In a new essay published Tuesday, Andreessen Horowitz General Partner David George dismissed fears of an “AI job apocalypse” as a “complete fantasy”—labeling it “unhelpful marketing, bad economics and worse history. ” He argued the concern stems from a logical error known as the “lump-of-labor fallacy, ” long debunked by economists, which wrongly assumes the economy has a fixed amount of work, so any automation or AI that takes over tasks must reduce human jobs. This essay is the most comprehensive articulation yet of a view the firm’s co-founders have expressed for months. Ben Horowitz noted on an earlier podcast that despite advances in AI since at least 2012 (starting with ImageNet revolutionizing computer vision), catastrophic job losses have not materialized. The core argument centers on the “lump-of-labor fallacy. ” George explains that human desires and needs are not static: when technology reduces costs of activities, people create new wants and jobs. John Maynard Keynes predicted nearly a century ago that automation would yield a 15-hour workweek, but people instead reinvented new activities and work. George cited historical examples: farm mechanization cut U. S. farm employment by a third in the early 20th century, but displaced workers moved into factories, offices, hospitals, and eventually software, while agricultural productivity soared. Electrification reshaped factories and doubled labor productivity, rather than destroying jobs. The spreadsheet, thought to kill bookkeeping jobs, instead created 1. 5 million more financial analyst roles than the million bookkeepers lost. Supporting this, Apollo Global Management Chief Economist Torsten Slok popularized the “Jevons Paradox, ” which says that reducing technology costs leads to increased demand and job creation. For instance, Microsoft Excel lowered costs of financial analysis, making such services accessible to many new businesses, thus increasing jobs. George noted that falling input costs, such as cheap fossil fuels, historically have led to expanded economic activity and new industries, like plastics, not job losses. Anthropic CEO Dario Amodei recently cited the same paradox when unveiling labor-impacting AI tools for Wall Street. Andreessen Horowitz also supports their historical and theoretical case with recent data. Various economic studies contradict doomsday predictions: An NBER working paper found AI adoption hasn’t significantly changed total employment; the Federal Reserve Bank of Atlanta found over 90% of firms saw no employment impact from AI over three years; a Census Bureau study showed modest employment changes balanced between increases and decreases; and the Yale Budget Lab concluded AI’s labor-market impact remains largely stable. The only exception was a Stanford study showing a 16% relative drop in employment among early-career workers (ages 22–25) in AI-exposed jobs since ChatGPT’s late 2022 release, though a16z argues this is complex, with some entry-level jobs increasing where AI is augmenting or neutral. Despite a16z’s strong case, notable critics dispute its premises. Economist Anton Korinek warns that if artificial general intelligence (AGI) is achieved, labor could become optional, unlike previous industrial revolutions.

The Carnegie Endowment categorized AI debate participants as the “alarmed, ” the “patient, ” and the “excited, ” placing a16z in the “excited” camp alongside co-founder Marc Andreessen. The “alarmed” and “excited” camps differ not only in facts but also in predictions about AI progress speed, firm adoption capacity, and new job emergence pace. What sets the current moment apart, critics say, is velocity. The alarmed fear rapid AI improvements, boosted by scaling laws and massive investment, could outpace historical precedents. OpenAI’s GDPVal benchmark found new AI models outperform humans on many tasks, with experts preferring AI responses 83% of the time in tested areas. Conversely, the “patient” camp, including Princeton computer scientists Arvind Narayanan and Sayash Kapoor, Nobel laureate Daron Acemoglu, and cognitive scientist Gary Marcus, argue that AI’s limitations, hallucinations, and integration complexities mean adoption will unfold over decades. Scale AI’s Remote Labor Index found the best AI systems in March 2026 could complete just 2. 5% of complex freelancer-level tasks to human standards, with minimal improvement since. Economist David Autor offers a nuanced “conditional optimist” stance: AI can help restore middle-skill jobs but cautions this is not a firm forecast, only a statement of possibility. Andreessen Horowitz’s optimism aligns with its financial interests—having invested billions across AI startups and infrastructure—making a cultural consensus that AI destroys jobs unfavorable because it invites regulation, slows adoption, and weakens consumer confidence critical for its portfolio companies. However, a conflict of interest doesn’t invalidate their use of real historical records and academic research. Carnegie notes most economists expect AI to produce only modest deviations from historical trends unless rapid capability growth triggers severe disruption. A crucial point a16z underemphasizes is the asymmetric risk if they are wrong. If optimistic, the labor market will reorganize as before, creating new roles; but if the alarmists prove correct and policy is shaped by overconfidence, millions of displaced workers may face inadequate support and retraining infrastructure. Paradoxically, the Yale Budget Lab observed that AI-driven productivity gains, while addressing a $39 trillion national debt, could simultaneously cause widespread worker displacement, exacerbating inequality. Public perception reflects rising anxiety: a March Quinnipiac survey found 70% of Americans now believe AI will reduce human job opportunities, up from 56% the previous year. Whether this fear is misguided or a true intuition about unprecedented change remains an open question that historical analogy alone cannot resolve. Andreessen Horowitz declined to comment on the essay.