When AI Saves the Company Time, Should Workers Get a Raise?
Apollo researchers matched U.S. pay and employment data to Claude-use scores across 321 occupations. In the 11 jobs above their high-exposure cutoff, real wage growth was 6.7 percentage points slower after 2023 than in less-exposed work. They did not find a statistically significant employment effect. That is a warning, not a verdict. The wage estimate shrinks to 1.89% with a broader cutoff, comes from one AI provider’s usage, and does not track which companies actually adopted AI. Still, it points at the labor story that is easiest to miss. Layoffs make headlines. A missing raise just arrives as the same paycheck. If a company reports that AI saved thousands of hours, what should employees reasonably expect to see: higher pay, a shorter week, protected headcount—or simply a larger queue?
Comments
Start with take-home pay, not the number of hours on the schedule. A four-day week is not a benefit if an hourly worker loses Friday’s wages. And a raise is thin comfort if the same shift quietly absorbs more customers. Put one rule in writing before rollout: verified savings cannot leave the worker with less money or a bigger queue for the same paycheck. Then ask the people doing the job whether they want the gain as paid time, pay, or more hands.
One distinction before “verified savings”: Apollo did not measure hours saved at any company. It grouped occupations by a Claude-use score, then compared wage trends after 2023. The 6.7-point gap covers 11 occupations above the paper’s chosen cutoff; at 0.4 it falls to 1.89%, and at 0.6 it is not statistically significant. So the paper is a reason to ask the pay question, not proof that a particular employer owes 6.7%. For a rollout, keep the local arithmetic simple: paid hours, take-home pay, queue size and output before and after. If management says AI saved time, workers should see those numbers before anyone calls it a gain.
One more trap: freeze the baseline before anyone knows which version makes the rollout look cheap. If “before” was an understaffed month, AI can look productive by comparison with a mess workers were already absorbing. Record volume, staffing, errors, cleanup and after-hours time; let workers challenge the comparison. Then argue about where the gain goes. Otherwise “verified savings” means finance picked the counterfactual.
Ask before the rollout, not after the savings deck. Once the queue has grown, ‘pay, time or more hands?’ is a choice about how to repair work people are already doing. Put the benefit and the baseline in writing before the pilot. Otherwise every saved hour gets counted for the company and argued away at payroll.