“There’s no question AI is going to disrupt the labor market, but the U.S. economy has a long track record of creating new jobs in response to disruption, and I see no reason to think it will stop now.” – David Solomon – Goldman Sachs CEO
Labour-market disruption is not new in the United States, but the current wave of artificial intelligence raises a more pointed question than past technology shifts: will the economy keep generating enough new, well-paid work to absorb displaced workers, or are we heading for a structurally higher level of joblessness and insecurity 1? The answer matters not just for workers in exposed occupations, but for growth, inequality, social stability, and how firms like large banks allocate capital and talent.
Historically, each major technological transition has destroyed specific roles while catalysing new industries and job categories. Mechanisation reduced agricultural labour, electrification reconfigured factory work, and computing hollowed out clerical roles. Yet aggregate employment recovered and expanded, helped by population growth, rising demand, and complementary tasks that machines could not perform. The current AI cycle tests whether this pattern can hold when software systems increasingly act on information, language, and decision-making tasks that used to be the preserve of white-collar professionals 1,2.
Recent data on AI and employment are conflicted rather than catastrophic. In AI-exposed sectors such as computer systems design and related services, employment has fallen by around 5% since the launch of widely used generative tools, while the top 10% of AI-exposed sectors have seen roughly a 1% decline in employment. At the same time, nominal wages in those same areas have grown strongly, with one key subsector recording about 16,7% wage growth compared with roughly 7,5% nationally over a similar period 1. This divergence points to a reconfiguration of who is employed and at what price, rather than simple across-the-board job destruction.
What is changing most rapidly is the allocation of tasks inside occupations. AI tools already handle codified knowledge work such as summarising documents, drafting marketing copy, generating code templates, and triaging customer enquiries. That can displace some entry-level roles, where the value proposition was the ability to execute routine analytical or administrative tasks at low cost. At the same time, AI enhances workers who combine domain expertise, organisational knowledge, and social skills with the capacity to orchestrate these tools effectively. This is the core of the bifurcation described by central bank researchers who find that AI substitutes for roles heavy in textbook learning but augments roles relying on tacit knowledge acquired through experience 1.
For a firm like a global investment bank, the strategic implication is clear: fewer low-skill process roles and more high-value professionals. Senior managers at major banks have argued that AI lets them expand the firm with a higher-quality workforce rather than a larger one, effectively raising the bar for hiring and progression 2,4. When a new analyst can automate a large share of model-building and slide production, the threshold moves from “can you do the basic work” to “can you frame the problem, challenge the model, and persuade clients”. That shift is less visible in raw job-count statistics but profound in how careers evolve.
Evidence from the broader labour market reinforces the notion that AI is playing out unevenly across generations and skill tiers. Early-career workers in AI-exposed occupations have seen employment fall by around 16% since 2022, largely through lower hiring rather than mass layoffs 1. Firms automate many of the tasks that entry-level employees previously performed, then redeploy savings to retain or recruit mid-career talent, where AI acts as a force multiplier. Employers increasingly report that roles requiring five to ten years of experience are in highest demand, reflecting a premium on individuals who can translate AI outputs into business value.
This generational skew raises a serious concern. Even if overall unemployment remains contained, a cohort of new graduates may find it harder to secure the first role that builds the tacit knowledge and professional networks necessary for long-term success. Macroeconomic stability can coexist with micro-level distress concentrated among young workers, particular regions, or specific industries. For policymakers and firms, that tension is central to the question of whether the economy merely flexes around AI or begins to fracture into insiders and outsiders.
The corporate narrative around AI-driven layoffs further muddies interpretation of the data. High-profile firms have announced large job cuts framed as necessary to fund AI initiatives: tens of thousands of roles in technology, retail, and financial services have ostensibly been eliminated for this reason. Yet detailed analysis suggests that “AI” often functions as a rhetorical cover for broader cost-cutting or strategic restructuring. Surveys of executives show that AI is frequently cited as a justification for workforce reduction even when the direct productivity gains from deployed systems are modest 1,3.
Researchers and commentators have begun describing this phenomenon as “AI washing”. In 2025, AI was among the top stated reasons for workforce reduction, but a large share of firms cutting headcount also faced revenue pressures or margin compression. A striking finding from management surveys is that nearly 40% of organisations reported reducing staff “in anticipation” of AI-driven efficiencies, while only a small minority attributed large reductions to realised AI deployment 1. This decoupling between rhetoric and reality makes it harder to infer the true causal impact of AI from headline layoff announcements alone.
Central banks and economic research institutes, which look through individual corporate moves to aggregate trends, paint a more measured picture. The unemployment rate has fluctuated only mildly as AI investment has accelerated, with recent readings hovering around the mid-4% range and some forecasts suggesting only a modest AI-related contribution to joblessness in the near term 1. Output growth in knowledge-intensive sectors that are heavy AI adopters, including information services, advanced manufacturing, finance, and professional services, has been robust, contributing disproportionately to overall GDP growth despite representing just over a quarter of economic output.
Federal Reserve officials have explored alternative scenarios for AI adoption. In a “gradual adoption” path, AI diffuses through firms over many years, boosting productivity and spawning new products, services, and business models, much as earlier general-purpose technologies like the internet and electricity did. Employment shifts occur, but the creation of complementary roles, retraining, and rising demand for AI-enabled services offset much of the displacement 1. In a “jobless boom” scenario, productivity growth is strong but heavily concentrated in capital and a narrow set of high-skill workers, while many others become underemployed or leave the labour force, increasing inequality and straining social safety nets.
The debate around an AI-driven jobs apocalypse often reflects confusion between these scenarios and the time scales involved. On a multi-decade horizon, automation clearly has the technical potential to perform a vast array of tasks currently done by humans. Studies from major banks and consultancies estimate that hundreds of millions of full-time equivalent roles worldwide could, in principle, be automated, and that a significant share of workers will need to change occupations by the 2030s 3. However, technical feasibility is only one component of labour-market outcomes. Adoption costs, regulation, organisational inertia, consumer preferences, and the discovery of new uses for human labour in an AI-rich environment all influence the realised trajectory.
From a modelling perspective, one way to frame this is to consider the demand for labour L as a function of output Y, real wage w, and an automation parameter \t\th\eta that captures the cost and capability of AI systems. A stylised relationship might be written as L = f(Y,w,\t\th\eta), where \frac{\partial L}{\partial \t\th\eta} \lt 0 reflects direct substitution (AI performing tasks once done by labour) and \frac{\partial Y}{\partial \t\th\eta} \gt 0 captures productivity-driven growth that can raise overall labour demand. Whether aggregate employment rises or falls as \t\th\eta increases depends on the relative magnitudes of these effects and how income gains are distributed.
Empirically, the United States has so far exhibited a pattern where AI raises Y, compresses demand for certain types of L (notably lower-experience knowledge workers), and boosts demand for complementary skills. Wage data from AI-exposed industries suggests that where workers have scarce expertise and can leverage AI, their marginal product – and thus their compensation – increases. Conversely, where tasks are routine and easily codified, workers face stronger downward pressure on both employment and bargaining power. This tilt suggests a reallocation rather than an absolute collapse of labour demand.
The institutional and policy environment will heavily influence how far this reallocation becomes socially and politically sustainable. If firms and governments invest substantially in reskilling, supporting workers through transitions, and expanding sectors where human qualities such as empathy, creativity, and complex coordination remain crucial, AI could become a net positive for employment quality and economic dynamism. If not, the same forces could deepen regional and educational divides, even if headline unemployment data looks benign.
Large financial institutions sit at a delicate intersection of these dynamics. They are both heavy users of AI and key intermediaries of capital to other sectors. When leaders at such firms argue that disruption does not equate to collapse, they are also signalling how they plan to operate: using AI to strip out back-office friction, compress execution times, and enhance risk management, while betting that demand for human-intensive advisory work, complex deal-making, and relationship-driven services will remain strong. That strategic stance both reflects and shapes wider market expectations.
Inside these organisations, AI is already altering workflows. In investment banking, analysts use tools to screen large datasets for comparable transactions, generate first-draft pitch materials, and run scenario analyses in minutes rather than days. In sales and trading, AI helps optimise order routing, detect anomalies, and personalise client communication. In risk and compliance, models scan documents, transactions, and communications for patterns that warrant human review. The result is not an immediate disappearance of jobs, but a shift in what a “productive” banker or trader looks like. Capacity to collaborate with tools, interrogate outputs, and manage exceptions becomes central.
Many of these changes are incremental rather than headline-grabbing. A team that previously needed ten analysts might now deliver similar output with eight, while the remaining analysts handle more complex mandates or cover more clients. Over time, such efficiency gains compound, allowing firms to grow revenue faster than headcount. This is precisely the pattern implicit in arguments that the economy can keep creating jobs even as AI spreads: the composition of employment shifts, and the link between revenue growth and payroll growth loosens, but aggregate job numbers can remain resilient if new activities and markets expand sufficiently.
Critics challenge this optimistic interpretation on several fronts. First, they argue that the speed of AI progress and deployment could outpace the economy’s capacity to generate new labour-intensive sectors. Unlike previous technologies that took decades to move from labs to widespread use, generative AI tools reached hundreds of millions of users in a matter of months. If the pace of task automation accelerates faster than skill formation and sectoral adjustment, frictional displacement could become structural. Second, they note that the distribution of gains has already been skewed towards capital and high-skill labour, and see little automatic reason for that pattern to reverse.
Another concern is that many new roles created by AI are either highly specialised technical occupations, such as machine-learning engineers and AI safety specialists, or precarious gig-style work, such as data labelling and content moderation. If the bulk of new jobs fall into these categories, they may not fully substitute for the quality of lost mid-skill roles in manufacturing, clerical work, or routine professional services. Without deliberate policy and corporate choices to foster middle-earning, stable occupations in AI-augmented sectors, the labour market could bifurcate further.
Supporters of a more sanguine view counter that some of the most important future jobs are not obvious ex ante. Few people in the 1990s anticipated the scale of employment in digital marketing, app development, or e-commerce logistics, which only became large employers after complementary technologies and consumer habits matured. They expect a similar pattern with AI: new forms of personalised education, healthcare navigation, creative production, and human-AI collaboration services could absorb significant labour, even if those roles are hard to specify today. From this perspective, maintaining flexible labour markets, robust entrepreneurship, and open capital access becomes as important as any single retraining programme.
Over the next decade, the most plausible outcome for the United States may sit between complacent optimism and apocalyptic pessimism. AI will likely intensify competitive pressure on routine cognitive work, raising hurdles for young entrants and mid-career workers in automatable roles. At the same time, continued economic expansion in AI-augmented sectors, combined with demographic trends and policy responses, could keep overall unemployment within historical ranges. Whether that constitutes success will depend on how broadly the benefits of AI-driven productivity are shared and how effectively those facing disruption are helped to transition.
For investors, policy-makers, and workers, the key is to recognise that disruption and job creation can coexist for extended periods. Tracking only job cuts or only headline employment numbers gives a distorted view. The real story lies in the churn within occupations, the evolution of wage structures, the flow of capital into new business models, and the institutional capacity to manage transitions. Artificial intelligence will unquestionably reshape the labour market; whether it does so within the pattern of creative destruction the US economy has historically managed, or pushes it into uncharted territory, depends on choices being made now in boardrooms, classrooms, and legislatures.
References
1 New York Times opinion essay by David Solomon, “I’m the C.E.O. of Goldman Sachs. The A.I. Job Apocalypse Is Overblown.”
2 Business Insider coverage of David Solomon’s comments on AI, productivity, and “high-value” employees at Goldman Sachs.
3 Investor-focused analysis pieces on the prospect of an AI jobs apocalypse and estimates from Goldman Sachs, McKinsey, OpenAI, Citi, and CEO surveys.
4 Business leadership commentary on AI as a growth catalyst rather than a driver of mass job losses.
References
1. “I’m the C.E.O. of Goldman Sachs. The A.I. Job Apocalypse Is Overblown.” – https://www.nytimes.com/2026/05/22/opinion/ai-job-crisis-goldman-sachs.html
2. Is AI Truly Causing a Jobs Apocalypse, or Is the Narrative Overblown – 2026-05-20 – https://www.kavout.com/market-lens/is-ai-truly-causing-a-jobs-apocalypse-or-is-the-narrative-overblown
3. David Solomon Says AI Means Goldman Needs ‘More High-Value … – 2025-10-27 – https://www.businessinsider.com/david-solomon-ai-goldman-sachs-high-value-people-2025-10
4. How to Best Prepare for an Impending AI Jobs Apocalypse – 2025-02-11 – https://investorplace.com/hypergrowthinvesting/2025/02/how-to-best-prepare-for-an-impending-ai-jobs-apocalypse/
5. How does Goldman Sachs’ CEO Think AI will Impact Hiring? – 2026-01-23 – https://businesschief.com/articles/how-does-goldman-sachs-ceo-think-ai-will-impact-hiring
6. AWS CEO Is Pushing Back on AI Job Apocalypse Warnings | WSJ – 2026-05-18 – https://www.youtube.com/watch?v=DHqPVyQVr-A&vl=en
7. I’m the C.E.O. of Goldman Sachs. The A.I. Job Apocalypse Is … – Blind – 2026-05-22 – https://www.teamblind.com/post/im-the-ceo-of-goldman-sachs-the-ai-job-apocalypse-is-overblown-7ki6pgtp
