Anxiety and Uncertainty

Jessica Chen has been a graphic designer in San Francisco for twelve years. Logos, branding, illustration, clients from scrappy startups to Fortune 500 boards. She is good at it, her portfolio proves it, and until recently her livelihood felt as solid as anyone's in a precarious profession.

Then the questions started. In 2025, clients began asking whether she used AI to speed things up. A few asked for "AI-generated concepts" as a starting point. One or two mentioned they'd been playing with Midjourney themselves over the weekend, and wondered—politely, almost apologetically—why they should pay her rates when a machine could spit out "something similar" in the time it takes to make coffee.

Jessica does use AI now. As a tool, not a replacement: she generates concepts faster, iterates more, explores directions she wouldn't have had time for before. It has made her more productive. It has also made her afraid. Because underneath every client's casual mention of AI she hears a question they haven't quite asked yet: Why do we need you? She lies awake doing the math on how long before the answer is "we don't"—not because her work got worse, but because the alternative keeps getting better, faster, and cheaper.

This is AI anxiety, and it is not the cinematic fear of robot overlords. It is the quieter, more corrosive worry that the thing you trained for, the thing that pays your rent and tells you who you are, is quietly being done better by software. Not someday. Soon. Jessica is one worker. The striking thing about the data is how many others describe the exact same feeling—and how hard it is, once you look closely, to say precisely how many.

What the Numbers Actually Say

Open any 2025 write-up of AI and worker wellbeing and you will trip over a statistic within the first paragraph. Thirty-eight percent of workers worry AI will make some or all of their job duties obsolete. Eighty-nine percent are concerned about AI's impact on their job security. Seventy-one percent fear AI could permanently eliminate jobs. The numbers are real, but the gap between 38% and 89% is wide enough to drive a truck through, and understanding why that gap exists is the first step to reading any of these figures honestly.

The 38% figure comes from a July 2025 survey by the American Psychological Association, which asked workers a fairly narrow question: do you worry AI will make some or all of your job duties obsolete (APA, 2025). The 89% figure comes from a separate employer-focused survey that asked a much broader question—are you concerned about AI's impact on your job security—and then let respondents grade that concern: 47% slightly worried, 34% moderately worried, 8% extremely worried (Resume Now, 2025). Stack those two numbers next to each other and they look contradictory. They aren't. They are measuring different things.

The variation across studies is not noise; it is almost entirely an artifact of how the question is framed and what threshold counts as "anxious."

Read the table top to bottom and the "wide variation" dissolves into something almost orderly. When you ask whether people feel any flicker of worry, nearly everyone says yes—which tells you AI is salient, but not that the workforce is in crisis. When you ask people to commit to a strong prediction or to rate their distress as severe, the numbers fall to a fifth or a tenth of workers. Both ends of the range are true. The 89% describes the breadth of a mood; the 8-to-24% describes the depth of genuine distress. Confusing the two—treating "everyone is a little concerned" as "everyone is in acute distress"—is the single most common error in the popular coverage, and it inflates a real problem into a moral panic.

The anxiety is not free-floating, either. It sits on top of visible change. One 2025 analysis estimated that employers had already cut more than 76,000 jobs in moves attributed to AI adoption (Careerminds, 2025)—not in one dramatic wave, but steadily, across industries, as tasks were absorbed and roles compressed. Workers have watched colleagues quietly transition out, job postings turn pickier, and the bottom rung of the career ladder—the entry-level analyst job, the junior copywriting gig—thin out into what feels like an automation gap. The fear tracks something moving in the world, even if the worker overestimates its speed.

A Different Kind of Dread

To understand why AI anxiety feels distinct, it helps to compare it with the older, more familiar fear of losing a job. Traditional displacement is a discrete event with a clear shape. A plant closes. A department is dissolved. An employer files for bankruptcy. It is brutal, but it is legible: there is a date, a cause, a before and an after. And crucially, a discrete rupture gives a person something to do. You can grieve the old job, rage at the decision, file for unemployment, retrain, and tell a coherent story—"the plant closed, so I moved on." Psychologists have long known that clear events, however painful, are easier to process than ambiguous ones, precisely because they permit an ending.

AI displacement, for most workers, offers no such ending. It arrives as erosion rather than rupture. A task gets automated here, a decision gets "assisted" there, judgment is supplemented rather than overridden, and the worker remains employed the whole way through—useful but slightly hollowed out, competent but no longer distinct. Therapists report that the fear they hear most is not of being fired but of becoming obsolete: the creeping sense that what you do, what you trained for, is being done better by a machine while you're still sitting at your desk (CNBC, 2026). There is no date to circle, no event to grieve, no moment at which you are permitted to say "that's over now." You cannot mourn a loss that hasn't finished happening, and you cannot fight an adversary that advances through a thousand small adjustments rather than a single pink slip.

This is why the threat feels not merely economic but, to borrow a heavier word, ontological. When machines automated physical labor in earlier industrial waves, they took over work that few people had built their identities around lifting or hauling; the loss was to the wallet. AI reaches instead into the cognitive and creative domains that workers had understood to be the distinctly human part of themselves—writing, designing, analyzing, judging. When a copywriter watches a model produce serviceable prose in seconds, the injury is not only to her income but to her sense of what made her her. The threat to employment is frightening. The threat to identity is disorienting, because it unsettles not just how you earn but who you get to be.

Four Mechanisms, One Compounding Burden

Researchers who study AI anxiety have converged on roughly four psychological pathways through which the mood translates into measurable strain—and the important point is that they do not act in isolation, they compound.

The first is cognitive overload. Continuous AI integration demands that workers keep learning new tools, adapting workflows, and running to stay in place. Each new system is individually manageable; the relentless sequence of them is not. The demand drains the same finite pool of cognitive resources people rely on to focus, process information, and make decisions, so the worker ends the day depleted before any anxiety about the future even enters the picture.

The second is perceived lack of control. Occupational psychology has found for decades that the sense of powerlessness—not the objective severity of a threat, but the feeling that its outcome lies beyond your reach—is among the strongest predictors of chronic stress. AI displacement is almost designed to maximize this feeling. It is driven by forces at the scale of the whole economy, decided in boardrooms the worker never enters, and it offers no clear target to push against. You cannot out-argue a productivity trend.

The third is anticipatory rumination: the mind rehearsing the feared event over and over, replaying worst-case scenarios and triggering real physiological stress in response to a threat that hasn't arrived. Ordinary grief or acute stress tends to fade as circumstances resolve. Anticipatory rumination does the opposite—it persists precisely because the threat stays ambiguous, keeping the worker in a state of chronic low-grade alarm with no resolution in sight.

The fourth is emotional resource depletion, which is what you get when the first three run long enough. Overload, powerlessness, and unrelenting uncertainty gradually exhaust the psychological reserves a person draws on to stay resilient. Work that once felt engaging turns draining; small setbacks provoke outsized reactions; the passion that made the job meaningful thins out (ScienceDirect, 2025). Mental health researchers warn that these dynamics, sustained over time, can move beyond ordinary stress into clinical anxiety, or sharpen existing symptoms in workers already vulnerable.

graph TD
  A[Continuous AI change] --> B[Cognitive overload]
  A --> C[Perceived lack of control]
  A --> D[Anticipatory rumination]
  B --> E[Emotional resource depletion]
  C --> E
  D --> E
  E --> F[Burnout / clinical anxiety risk]
  C -.reinforces.-> D

The compounding is the crux. Any one mechanism is survivable. It is their interaction—exhaustion feeding powerlessness feeding rumination feeding more exhaustion—that turns a reasonable worry into a burden capable of doing lasting harm.

The Generational Puzzle

One finding recurs across surveys and, at first glance, looks backwards. Younger workers report far more AI anxiety than older ones: 24% of workers aged 18 to 34 rate their job-loss concern at 8 or higher on a 10-point scale, against just 10% of those 55 and over (Great Place To Work Canada, 2025). Yet historically, it is older workers who have faced the steeper displacement risk—harder to retrain, more expensive to keep, more likely to be pushed out in a restructuring. Why would the group with more runway be the more frightened?

The answer is that the older worker's advantage in a discrete layoff becomes a disadvantage in the face of a gradual transformation. Displacement risk and displacement dread are governed by different clocks. A 60-year-old in an automatable role faces genuine danger, but over a short horizon—a few years to retirement, after which the question is moot. A 25-year-old faces four decades of working life in which the ground keeps shifting, and every skill investment is a bet on a future no one can read. The relevant quantity for anxiety is not "how likely am I to be displaced this year" but "how much of my career is exposed to an unknowable trajectory," and on that measure the young are maximally exposed.

Career stage sharpens it further. Younger workers are still building the professional identity, the network, and the accumulated expertise that older workers already own outright. AI threatens those foundations before they have set. An established worker has reputation and relationships as buffers; a new one has a résumé and hope. Financial vulnerability then turns the screw: student debt, high rents, thin savings, and for many millennials and Gen Z the memory of entering the workforce during a prior crisis, so that AI anxiety lands not on a clean slate but on top of economic stress that was already there.

None of this makes older workers immune—those in data entry, administrative support, and routine analysis face real risk. But they tend to voice it differently: less as existential dread about the future of a whole profession, more as practical concern about re-employment if they're let go close to retirement. Age changes the form of the anxiety more than its underlying logic. This raises a genuinely open question the data cannot yet settle: as today's anxious 18-to-34 cohort ages into career maturity under continued AI advancement, does the accumulated strain compound into a lasting mental health burden, or does familiarity breed adaptation? We don't know, and honest analysis should say so.

The Employer Response Gap

Faced with an anxious workforce, many employers reached for the tools closest to hand: they expanded mental health benefits. Employee assistance programs grew, therapy access improved, mindfulness apps and stress-management resources were rolled out at scale. These are not worthless. But they treat the symptom and leave the cause untouched—and workers notice. Offering therapy for AI anxiety while simultaneously automating roles and trimming headcount is a bit like handing out umbrellas indoors while you take the roof off. When benefit expansions arrive in the same quarter as a restructuring announcement, they read as performative: an acknowledgment of stress that stops well short of any commitment to stability.

The sharpest failure point is transparency. When 54% of workers say their employer is only "somewhat transparent" about its AI plans (Resume Now, 2025), the silence itself becomes a generator of anxiety, separate from and additional to whatever AI actually does. Not knowing which roles are under review, on what timeline, or what options exist forces workers to fill the vacuum with speculation—and speculation is almost always worse than the truth, because it runs to the worst case and never resolves. A worker told plainly that her role will change in eighteen months, with retraining support attached, can plan. A worker who senses something coming but has been told nothing can only ruminate. This is the transparency gap in a sentence: uncertainty about AI deployment compounds the anxiety that actual AI deployment creates.

A minority of companies have done better—internal mobility programs, proactive reskilling, phased transitions instead of abrupt cuts—and where they exist, the evidence suggests these measures genuinely reduce anxiety, largely by restoring a sense of agency and forward visibility. But they remain the exception. The dominant pattern is reactive: deploy the AI, watch the disruption, add the wellness resources afterward. Responding to the structural source rather than its psychological expression would mean treating worker wellbeing as an input to deployment decisions rather than a bandage applied once the damage is done.

Adaptation and Its Limits

Workers are not passive in the face of all this, and their responses reveal both what individual effort can accomplish and where it hits a wall. The most common move is upskilling—learning the AI tools, acquiring new technical competence, developing capabilities that complement automation rather than compete with it. Workers who come to treat AI as a productivity multiplier, rather than a purely existential threat, do report lower anxiety and higher satisfaction. The catch is that rapid development keeps moving the target: the skill that looks AI-proof today may not be next year, so the worker is running a race with no visible finish line.

A second strategy is role diversification—expanding into the parts of the work that still demand distinctly human judgment. Designers move toward brand strategy, writers toward editorial judgment and audience development, analysts toward the consultative framing of findings rather than their mere production. This has real merit, but it presumes time and financial slack that not every worker has. A third, underrated response is deepening professional relationships: positioning yourself as a trusted advisor whose judgment and context are valued, rather than a production resource whose output can be metered. Advisors are harder to automate than executors.

Every one of these helps. None of them dissolves the underlying uncertainty. They buy degrees of agency and resilience, not guarantees, and the continuous, endpoint-less nature of the adaptation is itself a psychological cost that no amount of individual cleverness can retire. This is the honest boundary of self-help here: individual strategies work on the worker's exposure, but the source—an open-ended transformation driven by forces above the level of any single employee—is structural, and structural problems are not solved one anxious person at a time.

How Much Should We Trust the Data?

A responsible reading of all this has to reckon with what the evidence cannot do. Nearly every figure quoted here comes from self-reported surveys, and self-report carries known biases that mostly point in opposite directions, which is part of why studies disagree. A worker asked by her own employer whether she fears AI has every incentive to understate it—admitting anxiety can read as admitting fragility. The same worker asked by an independent pollster, primed by months of alarming headlines, may round her vague unease up to "concerned." Survey wording, who is asking, and the ambient news cycle all move the number before any real feeling is measured. Most of these studies do not correct for these effects, which is why the honest thing to report is a range, not a point estimate.

The deeper limitation is causal. Very few studies can show that AI exposure specifically produces measurable anxiety, as opposed to general economic uncertainty, pre-existing mental health vulnerability, or a media environment saturated with job-loss stories. Workers worried about AI are often also worried about rent, recession, and a dozen other things, and disentangling AI's unique contribution from that background hum is genuinely hard. The correlations are robust; the clean causal isolation is not yet there. And we remain largely ignorant about who is most vulnerable—whether any observable trait reliably predicts who adapts and who suffers sustained harm. Age gives a partial signal; beyond it, the predictive picture is thin.

This matters for how we should respond, because it bears on a normative line the book should draw carefully. Some AI anxiety is an appropriate response to genuine risk—the graphic designer watching her market compress is reading the situation correctly, and treating her worry as a disorder to be medicated would pathologize a rational assessment. Other anxiety tips into something disproportionate, a rumination loop feeding on threats that outrun the evidence. Policy and clinical practice have to hold both at once: take the legitimate concern seriously enough to address its structural causes, without medicalizing a reasonable fear into an individual pathology. The failure mode in both directions is real—dismiss the worry as irrational and you abandon people to genuine harm; treat every worried worker as a patient and you relocate a political-economic problem into the therapist's office, where it cannot be solved.

Which leaves the question of who is actually responsible for the psychological costs of this transition. Workers have the least leverage and are asked to carry the most; their adaptation strategies matter but cannot reach the source. Employers hold real leverage—transparency, phased transitions, genuine reskilling—and the strongest claim that they ought to use it, to the point where a persistent failure to be transparent about deployment plans starts to look less like a communication lapse and more like a foreseeable harm to worker wellbeing that regulation might eventually address. AI developers set the pace that resets the anxiety baseline but sit furthest from the affected worker. The state holds the broadest tools—safety nets, retraining infrastructure, labor protections—but moves slowly. No single actor can shoulder it alone, and the temptation to pin the whole burden on the individual worker, who has the least power to change any of it, is both the easiest response and the least defensible one.

Summary

AI anxiety—the chronic, diffuse worry that one's skills, role, or profession may be quietly rendered obsolete—became one of the defining workplace mental health stories of the mid-2020s. Its central data puzzle is the enormous spread in reported figures, from 38% to 89%, and the resolution is that these numbers measure different things: the high end captures how broadly AI weighs on workers' minds, the low end (8–24% at severe levels) captures how deeply it distresses them. Both are true; conflating breadth with depth is the most common way the statistics get misread.

What distinguishes this anxiety from older job-loss fear is the absence of a rupture. AI arrives as gradual erosion, offering no event to grieve and no ending to reach, and it reaches into cognitive and creative work in a way that threatens identity, not just income. Four psychological mechanisms—cognitive overload, perceived lack of control, anticipatory rumination, and emotional resource depletion—compound into a burden capable of tipping into clinical harm. Younger workers carry the heaviest load, not because they face more immediate displacement, but because more of their unwritten careers is exposed to an unreadable trajectory.

Employer responses have mostly treated symptoms; the structural levers—above all transparency about deployment plans—remain underused, and the silence itself amplifies the anxiety. Individual adaptation helps at the margins but cannot resolve a structural problem one worker at a time. And the evidence, honestly read, is a set of robust correlations built on self-report, not clean causal proof—enough to take the distress seriously, not enough to say precisely who will be harmed or to pin the cost on any single actor. The appropriate response treats the anxiety as often rational, addresses its structural sources, and resists the easy move of relocating a political-economic problem into the individual worker's psyche.

Key Takeaways

1. The 38–89% range is a measurement artifact, not a contradiction. High figures count any concern; low figures (8–24%) count severe, self-rated distress. Read the question wording before trusting the number.

2. AI anxiety differs from job-loss fear by lacking a rupture. Gradual erosion offers no event to grieve and no ending to reach, which is precisely what makes it hard to process—and it threatens identity, not just income.

3. Four mechanisms compound. Cognitive overload, perceived lack of control, anticipatory rumination, and emotional resource depletion interact, turning a reasonable worry into a burden that can become clinical.

4. Younger workers are more anxious because more career is exposed, not because they face more immediate displacement. Time horizon, unformed professional identity, and financial fragility explain the 24%-versus-10% gap.

5. The employer gap is structural. Mental health benefits treat symptoms; transparency, phased transitions, and real reskilling treat causes—and the 54% "only somewhat transparent" silence is itself a source of anxiety.

6. The evidence is correlational and self-reported. It justifies taking the distress seriously but cannot yet isolate AI as the cause or predict who will be harmed. The right posture treats much of this anxiety as rational, fixes its structural sources, and refuses to pin the cost on the least powerful actor.

Sources

• Workplace Mental Health Trends 2025 | Mindshare Partners
• The impact of AI anxiety on employees' work passion | ScienceDirect
• Mental health in the "era" of artificial intelligence | Frontiers in Psychology
• Impact of AI workplace anxiety on life satisfaction | PMC
• Workplace Anxiety in 2025 | Great Place To Work Canada
• The social anatomy of AI anxiety | Frontiers in Psychiatry
• Gen Z's AI Anxiety | FAS Psych
• Therapists say workers anxious about AI | CNBC
• AI Disruption: 9 in 10 Workers Fear Job Loss | Resume Now
• AI job anxiety in 2025: Professions most worried | Careerminds
• New Poll: 7 in 10 Americans Fear AI Job Loss | eWeek

Last updated: 2026-08-23