AI’s Effect on Workers Shows Up in Wages, Not Jobs

Artificial intelligence is not cutting American jobs at scale — at least not yet. According to The New York Times, the technology’s earliest measurable effects on the workforce are surfacing in wage data instead, as employers appear to be absorbing AI’s productivity gains by slowing pay growth rather than eliminating positions. The full picture, however, is more complicated: the pay compression is real, unequally distributed, and unfolding against a labor market that still looks remarkably healthy on the surface.

The Central Finding: Slower Wage Growth, Steady Employment

The evidence traces largely to a July 30 white paper by Apollo Global Management chief economist Torsten Slok and economist Sania Edlich, which tracked wage and employment outcomes across 321 occupations from 2015 to the present. As Apollo summarizes, workers in occupations highly exposed to AI have seen real wage growth roughly 6.7 percentage points slower than comparable low-exposure workers since 2023 — the year ChatGPT went mainstream — while employment levels in those same occupations remained statistically unchanged.

The paper’s title captures the paradox bluntly: “AI Lowers Wages But Doesn’t Cut Jobs.” Its conclusion is that companies are capturing AI’s productivity benefits through wage compression rather than workforce reduction.

“And the wage that you’re getting paid is where we’re seeing all the action,” Edlich told The New York Times.

Who Feels It Most

The pain is not evenly shared. According to Business Insider’s analysis, the wage effect is harshest at the bottom of the pay scale. Workers in the lowest wage quartile saw real earnings growth fall 10.7% since 2023; the second quartile declined 5.4%, and the third 4.0%. The highest-paid quartile showed no statistically significant effect at all.

The authors flagged one especially dramatic figure — a 24.3% decline for service workers — as resting on a small subsample that should be read with caution.

Specific occupations with high AI exposure and notable wage erosion between 2022 and 2024 include computer programmers (−6.1%), statistical assistants (−5.4%), software quality assurance analysts and testers (−2.9%), and database architects (−2.7%), per the same reporting.

How the Study Was Done — and Its Limits

Rather than measuring how much AI could theoretically disrupt a job, the Apollo team used the Anthropic Economic Index — which captures actual AI usage through real model interactions — and matched it to Bureau of Labor Statistics wage and employment data. That methodological choice is a key reason the findings carry weight.

But the authors are candid about the caveats. As Dataconomy reports, exposure measurement relies on data from a single company, only a fraction of occupations were included, and shifts in how jobs and industries are classified over time complicate any long-run comparison.

A Hiring Squeeze Beneath the Headlines

Wage data is not the only signal. Revelio Labs’ AI Labor Market Tracker, launched July 28, found that jobs most exposed to AI have grown roughly 4% slower than the least-exposed roles since ChatGPT’s release, with a steeper drop for younger workers. At companies actively adopting AI, headcount grew 27% more overall — but senior roles expanded 31% while junior roles grew just 6%.

That entry-level squeeze is one of the most consequential threads. Slower hiring for junior workers threatens to erode the traditional apprenticeship pipeline through which early-career employees once built expertise.

The Counterpoint: A Two-Track Market

Not everyone is losing ground. PwC’s 2026 Global AI Jobs Barometer, which analyzed more than one billion job ads across 27 countries and territories, describes a “two-track” labor market. Workers with AI skills now command an average wage premium of 62%, up from 57% a year earlier. “Professionalised” roles — where AI amplifies expert judgment — are growing twice as fast and paying 42% faster salary growth than roles that AI simply makes easier to perform.

In other words, AI is compressing pay for some workers while simultaneously bidding up the value of others. The result is not a uniform story of decline, but a widening divergence.

Why Workers Are Anxious Anyway

Public sentiment is running ahead of the employment data. A Gallup survey fielded August 3–24 found that 27% of U.S. workers — a record high — now worry that technology could make their jobs obsolete. That is up seven percentage points in a year and roughly double the 13% recorded when Gallup first asked in 2017. Among workers under 45, technological displacement now ties benefit cuts as the top job concern.

The irony is that AI’s overall effect on employment remains modest. The latest jobs report showed solid hiring and a steady 4.1% unemployment rate. What is shifting is not whether people have jobs, but what those jobs pay — and whether the next generation gets a foothold at all.

Some analysts also suspect the disruption is partly hidden. As Dataconomy notes, former IBM chief human resources officer Diane Gherson has argued that companies may quietly reduce hiring in high-attrition, lower-wage roles without announcing layoffs, and that severance costs recorded as one-time restructuring can make workforce reductions appear less severe than they are.

The Bigger Picture

The wage-compression findings land within a longer trend: U.S. workers now take home their smallest share of economic output since the Bureau of Labor Statistics began tracking it in 1947. If AI is accelerating that shift, the policy stakes are considerable.

Apollo’s authors frame the challenge plainly. The critical question, they argue, is not whether AI will reshape the labor market — it will — but how, and specifically how to support lower-wage workers who have the fewest resources to weather the transition. Slok has suggested the economy is becoming more dynamic rather than contracting, pointing to record business formation, while conceding that the productivity benefits remain unproven outside the largest technology firms.

What to Watch

Three signals will determine whether today’s quiet adjustment becomes tomorrow’s crisis. First, whether the wage gap between AI-exposed and insulated occupations continues to widen or stabilizes as adoption matures. Second, whether entry-level hiring recovers — the junior-role slowdown is the clearest early warning of a pipeline problem. And third, whether the productivity gains AI promises ever show up broadly in corporate margins, or remain concentrated among a handful of “super-star” firms.

For now, the data tells a subtler story than the loudest voices on either side suggest. AI’s first measurable mark on the American workforce may not be a pink slip. It may be a raise that never arrives. As one market analysis put it, a worker who would have gotten 4% and instead gets 2% still has a job — and still feels poorer every year.