AI Labor Disruption: Still Modest, but Increasingly Visible

Research evidence excerpt

AI Labor Disruption: Still Modest, but Increasingly Visible

FoundationM

Executive Summary

We update our AI labor disruption tracker with data through 1H26. Introduced in April, the

tracker combines labor-market microdata, worker-flow statistics, task reallocation

measures, payroll employment data, job postings, and company transcript analysis to

monitor emerging evidence of AI-driven labor-market change. The framework evaluates

outcomes across occupations and industries with different degrees of AI exposure while

accounting for differences in historical cyclical behavior. We use Felten et al (2021)

measure of AI exposure in our main results, but we also include an appendix with

robustness checks with an alternative measure developed by our Morgan Stanley

Thematic team.

This note focuses on what has changed since our initial assessment. Readers interested in

the construction of the tracker, AI exposure measures, and robustness tests should refer

to our earlier report, AI Labor Disruption: More Micro Than Macro So Far.

As we mentioned in our earlier note, this analysis still suffers from an "omitted variables

problem" in econometrics parlance. Other shocks — such as monetary policy, tariffs,

immigration policy changes, or post-Covid hiring dynamics — may affect occupations

differently and contribute to the patterns we observe across exposure groups. However, a

persistent pattern that strengthens as AI adoption increases would help separate signal

from noise. The additional data from 1H26 appear increasingly consistent with that view:

disruption still seems modest, but the signs are becoming clearer and harder to ignore.

The most important change since our previous update is that labor-market disruption

is becoming more visible in household survey microdata.