Why the AI Job Boom Looks a Lot Like Past Tech Hype
Peter Cappelli, the George W. Taylor professor of management at Wharton, believes today’s artificial-intelligence frenzy is repeating a familiar mistake. He points to the mid-2010s, when experts confidently predicted that self-driving trucks would soon eliminate truck drivers. In practice, that never happened.
“The predictions ignored basic realities,” Cappelli said. “Trucks still need fuel, deliveries still need to be made, and when problems arise, a human has to step in. Once you add a human, the whole cost-saving logic starts to collapse.”
That same gap between what is technically possible and what is economically practical now defines much of the AI conversation. Cappelli, who recently partnered with Accenture to study AI’s impact on jobs, argues that technology vendors and AI boosters focus on what systems can do in theory, not what it takes to deploy them in real organizations.
AI Works—but It Is Not Cheap
Cappelli highlights a Harvard Business Review case study on Ricoh, an insurance claims processor, to show how AI actually plays out inside companies. Claims processing is exactly the kind of repetitive administrative work AI is supposed to automate. Yet the first attempt was startlingly expensive.
Large language models could perform the task, but at roughly three times the cost of human workers. Ricoh spent about $500,000 on outside consultants and needed a year-long project team of six just to make the system operational. Even after optimization, the company was paying around $200,000 per month in AI fees—more than it had previously spent on payroll for that work.
The result was not mass layoffs. Ricoh reduced headcount only from 44 to 39 people. Humans were still required to resolve exceptions, chase down missing information, and fix problems created by AI output. What changed was productivity: the unit became about three times more efficient.
Ricoh executives say the project did pay off. It reached break-even in under a year and cut overall costs by roughly 15%, even without major job reductions. Employees now spend less time on repetitive work and more time on judgment-heavy tasks like quality control and customer service.
The Real Cost Is Organizational Change
Cappelli saw similar patterns at firms such as Mastercard, Royal Bank of Scotland, and Jabil. Productivity improved, but only after large investments of time, money, and organizational effort. He found little evidence of immediate job cuts.
What is missing from most boardroom conversations, he argues, is recognition that AI is not just a technology project—it is a massive organizational change initiative. Workflows must be mapped, jobs broken into tasks, and employees trained to work alongside AI systems. The people who already do the work must be involved, because they understand where processes fail.
Instead, many executives are engaging in what critics call “AI washing”—announcing AI initiatives for appearances rather than substance. A Harris Poll in early 2025 found that 74% of CEOs feared losing their jobs if they could not show AI progress, and 35% admitted that their initiatives were mostly for optics.
Cappelli believes this pressure leads to rushed, performative deployments that ignore the hard work required to make AI actually productive. Markets reward the appearance of transformation, even when little has changed.
A Slow, Expensive Reality
Cappelli predicts a slow learning curve. As CFOs confront the real costs of building and maintaining AI systems, enthusiasm will become more cautious. Unlike past decades, when labor was cheap and organizations could grow easily, AI requires sustained investment in people, processes, and management.
The bottom line is not that AI will fail. It is that meaningful success will look far less dramatic than the hype suggests. Productivity will rise, but slowly. Jobs will change more than they disappear. And the biggest challenge will not be the software—it will be the organization.