We’re convinced that AI holds the promise of true abundance. But, alas, it isn’t as simple as that.

For the past 150 years, the single most important driver of economic growth has been productivity change. AI now offers the potential to turbocharge the process, increasing the pace to rates well beyond historical norms. Indeed, a study on customer support services found that AI assistance rapidly boosted worker productivity by an average of 15 percent.

That’s just one study, of course. But many experts following AI think it is indicative of where we’re headed. Estimates of the cumulative impact on GDP in the next decade run as high as 7 percent, adding trillions to national income. And as researchers at the Federal Reserve Bank of Dallas note, the potential for change is virtually limitless. In a benign future scenario in which generative AI surpasses human intelligence, machines may eventually produce everything — effectively solving the fundamental economic problem of scarcity. 

But here’s the rub — well, one rub. If history and economic projections are to be believed, the gains from AI will not be evenly distributed. AI could thus pose challenges similar to those created in the past few decades by the digital economy, including severe near-term dislocations in job markets and widening income inequality.

The Emerging Consensus

To be sure, some economists remain skeptical about AI’s potential for transformative impact on both productivity and jobs, pointing to the limited effects on hiring and unemployment rates thus far. A recent study by economists from the Budget Lab at Yale found that AI has not caused changes in the labor market that would differ dramatically from those resulting from past major technological advancements — everything from the telephone to the internal combustion engine. This leads to the conclusion that while AI may be rapidly changing the way we work, widespread effects are likely to take several years to materialize, if they occur at all.

However, other sources suggest early signs of major disruption, particularly among entry-level and early-career workers, are already apparent. Between October 2024 and October 2025, new job postings on Indeed fell by more than 13 index points (from a base in February 2020 = 100). While the number of job postings rebounded slightly in November, the data for December showed another drop, hinting at a secular trend in the labor market. Admittedly, multiple factors may be at work here. However, AI’s influence is evident beyond job boards.

A recent Stanford study found that workers aged 22 to 25 in the occupations most exposed to generative AIexperienced a 16 percent drop in employment compared to both their peers in less-exposed fields and to experienced workers in their own occupations. Such findings lead to an emerging near-consensus that unemployment could rise by about 10 percent as a result of the broad diffusion of AI.

There’s a long history of rapid sectoral job displacement due to technological change ranging from textile workers in the early 19th century to bank tellers in the 21st. But unlike past technological shifts, AI appears to disproportionately affect white-collar roles. Research by JP Morgan concludes that non-routine cognitive workers — think scientists, engineers, and lawyers — face rising unemployment risks. The research further warns that “a much larger unemployment risk and anemic recovery prospects for workers in non-routine cognitive occupations might cause the next labor market downturn to look pretty dismal.”

A Microsoft study reinforces this conclusion, finding that knowledge work and communication-focused roles are most susceptible to AI integration. This includes occupations that have, until very recently, been seen as well-insulated from technological disruption — most notably, software development.

Black Swan Scenarios

Some experts predict even worse outcomes. In a recent 60 Minutes interview, Anthropic CEO Dario Amodei cautioned that AI could drive unemployment up to 10 to 20 percent within a decade. Other economists echo these concerns, projecting unemployment spikes of up to 20 percent.

History offers cautionary parallels. The rise of dominant digital platforms in travel, retail and entertainment coincided with steep declines in incumbent businesses, particularly small and local operators. Companies such as Uber, Amazon and Netflix leveraged vertical integration and “blitzscaling” strategies to dominate multiple industries, often initially prioritizing market share over profits. This left traditional workers — such as taxi drivers — jobless or trapped in insecure, low-wage roles.

Could AI have similar effects on white-collar professions? Or would productivity gains offset these risks by boosting wages in remaining jobs and creating whole new job categories — as conventional wisdom holds was true in the past? Doesn’t history show that technological progress ultimately benefits society, with only short-term pain for a few occupations?

Perhaps. But as Amodei notes, AI’s pace of change is likely faster than any previous technological revolution. Such rapid change could create struggles in the labor market, especially because — as Nobel laureates Daron Acemoglu and Simon Johnson argue — automation raises wages only when the new technology raises the marginal productivity of labor and labor markets are sufficiently competitive. 

Moreover, structural factors in AI-adjacent markets amplify the risk that gains will be concentrated. As Susan Athey and coauthors argue in a recent National Bureau of Economic Research paper, while AI can deliver broad welfare gains as measured in output per capita, “monopoly power in AI depresses welfare and can harm both skilled and unskilled workers.” This is troubling given today’s market concentration. NVIDIA controls over 90 percent of data center graphics processing units, while OpenAI, Microsoft and Google account for more than 85 percent of the generative AI chatbot market, and Microsoft, Amazon and Google dominate AI cloud services.

• • •

The message here ought to be plain. Unless we prepare for it, AI-driven disruption could be severe. And current policies — or rather, lack of policies — leave society highly vulnerable to these risks.