Social Comparison and Status

Claire Hoffman had spent six years becoming the person her company called when a campaign was in trouble. A senior marketing strategist at a consumer-goods firm, she knew which clients needed a phone call before a difficult email, which product launches would stumble in which regions, and why the campaign that tested beautifully in a focus group would fall flat in the wild. None of this was written down anywhere. It lived in her judgment, and for most of her career that judgment was the thing her status rested on.

Then, in early 2025, the company hired three strategists in their mid-twenties. They were fluent with generative AI in a way Claire was not—drafting campaign concepts in an afternoon, spinning up sentiment analyses that would once have needed a data team, producing polished decks faster than anyone could sketch a rough one. In meetings they narrated their process out loud: what the model suggested, what the AI flagged. Leadership was impressed. At her next performance review, Claire's manager praised her strategic instincts and then, almost as an aside, mentioned that promotions would increasingly favor "AI-native thinking." The praise was real. So was the demotion buried inside it.

Claire uses AI too—for research, for first drafts, for brainstorming. She is not incompetent; she is simply less fluent than people fifteen years younger, and she can feel that gap being converted into a verdict about her worth. What she is experiencing has a name in the emerging research literature: AI-based status anxiety, the perception that professional standing now rests on demonstrable AI skill more than on domain expertise, and that failing to perform that fluency means sliding down the organizational hierarchy regardless of what one actually contributes. Her story is not unusual. It is a preview of a status system being rewritten across knowledge work, and the rewrite is no longer only informal. It has begun to appear, in black and white, in the criteria organizations use to decide who gets in the door and who gets ahead.

The Number That Redefines Value

The clearest evidence that this is a structural change and not a mood comes from PwC's Global Workforce Hopes and Fears survey, which polls tens of thousands of workers and business leaders worldwide each year. In its 2025 findings, two figures stand out. Sixty-six percent of leaders said they would not hire someone who lacks AI skills. Seventy-one percent said they would rather hire a less experienced candidate who has AI skills than a more experienced candidate who does not.

Sit with the second number for a moment, because it is the more radical of the two. For most of the history of professional work, experience was the master signal—the thing that, all else equal, broke the tie. Employers paid for it, promoted on it, and deferred to it precisely because it stood in for a bundle of hard-to-observe qualities: judgment, reliability, pattern recognition earned the slow way. What the 71% figure records is a majority of leaders now willing to invert that ranking outright. Faced with a direct trade-off—seasoned judgment versus tool fluency—they choose the tool fluency. The 66% figure completes the picture: for two-thirds of leaders, the absence of AI skills is not a gap to be trained around but a disqualification.

These numbers matter less as facts about any individual hire than as evidence about the formalization of a new hierarchy. Status anxiety of the kind Claire feels is easy to dismiss as insecurity—the eternal grumbling of the experienced about the young. But when two-thirds of leaders encode "no AI skills, no offer" into their hiring, and seven in ten rank fluency above experience when the two conflict, the anxiety is not a misperception. It is an accurate reading of a valuation system that has already changed. Professional worth is being redefined at the level where such things become consequential: the job description, the shortlist, the promotion committee. The old currency has not become worthless, but it has been formally repriced against a new one, and the exchange rate is set by people who write the offers.

Five Pillars, All Shifting at Once

To see why this repricing is so destabilizing, it helps to name what professional status in knowledge work traditionally rested on. Five dimensions did most of the work: years of experience, domain expertise, institutional knowledge, relationship networks, and a track record of results. AI does not erode these one at a time. It undercuts all five simultaneously, which is part of why the ground feels like it is moving rather than tilting.

Experience loses some of its scarcity value when a model can, in seconds, surface information that a professional spent years accumulating. Domain expertise is relativized when an AI system offers instant, plausible-sounding analysis across fields the expert took a decade to master. Institutional knowledge—the "we tried that in 2019 and here's why it failed" that lived in senior heads—becomes less load-bearing when organizations can archive and query their own history on demand. Relationship networks, long a quiet source of power, are increasingly supplemented and in some roles supplanted by AI-mediated outreach and communication tools. And even a track record is quietly discounted if it was built with methods now seen as pre-AI, as though the results were achieved on the wrong equipment.

None of these erosions is total. A model that retrieves a fact is not the same as a professional who knows which fact matters, and we will return to that gap. But status is comparative and perceptual, and perception moves faster than reality. What each of the five pillars loses is not its actual usefulness so much as its legibility as a status marker. The signals that used to say "defer to this person" have been jammed, all at once, by a technology that mimics their outputs without possessing their substance.

Why the Comparison Cuts So Deep

Humans are relentless comparers. Leon Festinger's 1954 theory of social comparison described the drive as fundamental: in the absence of objective standards, we gauge our abilities and worth by measuring ourselves against others, and we care most about comparisons with people close to us. Social media industrialized this instinct, making comparison constant and public. AI intensifies it again—not by adding another highlight reel, but by making performance differences at work suddenly measurable, immediate, and stark.

What makes the AI comparison uniquely disorienting is the ambiguity of its source. When a colleague generates in an afternoon what used to take you three days, the gap is real and visible. But its cause is unclear. Are they smarter? More creative? Or simply more practiced with a tool you have not yet internalized? Festinger's theory predicts that comparisons are most useful, and least threatening, when we can attribute the gap to something specific—effort, training, talent. AI scrambles that attribution. The gap could be trivial (a prompt technique learned in a weekend) or profound (a capability you lack), and you cannot easily tell which. That uncertainty is what turns an ordinary performance difference into a crisis of self-assessment. Instead of concluding "I need to learn that tool," the anxious mind slides toward "I am the kind of person who is behind." The comparison stops being about a skill and starts being about the self.

This is precisely what distinguishes AI-based status anxiety from the more familiar fear of losing one's job. Fear of job loss is about the future and about the organization's decision; status anxiety is about the present and about one's standing among peers. A worker can be in no danger of layoff and still feel the daily erosion of being outranked on the criterion that now counts. And unlike general workplace anxiety—which attaches to deadlines, workload, difficult managers—AI status anxiety attaches to a comparison whose terms keep changing and whose fairness is impossible to verify. The comparisons also refuse to stay in the office. Workers now measure their fluency against curated "AI-native" professionals on networking platforms, against breathless media portraits of what everyone else can supposedly do, and against the accelerating capabilities of the models themselves. Every new benchmark ratchets the standard upward, so the sense of falling short becomes self-renewing.

Whose Work Was That?

Layered on top of comparison is a second, quieter destabilizer: uncertainty about ownership. The impostor phenomenon—the nagging sense that one's success is undeserved and exposure is imminent—is well documented among high achievers. AI introduces a genuinely new variant, and it is worth being precise about why.

Ordinary impostor feelings involve doubting a competence you in fact possess. The AI version involves a real, structural ambiguity about where your contribution ends and the model's begins. A strategist who drafts a concept with AI assistance cannot always say how much of the good idea was hers. A lawyer who used a model to surface precedents may wonder whether the sharp brief reflects her legal judgment or the system's pattern-matching. This is not irrational self-doubt to be coached away; it is an accurate perception of a blurred boundary. And because professional confidence normally rests on being able to point at your work and say I made that, the blurring strikes at the foundation. The more capable the model, the harder it becomes to claim unambiguous authorship of anything, and the wider the gap grows between the competence others attribute to you and the competence you feel you can honestly own.

The Treadmill That Never Stops

If comparison and ownership describe the psychology, the skill treadmill describes the tempo. Researchers use the term for the specific experience of continuous, accelerating learning pressure that AI imposes—a demand to stay current that is never finished and never can be. This is not the same as ordinary professional development, and the difference is psychological, not just quantitative.

Normal upskilling has a shape: you identify a gap, you close it, you reach a plateau where the new skill becomes stable competence, and you rest there for a while. That plateau is where mastery lives, and mastery is a major source of professional confidence and identity. The skill treadmill removes the plateau. A tool that confers advantage this quarter is a baseline expectation the next; a workflow mastered at real cost is partly obsoleted by a new model release or a platform your employer now considers standard. There is no arrival, only the running. What this denies workers is not merely time but the feeling of ever being good enough—the settled competence around which a stable professional self is built. The resulting anxiety is therefore not episodic but chronic, a background hum of the AI-saturated workplace, and it is qualitatively different from the temporary stretch of learning a defined new system.

The Trust Paradox

Here the story takes an ironic turn that ought to worry organizations more than it does. The prestige attached to AI fluency does not reliably track the quality of AI-assisted work—and can invert it.

Deep domain expertise is exactly what allows someone to catch an AI system's failures: the hallucinated statistic, the confident but context-blind recommendation, the assumption that would never survive contact with how this particular organization actually works. But in a workplace where fluency confers status, exercising that critical judgment becomes socially costly. When an experienced worker questions a slick AI-generated analysis, the challenge is easily read not as expertise doing its job but as defensiveness—the old guard protecting outdated methods. Meanwhile a worker who presents AI output confidently and uncritically can read as forward-thinking, even when the output is wrong. Fluency raises status; skepticism lowers it; and accuracy is decoupled from both.

The perverse consequence is that the status system disincentivizes precisely the domain expertise an organization needs to evaluate AI outputs safely. It rewards enthusiasm over scrutiny at the exact moment scrutiny matters most. This is a real organizational risk, not a hypothetical one: trust is load-bearing in good decision-making—PwC's data links high trust in managers to sharply higher motivation, with workers who trust their direct managers reporting themselves 72% more motivated than those with the lowest trust. Introducing a new axis of credibility that does not align with reliability erodes that foundation. An organization that cannot tell "good with AI" apart from "produces good work with AI" is quietly dismantling its own quality control.

What the Metrics Cannot See

The trust paradox has a measurement cousin. As AI tools embed themselves in workflows, the metrics that evaluate performance shift toward what AI-assisted work makes easy to count—and away from forms of value that were always hard to count and are now, by contrast, nearly invisible.

What AI-oriented metrics tend to capture	What expert judgment tends to provide
Speed and volume of output	Relationship depth and earned client trust
Data-driven analysis at scale	Strategic intuition and contextual judgment
Short-term, measurable performance	Long-term thinking and organizational memory
Process efficiency	Institutional knowledge and cultural nuance

The right-hand column is where much of an experienced worker's value actually lives, and almost none of it lands on a productivity dashboard. Worse, these contributions resist causal attribution. When a relationship keeps a client from walking, or a strategist's instinct kills a campaign before it insults an audience, the expertise that prevented the disaster is rarely credited—because the disaster never happened. The absence of harm is invisible, and invisible value cannot be measured, and unmeasured value looks, on a dashboard, like no value at all. Workers whose primary contribution is tacit therefore face a structural disadvantage in evaluations optimized for quantifiable, AI-assisted throughput. Over time an organization can find itself producing high volumes of fast, data-driven work while quietly starving the judgment needed to tell whether any of it is good.

The Uneven Weight of the Treadmill

The skill treadmill is not experienced equally, and this is where individual anxiety connects to structural inequality. Continuous upskilling costs time, energy, money, and bandwidth—resources that are themselves unevenly distributed. A worker with caregiving responsibilities, a chronic health condition, a second job, or simply no disposable income for courses and no slack in the day for self-directed tinkering runs the same treadmill at a steeper incline. The people with the least capacity to keep pace are frequently the people who most need their existing status to hold.

The result is that a technology often marketed as a great equalizer functions, on this axis, as an amplifier of existing disadvantage. Those with time, money, and cognitive slack pull ahead; those without fall behind, and the gap is then read as a difference in merit rather than in circumstance. The treadmill reinforces the socioeconomic hierarchy it sits on top of rather than flattening it.

Nowhere is this clearer than along generational lines. Younger workers arriving with native fluency carry advantages beyond the technical: they are perceived as adaptable, future-oriented, tech-savvy—high-status attributes in organizations anxious about being left behind. Those perceptions shape which assignments they get, who mentors them, and how fast they rise. Older workers, however deep their expertise, risk being cast as traditional, slow, or resistant if their AI use is not visible and enthusiastic. And here the dynamic becomes self-fulfilling in a genuinely corrosive way: workers labeled as behind receive fewer chances to demonstrate otherwise, the resulting gaps—often caused by less access and less practice, not less ability—are taken as proof of the original label, and the label hardens. Status loss of this kind is well established in the psychological literature as a driver of stress, depression, and diminished commitment. What makes the AI version especially cruel is its invisibility in the official numbers: a worker can be objectively effective and still be losing the informal contest that decides who gets the sponsorship, the stretch project, the promotion.

Whether this generational divide widens or stabilizes over the next five to ten years is genuinely uncertain, and the honest answer is that the current dynamics point toward widening before any equilibrium. Advantages compound: the AI-fluent young get the high-visibility work, which builds more skill and more status, which gets them more of the work. Disadvantages compound in mirror image. Left alone, such feedback loops do not self-correct; they run until something external interrupts them.

Are Better Frameworks Emerging?

There is a version of this story in which organizations notice the problem and fix their instruments—building evaluation frameworks that credit both AI-assisted productivity and the complementary human expertise that makes AI output usable. Is that happening?

The evidence is thin but not absent. Research on human-AI complementarity consistently finds that the best outcomes come from combining machine and human strengths rather than substituting one for the other: models excel at processing structured data at volume, generating options fast, and applying rules consistently, while human experts excel at judgment under ambiguity, contextual interpretation, relationship management, and ethical reasoning. A growing number of organizations articulate this in principle. Far fewer have translated it into how they actually measure and reward people. The obstacle is not conceptual but practical: tacit expertise is, by definition, hard to make legible, and building metrics that surface "the crisis that didn't happen" requires deliberate, expensive effort against the path of least resistance, which is to count what the software already counts for free. Until measuring judgment becomes as cheap as measuring throughput, the default will keep favoring throughput. What prevents better frameworks from becoming standard, in other words, is not that no one wants them but that the easy metrics are already installed and the hard ones must be built by hand.

What Is Owed to Those Left Behind

This raises a question the productivity framing tends to skip: what does an organization owe a worker whose status is declining not because she failed but because the criteria changed underneath her? Claire did not choose to be born fifteen years before generative AI became a workplace standard. She did nothing wrong. Yet she is being structurally disadvantaged by a shift she did not initiate and cannot fully control.

Standard employment ethics has language for this, though it is rarely applied here. The principle that those who impose a change bear some responsibility for its predictable harms is familiar from how we think about plant closures, offshoring, and other structural shifts that harm workers through no fault of their own. When a firm reorganizes and a role becomes obsolete, decent practice—and in many jurisdictions, law—recognizes obligations to retrain, redeploy, or at minimum not to penalize the displaced for a change they did not choose. AI-driven status shifts fit the same moral template even though they rarely trigger the same formal protections, because the harm arrives as gradual devaluation rather than a pink slip. An organization that reaps the efficiency gains of an AI transition while offloading the entire cost of adaptation onto its most experienced workers is engaged in a quiet transfer of risk from the institution to the individual—and the individual, in this case, is the one with the least capacity to run the treadmill.

There is also a sharper, legally testable question hiding inside the 66% figure. Should an organization be required to show that its AI-fluency requirements are genuinely tied to job performance, rather than functioning as proxies that systematically screen out experienced—which often means older—workers? The frameworks to ask this already exist. Under U.S. anti-discrimination law, a hiring criterion that disproportionately excludes a protected group (age, under the Age Discrimination in Employment Act) can be challenged as disparate impact, and the employer must then defend it as job-related and consistent with business necessity. Comparable equal-treatment principles operate in the EU and elsewhere. The U.S. Equal Employment Opportunity Commission has issued guidance warning that AI-based selection tools can produce exactly this kind of unlawful adverse impact. On paper, then, a blanket "no AI skills, no hire" rule that happens to filter out workers over fifty is already legally vulnerable if the skill is not demonstrably necessary for the job.

Whether those frameworks are adequate in practice is another matter, and here honesty requires admitting they are strained. Disparate-impact cases are hard to bring, require the plaintiff to surface statistical patterns an individual rarely has access to, and move far slower than the technology. Proving that "AI fluency" served as a pretextual proxy rather than a real requirement is genuinely difficult when the tools change every quarter. The doctrine exists; its enforcement machinery was not built for a criterion this fast-moving and this easy to dress up as necessity.

This is where the policy stakes come into focus. If AI fluency hardens into a prerequisite for advancement—one unevenly distributed across generations and income levels—it risks entrenching existing disparities rather than dissolving them into the fluid meritocracy technology optimists promise. Three responses follow. Training access has to be treated as a matter of equity, not left to whoever already has time and money, or the treadmill will keep sorting people by circumstance. Workplace-equity enforcement needs sharper, faster tools for testing whether AI requirements are genuine or exclusionary. And technology-adoption governance should recognize that how quickly and how bluntly an organization reprices human value is itself a choice with distributional consequences, not a neutral fact about the technology.

How Much of This Do We Actually Know?

A book committed to honesty about uncertainty has to draw a firm line here between what the evidence supports and what remains conjecture, because the temptation to overclaim in this territory is strong.

Start with the psychological harms. Several 2025 studies do find that AI-related job anxiety negatively predicts life satisfaction, with the effect substantially mediated by negative emotions, and workplace mental-health surveys report striking figures—on the order of 90% of U.S. workers reporting at least one mental-health challenge and around half reporting moderate-to-severe burnout, depression, or anxiety. But the quality of this evidence must be stated plainly. Much of it is cross-sectional and self-reported: it captures correlation at a single moment, not causation over time. Disentangling AI-specific status anxiety from the general economic uncertainty of the mid-2020s—inflation, layoffs, restructuring—is genuinely hard, and few studies have done it cleanly. The mediation findings are suggestive and theoretically coherent, but they are not the same as longitudinal proof that AI status pressure causes burnout and depression as opposed to travelling alongside a broader anxiety. The direction and mechanism are plausible and consistent with well-established social-comparison theory; the causal claim is not yet nailed down. We are, in the book's terms, mostly in the register of expected here, not known.

The same caution applies to the trust paradox and organizational performance. The claim that prioritizing AI fluency over domain expertise produces worse outcomes is theoretically well-motivated—it follows directly from the complementarity research and from what we know about the value of critical scrutiny—but direct empirical demonstration is scarce. What would settle it are studies tracking teams over time, comparing those that reward fluency against those that preserve expert oversight, and measuring downstream error rates, costly mistakes, and decision quality. Those studies are only beginning to be done. Until they exist, the trust paradox is a well-reasoned hypothesis with strong theoretical support, not an established empirical law, and it should be read that way.

Finally, the largest unknown: whether this whole status transition is a permanent feature or a transitional spasm. History offers a genuine reason for hope alongside the anxiety. Earlier general-purpose technologies—electricity, the personal computer, the spreadsheet, the internet—each produced a phase in which fluency was a scarce, status-conferring differentiator, followed by a phase in which the skill became an unremarkable baseline that no longer sorted people. Spreadsheet proficiency was once a résumé line; now it is assumed and invisible. If AI follows that pattern, today's fluency premium erodes as the tools become ambient and as the cohorts entering the workforce all arrive fluent, at which point domain expertise reasserts its value on top of a fluency floor everyone shares. But two things make the analogy uncertain. The timeline is compressed—earlier transitions unfolded over decades that let institutions and workers adapt, whereas AI is repricing skills in quarters, which may deny the workforce the time that normally lets a new equilibrium form gently. And the technology itself keeps moving, so the "baseline" may never hold still long enough to become a floor. Whether AI fluency settles into an assumed baseline within, say, five to ten years, or whether the treadmill's acceleration prevents any equilibrium from forming at all, is not something current evidence can decide. It is one of the real open questions, and pretending otherwise would be its own kind of dishonesty.

Summary

The valuation has been formally rewritten. PwC's 2025 data—66% of leaders would not hire someone lacking AI skills, and 71% would take a less experienced candidate with AI skills over a more experienced one without—shows that professional worth is being repriced not just in perception but in hiring and promotion criteria. Experience, long the tie-breaking master signal, has been formally ranked below tool fluency by a majority of leaders.
AI-based status anxiety is distinct. It is about present standing among peers, not future job loss, and it fastens onto a comparison whose terms keep shifting and whose fairness cannot be verified.
All five status pillars shift at once. Experience, domain expertise, institutional knowledge, relationship networks, and track record are simultaneously relativized—less in real usefulness than in their legibility as status markers.
Comparison and ownership destabilize the self. Festinger's social comparison theory explains why gaps of ambiguous source (tool fluency versus innate ability) are so corrosive, and AI-assisted work blurs the boundary of authorship that professional confidence normally rests on.
The skill treadmill removes the plateau. Unlike ordinary upskilling, it offers no stable mastery to rest on, making the anxiety chronic—and it falls hardest on workers with the least time, money, and slack, amplifying existing inequality.
The trust paradox inverts quality and status. Fluency can raise standing even when the output is worse, discouraging exactly the expertise needed to catch AI errors—a real organizational risk, not a hypothetical one.
Metrics see throughput, not judgment. Tacit expertise, relationship depth, and prevented disasters do not appear on dashboards, structurally disadvantaging the workers who provide them.
The generational divide is widening, not stabilizing—for now. Advantages and disadvantages compound through self-fulfilling perceptions, and such feedback loops do not self-correct without intervention.
Obligations and legal tools exist but are strained. Employment ethics recognizes duties to those harmed by changes they did not choose; disparate-impact and age-discrimination law can in principle challenge AI-fluency requirements that are not job-related—but enforcement is slow and ill-suited to a criterion this fast-moving.
The evidence is partial and honest labeling matters. The link between AI status anxiety and burnout or depression is correlational and hard to separate from general economic uncertainty; the trust paradox is well-reasoned but not yet empirically proven; and whether AI fluency settles into an invisible baseline—as spreadsheet skills did—or keeps accelerating past any equilibrium remains a genuine open question, made harder by a timeline measured in quarters rather than decades.

Sources

Last updated: 2026-09-03

V2 (in progress) Previous: V1