Why more AI may also mean more clinicians

Clinical takeaway: Automating clinical tasks does not necessarily mean automating clinical jobs. As AI absorbs documentation and analysis, the human work around it may gain value. 

Predictions that AI will one day thin the ranks of clinicians have moved from speculation to specifics. Models now outperform physicians on some complex clinical tasks, and federal health officials are moving quickly to deploy medical AI, including agents that offer therapy and prescribe medication. The specialties most often named as exposed include radiology, pathology, psychiatry, and primary care. 

The case for displacement was made bluntly in JAMA last month, in a piece arguing that AI alone may soon deliver better care than physicians or physician-led teams at core cognitive tasks. The standard rebuttals appeal to technology underdelivering outside benchmarks, regulation slowing deployment, and patients preferring human care. Writing in a New England Journal of Medicine Perspective, Weill Cornell Medicine physician and health policy researcher Dhruv Khullar grants those objections some merit but sets them aside for a different kind of argument. He argues that when medicine gets more efficient, more people get treated, and more hands are needed to treat them. 

Khullar builds first on the notion that technological advances which make a resource more efficient to use tend also to increase its use. Steam engines that burned less coal drove coal consumption up, and medicine has repeated the pattern. Cataract surgery and joint replacement became faster, safer, and easier to recover from. More patients got those procedures as a result. Cheaper genetic testing and MRI have coincided with wider use of both. Radiology is the sharpest version: a decade after predictions that the specialty would be automated away, the US has more radiologists than a decade ago. 

Then, he targets the assumption that there is a fixed amount of clinical work to divide between humans and machines. Demand already exceeds supply across parts of medicine; roughly one quarter of Americans live in primary care shortage areas. In the near term, Khullar argues, useful AI agents will likely extend clinicians rather than substitute for them. Longer term, the work itself likely changes. Early 20th-century physicians could not have foreseen radiologists, and midcentury oncologists could not have imagined managing immunotherapies. AI may create demand for clinician capabilities that do not yet exist. 

Finally, Khullar separates individual tasks from the actual jobs built around them. AI increasingly automates discrete clinical work such as visit documentation and ECG analysis, but practicing medicine involves many interdependent steps from interpreting results to negotiating treatment plans. An error at any step can undermine the whole. High-stakes care will keep clinicians in the loop supervising for safety and trust. If AI absorbs some tasks, the value of the human work around them may simply rise. Khullar does grant that this dynamic can cut the other way: some health care jobs may be replaced. Adoption of even superior autonomous systems can also stall on public discomfort, as it has with self-driving cars. 

Assumptions about the size and shape of the future clinical workforce shape decisions already in motion including training program capacity, payment policy, biomedical investment, and health care infrastructure funding, along with what the next generation of clinicians expects from the field. Khullar warns that predictions can be self-fulfilling. If people believe displacement is coming, they may act in ways that bring it about. 

"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," summed up Khullar, MD, MPP, an associate professor of population health sciences who is also a hospitalist at NewYork-Presbyterian/Weill Cornell Medical Center. 

Source: Khullar D. (2026 Sep 12) N Engl J Med. Artificial Intelligence and the Future of the Clinical Workforce