Goldman warns a recession could accelerate AI job losses despite limited disruption so far

AI could temporarily displace up to 9% of workers, but the pace may depend heavily on the economy. In an October 22 research note, Goldman Sachs economist Joseph Briggs found that past waves of automation caused less labor disruption during healthy expansions—and accelerated during downturns. A recession could make AI-related job losses worse.

The estimate assumes AI eventually raises productivity by 15% in the U.S. and other developed economies, according to Fortune’s account of the note. It measures workers displaced from existing roles, not a permanent increase in unemployment. Goldman’s baseline still anticipates a gradual rise in disruption if economic growth continues, rather than an abrupt employment shock.

Displacement is not the unemployment rate

Briggs’s calculation draws on the historical relationship between technology-driven productivity gains and job losses. Each 1% productivity increase raised the cumulative gross job-loss rate by 0.5 to 0.6 percentage points. Applying the upper end to Goldman’s assumed 15% productivity boost produces the 9% estimate. That is a forecast built from past experience, not a count of jobs already lost to AI.

Three measures in the analysis capture different parts of the adjustment:

• Gross displacement counts workers leaving or losing their existing positions. Some may find another job quickly; others may leave the labor force.
• Unemployment captures a smaller, temporary effect. Historically, each 1% technology-driven productivity gain raised unemployment by 0.3 to 0.4 percentage points at its peak, with the effect fading after two years.
• Slower hiring is a separate pressure. Briggs estimates it could weigh on employment by another 3%, alongside the displacement forecast.

The headline displacement figure also predates this note. In a Goldman Sachs podcast published July 2, Briggs described the transition as unfolding over about a decade. He expected enough new jobs eventually to absorb displaced workers, while identifying the pace of job creation as the key near-term uncertainty.

Recessions compress the adjustment

The October note’s timing argument rests partly on the history of routine work, including clerical and factory-machine jobs. Briggs found that these jobs’ share of U.S. employment stayed stable during expansions but dropped sharply in recessions, including those of the early 1980s, 2001, 2008 and 2020. The share did not return to its previous level afterward.

The proposed mechanism is straightforward: when business slows, companies have more incentive to automate work and reduce labor costs. Across 25 occupations disrupted by earlier technologies, employment fell an additional 4% over five years for every percentage-point increase in the national unemployment rate. The economic backdrop affected the severity of the adjustment, not just its timing.

Those occupations included photo processors, movie projectionists, proofreaders, file clerks and bank tellers. Employment fell by an average of 25% in the five years after declines began. Briggs cautioned that these were likely upper bounds for AI because he selected occupations hit particularly hard by technology. Most of their contraction came from workers losing or leaving jobs, rather than reduced recruitment.

Today’s pressure remains modest

For now, Goldman sees a much smaller effect. Employment growth in the industries most exposed to AI ran about 10,000 jobs below its long-term trend over the prior three months. Increased hiring in AI-related construction largely offset that shortfall, according to Investing.com’s account on Yahoo Finance. AI-related layoff announcements also fell sharply from their spring peak.

Across service industries, the analysis found a statistically significant association between AI adoption and higher unemployment. Its relationship with weaker employment growth was less pronounced. These are emerging signs of pressure, rather than evidence of an economy-wide employment shock.

Briggs offers two explanations for the limited impact: only 23% of U.S. establishments have incorporated AI into regular workflows, leaving much of its potential productivity effect ahead; and the resilient economy has reduced the incentive for aggressive labor-cost cuts. His warning is conditional: a slowdown could change those incentives and make automation amplify an already weakening labor market.