Goldman Sachs says AI could displace about 9% of workers, roughly one in 11, if it lifts productivity by 15%. The estimate comes from a new note by Goldman economist Joseph Briggs, reported by Fortune and Investing.com on October 22, 2026.
The headline number is not new. What is new is the method, and a finding that matters more than the percentage: according to the note, disruption from automation stays small while the economy is healthy and speeds up when it slows. This post unpacks the arithmetic, separates scenario from forecast, and lists what to track. We did not read the Goldman note itself, so every figure below is as reported by those outlets.
TL;DR: the Goldman AI jobs note
| Question | Answer |
|---|---|
| Who wrote it? | Goldman Sachs economist Joseph Briggs |
| Headline | A 15% AI productivity gain could displace up to about 9% of workers |
| Extra effect | Slower hiring could add roughly a further 3% hit to employment (Fortune) |
| What is measured? | Gross job losses, not the unemployment rate |
| Unemployment effect | Each 1% productivity gain historically raised peak unemployment 0.3 to 0.4 points, fading after about two years |
| Key finding | Displacement is limited in healthy economies and accelerates in downturns |
| Is it a forecast? | A scenario from historical analogies, not a dated prediction |
How does Goldman get from 15% to 9%?
Per Fortune, the note reviews decades of data on technology-driven labor shifts. Historically, each 1% gain in productivity raised the cumulative gross job-loss rate by 0.5 to 0.6 percentage points. Take the top of that range, 0.6, and multiply by an assumed 15% AI productivity gain in the US and other developed economies, and you get 9%.
Investing.com describes three methods that point the same way:
- Economy-wide. Each 1% productivity rise has caused a temporary 0.3 to 0.4 point increase in unemployment, driven by higher job losses and a smaller 0.2% drop in hiring.
- Industry-level. Results are similar, with larger job-loss effects in the late 1990s and after bigger productivity gains.
- Occupation-level. Across 25 occupations hit by major technological disruption, such as photo processors, proofreaders, bank tellers and file clerks, employment fell about 25% on average over five years. The 9% is therefore a stack of assumptions: that AI delivers a 15% productivity gain, that the future looks like past technology shocks, and that the top of the historical range applies.
Why does the note focus on gross job losses?
The unemployment rate hides a lot. Workers who leave the labor force, retire early or move to another employer do not show up as unemployed. Gross job losses count the separations directly, which is why the number looks large next to the modest unemployment effect. Historically, the unemployment bump from a 1% productivity gain faded to zero within about two years, so a 9% gross displacement does not translate into 9% unemployment.
That distinction is easy to lose in headlines. "Displaced" here means a person's job went away, not that they stayed without work. Earlier this year Briggs described the same roughly 9%, or 15 million US workers over a decade, as workers who would be "reallocated" to new positions, according to Fortune. Thursday's note uses the harder word.
An outlined hand rearranges green blocks within a cream briefcase, illustrating jobs being reshaped by AI
Why does a recession change the picture?
The most useful finding is about timing. Fortune reports that the share of workers in routine jobs stayed stable during expansions and dropped sharply in recessions, namely the early 1980s, 2001, 2008 and 2020, without recovering afterward. Across the 25 occupations, each point of rising unemployment added about 4% extra employment loss over five years.
Briggs's conclusion, as quoted: labor market displacement is "more severe when the economic backdrop is less favorable," and AI-related disruption "could well amplify the broader business cycle." Goldman economists David Mericle and Pierfrancesco Mei made a similar point in October 2025 about "jobless growth" after the dot-com bust.
The implication is uncomfortable. Companies hold back on cuts when demand is strong and use a downturn as cover to automate. If a slowdown arrives while AI tools are mature, layoffs that were deferred could arrive together.
Why is the effect quiet right now?
Fortune lists several reasons the impact looks small in 2026:
- AI-attributed layoffs in Challenger's tracking have fallen sharply from their spring peak.
- Only 23% of US establishments have built AI into regular workflows.
- The economy and labor market have stayed resilient, so companies had less incentive to cut labor costs.
- Job growth in the most AI-exposed industries has run about 10,000 below trend over the past three months, roughly offset by AI-related construction hiring.
That matches adoption data we have covered. Gartner's survey found only a small share of CIOs reporting significant AI value, and a Monday.com and Nielsen survey found managers struggling to evaluate AI-assisted work. Slow adoption is part of why the displacement has not yet arrived at scale.
How does this fit the rest of the AI jobs debate?
Voices span a wide range. Dario Amodei warned in 2025 that AI could eliminate half of entry-level white-collar jobs, then in May 2026 leaned on the Jevons paradox argument, as Fortune notes; our post on Jevons paradox and AI demand covers that line of thought. On the evidence side, Anthropic's review of 56 randomized studies on worker retraining asks what actually helps displaced workers. On the corporate side, Barclays' 5,000 job cuts show how banks frame AI as a contributing factor, and our record of tech layoffs cited to AI tracks the announcements.
Goldman's contribution is a historical anchor. It does not say AI is different. It says that if AI behaves like earlier technologies and delivers a big productivity gain, here is the scale and here is when it bites.
What should you take from this?
If you are a worker. Routine, clerical and process-driven roles carry the highest historical risk, according to Fortune's list of clerks, secretaries, machine operators and mechanics. Skills that sit on top of AI tools, rather than doing what the tool does, hold up better. Our guide to AI-ready skills for 2026 and 2027 is a practical start.
If you run a team. The note frames displacement as a choice made against a backdrop. Plan reskilling before a downturn, when labor markets absorb people more slowly.
If you read forecasts. Check what is being measured. Gross losses, net losses, unemployment and hiring slowdowns are four different numbers, and the 9% plus 3% presentation can confuse them. Fortune itself notes that it does not explain how the pieces add up to about one in 11.
What does a worked example look like?
A quick sensitivity check shows how fragile a single headline is. Using the ratio Fortune describes, where each 1% of productivity adds 0.5 to 0.6 points of cumulative gross job losses, the arithmetic scales linearly:
| Assumed AI productivity gain | Gross job losses at 0.5 | Gross job losses at 0.6 |
|---|---|---|
| 5% | 2.5% | 3.0% |
| 10% | 5.0% | 6.0% |
| 15% | 7.5% | 9.0% |
| 20% | 10.0% | 12.0% |
The table is our own back-of-envelope scaling of the reported ratio, not a Goldman output. It shows why the 9% should be read as the top of the 15% case: the low end of the same range gives 7.5%, which is close to the 6% to 7% central figure Goldman published in August 2025 with a range of 3% to 14%.
It also shows what drives the answer. A faster-than-expected productivity boom raises the number quickly, while slow adoption lowers it just as quickly. That is why the 23% share of establishments using AI in regular workflows matters more than any single forecast: the productivity gain has to show up in real work before it can displace anyone.
Limits of the estimate
- Assumed gain. The 15% productivity figure is an input, not a finding. If AI delivers 5%, the same math gives about 3%.
- Historical analogy. Bank tellers and file clerks were displaced by technologies with narrower scope than general-purpose AI. The past may under- or overstate.
- Secondary reporting. We relied on Fortune and Investing.com. The Investing.com summary itself says to check the original before relying on specific figures.
What to watch next
- AI-attributed layoffs. The Investing.com summary says tracking them could serve as a real-time signal.
- Hiring rates in exposed occupations. Hiring slowdowns come before layoffs.
- A cyclical slowdown. The note's central warning is about what happens if one arrives.
- Goldman's follow-ups. The full note and any revisions to the 15% assumption will move the number.
The takeaway: 9% is a scenario about a world where AI delivers a large productivity boom. The more actionable claim is that the pain is back-loaded into bad economic times. Figures here reflect press reports of Goldman Sachs research as of October 22, 2026.