Goldman Sachs on the AI Job Apocalypse: Real Pain, Not Mass Unemployment

⬅️ Back to Articles

Goldman’s June 2026 Top of Mind report is titled “An AI Job Apocalypse?” The question mark matters. Three economists disagree on how bad it gets.

  1. Joseph Briggs expects more than 9% of the U.S. labor force (about 15 million workers) displaced over a ten-year AI transition, then reabsorbed as new jobs appear. Neil Thompson of MIT calls AI a “rising tide,” not a “crashing wave,” because most jobs are task bundles only partly automatable. Daron Acemoglu expects a small net negative hit over five years, worse later if investment keeps favoring replacement over complementing workers.

  2. Briggs’s math is less cinematic than the title. Goldman assumes a 15% productivity uplift from full AI adoption. Historical patterns imply 6-7% displacement (3-14% range), with up to 9% plausible. Spread over ten years, peak unemployment might rise under 1 percentage point. Painful. Not apocalyptic.

  3. Exposure scores alone mislead. Customer service reps and interior designers look similarly exposed, but designers need unstructured work and physical presence. Pair exposure with an IMF complementarity index: telephone operators, insurance claims clerks, and bill collectors face high substitution risk; education workers, judges, and construction managers face high augmentation potential.

  4. Since ChatGPT launched, substitution is winning narrowly. Elsie Peng estimates substitution cut monthly payroll growth by roughly 25,000 jobs and pushed unemployment up 0.16 point. Augmentation added about 9,000 jobs a month. Net drag: 16,000 jobs a month, 0.1 point on unemployment. Younger, less-experienced workers took the hit. Peng notes the true drag is probably smaller once you count data-center construction and productivity-driven demand.

  5. The new-graduate panic may be ahead of the evidence. College grad unemployment was 2.7% versus 2.1% pre-pandemic, but Rindels and Mei are not convinced AI is the driver yet. Disruption clusters in software publishing, data processing, and call centers. Grads still sit in high-adoption industries with automatable tasks, so near-term risk is real.

  6. Acemoglu’s line stuck with me: “No general law of economics says that job creation must match job destruction.” Pierfrancesco Mei finds displaced tech workers lose more than 3% in real wages and take a month longer to find work. Ryan Hammond adds that markets cannot price any of this yet: only 2% of S&P 500 firms tied AI productivity gains to earnings in Q1 2026.

The takeaway: Treat “job apocalypse” as marketing and “small net drag, unevenly distributed” as the base case. If you are early career, watch substitution-exposed clerical work. Build toward augmentation-heavy skills: judgment, physical presence, accountability.

Related TMFNK Content

Crepi il lupo! 🐺