Khullar does not deny that clinical roles will change, or that some specific jobs may disappear as machines take over discrete tasks. His claim is subtler and, in some ways, more provocative: that the aggregate effect of AI on clinical employment is likely to be expansionary rather than contractionary. “But economic theory and history offer distinct reasons to believe that in the long run, adoption of AI agents—even models capable of performing some forms of cognitive work—could lead to an expansion, rather than a contraction, of the clinical workforce,” he wrote. The argument rests not on optimism about technology for its own sake, but on well-documented patterns in how efficiency gains reshape demand for human labor.

The first pillar of that argument is a phenomenon economists call Jevons paradox. In its original nineteenth-century formulation, the paradox described how improvements in the efficiency of coal use in Britain led to greater, not lesser, total coal consumption, because cheaper energy unlocked new uses across the economy. Khullar points to two surgical examples that illustrate the same dynamic in modern medicine: cataract surgery and joint replacement. Both procedures have become dramatically more efficient over recent decades, reducing the clinical effort required per operation, shortening recovery times, and improving safety. The result was not a shrinking demand for surgeons and operating room staff. Instead, as the procedures became faster, safer, and less burdensome for patients, far more people became candidates for them, and the total volume of care—and the workforce needed to deliver it—grew.

Khullar reasons that AI may exert a similar effect across other health care services, particularly if AI systems are priced near their marginal cost and thereby reduce the overall expense of delivering care. When the cost of a service falls, demand tends to rise, and health care is no exception. An AI-assisted diagnostic process that cuts the time a physician spends per case does not necessarily eliminate physician jobs; it can make each consultation cheaper and more accessible, drawing in patients who previously went unserved or underserved. In a country where millions of people lack adequate access to primary care, mental health services, and specialty evaluations, the latent demand for clinical attention is enormous. Efficiency, in this framing, is the mechanism by which unmet need becomes delivered care—and delivered care requires professionals.

The second pillar of the argument targets a persistent cognitive trap in workforce forecasting: the “lump of labor” fallacy. This is the mistaken assumption that there is a fixed quantity of work in an economy, so that any task performed by a machine must subtract from the pool available to humans. History repeatedly contradicts this view. The type and amount of work change over time in response to technological advances, and medicine is perhaps the clearest example. The discovery of antibiotics did not put physicians out of work; it created entire fields of infectious disease practice, intensive care, and public health. Imaging technologies did not eliminate doctors; they created radiology as a discipline and multiplied the conditions that could be identified and treated.

Khullar extends this logic to AI, arguing that the technology may allow clinicians to prevent or treat conditions in ways that are not currently possible or even imaginable. “AI may increase the demand for new professional capabilities by creating new treatments, care modes, and opportunities for specialization,” he wrote. If AI accelerates drug discovery, enables earlier detection of disease, or supports novel modes of remote and continuous care, each of those advances generates new clinical questions, new treatment pathways, and new roles for trained professionals to supervise, interpret, and integrate the technology into patient care. The workforce of an AI-rich health system may look different from today’s, but different does not mean smaller.

The third pillar addresses the granular structure of medical work itself. Delivering high-quality care involves far more than successfully executing a list of discrete tasks. The practice of medicine depends on numerous skills and many interconnected steps, from interpreting test results to developing and negotiating treatment plans with patients who have competing priorities, comorbidities, and preferences. Khullar invokes what economists call “O-ring theory,” a framework named for the faulty rubber seals that caused the Space Shuttle Challenger disaster. The theory holds that when production consists of many sequential steps, a single error at any step can seriously undermine the final outcome, and that the value of reliability at every stage rises accordingly.

That insight has particular force in health care, where many situations are high-stakes and the cost of a single catastrophic error is measured in lives, liability, and public trust. Even if an AI system performs individual cognitive tasks with superhuman accuracy, the chain of care that surrounds those tasks—verification, contextual judgment, communication, accountability, and the human relationship at the center of medicine—will continue to require clinicians, for reasons of both safety and trust. The practical implication is a distinction that Khullar emphasizes throughout the Perspective: automating tasks does not necessarily mean automating jobs. “In fact, if AI automates some clinical tasks, the value of nonautomated, human tasks may increase,” he said. As the routine components of a clinician’s day are handled by machines, the residual human components—those requiring empathy, negotiation, ethical judgment, and accountability—may become more valuable, not less.

The technical literature on automation in other industries offers a consistent picture of this dynamic. When automation decomposes a job, it typically reallocates human effort toward the tasks that machines handle worst: those involving ambiguity, dexterity, interpersonal connection, and responsibility for outcomes under uncertainty. In medicine, where the product is not a widget but a therapeutic relationship embedded in a complex biological and social context, the share of work that resists full automation is unusually large. An AI agent may draft a differential diagnosis or summarize a chart in seconds, but deciding how to present a terminal diagnosis, weighing whether an elderly patient with multiple conditions should undergo surgery, or earning the trust of a patient hesitant to accept treatment are tasks that remain stubbornly, and perhaps permanently, human.

None of this means the transition will be painless or evenly distributed. Khullar acknowledges that some types of health care jobs could be replaced, and clinicians in highly exposed specialties are right to scrutinize how their daily work maps onto what current AI systems can do. The composition of the workforce may shift, with new roles emerging around AI oversight, data quality, and hybrid human-machine care models, while some traditional task bundles are redefined. But the central claim of the Perspective is that the standard doom-laden forecast rests on flawed assumptions: that demand for care is fixed, that efficiency reduces total employment, and that a job is nothing more than the sum of its automatable parts. Economic theory and the historical record of medical technology suggest otherwise. If AI makes care cheaper, safer, and more capable, the United States may need more clinicians than ever—not to compete with the machines, but to do the human work the machines make possible.

Subject of Research: The potential impact of artificial intelligence on the future size and composition of the clinical workforce in the United States.

Article Title: How will AI impact the future of the clinical workforce?

Article References: How will AI impact the future of the clinical workforce?. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: artificial intelligence, clinical workforce, health policy, Jevons paradox, lump of labor fallacy, O-ring theory, New England Journal of Medicine, Weill Cornell Medicine, health care economics, medical automation, radiology, primary care