If you've felt whiplash reading headlines about AI and employment this year, you're not imagining it. Two separate pieces of research released in recent months — one from Goldman Sachs, one a large multi-country NBER working paper — reach conclusions that sound contradictory on the surface. Understanding why both can be true tells you more about the current job market than either headline alone.
In its report "Is AI Impacting Global Labor Markets?", Goldman Sachs economists Sarah Dong and Joseph Briggs found clear signs of AI-linked hiring weakness concentrated in specific, highly-exposed industries. Call center employment in the US is running meaningfully below its long-run trend, and software publishing, management consulting, and advertising services show similar — though smaller — gaps. Briggs has also pointed to rising joblessness among tech workers in their 20s since early 2024, noting the effect falls more heavily on less-experienced workers (People Matters).
Crucially, Goldman's own numbers also show the aggregate effect is still small: overall employment growth in the most AI-exposed industries has run only about 10,000 jobs below trend over a recent three-month period in the US economy — a real but narrow effect, not a labor-market-wide collapse.
A separate, large-sample NBER working paper, "Firm Data on AI," surveyed thousands of senior executives across the US, UK, Germany, Australia, and other economies. Its headline finding: executives report that AI adoption has had essentially no net effect on their firms' total employment over the past three years, and most expect only modest net effects — in either direction — over the next three. Beneath that flat aggregate number, though, the paper finds real churn: a substantial share of firms report AI driving both headcount increases in some functions and reductions in others, with the two often roughly offsetting (NBER Working Paper 34836).
These aren't contradictory once you notice they're measuring different things. Goldman is looking at specific, highly-exposed job categories — call centers, entry-level software roles — where the effect is concentrated and real. NBER is averaging across entire firms and all employee types, where AI-driven cuts in one function are frequently offset by AI-driven growth or hiring in another (sales enablement, AI governance, data operations, and similar newer functions).
The practical takeaway: don't evaluate your own risk from an aggregate "AI and jobs" headline. Ask where your specific function sits on the spectrum between "highly automatable, narrowly scoped" and "judgment-heavy, hard to fully delegate to a model" — that's a far better predictor than any single economy-wide statistic.