Bank of America's global research team released a report April 28 rejecting the narrative of an AI-driven jobs apocalypse, grounding the argument in 85 years of labor market data.
The core claim: 60 percent of existing U.S. jobs today did not exist in 1940. Data scientists, social media managers and cloud developers—professions barely present two decades ago—are now mainstream. Agriculture employed roughly 40 percent of Americans in the early 1900s; today it accounts for 1 percent.
The bank's economists distinguish sharply between exposure and elimination. Roughly 840 million jobs globally, or about one in four, face some generative AI exposure. But the granular breakdown matters: only 2.3 percent of global jobs possess genuine automation potential. Another 13 percent fall into an "augmentation" category, where AI enhances rather than replaces workers.
Historical precedent supports this bifurcation. When automated teller machines proliferated in the 1970s and 1980s, lower operating costs enabled banks to open more branches. Tellers redeployed into sales and customer service roles. Total teller employment rose. Word processors shifted clerical workers toward coordination work. Excel expanded accounting departments. E-commerce did not gut retail employment—the U.S. maintains 15 to 16 million retail workers today, comparable to the 1990s.
The disruption will not be evenly distributed. Professional and financial services sectors stand to benefit most from AI integration. Repetitive roles in customer service, IT and administration face the highest substitution risk. Younger workers, women and highly educated individuals—concentrated in white-collar, language-intensive roles—face the steepest exposure.
Economist David Oks has challenged the ATM parable in a widely read analysis, arguing the story tells only half the truth, particularly since the early 2000s when check-uploading capabilities accelerated branch consolidation. The debate over whether historical job-creation patterns will hold in the AI era remains unresolved.
