AI wage compression: where your productivity gain lands

The AI jobs debate has been stuck on the wrong question for two years. Everyone argues about how many jobs get eliminated. New research from Apollo Global Management says the measurable effect so far isn't headcount at all — it's pay. Jobs most exposed to AI saw real wage growth fall by an average of 6.7% after 2023, while the employment effect was, in the researchers' word, not detectable. That's AI wage compression, and it's a very different management problem than layoffs.

What actually happened

Apollo chief economist Torsten Sløk and analyst Sania Edlich tracked wage and employment data across 321 occupations using Bureau of Labor Statistics figures from 2022 to 2024, and measured each occupation's AI exposure using Anthropic's Economic Index — which infers exposure from the tasks people actually bring to AI tools, not from a theoretical list of what a model could do. Roughly 5.8 million US workers sit in the high-exposure bucket.

The distribution is the finding. Per Axios and coverage of the paper, the top of the income ladder showed no significant effect, while service occupations saw earnings growth fall by roughly 24% since 2023. The conclusion the authors draw is blunt: companies are capturing AI productivity gains through wage compression rather than workforce reduction. Two caveats worth stating plainly — this is BLS data through 2024, so it's a rear-view measurement of the early ChatGPT era, and correlation across occupations isn't proof of mechanism.

Why AI wage compression matters for your business

If you employ ten people instead of ten thousand, this isn't an abstraction — it's a decision you're going to make without noticing you made it. You automate the quoting workflow. Your ops person's job gets 30% easier. Then next spring, when raises come up, the productivity gain quietly becomes margin because nobody explicitly decided otherwise.

That default has a cost at your scale that it doesn't have at Amazon's. The person who now runs three times the throughput knows exactly what changed, has a market for that skill, and will leave for a shop that priced it. Small businesses don't survive that trade the way a Fortune 500 absorbs it.

So decide on purpose. Measure what the automation actually returned — hours per week, error rate, jobs quoted — before anyone's compensation conversation, not after. Name where the gain goes: margin, more volume with the same team, or pay. Any of the three is defensible; drifting into one by accident is not. And watch the roles you're quietly deskilling. If a job becomes "review what the model produced," you've changed what you're paying for, and that shows up in retention before it shows up in a report.

The macro story is that AI is showing up in paychecks rather than pink slips. The operator story is that the gain lands somewhere by default, and default is a choice you didn't make.

Sources: Axios, Fortune via Yahoo Finance.