A report attributed to Goldman Sachs suggests AI’s effect on US employment may be spreading beyond tech. Greater industry exposure to AI is increasingly associated with higher unemployment in finance, telecommunications, technical services, and administration. That’s a shift in the evidence, not proof that AI caused those job-market changes.
The findings are one of two new sources of data examined in reporting by the Financial Times. Despite a visibly weakening job market and widespread anxiety about AI automation, empirical evidence definitively linking the two has been hard to find. These analyses offer more concrete signals, but come with substantial caveats.
The Goldman Sachs document, which appears to be an internal report, began circulating on social media the day after US-based financial researcher Eric Basmajian published his own analysis. The apparent Goldman Sachs report warned that the “AI labor market impact” was extending beyond tech work. That field has already been hit hard as generative AI has become a common tool for writing software code.
Basmajian’s analysis examined productivity in sectors with high AI adoption. In his social media post, he wrote that “AI productivity gains are showing up” in jobs data, while qualifying that only about 15 percent of jobs seem to be affected so far.
“The three heaviest AI adopters (information, professional services, finance) went from a 2.2 percent productivity trend to 5.0 percent since 2022,” Basmajian explained. He pointed to incremental workplace productivity gains coinciding with high AI adoption, without establishing that AI caused them. The relationship between AI exposure and unemployment in the apparent Goldman Sachs report seems to reinforce his assertion, with an apparent upward trendline in at least some parts of the labor market.
But productivity is the crucial wrinkle. The Financial Times cautioned that the metrics used here don’t necessarily reflect quality. Higher measured productivity needn’t mean AI has sent productive output and revenue soaring; it may instead reflect fewer people on the payroll. Those are very different outcomes, even if both can produce a stronger-looking productivity number.
There’s also a geographic limit: these data concern the United States, not a worldwide wave of AI-driven job losses. Broader post-pandemic economic troubles remain relevant when interpreting the US labor market. As the Financial Times put it, the “link that was visible in the US between AI usage on the one hand, and productivity growth and falling employment on the other, is conspicuously absent elsewhere.”
For those warning about AI-driven job losses, the evidence outside software work gives them more to examine, but these correlations don’t establish the sweeping AI labor upheaval some fear. The unresolved question is whether AI is driving those employment changes or tracking alongside other pressures on US employers.
