Since the widespread emergence of generative language models, public discourse has been dominated by warnings of rapid mass unemployment. An empirical investigation conducted by Apollo Global Management, led by Chief Economist Torsten Slok and Sania Edlich, now provides a much more nuanced view of market realities. Instead of abrupt workforce reductions, exposed professions are currently experiencing a noticeable deceleration in real wage growth.
The research team examined a sample of 321 standardized occupations across the United States. Rather than relying on theoretical substitution matrices, the authors leveraged telemetry data from the Anthropic Economic Index, which measures real user interactions with the Claude foundation model. This method allowed the economists to capture the actual workplace adoption of AI systems and evaluate it against official compensation trends.
The findings contradict claims of near-term employment collapse in highly exposed fields, revealing instead a substantial dampening of income gains. In occupations with heavy day-to-day AI utilization, real wage growth post-2023 expanded 6.7 percentage points slower compared to low-exposure occupations. While labor demand appears broadly stable, the bargaining power of workers during compensation reviews is being curbed by the availability of automated tools.
This wage compression disproportionately affects workers in lower earnings brackets and service roles. In the lowest income quartile, real wage expansion lagged behind peer groups by 10.7 percentage points. Service occupations experienced an even steeper slowdown, recording a 24.3 percentage point shortfall in wage growth. Specialized white-collar and management positions were not entirely insulated either, enduring a 4.1 percentage point growth deceleration.
From a macroeconomic perspective, the study quantifies the aggregate financial consequences across the American labor pool. Approximately 5.8 million workers in the United States are currently affected by this deceleration in compensation growth. Cumulatively, this divergence represents an estimated 28 billion dollars in unrealized wage expansion that did not reach household balance sheets.
These empirical observations indicate that generative software continues to augment individual subtasks rather than eliminating complete occupational categories. While manual labor and in-person roles remain largely shielded from these trends, knowledge workers face growing economic adjustment pressures. Future labor policy debates will therefore need to concentrate far more on distribution dynamics and wage trajectory compression rather than headcounts alone.