director, telling them it will be fine. The first gives them nothing to act on. The second is a promise that director has no authority to make, and they can tell.

I want to make the case that the second failure is the expensive one, and that it is the one leaders keep choosing. Reassurance you cannot underwrite is the most costly thing you can offer an anxious team, because it spends the only asset you will need later.

The fear is not the problem. It is the data.

Start with whether that worry is irrational, because almost every corporate response assumes it is. It is not.

Alongside that, 18% of US employees say their job is likely to be eliminated within five years, rising to 23% at organizations that have actually implemented AI, and that 18% is up from 15% a year ago (Gallup). Mercer has the share of employees naming AI-driven job loss as a fear going from 28% in 2024 to 40% in 2026 (Mercer, a disclosed survey of roughly 12,000 people).

So the fear is rising, and the cuts underneath it are real and accelerating. Treating that worry as a morale problem to be managed down means discarding the most accurate instrument you have for finding out where your rollout is about to fail.

The honest version is worse in one direction and better in another

Being accurate about the fear also means being accurate about its shape, and here the evidence cuts against the panic.

MIT FutureTech had experienced workers run more than 17,000 evaluations across over 6,000 text-based tasks drawn from the Department of Labor's own O*NET taxonomy. They found little evidence of capability jumping abruptly across narrow sets of work, which is the crashing wave everyone pictures, and substantial evidence of a rising tide instead: continuous, broad improvement (MIT FutureTech). On tasks that take a person about an hour and a half, models succeeded roughly 60% of the time in mid-2024 and above 70% by late 2025. Meanwhile Goldman Sachs, working from Census Bureau data, put US establishment AI adoption at 22.4% in August 2026, with firms expecting a little under 26% within six months (Goldman Sachs). The Challenger cuts are also concentrated rather than general, with technology the dominant sector.

Put those together and you get something more useful than either camp is selling. Their whole occupation almost certainly does not disappear next quarter. Individual tasks inside it move continuously, in different directions, and nobody sends an announcement when they do. That is a harder thing to plan against than an apocalypse, and it is the truth.

Why "just learn the tools" is not an answer to the question being asked

The standard prescription is training. Get fluent, stay valuable. It is decent advice and you should follow it. What it is not is a treatment for the fear, and the evidence on that point is the most uncomfortable number in this essay.

Fluency does not settle the question. It sharpens it. The people who use these systems every day are the ones best positioned to see what they will be able to do in eighteen months, and what they see makes them more worried rather than less. Anyone selling upskilling as an anxiety program is selling something that produces the opposite result, and the research saying so is public.

This matters for what you say on Monday. If you answer "am I still valuable here" with "we are investing in training," you have answered a different question, and they will notice that you dodged.

Nobody is funding the part that breaks

There is a gap between what companies buy for AI and what makes AI land, and it is wide enough to see from orbit.

The consequences show up in numbers nobody wants on a board slide. Mercer has employees thriving at work falling from 66% in 2024 to 44% in 2026, below pandemic levels (Mercer). Gallup has global engagement down to 20%, the first time it has recorded two consecutive annual declines and the lowest reading since 2020 (Gallup).

The lever is the manager, and the manager has been handed nothing

Here is the finding that should reorganize how rollout money gets spent.

The single largest multiplier on an AI investment is a manager actively backing it, and two thirds of managers are not doing that. Not because they are obstructive. Because nobody gave them anything to say. They were handed a license count, a training deadline, and a deck, and then asked to stand in front of people whose livelihoods are in question and be convincing about a future the company has not decided yet.

So they reassure. It is the only move available when you have no information and a room full of frightened people. And it is the move that costs the most.

What to say instead

The alternative is not brutality and it is not a script. It is being specific about four things, one of which most leaders skip.

Notice what none of those require. No budget, no platform, no vendor, no survey. They require a manager willing to say "I do not know" out loud, which is a thing most organizations have spent twenty years training out of them.

What cuts against this

Three honest objections, and I would rather raise them than have you find them.

The rising tide finding undercuts the urgency. If capability improves gradually and broadly, and if fewer than a quarter of US establishments have adopted AI at all, then maybe this is ordinary technological churn with an unusually loud press cycle, and the right response is patience rather than a program. That reading is available from the same sources I used above. My answer is that gradual change still produces the 112,713 announced cuts and the 40% fear number, and a manager standing in front of a worried team this month cannot wait for the average to resolve. But the objection is real and I will not pretend it is not.

Telling people to stop reassuring is convenient for me. I wrote a book about leading through this and I sell an engagement built on a diagnostic. An argument whose conclusion is "your managers need better language for a hard conversation" happens to point at what I do. Weigh that. The defense I would offer is that everything above is sourced to Gallup, Mercer, Challenger, MIT and Goldman rather than to my own numbers, and that the four things to say cost nothing and need nobody's help.

Some of what circulates here is not usable. The figures that would have made this essay easier are the ones I left out.

The position, in one paragraph

Employee fear about AI mostly tracks reality, and the people most fluent with these tools are the most worried, so fluency is not the cure. The gap between what companies spend on AI capability and what they spend on human absorption is where the investment quietly dies, and the manager is both the largest available multiplier and the person handed the least to work with. The most valuable thing a leader can do is stop offering comfort they have no authority to deliver, and start being specific about what is decided, what is not, and when they will know. That is harder than reassurance and it is the only version anyone will believe.