The Federal Reserve Bank of Dallas has published data showing that artificial intelligence is simultaneously raising wages for experienced workers and suppressing them for entry-level employees, a split that challenges both optimistic and pessimistic narratives about AI’s labor market impact.
In a research note published February 24, Dallas Fed economist Scott Davis examines employment and wage trends across 205 occupations since ChatGPT’s release in fall 2022. The headline numbers tell a contradictory story. Employment in the computer systems design sector has declined 5 percent since late 2022, and among the 10 percent of sectors most exposed to AI, total employment fell 1 percent. Yet wages in those same AI-exposed sectors grew 8.5 percent, outpacing the national average of 7.5 percent. Computer systems design wages rose 16.7 percent.
The puzzle is how employment can fall while wages rise. Standard labor economics says automation should push both down. Davis argues the answer lies in the distinction between codified knowledge and tacit knowledge.
Codified knowledge is what you learn from textbooks. Tacit knowledge is what you gain from experience. AI systems can replicate the first but not the second. The implication is that AI substitutes for entry-level workers whose jobs consist largely of codifiable tasks, while it complements experienced workers whose value comes from tacit understanding.
The data supports this. Davis uses Bureau of Labor Statistics wage estimates to calculate an “experience premium” for each occupation, the percentage difference between experienced and entry-level wages. Occupations most exposed to AI tend to have higher experience premiums. The median experience premium across all occupations is 40 percent, but it ranges from under 10 percent for fast-food cooks and ticket agents to over 100 percent for lawyers, insurance underwriters, credit analysts, and marketing specialists.
When Davis regresses post-2022 wage growth against AI exposure and experience premium, the interaction term reveals the divergence. For occupations with a zero percent experience premium, increased AI exposure is associated with a 0.28 percentage point reduction in wage growth. These are jobs where both entry-level and experienced workers are easily substituted by AI. For occupations in the 90th percentile of experience premium, increased AI exposure is associated with a 0.2 percentage point increase in wage growth. AI substitutes for entry-level workers in these fields but complements experienced ones.
The employment data shows the same pattern concentrated on the young. Davis cites research by Stanford’s Erik Brynjolfsson, Bharat Chandar, and Ruya Chen showing that the employment decline in AI-exposed sectors is particularly pronounced for workers under 25. Employment totals for older workers have not declined. A companion piece by Dallas Fed economist Tyler Atkinson argues this is not driven by layoffs but by a low job-finding rate for young workers entering the labor force.
This is not a story about mass unemployment. It is a story about a broken entry-level job ladder in AI-exposed fields. The traditional model of white-collar career progression involves taking an entry-level job out of school, doing codifiable tasks, and slowly learning tacit knowledge on the job. Davis argues that AI is making this method of employee development cost-ineffective, at least in the short run.
The research draws on a framework developed by economist David Autor and coauthors, who use natural language processing to connect U.S. patent applications to occupation descriptions. Patents for augmenting products increase labor demand. Patents for automating products reduce it. Autor and Neil Thompson model jobs as bundles of tasks, where the same AI innovation might automate the expert components of one job while automating the routine components of another, thereby enhancing the value of worker expertise by freeing time for higher-value work.
The Dallas Fed data is early but consequential. It suggests that the aggregate wage and employment numbers mask a fundamental restructuring of returns to experience. Young workers with primarily codifiable knowledge face a challenging job market. Experienced workers in high-experience-premium occupations see their wages rise. The net effect is a flattening of aggregate wage growth in AI-exposed sectors, but the distribution is anything but flat.
For AI builders, the implication is uncomfortable. The technology is not simply a tool that makes all workers more productive. It is a technology that selectively augments some workers while substituting for others, and the dividing line is experience. The same model that helps a senior lawyer draft contracts faster may make it harder for a new graduate to get hired as a junior associate.
Davis is careful to note that this dynamic is not sustainable in the long run. Leaving new employees off the job ladder means the pipeline of experienced workers eventually dries up. AI adoption will require rethinking how entry-level employees gain experience on the job. That rethinking has not yet begun.