AI and jobs: Still in an ATM phase

Commentary by Adam Schickling, Vanguard Senior Economist.

On our “Market views” tab: Rotation away from large-cap tech accelerates

Every wave of technological change seems to arrive with a familiar prediction: This time, jobs are going away for good. The effect of automated teller machines (ATMs) on the bank teller profession reveals a more nuanced reality.

When ATMs became widespread in the 1980s, many people assumed bank tellers would soon become obsolete. They were only partially correct.

The number of tellers needed at individual branches did decline to some degree as ATMs automated routine tasks. Yet the broader employment outcome was less stark than feared. By lowering operating costs, ATMs made it economical for banks to open more branches. As a result, total U.S. bank teller employment remained broadly stable from 1980 through 2010. For a mid-career teller in the 1980s, the ATM posed far less of a threat to employment than many forecasts suggested.

In fact, just as we expect artificial intelligence to transform the labor market, the expansion of retail banking created demand for a wider range of occupations. Banks hired more loan officers, credit analysts, personal bankers, and fraud and risk specialists. The work performed inside a branch moved up the skill-value chain. Branches became less about processing transactions and more about managing customer relationships.

The real disruption came later. Beginning around 2010, mobile banking changed the equation. Unlike the ATM, which automated a task, mobile banking largely automated the entire trip to a bank. Customers no longer needed to visit a branch for many everyday banking activities. By 2025, only 9% of bank customers said branches were their primary banking channel, compared with 36% in 2007.¹ Bank teller employment fell accordingly.

Importantly, this transformation was not driven by technology alone. The Electronic Signatures in Global and National Commerce Act of 2000 gave electronic signatures the same legal standing as ink signatures, helping to enable fully digital banking experiences and accelerate the shift away from in-person transactions.

The lesson is that isolated task automation rarely results in large-scale job losses, except in occupations built around a very narrow set of activities. (There aren’t many switchboard operators left.) More often, meaningful disruption occurs when technologies are combined with new workflows, business models, and institutional changes that fundamentally alter how work is organized.

The disruption caused by mobile banking included the creation of entirely new forms of employment: cybersecurity analysts, digital product managers, payment-platform engineers, and data-platform operators.

This history offers a useful lens for understanding today’s debate around AI. If AI becomes a general-purpose technology like electricity and the personal computer before it—as developments increasingly suggest—it will enable products, services, and industries that we have not yet envisioned. In short, fears of widespread job loss are likely overblown.

Since ChatGPT’s arrival in late 2022, many people have argued that AI will quickly eliminate large numbers of white-collar jobs. Nearly four years later, the labor market tells a different story. Occupations with the greatest exposure to AI have not experienced widespread employment declines. Employment growth in highly exposed occupations has generally kept pace with—or exceeded—that of less exposed occupations. Layoff rates remain low, and although hiring has slowed, the slowdown has been broad-based rather than concentrated in AI-intensive fields.

Today’s large language models may be reminiscent of the ATMs of the 1980s—powerful tools that automate certain tasks but augment many more, making workers more productive and leaving the broader structure of work largely intact. More significant labor market disruption may require something closer to the shift from ATMs to mobile banking: a deeper reconfiguration of business processes, organizational structures, and customer interactions that reshapes the role of workers rather than removing them from the equation.

The history of technological change suggests capabilities alone rarely determine employment outcomes. What matters more is how organizations redesign work around those capabilities.

AI may ultimately transform the labor market, just as mobile banking transformed retail banking. But the evidence today suggests we remain closer to the ATM phase than the mobile banking phase.