Goldman Sachs Exchanges
2026-07-06 · Hosted by Allison Nathan · Goldman Sachs
Executive Summary
Goldman Sachs’ Allison Nathan convenes three experts — Goldman’s Joseph Briggs, MIT’s Neil Thompson, and MIT’s Daron Acemoglu — to debate whether AI will trigger a “job apocalypse.” Briggs forecasts that AI-driven productivity gains (a 15% uplift at full adoption) will displace roughly 9% of US workers (~15 million) over a 10-year transition, but expects job creation to fully reabsorb them, keeping annual unemployment increases below one percentage point. Thompson emphasizes that AI capability alone doesn’t automate jobs — adoption depends on information access and cost-effectiveness — and frames the transition as a manageable “rising tide” rather than a “crashing wave.” Acemoglu is more cautious, expecting net job losses of less than 2–4% near term but warning of larger displacement over 10–15 years if AI investment keeps prioritizing worker replacement over complementarity, with rising labor income inequality as the likely result. The overriding theme: significant labor churn is coming, but its ultimate scale hinges on the pace of new job creation and where AI investment flows.
Key Stories & Changes
1. Goldman’s Baseline: 9% Displacement, but No Permanent Job Loss
Goldman’s Joseph Briggs sees AI’s current imprint in a few sectors — tech, management consulting, graphic design — creating a 10,000–15,000 drag on month-over-month job growth, still a narrow shock with no broad economic impact yet.
Under Goldman’s baseline of a 15% productivity uplift at full AI adoption, applying historical productivity-to-displacement elasticity yields ~9% of US workers reallocated during the transition — equivalent to ~15 million workers displaced.
Briggs stresses the displacement plays out over a 10-year period; spread out, the unemployment-rate increase in any single year would likely be less than one percentage point.
Job creation offsets: ~85% of job growth over the last 80 years came from technology creating new positions; the US labor market already sees ~30 million jobs created and ~29 million destroyed annually. Even a 5% acceleration in new-job creation would more than reabsorb displaced workers.
Briggs explicitly rejects the tech-commentator view of permanent mass unemployment, arguing it fixates on job loss and ignores job creation.
2. Thompson’s “Rising Tide”: Capability ≠ Automation
MIT’s Neil Thompson argues AI capability is only the first of several steps to automation — you must also supply all the right information (often blocked by privacy/records hurdles, e.g., medical data) and confirm it’s cost-effective to run.
Adoption therefore lags capability: large businesses automate before small ones, and high-value tasks get done before a long tail that “probably will take a long time or not happen.”
Expert vs. inexpert task framing: automating an inexpert task (e.g., GPS for taxi drivers) lowers wages but expands the workforce; automating an expert task (e.g., spell-check for proofreaders) shrinks the workforce but raises pay for those remaining.
Thompson contrasts “crashing waves” (workers suddenly washed away) with “rising tides” (gradual, visible change) and says research points to the rising-tide dynamic — meaning workers and businesses can see AI coming and manage the transition.
3. Acemoglu’s Caution: Small Net Losses Now, Bigger Risk Later
MIT’s Daron Acemoglu expects to see more layoffs or hiring slowdowns by 2027 in AI-exposed jobs, but forecasts net job losses of less than 2–4% over the next five years — far below alarmist predictions.
Most vulnerable now: cognitive, routine tasks — customer service reps and back-office work — which he estimates total ~8–9 million US workers.
The constraint on faster AI spread is the lack of reliable, easy-to-use applications built on foundation models (a “Microsoft Office version of AI”); today’s reliance on individual prompt engineering is inconsistent and time-consuming.
Coding is the exception — models are already capable, and software engineers are AI experts who can prompt and troubleshoot.
Over 10–15 years, Acemoglu warns of bigger net job losses if investment keeps favoring replacement over complementarity, and flags AI + robotics as the biggest wild card (physical work is ~50% of the US economy).
Trends Identified
1. Displacement Is Real, Permanence Is Contested
All three experts agree AI will displace a meaningful number of workers, but diverge sharply on permanence. Briggs sees churn fully offset by new job creation over a decade; Acemoglu warns that job creation has not matched destruction since the late 1970s and there is “no general law of economics” guaranteeing it will. This is the central analytical fault line — the optimistic case rests entirely on the historical record of technology generating new work repeating itself.
2. Adoption Friction as the Real Governor
A recurring theme across Thompson and Acemoglu is that AI’s labor impact is throttled not by capability but by adoption barriers — information access, cost-effectiveness, and the absence of reliable turnkey applications. This reframes the timeline: capabilities improve fast, but deployment is gated, meaning the labor shock arrives gradually and unevenly rather than all at once.
3. Distributional Effects and Inequality
The experts connect automation to inequality via which tasks and workers are hit. Acemoglu argues that because the vulnerable jobs (customer service, back office) are lower-to-middle paid rather than managerial, AI is likely to increase labor income inequality — mirroring the 1980s–2000s pattern where middle-class wages were replaced by lower-wage occupations. Thompson’s expert/inexpert framing shows the same dynamic can cut either way depending on the task automated.
4. Investment Direction Determines the Outcome
A forward-looking trend is that policy and capital allocation — not technology itself — will shape the long-run result. Acemoglu argues the “complementary path” is productive but under-invested, and that breakthroughs in AI+robotics or agentic AI for middle-management judgment tasks could dramatically expand AI’s reach. The outcome is presented as a choice, not a fate. —-
Sentiment Analysis
Overall Market Sentiment: Cautiously Measured
The dominant mood is analytical and de-alarmist: all three experts push back on “job apocalypse” narratives while acknowledging genuine, meaningful displacement ahead. The host herself ends unsure whether she is “more concerned or more comforted.”
Risk Factors Highlighted
Job creation failing to keep pace: If new-job creation doesn’t accelerate fast enough, near-term displacement could translate into elevated unemployment before reabsorption.
Investment tilted toward replacement: Acemoglu warns that under-investment in the “complementary path” could produce bigger net job losses over 10–15 years.
Rising labor income inequality: Because vulnerable jobs are lower-to-middle paid, AI displacement is likely to widen income inequality.
AI + robotics breakthrough: A wild card that could dramatically expand AI’s reach into physical work (~50% of the US economy).
Concentrated sectoral shocks: Customer service, back-office, and coding roles face outsized near-term exposure (~8–9M workers in routine cognitive roles).
Adoption outpacing worker adjustment: If the “rising tide” rises fast, even a visible transition could strain workers’ ability to adapt.
Structural weakness for non-college workers: Job creation has lagged for workers without a college degree since 1980, a pattern AI could reinforce.
Agentic AI accelerating middle-management displacement: Advances in judgment-task automation could put middle managers in the crosshairs within a year or so.
This episode was covered in today’s The Market Signal — 2026-07-06, a cross-source synthesis of multiple podcast reports.