AI’s biggest labor-market impact so far isn’t unemployment. It’s smaller paychecks. That’s the finding from a new Apollo Global Management white paper, covered by Hacker News, which tracked wage and employment data across 321 US occupations to measure what AI adoption has actually done since ChatGPT went viral in late 2023.
The headline number: jobs with the highest exposure to AI saw an average 6.7% decline in real wage growth after 2023. Meanwhile, the effect on overall employment was not even “detectable,” according to analyst Sania Edlich and Apollo chief economist Torsten Sløk. So the machines aren’t firing people yet. They’re quietly capping how much those people earn.
What stands out here is who’s absorbing the hit.
The pay squeeze lands on the bottom
Apollo’s data shows AI cutting deepest into the lowest earners, not the highest:
- Service workers: average 24.3% decline in earnings growth since 2023
- Bottom 25% of earners: wages down 10.7% over the same period
- Highest-paid workers: no “significant effect” observed
That pattern matters because it points the other way from a lot of AI hype. The fear was that white-collar knowledge workers would get automated first. Instead, the early wage pressure is showing up among lower-paid roles, with real implications for income inequality.
How they measured it
The researchers paired occupational and wage data from the Bureau of Labor Statistics with Anthropic’s Economic Index, which estimates how exposed a job is to AI based on the share of its tasks people have actually performed using Anthropic’s tools. Comparing BLS figures from 2022 to 2024, they lined up wage changes against that exposure score.
A few results cut against the trend, and the authors were honest about it:
- Radio DJs and broadcast announcers saw real wages crater 52%, despite low AI exposure. That looks like broader industry decline, not automation.
- Personal finance advisors, with more than a third of tasks exposed to AI, saw wages grow 8.4%.
- Administrative law judges and hearing officers, around 30% task exposure, saw pay jump 17.5%.
So exposure alone doesn’t decide your paycheck. Industry health, skill demand, and how AI slots into a role all bend the outcome.
Why this matters for you
About 5.8 million US workers hold roles that are highly exposed to AI, Apollo estimates, and the authors expect that figure to “grow substantially” as adoption spreads across corporate America.
Here’s the practical read. If your job is exposed to AI, the near-term risk isn’t a pink slip. It’s stalled wage growth while your employer captures the productivity AI adds. UPenn economist Ioana Marinescu has noted that wages tend to take a hit once roles hit around 37% of intelligence tasks being automated. That’s a useful threshold to watch for your own field.
What you can do with that:
- Track your task mix. Figure out what share of your daily work an AI tool could plausibly do today. The closer to that ~37% line, the more wage pressure to expect.
- Move up the value chain. Roles like the finance advisors and judges above held their wages by owning judgment, trust, and accountability that AI can’t hand off.
- Treat AI as leverage, not a threat. Workers who use it to do more, faster, are better positioned to argue for the raise the average worker is losing.
The history isn’t gentle either. A Goldman Sachs analysis cited in the report found workers displaced from “technology-disrupted occupations” took an average 3% real pay cut when they found new work, and earned about 10 percentage points less over the following decade.
The limits
Apollo flagged real caveats. BLS job classifications shift over time, industries get recategorized, and collection methods change, all of which muddy a two-year comparison. This is correlation across occupations, not proof that AI directly caused each wage move. The radio DJ number is a clean reminder that other forces are in play.
Still, the direction is getting harder to wave off. The jobs-apocalypse framing may have been the wrong thing to fear. The quieter story, a slow erosion of raises concentrated among workers who can least afford it, is the one the data now supports. Watch how the next round of BLS figures lands. You can read the full breakdown at the original source.