The human cost of AI: has anybody told them?

Start here

The sentence, and what was underneath it.

Thursday afternoon, near the end of a one-to-one that was almost over. Laura runs customer support for a company of about two hundred and forty people. She had been going through her numbers, which were good, and then she asked whether she should be learning something new in the evenings.

I asked her why she was asking. She said her team had been given a set of AI tools in March, the handling time had come down, and since then nobody above her had mentioned it again. Not in the monthly update, not in the all-hands, not once. Her read on that silence was simple. If it were going well for her, somebody would have said so.

She was wrong about the reason. Her CEO had not said anything because he did not yet know what the numbers meant and did not want to promise something he might have to take back. Reasonable, careful, and from where Laura sits, indistinguishable from a decision that has already been made about her.

An open office where people work alongside their agents

That is the whole of this page. The cost of AI inside a company is rarely the redundancy round that people picture. It is the six months before anybody says anything, when your steadiest people are quietly working out where they stand and doing it with no information at all.

Composite. The details are real, the person is not one person.

Sound familiar

Three things people say, and what they are actually asking.

None of these arrives as a complaint. They arrive at the end of a meeting, in a half-joking tone, which is exactly what makes them easy to answer badly.

What you hear

“I suppose I should be worried about this thing.”

What they are asking

Is my job on the list. They ask it lightly so that a light answer is possible, and a light answer is the one thing that will not settle it. If you laugh and move on, they will read the laugh as an answer and it will be the wrong one.

What you hear

“It saves me a lot of time, to be fair.”

What they are asking

Whether being faster is safe to admit. Someone who has halved a task is now weighing up whether saying so gets them praise or gets the task counted differently next quarter. Until you answer that, the honest numbers stay in their head.

What you hear

“Whatever you decide is fine by me.”

What they are asking

Nothing, and that is why this is the one to worry about. It sounds like agreement and it is usually withdrawal. Somebody has worked out that their opinion has no effect on the outcome, so they have stopped spending it. Gallup sees the same thing in its measure of whether people feel their opinions count, which slipped again this year.1

The human cost of AI is what people carry while nobody explains what the change means for them. Employees whose manager actively supports how the team uses AI are engaged at 48%. Where that support is missing, 30%.1

Before you scroll on

In the first half of 2026, 31% of American employees were engaged at work, the same as 2025. Since 2020 that is roughly eight million fewer engaged people.1
Handing out the tools does almost nothing on its own. What moves the number is a manager who talks about it: 48% engaged against 30%.1
Put three things together, regular use, a clear plan and that manager, and engagement reaches 53%.1
Only 1% of people who had been laid off named AI as the main reason.2 The fear runs far ahead of the cause, and the fear is the part costing you money.
People who never touch AI made up 62% of the laid-off group and 50% of those still working.2 Saying nothing is not free for them either.

You are reading this because

Somebody asked you, half joking, whether they should be worried
You rolled out the tools and the mood got heavier rather than lighter
Two people you rely on have gone quiet since the announcement
You have put off the conversation because you do not have the full answer yet

What you take away

The one conversation that moves this further than any tool you can buy
Three questions, matched to where you actually are right now
The research, with the sample sizes and what it cannot prove
What to say in the version where you do not know yet

Defining it

People can handle bad news. What they cannot do is guess.

Every company I work with has had some version of the same six months. The tools arrive, a few people get much faster, the leadership team waits until it understands what that means, and in the meantime everybody else writes their own explanation.

The explanations people write are worse than the truth, every time. Somebody who is quick with the tools decides they are making themselves replaceable. Somebody who is slow with them decides they are already on a list. Both of them keep it to themselves, and both of them start spending part of their week on a question that has nothing to do with the work.

Silence is not neutral. People fill it, and they fill it with the worst version. Paul Musters

There is a second thing underneath it, and it is the one that surprises people. The change is not only about jobs. Someone who has done a job well for nine years knows exactly what good looks like in that job. Move half of it to a machine and that knowledge stops being worth what it was, and there is no announcement for that. It is a loss, and treating it as resistance to change is the fastest way to lose the person.

So the human cost has two halves. One is about security, and you can answer that with information. The other is about standing, and you can only answer that by telling somebody what you now need them for. Most companies attempt the first and skip the second.

The numbers

Access changes nothing. The conversation changes a lot.

Gallup measures American employee engagement continuously, and in July it published what it is seeing as AI use spreads. In the first half of 2026, 31 in a hundred employees were engaged at work, which is where the number has sat since 2024. Each point is about 1.6 million people, so the drop from the 2020 peak works out at roughly eight million people.1

Buried in that flat line is the finding worth your afternoon. Giving people AI, on its own, barely moves engagement. What moves it is whether anybody has talked to them about it.

Manager does not actively support how the team uses AI

30%

Manager actively supports how the team uses AI

48%

That manager, plus regular use, plus a clear plan for how AI fits the work

53%

The dotted line is the American average of 31%. Bars run to 60%.

Share of employees who are engaged at work, among organisations that have adopted AI.1

Eighteen points between the first bar and the second. No new software, no budget line, no programme with a name. The difference is a manager who brings it up in a normal week and says where AI helps, where it does not, and what counts as good work now.

1%

of people who had been laid off named AI or automation as the main reason, in their own words

2

62%

of laid-off workers used AI once a year or less, against 50% of people still in work

2

3×

the layoff rate for technology workers who used AI less than monthly, compared with those who used it at least monthly

2

Those three deserve to be read together, because separately each one gets used to win an argument. Almost nobody is being replaced by AI today, and being the person who never touches it is still the riskier place to stand. Both are true. Gallup is careful to add that restructuring and cost-cutting may carry some AI influence that the people involved were never told about.2

What that means for the conversation you are avoiding is oddly cheerful. You do not have to promise anybody that nothing will change. You have to tell them which way the risk actually runs, and it runs towards standing still.

One more number, for the managers in the middle

Gallup notes that since 2020 the sharpest falls in engagement have been among younger workers, hybrid workers and managers.1 The layer you are relying on to hold this conversation is the layer that has been hit hardest. Manager support for AI did rise four points between the first and second quarter of this year, which is the most encouraging line in the whole report.1

A caution on all of it. These are self-reported answers from American employees, collected at one moment, and Gallup describes engagement differences by AI condition as associations rather than proven causes. Some of the gap will be the kind of company that adopts AI in the first place. Read it as a strong signal about where to spend your attention.

Why nobody notices

Silence is never neutral. People fill it in themselves.

Waiting until you are certain feels responsible

You do not want to say something in March that you have to withdraw in June. So you wait for the numbers, the pilot, the board meeting. Every week of that waiting reads downstairs as a decision that has already been made and is being kept from them. The honest sentence, “here is what we know, here is what we do not, I will tell you in six weeks”, costs you nothing and is the one almost nobody says.

The person who should say it is the most stretched person you have

Gallup finds manager support to be the factor most strongly linked to engagement as work changes, and also reports that managers are among the groups whose own engagement has fallen most since 2020.1 That is the whole problem in two sentences. You are asking someone who is already carrying more than they were to explain a change that nobody has explained to them.

You cannot ask a manager to bring calm to a room that nobody has brought calm to them about. Paul Musters

Announcing a tool is easy and naming what stays human is hard

Every leadership team can announce a rollout by Friday. Saying which judgement calls will still belong to a person in two years means committing to something you might be wrong about. So the first happens and the second waits, and people take the silence on the second as its answer.

Level 2 is where the tools arrived and the conversation did not, and it is where that eighteen-point engagement gap does the most damage.

01Campfire60%

02Wild West25%The gap bites hardest here

03Blueprint10%

04Engine4%

05Ecosystem1%

Share of companies per level. The five AI Culture Levels we use with clients, from everyone doing their own thing to AI being part of how the company runs.3

Where is the company as a whole?

Three minutes, anonymous, six questions. Your level, what it is costing, and what shifts one step up.

Do the Culture Level scan

A check that works

Thirty minutes, one person, three questions.

This is a conversation rather than a survey, and it works best with somebody who is doing fine. The person in trouble will tell you eventually. The steady one who has quietly started looking is the one you find out about from their resignation.

Pick where you are, and the three questions below change with it. Underneath each one is what a healthy answer sounds like and what to listen for when it is not.

Open with“We put those tools in a few months ago and I have not come back to it since. I want to know how it has actually landed with you.”

What did you think we meant by it? Good sign: they describe roughly what you intended, in their own words. Listen for: a version of your announcement that is darker than anything you said.
What have you stopped doing since we brought it in? Good sign: a concrete task, named, with an opinion about whether it was a good trade. Listen for: nothing has changed, said quickly. Either it did not land, or it did and they are not saying.
What would you want me to say at the next all-hands? Good sign: a specific request, even an awkward one. Listen for: whatever you think is best. That is withdrawal wearing a polite jacket.

The first question does most of the work. What people believe you meant is usually a long way from what you said, and you cannot correct a version you have never heard out loud.

Do three of these in a fortnight and you will know more about how the change has landed than any engagement survey will tell you. Take notes and repeat one line back to the person a week later, because that is the part that proves you were listening rather than collecting.

One thing I have learned running this badly a few times. Do not bring a laptop and do not open with the numbers. The moment it looks like data collection, you get the answers people give to data collection.

Where it sits

Three conversations tell you a lot. Thirty need something underneath them.

The conversation works and it does not scale. Past a certain size you are relying on twelve managers to each hold a version of it, and you have no idea which of them did.

An Operating Profile in use. Personality type and AI level in one profile, with the agents that fit it.

We take two readings per person: the way somebody thinks and works, and where they are with AI. Those two together tell a manager what this particular person needs to hear. The one who talks things through in a group needs a different conversation from the one who wants the change on paper a week beforehand, and getting that wrong is how a well-meant announcement lands badly.

It also answers Laura's question with something better than reassurance. She gets to see where she actually sits, what the next step looks like for someone who works the way she does, and which agents fit her week. That is a far more settling thing to hold than a promise from a CEO who is himself still working it out.

About emaho

emaho measures one Operating Profile per person: personality type and AI level in a single profile. On that we build a personal set of AI agents that fit how that person works, inside the tools they already use. Fifteen minutes to complete, first profile free, built for companies between 20 and 500 people.

See your own profile, free Find your AI Culture Level

Fifteen minutes per person. No credit card, no strings.

The move

One conversation, and who is in the room.

There is no programme here. The thing that moves this is a conversation that already fits in a week you already have, held by somebody who already has the relationship.

Who is in the room

Their manager, not you. Gallup's finding is specific about this: the manager is how people experience the whole company, and manager support for AI use lines up with engagement more closely than anything else measured.1 A CEO doing twelve of these looks impressive and changes less, because the person will hear it as a special event rather than as how things now work.

That puts the real job on you one step earlier. Have the conversation with your managers first, in the same shape, so they are not asked to bring calm to a room that nobody has brought calm to them about.

What you open with

Say what you know, say what you do not, and give a date for the next update. The sentence that does the work is the second one. “I do not know yet what this means for the size of this team in two years, and I will tell you in October what I have learned by then” is worth more than any reassurance, because people can check it in October.

What you leave out

Leave out the promise that nobody's job will change, because everybody in the room can see that it already has. Leave out the productivity numbers, which sound like a case being built. And leave out the word transformation. People hear it as something happening to them.

And then again in six weeks

One round of conversations reads as a campaign. Two rounds, six weeks apart, with the second one referring back to what somebody said in the first, reads as how this company works now. The second round is where the honest answers arrive, and it is the one that gets dropped when the quarter gets busy.

What did they say when you asked?

If you run one of these and the answer surprises you, tell me what you heard. I have had this conversation in a lot of companies and I can usually say what is underneath it. No sales call attached. Just an answer.

Message me on WhatsApp

The cost of waiting

People fill in the answer themselves, and they fill it in worse.

The visible cost is the resignation you did not see coming. Somebody good, somebody who never raised anything, who spent four months privately concluding that this place had stopped needing what they were best at. The exit conversation is polite and tells you almost nothing, because by then they have made their peace with the story they wrote.

Underneath sits the slower one. Eighteen points of engagement separate a team whose manager talks about this from a team whose manager does not.1 That gap does not show up as anything you can name. It shows up as work being handed over slightly later, questions being asked slightly less often, and a company that feels heavier to run than it did.

And there is one that only arrives later. The people who never got comfortable with any of this made up 62 in a hundred of the laid-off group, against 50 in a hundred of those still working.2 Every month of silence is a month in which the people most at risk are the least likely to have been told, and they are usually the ones who have been there longest.

Asked and answered

What people ask about the human side of this

What is the human cost of AI at work?

It is what employees carry while a company changes how work is done and nobody explains what that means for them. It shows up as disengagement rather than as complaints: people spend part of their week working out where they stand, stop volunteering opinions, and leave without ever raising it. Gallup's July 2026 data puts U.S. engagement at 31%, flat since 2024, which is around eight million fewer engaged employees than in 2020.

Is AI causing layoffs?

Rarely as the stated reason, so far. In Gallup's first-quarter 2026 data, 1% of workers who had been laid off named AI or automation as the primary cause when asked in their own words. Most named restructuring, cost-cutting or the elimination of their role, and Gallup notes those explanations may carry AI's influence without the worker being told so. About 21% of U.S. employees said their employer was reducing its workforce, against 34% who said it was hiring.

Are employees more likely to be laid off if they do not use AI?

In the American data, yes. Gallup found that people who use AI once a year or less made up 62% of laid-off workers against 50% of those still employed, and the pattern held after accounting for age, education, industry and time since layoff. Inside the technology industry the gap was sharpest: workers who used AI less than monthly were three times as likely to have been laid off as those who used it at least monthly.

How do you find out how your organisation is handling the human side of AI?

Ask people rather than count licences, and then read the organisation around them. emaho places companies on five AI Culture Levels, from everyone quietly doing their own thing to AI being part of how the place runs. Tools arriving without a plan or a conversation is Level 2, where about a quarter of organisations sit, and it is where Gallup's 18-point engagement gap between supported and unsupported teams bites hardest. The scan is six questions, three minutes and anonymous.

Does giving employees AI tools improve engagement?

Barely on its own. Gallup found engagement six points higher in organisations that have adopted AI, and part of that is which industries adopt it. What moves the number is the conditions around the tools: employees whose manager actively supports the team's AI use are engaged at 48%, against 30% where that support is missing. Add regular use and a clear plan for how AI fits the work and it reaches 53%.

Why does manager support matter more than the AI tools themselves?

Because the manager is how people experience the whole company, and they are the one who says where AI helps, where human judgement still decides, and what good work looks like now. Gallup found manager support to be the factor most strongly associated with engagement as work changes, an 18-point difference. Without it, employees interpret AI expectations on their own, and most of them interpret silence as bad news.

How do you know what each person actually needs to hear about AI?

You measure the person rather than the mood of the room. emaho builds one Operating Profile per person: personality type and AI level in a single profile. Somebody who thinks out loud in a group needs a different conversation from somebody who wants the change on paper a week beforehand, and a manager holding twelve of these cannot guess that twelve times. It also hands the person something concrete instead of a reassurance.

Why do people go quiet after an AI rollout?

Because they are working out what it means for them and have decided that asking is risky. Two questions sit underneath the silence: whether their job is on a list, and whether admitting they are faster will be rewarded or used to recount their workload next quarter. Until a manager answers both out loud, the honest numbers stay in people's heads and the quiet reads as agreement.

What should a leader say about AI when they do not know the answer yet?

Say what you know, say what you do not, and give a date. A sentence like “I do not know yet what this means for the size of this team in two years, and I will tell you in October what I have learned” is worth more than reassurance, because people can check it in October. Waiting for certainty reads downstairs as a decision that has already been made and is being kept from them.

What does emaho do about the human cost of AI?

emaho turns the change into something people own rather than something happening to them. Everyone gets an Operating Profile, personality type and AI level in one profile, and on that we build a personal set of AI agents that fit how that person works, inside the tools they already use. That answers the question underneath most of the anxiety, which is what this company still needs you for. The first profile is free, it takes fifteen minutes, and it is built for companies between 20 and 500 people.

What does “whatever you decide is fine by me” usually mean?

Usually that the person has concluded their opinion does not affect the outcome, so they have stopped offering it. It sounds like agreement and it is closer to withdrawal. Gallup tracks a related measure, whether employees feel their opinions count, and reports it slipping again in 2026 alongside a fall in feeling connected to the organisation's purpose.

Is AI anxiety at work justified by the evidence?

Partly, and it points in an unexpected direction. Direct displacement is rare so far, with 1% of laid-off U.S. workers naming AI as the primary reason. The measurable risk sits with people who never use it: AI non-users were 62% of laid-off workers against 50% of the employed. So the honest message is not that nothing will change, it is that standing still is the riskier position.

Which employees are hit hardest when work changes with AI?

Gallup reports that the sharpest engagement declines since 2020 have been among younger workers, hybrid workers and managers. Managers are the difficult case, because they are both the group most affected and the group everything else depends on: manager support is the strongest lever in the data, and it is being asked of people whose own engagement has fallen furthest.

Nearby

Landing on the same people

Each one is a different part of what the change costs the humans doing the work.

23

Managers absorb it

The layer you are asking to hold this conversation, and what it is already carrying that nobody counted.

Read this one 06

Usage gap

Why the leadership team's picture of who uses what is reliably wrong, and the two-minute way to find out.

Read this one 03

No juniors to develop

The same silence, one layer down, where the people affected have not started yet.

Read this one

All 25 AI challenges, and what to do about each

Twenty-five things that break inside a company once people start using AI, each with the research behind it and the level where it starts to bite. This one starts to bite at Wild West, and 8 of the others start there too.

Open the full set

The next move

Give people something to hold, not a reassurance.

Fifteen minutes each, and everybody knows where they stand with AI and what their next step looks like. That is a better answer to Laura's question than anything you can say from a stage.

Answer Laura's question with something concreteWhere somebody stands and what their next step is beats any reassurance about nobody losing their job.
Let people read their own result firstFifteen minutes on their own, and the news arrives from the page rather than across a desk.
Say the part you do knowMost of the fear comes from the silence around the plan rather than from the plan.

Start with your own profile, free Ask me anything first Plan a 30-minute call

First profile free · no credit card · built for companies of 20 to 500 · you decide what your team gets to see

Not ready to put your team in anything yet? Start with the level of the company instead. The Culture Level scan is six questions, three minutes, and asks nothing of you.

Paul MustersFounder of emaho

Paul Musters

Fifteen years of leadership and team development in Dutch scale-ups. That practice now sits in software: one Operating Profile per person, with agents that actually fit. He writes these pages from what he runs into with clients, not from a research summary.

LinkedIn · paul@emaho.world · WhatsApp

Sources and numbers used on this page

Gallup, Employee Engagement Remains Flat as AI Adoption Accelerates, by Jim Harter, published 21 July 2026, with analysis contributed by Sangeeta Agrawal. Gallup Panel data on U.S. employees. Source of: 31% of U.S. employees engaged in the first half of 2026 and 18% actively disengaged; a recent high of 36% in 2020 falling to 31% in 2024; each percentage point representing approximately 1.6 million full- or part-time U.S. employees, so roughly 8 million fewer engaged employees since 2020; an estimated $2 trillion in annual lost productivity from employees who are not engaged or actively disengaged; 49% strongly agreeing they know what is expected of them, up from 47%, against a peak of 61% in 2015; engagement six points higher in AI-adopting organisations; eight points higher for weekly AI users within those organisations; a 15-point higher engagement rate where the organisation provides a clear plan for integrating AI; 48% engagement where the manager actively supports the team's AI use against 30% where employees do not say this; 53% where frequent use, a clear plan and active manager support are all present; manager support for AI rising four points from the first to the second quarter of 2026; sharpest declines since 2020 among younger workers, hybrid workers and managers; and, on productivity, 17% of weekly AI users giving the highest possible rating, 24% against 5% by frequency of use, 33% against 9% by manager support, and 50% against 17% where all supportive conditions are present. Gallup presents these differences as associations and notes that part of the gap may reflect which industries adopt AI.
Gallup, U.S. Workers Continue to Report Downsizing, by Mary Page James and Ryan Pendell, published 17 June 2026, covering the first quarter of 2026. Source of: about 21% of U.S. employees reporting that their employer is letting people go, against 34% reporting hiring and 45% reporting no change; 1% of currently laid-off workers naming AI or automation as the primary reason in their own words; 62% of laid-off workers being AI non-users, meaning use once a year or less, against 50% of currently employed workers; 28% of employed workers being frequent AI users against 22% of laid-off workers; technology workers who used AI less than monthly being three times as likely to have been laid off as those who used it at least monthly; and 13% of laid-off workers previously working in technology against 6% of the employed workforce. Gallup notes that reasons such as restructuring and cost-cutting may reflect AI's influence even where workers were not told so.
emaho AI Culture Levels. Level 2, Wild West, is daily individual use with no shared standard and holds about a quarter of organisations. Share of organisations per level is emaho’s own estimate, calibrated against BCG 2025 and McKinsey 2025 rather than measured directly.

Numbers are quoted as published. Both surveys cover U.S. employees. Laura and the conversations described are composites drawn from client situations rather than transcripts, and the names are not real.