Goldman's AI Jobs Forecast: The Wave Waits for a Recession
Goldman Sachs sees AI displacing 9% of US workers — 15 million people — but its research says the layoffs wait for a recession. Prepare before the cycle turns.
Goldman Sachs Research now estimates that up to 9% of US workers — about 15 million people, roughly one in eleven — will be displaced from their jobs during the AI transition, under its baseline scenario of AI eventually lifting productivity by 15% in developed economies (Goldman Sachs Research, June 2026). But the number is not the news. In a note published October 22, 2026, Goldman economist Joseph Briggs reached a sharper conclusion about timing: the AI layoff wave is not missing — it is waiting, and history says it waits for a recession (Fortune, October 22, 2026).
What the 9% forecast saysPermalink to “What the 9% forecast says”
The forecast has a history worth knowing, because each revision tells you what changed in Goldman's model rather than in the world:
| Vintage | Estimate | What moved |
|---|---|---|
| August 2025 | 6–7% of US workforce displaced (range 3–14%) | Briggs and Sarah Dong, same 15% productivity assumption |
| June 2026 | ~9%, or 15 million workers over a decade | Historical displacement data, raised on the Exchanges podcast |
| October 2026 | 9% displaced, plus 3% slower hiring — one in eleven workers | Reframed from "reallocated" to "displaced," with cyclical timing analysis |
Every version rests on the same chain of reasoning: if AI eventually delivers a 15% productivity lift, and past technology-driven productivity gains displaced workers at a known historical rate, then multiplying that rate by 15 gives roughly 9% of workers displaced over the transition (Fortune).
Two qualifiers keep the number honest. First, this is gross displacement, not unemployment. Briggs puts it directly: even with nine percent of workers displaced, "the unemployment rate increase in any given year would be less than one percentage point," because the transition is spread over a decade and displaced workers find new positions (Goldman Sachs Research). Second, the long-run record favors reabsorption: about 85% of US job growth over the last 80 years came from technology creating new kinds of positions, and the economy churns roughly 30 million jobs created and 29 million destroyed every year. Goldman calculates that even a 5% acceleration in the pace of new-job creation would be enough to reabsorb everyone AI displaces (Goldman Sachs Research).
The catch is in the phrase "any given year." A decade is long enough for the math to work and short enough for the worst two or three years of it to break the people who live through them unemployed.
The real finding: the wave is cyclicalPermalink to “The real finding: the wave is cyclical”
What makes the October note different is not the size of the forecast but its shape. Briggs went back through decades of data on how technology has replaced workers and found that "automation-related labor disruption is limited when the overall economy is healthy but accelerates during cyclical slowdowns" (Fortune).
The evidence is a single line on a chart: the share of Americans working routine jobs — clerks, secretaries, machine operators, mechanics. That share holds flat through expansions and drops in steps at the recessions of the early 1980s, 2001, 2008, and 2020. It never climbs back (Fortune). Recessions, in other words, are when companies finally execute the automation they have been piloting. Goldman's own US economists David Mericle and Pierfrancesco Mei made the same argument a year earlier: "the full consequences of AI for the labor market might not become apparent until a recession hits."
The mechanism is not mysterious. Companies use downturns to restructure — layoffs in a recession read as a response to the recession, not as a bet on software. Quantitatively, Briggs found the hit to unemployment from a technology-driven productivity gain was 0.1 to 0.2 percentage points larger when the overall jobless rate was rising, and across 25 occupations that past technology upended, employment fell an extra 4% over five years for every point the national unemployment rate climbed (Fortune). His uncomfortable conclusion: "a cyclical slowdown would likely accelerate AI-related automation and job losses, thereby amplifying the broader business cycle."
A forecast, not a date
Goldman is not predicting a recession, and the 9% figure is scenario math on top of a 15% productivity assumption — not a count of scheduled layoffs. What the research does establish is the shape of the risk: diffusion in good times, concentration in bad ones. That shape is what a preparation plan should be built around.
Why it is quiet right nowPermalink to “Why it is quiet right now”
If a wave this size is coming, why did the AI-layoff panic that peaked in spring 2026 fade? Briggs offers two answers, and neither is reassuring.
The first is adoption: only 23% of US establishments have built AI into regular workflows, so most of the productivity gain — and most of the displacement it enables — is still in the future (Fortune). The second is the cycle itself: with the economy resilient through 2026, "companies may have had less incentive to focus on aggressively cutting labor costs." The current numbers confirm a small headwind, not a wave — AI-attributed layoffs in Challenger's tracking have pulled back sharply from their spring peak, and 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 AI-related construction hiring (Fortune).
Read together, those two answers say: the calm is cyclical, and the technology is still compounding. Quiet is not the same as cancelled.
What earlier waves did to the jobs they hitPermalink to “What earlier waves did to the jobs they hit”
The grimmest data in the note is occupational. Briggs examined 25 jobs that earlier technologies hit hard — photo processors, movie projectionists, proofreaders, file clerks, bank tellers — and found their employment fell 25% on average in the five years after decline began. He calls those cases "likely upper bounds," since he deliberately picked the worst-hit occupations (Fortune).
The composition of that shrinkage matters more than its size. Separation rates rose in 19 of the 25 occupations, by 3.8 points on average; hiring rates fell in only 13, and by less. People lost or left jobs faster than employers slowed hiring — which is why Briggs treats AI-attributed layoffs, not the unemployment rate, as the real-time signal to watch (Fortune).
He also flags why 9% could turn out to be low: displacement ran higher during the late-1990s and early-2000s productivity boom, "the period most widely viewed as comparable to today," and after unusually large productivity jumps. Goldman expects AI's boost to exceed most past ones (Fortune).
What this means for your careerPermalink to “What this means for your career”
Start with the distinction that keeps this forecast usable: exposure is not elimination. Most roles have automatable tasks; very few roles are entirely automatable, and the jobs that survive are rebuilt around judgment, verification, and ownership of outcomes. That larger frame — with the wage data on which side of the split pays — is in the pillar guide on whether AI will replace your job, and it holds here too. Analysts, for instance, are not disappearing; the routine share of their work is, which is exactly what the data-analyst exposure breakdown shows in detail.
The cyclical finding adds a time dimension the general AI-debate misses. If displacement concentrates in downturns, then your routine-task share is not a static risk profile — it is a fuse that a recession lights. The practical move is to measure it now: a structured AI skills assessment tells you which parts of your current work AI already covers and which adjacent skills close the gap.
I write this from a specific vantage point. My own move from sysadmin night shifts into HealthTech analytics happened in a healthy labor market — the hiring manager spent half the interview on monitoring dashboards I had nearly left off my resume, because the market had room to bet on a career changer. That is the luck of timing, and Goldman's research is a description of what happens to people who lack it: in a downturn, the same switch gets made under pressure, against thousands of others, at the exact moment companies are cutting rather than betting.
That is the case for moving early. Traecta — Your Personalized Career Roadmap maps the work history you already have against roles where your experience transfers and your routine-task share drops, then sequences the transition while the economy still rewards it — before the cyclical window compresses your options into whatever is left in a slump.
Use the calm before the cycle turnsPermalink to “Use the calm before the cycle turns”
If the wave is real and recession-timed, the rational response is not panic in either direction — neither "AI is overhyped" nor "it's too late." It is to treat the current expansion as prep time:
- Score your routine share honestly. List what you did last week and mark what a competent AI tool could have produced unsupervised. That fraction is your exposure; everything else is your platform to build on.
- Start the skill move now, not at the trigger. Transitions take months at best, and self-directed learning stalls in predictable ways — the common failure points are known and avoidable while you still have slack, not deadlines.
- Aim at AI-adjacent work, not away from AI. Roles that build, direct, or verify AI systems absorb the transition instead of competing with it; a structured route like the machine learning engineer roadmap is the archetype.
The forecast's most useful property is that it is falsifiable on a schedule. Watch AI-attributed layoffs as the real-time gauge Goldman suggests, and let a rising reading accelerate your plan rather than start it. Your Traecta career roadmap keeps the plan itself ready for that moment — sequenced, dated, and built on the skills you already have.
