The real lesson from history and economics is that AI will create more demand for human expertise, not eliminate it.
The Radiologist That Wasn't Replaced
In 2016, AI pioneer Geoffrey Hinton declared that people should stop training radiologists; deep learning would soon outperform them. Almost a decade later, demand for radiologists is at an all-time high. Dozens of AI tools can now detect diseases faster and more accurately than humans, yet radiologists haven't vanished. Why?
Part of the answer lies in regulation and liability—doctors still need to be in the loop. But the deeper reason is economic: cheaper scans mean more scans, and more scans mean more demand for complex diagnosis and treatment planning. Technology didn't replace the job; it expanded the market for it.
Jevons Paradox: The Engine of Hidden Demand
This phenomenon has a name: Jevons paradox. In the 1860s, economist William Stanley Jevons noticed that more efficient coal-burning engines led to more coal consumption, not less. The same logic applies to virtually every transformative technology. When containerization made shipping 90% cheaper in the 1960s, some dockworkers lost jobs, but global trade exploded, giving rise to entire industries in freight forwarding, logistics, and warehouse distribution.
Cloud computing followed the same arc. Infrastructure costs dropped tenfold in the 2010s, and traditional IT roles evolved into DevOps engineering and cloud architecture—jobs that manage systems at scales that previously didn't exist. More recently, as algorithmic improvements slashed the cost of AI inference, demand for GPUs skyrocketed, pushing Nvidia's stock to record highs. Efficiency gains reveal pent-up demand we didn't know we had.
What This Means for White-Collar Work
If the pattern holds, AI will make it cheaper and faster to analyze MRIs, draft legal documents, and write code. The result won't be fewer radiologists, lawyers, or engineers—it will be more demand for their expertise. Aaron Levie, CEO of Box, recently made the point bluntly: when the cost of work goes down, the demand for it goes up, and there is usually far more latent demand than we realize.
This does not mean jobs will stay identical. Many roles will shift from manual execution to supervising teams of AI agents. The nuclear engineer who once calculated safety margins by hand now runs simulations. The radiologist will review AI suggestions and handle edge cases. The job changes, but the human remains central.
Rote Tasks Disappear, Interesting Work Expands
Andre Karpathy, a co-founder of OpenAI, argues that AI will first transform jobs that are rote, require little context, and are forgiving of mistakes—think customer service and data entry. But even those jobs will often be refactored into manager or supervisor roles rather than vanishing entirely. At Y Combinator–backed companies, this is already happening. AOKA, an AI sales agent for plumbing and HVAC, frees customer service agents to handle higher-value client relationships. Tenor automates healthcare paperwork, turning admin staff into patient care coordinators.
The tasks AI takes over—impatient callers, repetitive forms—are precisely the parts of jobs that people dislike. The result is more engaging work, not unemployment. The internet decimated travel agents but created millions of digital marketing jobs. AI will produce a similar reallocation.
The Real Takeaway for Founders
If you are thinking about starting a startup, the evidence points toward a middle path. Do not be like economist Paul Krugman, who in 1998 compared the internet to a fax machine—underestimating the change is a dangerous mistake. But do not indulge in fantasies of fully automated luxury communism or imminent mass unemployment. AI is as big as or bigger than the internet; the future is being built right now by people who see what others miss.
The only real question is whether you will be one of them. Every great company starts with a founder who takes a leap based on conviction. The tools are here, the historical pattern is clear, and the demand is waiting to be unlocked. The permission slip is self-issued.