AI Is Eating Your Job. Here's the Data.

The numbers are more complicated than the headlines suggest. They are not, despite what some CEOs would have you believe, a clean story of robots replacing humans wholesale. They are something more structurally specific — and, in some ways, more troubling. The labor market is being reshaped by artificial intelligence, and the people bearing the sharpest edge of that reshaping are the youngest workers in the economy. But understanding what the data actually shows requires precision about what we are and are not measuring.

Let us begin with the most dramatic data point currently available.

A note on methodology: The employment and wage statistics in this article draw primarily from a February 24, 2026 analysis by Federal Reserve Bank of Dallas economist Tyler Atkinson. "AI exposure" in that research is measured using an index developed by Felten, Raj, and Seamans (2021), which scores occupations based on the degree to which AI capabilities overlap with the tasks those occupations require — think pattern recognition, language processing, and codified knowledge retrieval. The index is applied to 205 occupations and to industry-level employment data. "Under age 25" refers to the standard Bureau of Labor Statistics youth employment cohort. All wage figures cited are nominal (not inflation-adjusted) and cover the period from fall 2022 — the approximate release date of ChatGPT — through early 2026; this is the measurement window for every wage and employment statistic in this piece unless otherwise specified. Where the word "impacted" appears in IMF projections, it encompasses jobs that may be enhanced, made more productive, or replaced — not solely eliminated. These distinctions matter, and I will flag them as they arise.

One further caution: the Dallas Fed data, the IMF projections, and the Block layoff figures are not drawn from the same methodology or the same universe of workers. The Dallas Fed uses the Felten index applied to BLS occupation and industry data. The IMF uses its own macroeconomic modeling with a different definition of "impact." Block is a single corporate event. Treating these three as a unified, mutually reinforcing dataset would be a mistake. They are compatible signals pointing in a similar direction — not a single coherent proof.

Block's Bombshell

In February 2026, Jack Dorsey's payments company Block announced it would cut between 40 and 50 percent of its workforce — more than 4,000 jobs — and attributed the decision explicitly to AI adoption. Block Cuts 40% of Its Work Force Because of Its Embrace of A.I. This is one of the more prominent fintech firms in the world cutting nearly half of its people and saying, plainly, that AI made them unnecessary.

What made the announcement more significant was what Dorsey said next. He warned that most companies would follow Block's lead. Not some companies. Most. That is a statement worth sitting with — though it is equally worth noting that it is a statement, not evidence. CEO declarations about AI causation are, as we will get to, not always what they appear.

Block was not alone in early 2026. Over 30,000 tech jobs were cut globally in just the first two months of the year. Over 30,000 tech jobs cut globally in first two months of 2026: Report The layoffs are accelerating, not plateauing. But acceleration in headline numbers does not, by itself, establish AI as the cause of any particular cut.

Block is useful as a case study precisely because it is so explicit: the company named AI, the CEO named AI, and the scale is large enough to be significant. But a single high-profile event — even a dramatic one — is not a representative sample. Selection bias is real. The companies that make headlines when they cut workers and cite AI are not necessarily representative of the broader labor market. They are the loudest signal, not necessarily the clearest one.

Who Is Actually Getting Cut

The Federal Reserve Bank of Dallas published research on February 24, 2026 that clarifies something important: AI is not simply eliminating jobs across the board. ¹ It is doing something more structurally specific.

Since ChatGPT's release in fall 2022 through early 2026, total U.S. employment has grown approximately 2.5 percent. But in the computer systems design sector — one of the industries scoring highest on the Felten-Raj-Seamans AI exposure index — employment has fallen 5 percent over the same period. Across the top decile of AI-exposed sectors by that same index, employment has declined 1 percent. These figures come from BLS industry-level employment data analyzed by Dallas Fed economist Tyler Atkinson; the "top 10 percent of AI-exposed sectors" refers to the top decile of industries ranked by their Felten index score.

The decline is not falling evenly across age groups. Stanford University researchers Erik Brynjolfsson, Bharat Chandar, and Ruya Chen find that the employment decline in AI-exposed sectors is particularly pronounced for workers in the BLS under-age-25 cohort. Employment totals for older workers have not declined over the same period. The mechanism Atkinson identifies is not mass layoffs of existing employees. It is a collapse in the job-finding rate for young workers entering the labor force. New graduates are simply not getting hired. The door is closing before they get inside.

The theoretical explanation the research offers is precise: AI can replicate codified knowledge — the kind learned from textbooks, the kind entry-level workers are primarily hired to apply. It cannot yet replicate tacit knowledge, the understanding built from years of actual experience. An experienced software engineer or credit analyst carries something in their head that an AI system cannot easily extract or reproduce. A new graduate, largely, does not.

This is a structural hypothesis, not a proven causal chain. The correlation between AI exposure (as measured by the Felten index) and youth employment decline is real and statistically significant. Whether AI is the cause of that decline, or whether AI-exposed sectors are contracting for cyclical reasons that happen to coincide with the post-ChatGPT period, is a harder question. The Dallas Fed analysis is careful on this point; I will try to be equally careful.

The Wage Paradox

Here is where the data becomes genuinely interesting, and where lazy narratives about AI and jobs start to break down.

Even as employment falls in AI-exposed sectors, wages in those same sectors are rising — and rising faster than the national average. From fall 2022 through early 2026, nominal average weekly wages have grown 7.5 percent nationally. In computer systems design, they have grown 16.7 percent in nominal terms. Across the top decile of AI-exposed industries, nominal wages are up 8.5 percent. ¹

Two important caveats here. First, these are nominal figures. Over roughly the same period — fall 2022 through early 2026 — cumulative U.S. inflation ran approximately 8 to 10 percent, which compresses the real wage advantage considerably. On a rough real-terms basis, computer systems design's 16.7 percent nominal gain translates to something in the range of 6 to 9 percent real wage growth, still meaningfully above the national average — but the gap is narrower than the nominal figures suggest. Second, and more subtly, some of this wage growth may be a composition effect rather than a genuine wage increase for any individual worker: if lower-paid junior roles are eliminated while higher-paid senior roles persist, average wages rise mechanically, even if no one got a raise. The Dallas Fed analysis acknowledges this ambiguity explicitly. The wage data is suggestive, not conclusive.

What the data does show clearly is this: if AI were simply automating jobs wholesale, we would expect wages to fall alongside employment. Instead, nominal wages are rising in the most AI-exposed sectors. That pattern is more consistent with a story in which AI augments experienced workers — making them more productive, and allowing employers to partially share those productivity gains — than with a story of pure automation.

The result is a bifurcating labor market: experienced workers in AI-exposed fields appear to be doing better than average on wages; young workers trying to enter those same fields are doing worse on employment. The bottom is being sealed off.

The Executives Are Saying the Quiet Part Out Loud

This structural shift is being stated, directly, by the people running major corporations — though those statements require careful handling.

CEOs from Ford, Amazon, Salesforce, and JP Morgan Chase have all publicly proclaimed that white-collar jobs at their companies will soon disappear due to AI. Companies Are Laying Off Workers Because of AI's Potential—Not Its Performance That Harvard Business Review analysis surfaces something important: many companies are cutting workers based on AI's potential, not its demonstrated performance. The technology has not yet delivered the productivity gains being used to justify the headcount reductions. In a meaningful number of cases, the layoffs are a bet on a future that has not arrived.

This matters for how we interpret CEO declarations. When Jack Dorsey says AI drove Block's layoffs, that is a corporate framing, not an independent audit. When other executives make similar claims, those statements tell us something about how companies are presenting their restructuring decisions — and they may reflect genuine strategic choices — but they do not constitute proof that AI is the operative cause of any specific job loss. Corporate framing and economic causation are different things.

Sam Altman, CEO of OpenAI, has acknowledged this gap directly, confirming that some companies are engaging in what he called "AI washing" — blaming layoffs on AI when the actual drivers are unrelated. Oxford Economics has gone further, suggesting that AI layoffs may be, in some instances, "corporate fiction masking a darker reality" — restructuring decisions dressed up in the language of technological inevitability.

The right posture, then, is this: treat the Block announcement and the broader 30,000-job figure as meaningful signals about the direction of corporate behavior. Treat CEO declarations as data about corporate framing. Do not treat either as proof of AI causation. Cost-cutting cycles, post-pandemic corrections, shifting business models, and investor pressure on margins are all plausible contributing factors. AI is a genuine force in the labor market, but it is also, at this moment, a convenient narrative. The Dallas Fed's quantitative findings — grounded in BLS data and a systematic exposure index — are the stronger evidentiary foundation. The anecdotes are context, not proof.

The Tsunami Warning

The IMF is not known for hyperbole. Its managing director, Kristalina Georgieva, described AI as "a tsunami hitting the labor market," estimating that 40 percent of jobs globally — and 60 percent in advanced economies — will be "impacted" in coming years.

That word — "impacted" — requires unpacking, and the IMF's own framing makes the unpacking explicit. The 40 and 60 percent figures encompass jobs that will be enhanced or made more productive by AI, not just jobs that will be eliminated. Georgieva was direct about this: "The total impact on employment, surprisingly, in areas we study, is positive. But it is positive because of the increase of low-paying jobs." The 40 and 60 percent figures are projections derived from IMF macroeconomic modeling — not measurements of current conditions — and they carry significant uncertainty about time horizon, the definition of "impact," and the assumptions underlying the models. The IMF has not specified a precise year by which this scale of disruption is expected to materialize. These figures are best understood as a range of plausible disruption scenarios, not a forecast of mass unemployment.

What Georgieva did flag specifically — and this aligns closely with the Dallas Fed data — is the risk to young workers: "We also see automation eliminating entry-level jobs. So recent graduates, they worry, where is the job for me?" That convergence between the IMF's qualitative warning and the Dallas Fed's quantitative finding — both pointing to entry-level workers as the most exposed cohort — is worth noting, even though the two analyses use different methodologies and cannot be directly compared.

Alap Shah, CIO at Lotus Technology Management and co-author of the Citrini Research report that triggered an "AI scare trade" selloff in equity markets in late February 2026, has proposed that governments consider taxing AI windfall gains as a stabilization mechanism. The logic is macroeconomic: if AI-driven productivity gains accrue primarily to corporations while displaced workers lose income, consumer spending weakens, and that weakening feeds back into corporate revenues. Shah outlined a near-term risk scenario in which as much as 5 percent of white-collar workers could be cut within 18 months — a projection, not a measurement, and one that represents a risk case rather than a base case. The sectoral divide he described is sharp: chipmakers, data centers, and foundation model labs are likely beneficiaries of the AI buildout; insurers, banks, consumer discretionary platforms, and intermediation-heavy businesses face elevated disruption risk.

What the Data Actually Says

Strip away the executive pronouncements, the market volatility, and the competing narratives, and the data tells a coherent — if incomplete — story.

AI's labor market effects are real. They are, right now, most clearly visible among workers under 25 in the sectors scoring highest on established AI exposure indices, who are finding it increasingly difficult to get hired. They are also producing nominal wage gains for experienced workers in those same sectors — gains that, even after a rough inflation adjustment and after acknowledging the possibility of composition effects, appear to outpace national averages in at least some high-exposure sectors.

The experience premium — the wage gap between entry-level and experienced workers — is positively correlated with AI exposure across the 205 occupations in the Dallas Fed analysis. The more exposed an occupation is to AI, the higher that premium tends to be. AI appears to be increasing the returns to experience at the precise moment it is reducing the opportunities for young workers to gain that experience. That is a structural trap, and it is not a small problem.

What remains genuinely uncertain is how much of the current disruption is AI-driven versus cyclically-driven, how quickly the technology's capabilities will expand beyond codified knowledge into tacit knowledge domains, and whether new categories of work will emerge to absorb the entry-level workers currently being squeezed out. The data available as of late February 2026 cannot resolve those questions.

Whether this resolves through new categories of work, through policy intervention, or through something no one has yet named clearly — that remains unknown. What is known is that the labor market of 2030 will not look like the labor market of 2022. The data on that point is consistent. Whether it is unambiguous depends on how carefully you read it. ¹