Goldman finds AI job cuts wait for the next downturn. The quiet months are when you move

Goldman Sachs economist Joseph Briggs published a note on Thursday 22 October that puts a number and a timetable on AI and jobs. Fortune reported that, under Goldman's forecast that AI eventually lifts productivity by 15% in the US and other developed economies, up to 9% of workers could be displaced during the transition, with slower hiring weighing on employment by a further 3%. The number will get the headlines. The timing is the finding. Briggs went back through decades of data and concluded that automation-related disruption "is limited when the overall economy is healthy but accelerates during cyclical slowdowns." The cuts are waiting for the next downturn.

The quiet is a deferral

The obvious reading of 2026 is that the AI layoff scare peaked in spring and faded. Goldman's own gauges support the first half of that. AI-attributed layoffs in Challenger's tracking have pulled back sharply from their spring peak, and news coverage of AI job losses has fallen back to levels seen at the start of the year.

Look underneath and the work is already moving. Goldman finds job growth in the most AI-exposed industries has run about 10,000 below its long-run trend over the past three months, roughly offset by hiring for AI-related construction. Pantheon Macroeconomics' Samuel Tombs and Oliver Allen, cited by Fortune on 23 October, put AI-related job losses in the US in excess of 10,000 roles a month, concentrated in accounting, advertising, office administration and legal services.

Briggs gives two reasons the effect is small so far. Only 23% of US establishments have built AI into regular workflows. And a resilient economy has given companies less reason to cut labour costs aggressively.

Neither reason is permanent. Goldman's forecast assumes most of the productivity gain is still ahead. And the bank's case is that the next slowdown is where it lands.

Routine work leaves in steps, and the steps do not reverse

The chart at the centre of the Goldman note tracks the share of Americans in routine jobs: clerks, secretaries, factory machine operators, mechanics. Fortune reports that the share held steady through expansions and dropped sharply in the recessions of the early 1980s, 2001, 2008 and 2020. After each drop, it did not climb back.

The industry data says the same thing in a different unit. Across 25 occupations that past technology upended, employment fell an extra 4% over five years for every point the national unemployment rate rose. In those 25 jobs, which include proofreaders, file clerks and bank tellers, employment fell 25% on average in the five years after decline began. Briggs calls those figures likely upper bounds for AI, because he selected the hardest-hit cases.

The mechanism matters more than the magnitude. In 19 of the 25 occupations, separation rates rose, by 3.8 points on average. Hiring fell in 13, by 0.7 points. Most of the shrinkage came from people losing or leaving jobs, more than from employers hiring fewer. In 2025, Goldman's US economists David Mericle and Pierfrancesco Mei cited the leading explanation: companies use recessions to restructure and lay off workers in less productive areas.

Read that as an operator. The decision about which tasks in your role are routine gets made in a few weeks, during a slowdown, by someone looking at a cost line. The analysis that informs it is the one that already exists. If nobody has written it, the person under budget pressure writes it fast and broadly.

Your Next Move

The calm months are the only months in which you choose the terms. AI does the routine work. You do the thinking. Your job between now and the next slowdown is to make that split visible in your own role, on your own initiative.

1. Write your own routine list this month. Take last week's calendar and task log. Mark every task a competent AI workflow could do with your instructions. Move the marked tasks to AI now, and record the hours returned and where you spent them. A role that has already been restructured by the person doing it is harder to restructure from a spreadsheet.

2. Attach your name to the gain before it is attributed to a licence. If your organisation sets next year's plan before year end, propose owning one AI workflow for your team, with a before-and-after measure. Only 23% of US establishments have AI in regular workflows, by Goldman's count, so in most teams nobody yet owns this. A named owner of a measured gain is a harder line to cut than a seat on a licence.

3. Set a trigger. Briggs names AI-attributed layoffs as the real-time guide to disruption. Watch Challenger's monthly figures for your sector and your national unemployment rate. Decide now what you will have finished by the time either turns up. Then finish it before they do.