Everyone around you seems to have figured out AI.
That feeling is costing you more than the AI is.

Someone ten years younger than you shared their screen on Tuesday and did, in about twenty minutes, a piece of work you would have blocked out a full day for, and you said something generous about it, and you meant it, and you went home that evening carrying a weight you could not have described to anyone because nothing had actually gone wrong. You have done the courses, you have the tools open in a tab all day, and your company sent the email about being committed to supporting everyone through this transition, and still, somewhere underneath the competence you show the world, a quiet ledger has opened in which you keep coming up short.

Two studies published in the last three months put numbers to that ledger, and the numbers are more precise than the reassurance you have been offered. An 18-month study of 2,109 workers found that how often a person used AI at work did not predict their exhaustion at all, while the habit of measuring themselves against other people was the most consistent predictor of it. A second study, published last Friday, found that an employer's support eases the fear of losing your job by about a third and does not reach the deeper doubt about where your career is going. The tool is doing far less damage than the scorekeeping, and the reassurance is aimed at the wrong fear.

Does using AI at work actually make you more exhausted?

In the September 2026 issue of SSM - Population Health, Iina Savolainen, Atte Oksanen and colleagues at Tampere University reported on a panel of Finnish workers aged 18 to 65, built to match the national working population and followed across three waves, in the autumn of 2024, the spring of 2025 and the autumn of 2025. The study began with 2,109 people and kept 1,479 of them to the end, an average age of just under 43, half of them men, and it measured emotional exhaustion with five items from the Maslach Burnout Inventory, the standard instrument in burnout research, alongside how often each person used AI at work, how capable they felt with it, and how strongly they tended to compare themselves with others.

The frequency of AI use predicted nothing at all, because whether the researchers compared heavy users with light users, or tracked the same person as their own use rose and fell across the 18 months, the association with exhaustion was indistinguishable from zero, and among the people who used AI at work at all, between 42 and 53 percent were using it daily across the three waves, so this was hardly a sample that had stayed away from the tools. What did go with lower exhaustion was feeling capable with AI, which the authors treated as a personal resource in the way that control over your work or security in your job are resources, and a larger American study points the same way on the question of mood. In May 2026, Roy Perlis and colleagues at Massachusetts General Hospital and Harvard published in BMJ Mental Health an emulated trial drawn from 19,099 US adults, of whom 3,109 were followed up, and found that using generative AI several times a week or more had no causal relationship with later depressive symptoms once a person's starting mood was accounted for. The raw difference of two thirds of a point on the depression scale shrank to almost exactly zero after that adjustment, which the authors read as a sign that earlier cross-sectional surveys had most likely been picking up pre-existing differences between the people who used AI and the people who did not, and calling it an effect of the tools.

Why does watching colleagues use AI wear you down?

The Finnish team measured something called social comparison orientation, which is the settled habit of paying close attention to how you are doing relative to the people around you, and it turned out to be the single most consistent predictor of exhaustion in the entire study. People who habitually compared were more exhausted than people who did not, and, more tellingly, the same person became more exhausted during the periods when their own comparing rose, which is the within-person pattern that rules out the easy story that this is only about stable differences between people, although the authors are clear that their design cannot settle which comes first. In their words, the findings are consistent with the possibility that AI-mediated work environments, characterised by increased performance visibility and opportunities for comparison, may intensify strain among those workers who are particularly sensitive to social evaluation. In an exploratory analysis they are careful to flag as tentative, workers with a stronger comparison habit did become more exhausted in the periods when they used AI more heavily, an effect that did not appear in the sample as a whole.

Here is why this lands hardest on people with fifteen or twenty years behind them. You built your worth on judgment, on knowing which of the three plausible answers is the one that survives contact with the client, the regulator or the board, and judgment has always been invisible, slow, and impossible to demonstrate in a screen-share. What AI makes visible is speed, and speed is the one dimension on which a 28-year-old with six months of practice can now beat you in front of everyone, so the comparing that used to run quietly in the background of your career has been handed a scoreboard that updates in real time. You are unlikely to be measuring the right thing when you measure yourself against that colleague, and the research suggests the measuring itself, far more than the tool, is what has been draining you.

Your company says it supports you. Why does that never reach the part that hurts?

On 18 September 2026, João Matias and Joana Carneiro Pinto of the Universidade Católica Portuguesa published a study in Frontiers in Artificial Intelligence of 106 professionals working in organisations that had already integrated AI into their business, most of them in large companies, most of them degree-holders, and nearly nine in ten of them in organisations that had adopted AI within the previous two years. The study is small and cross-sectional, and the authors say so plainly, which is exactly why its central finding deserves a slow reading before anyone waves it away. AI-related insecurity explained 53 percent of the variation in how safe people felt in their jobs, a correlation of minus 0.73, which is about as strong as these relationships ever get in organisational research.

Then they tested what happens when the organisation is felt to be supportive. Perceived support did soften the link between AI insecurity and the fear of losing one's job, taking the slope from minus 0.58 to minus 0.38, which is a reduction of roughly a third, and which is presumably what the all-hands email and the reskilling portal are meant to achieve. Support had no measurable effect at all on the link between AI insecurity and career satisfaction, the sense that your working life is going somewhere you chose. The authors call this asymmetric buffering, and their open-ended responses explain it. Twenty-eight of the 68 people who wrote something wrote about anxiety and uncertainty over the future of work, and sixteen wrote specifically about the absence of clear communication, with one respondent putting it as hearing rumours about AI coming while no one really explains how it will affect their jobs, and another saying they feel that at any moment their role could be performed by a machine that is faster and cheaper. The authors put it bluntly: if employees feel excluded from strategic discussions, no amount of generic support will mitigate their role uncertainty.

Read those two studies together and the shape of your Tuesday evening becomes clear. The reassurance you were given was aimed at the fear of a layoff, and it took about a third off that fear, while the thing actually wearing you down all along was the scoreboard, together with the silence about where a person like you fits once the scoreboard is the only thing anyone can see. That silence tends to show up as the exhaustion of endlessly keeping up, as feeling more replaceable and more alone at the same time, or as a fraudulence that ignores everything on your CV, and each of those has its own research, and each is usually the same wound wearing a different coat.

You opened the ledger yourself, years before any of this, and all AI has done is hand it a scoreboard that updates in real time and a roomful of people who can read it over your shoulder.

How do you tell whether it is the AI or the comparing?

From the inside these feel identical, because both arrive as a tiredness with no obvious cause and a reluctance to open the tools you know perfectly well how to use, and what cannot be separated cannot be worked on, which is where the research offers a way in. If your exhaustion rises and falls with how much everyone else seems to be doing, while your actual exposure has barely changed, the problem is the ledger, and if you have never once sat down and worked out which parts of your role AI touches and which it cannot reach, the problem is a missing roadmap, which is a different job with a different fix.

That is what the free tools on this site are for. The AI Automation & Relevance Index gives you an honest, unhurried read of your actual exposure, task by task, which is very often smaller and more specific than the dread suggests, and having a number of your own is the first thing that quietens a comparison. The Cognitive Load & Bandwidth Index shows how much of your capacity the scorekeeping has already been eating, which tends to be the moment people realise the tiredness was never about the work. If the insecurity has started to feel like a career question rather than a tools question, the Career Friction & Alignment Audit separates friction that comes from the work from friction that comes from the place you are doing it, and if you find yourself measuring against peers in more areas than this one, the same habit runs the salary scoreboard too. All of them live on the free tools page, none of them stores your answers, and none of them is a verdict.

Where a real conversation comes in

Notice what the two studies are pointing at when you hold them side by side. The comparing is a habit, which means it predates AI and will outlast it, and it is a habit that responds to being noticed and named in a way that a technology never will. The roadmap is something your organisation has, on the evidence, failed to give you, and waiting for it to arrive is the same strategy as waiting for the scoreboard to start measuring judgment. Both of those are things a person can work on with another person, and neither is something a course, a portal or a reassuring email has ever fixed for anyone I have sat across from.

In more than three thousand conversations with professionals from Google, Meta, Apple and Fortune 500 companies, the version of this I hear most often is from someone senior who has never once said out loud that they feel behind, because saying it would confirm it. A coaching conversation is a place where that sentence can be said without it becoming a verdict, where the ledger can finally be looked at instead of carried, and where the roadmap for a person with your particular judgment can be drawn by the one person who actually has all the information, which is you. Coaching will not make the 28-year-old slower, and it cannot make your company communicate, and what it can do is stop the comparing from spending the capacity you need for the work only you can do.

If the Tuesday in the first paragraph was yours, and if you have been keeping that ledger quietly for longer than you would like to admit, that recognition is worth taking somewhere other than back to the next screen-share.

Common questions about AI, comparison and exhaustion at work