No Widespread Displacement, but the AI Employment Gap for Young Workers Has Widened to 19%

These are descriptive patterns, not causal estimates of the effect of AI. The timing and structure of the changes are suggestive, but the data alone cannot establish how much of the divergence was caused by generative AI rather than other forces affecting the labor market.

Beyond extending the data through mid-2026, the revised paper provides new evidence about the mechanisms that may be driving these patterns.

One distinction that appears important is between codified and tacit knowledge.

Employment has declined among young workers in occupations that rely heavily on codified knowledge: formal, standardized, documented knowledge that can be taught through education, textbooks, or written procedures. In contrast, employment has increased among experienced workers in occupations that rely more heavily on tacit knowledge acquired through practice, mentorship, and repeated exposure to real situations.

This distinction is consistent with a world in which generative AI is particularly effective at reproducing and applying knowledge that has already been encoded in text and other digital information, while experience-based knowledge remains harder to replicate.

We also find that women face greater AI exposure on average, an important source of heterogeneity that we intend to monitor as the data evolve.

We cannot yet answer that question definitively.

Several prominent alternative explanations do not appear sufficient to account for the pattern. The divergence remains when we exclude technology firms and computer occupations, when we control for exposure to interest-rate increases and remote work, when we include firms that enter or leave the sample, and when we use alternative measures of AI exposure.

There are also several reasons to think AI may be playing a meaningful role:

At the same time, there are important reasons for caution.

The gaps between more- and less-exposed young workers shrink when we account for education. Some differential trends are visible before the widespread use of generative AI. The estimated gaps are also larger in the ADP analysis sample than in national survey benchmarks.3

In addition, improvements to our data pipeline produced qualitatively similar raw patterns, but estimates that account for overall changes in firm hiring are directionally consistent while becoming more sensitive to specification choices. This raises legitimate questions both about how much of the pattern is actually caused by AI and about how well the results in the ADP analysis sample generalize to the broader economy.

For those reasons, we do not view this paper—or any single study—as definitive evidence of AI’s labor-market effects. A growing body of research has appeared since the first version of our paper, and the cumulative evidence across studies will ultimately be more informative than any one result.

We do not know whether the patterns documented here will accelerate, stabilize, or reverse.

That uncertainty is one reason we have invested in building the infrastructure to measure these changes continuously rather than relying on occasional snapshots.

The Stanford Digital Economy Lab recently launched the AI Economic Indicators, providing high-frequency measures of how AI is changing the economy. As part of that effort, our Canaries Dashboard will update the key results in this paper every month.

The goal is not to declare the labor-market effects of AI settled. It is to make them measurable.

If generative AI is beginning to reshape opportunities for particular groups of workers before its effects are visible in aggregate employment statistics, we want to identify those changes early, understand the mechanisms behind them, and track whether they spread.

That is, after all, what canaries are for.