← Back to feed

Goldman Sachs published a research report on July 2 titled “An AI Job Apocalypse?” — a question the bank’s economists answer with a qualified no, but with enough data attached to give pause.

Senior global economist Joseph Briggs estimates that more than 9% of the US workforce will be displaced during a 10-year AI transition, corresponding to roughly 15 million workers. The bank’s own labour model currently puts the measured drag at 16,000 jobs per month — elevated but far short of crisis-level.

The Optimistic Case and Its Conditions

The Goldman view is that displacement will be temporary. The historical parallel: the digital economy created nearly 15 million new jobs as it disrupted existing ones. Briggs expects AI to do the same — generating new occupations in AI oversight, system integration, and entirely new product categories that don’t exist yet.

GDP upside, if full adoption occurs: +15%. That figure reflects productivity gains across knowledge work, customer service, analysis, and decision support — the categories that absorb the most white-collar labour.

MIT’s Daron Acemoglu is more conservative, expecting 2-4% job losses over five years. His concern is narrower: current AI models replace routine office tasks more effectively than they augment workers who perform complex judgment-intensive work. Neil Thompson of MIT adds the deployment filter — capability in a lab setting does not equal reliability, private-data access, or workflow integration in practice.

Who Gets Hit First

The clearest near-term pressure is on:

• Customer support and call centre roles
• Back-office processing: claims, billing, data entry
• Entry-level analytical and coding tasks
• Research assistant functions

College graduates are disproportionately exposed. Occupations with higher educational attainment tend to have larger shares of tasks that AI can automate — the inverse of the historical pattern where automation displaced manual workers first. Goldman’s analysis finds that younger workers with degrees are entering a labour market where AI is absorbing the very tasks that traditionally taught judgment and built careers.

The Concentration Risk

Briggs acknowledges two scenarios that break the base case.

First: the displacement doesn’t spread over 10 years. If job losses concentrate in a 3-5 year window — which aligns with the current pace of enterprise AI adoption — the unemployment spike would be sharper and the labour market’s adaptive capacity would be tested more acutely.

Second: this time is structurally different. Prior automation waves (mechanisation, computing, the internet) created more jobs than they destroyed over multi-decade horizons. AI may be a more labour-replacing technology in ways that historical models don’t capture. Goldman is explicit that this question “will be hard to prove or disprove until the labour market holds up or doesn’t in the coming years.”

Current Signals

The July 2026 picture: AI is the top-cited reason for US job cuts for two consecutive months. College-educated workers are experiencing labour market softening at higher rates than workers in physical trades. Goldman’s 16,000 monthly displacement figure is a model output, not a direct measurement — the real mechanism is harder to isolate because companies rarely attribute headcount reductions to a single cause.

The report lands amid a broader debate about whether frontier AI is yet at the capability threshold required for mass displacement. Goldman says the threshold is being crossed now. Acemoglu says we’re still in the substitution-of-routine phase. The data to settle the argument won’t exist for several years.