People Keep Saying There’s a Shortage of Nursing Jobs. The Truth Is Actually More Subtle and Disturbing.

Work Short-Staffed The nursing job crisis is raising an uncomfortable question: What happens when hospitals trade nurses for A.I.? Getty Images Plus Sign up for the Slatest to get the most insightful analysis, criticism, and advice out there, delivered to your inbox daily. Imagine, for a moment, that you are a nurse. A cancer patient on your ward asks for pain medicine, and it’s easy enough to get it ready: Just go to the locked narcotics cabinet, punch in a code, draw the drug up into a syringe, and give it to them. But before you can do that, the wife of a different very sick patient calls to ask how her husband is doing. After that conversation ends, a transporter unexpectedly arrives to take another patient to a needed scan. The transporter is not allowed to wait more than five minutes, so the patient has to be ready now. Next, a consulting endocrinologist stops by to explain that a diabetic patient’s insulin is being switched to better control his blood sugars because regular doses of steroids are sending them sky-high. Twenty minutes have passed by the time the patient finally gets his pain relief, and a family member in the room jokes, “Well, we can’t expect you to do anything in a hurry, can we?” Every nurse has moments like this, and I was no exception. The shame of being mocked has lingered, as has the anger of being expected to do several time-sensitive activities simultaneously. No wonder many nurses, myself included, struggle with constant feelings of failure. And no wonder that amid such a challenging work environment, the U.S. is having trouble filling nursing jobs. According to the Health Resources and Services Administration, the country will experience a shortage of 267,330 RNs by 2028. And by 2029, 40 percent of nurses—1.6 million people—intend to leave the profession. This has led some to declare a nursing shortage, but the truth is actually much more subtle and disturbing. There are still plenty of registered nurses in the United States, enough to fill most empty positions, but there’s a shortage of jobs they’re willing to take. Why? Because most nurses find those demands and clinical responsibility impossible and inhuman. National Nurses United, the country’s largest professional association of RNs, blames this on “the industry’s systematic failure to invest in safe, quality, human-to-human patient care.” In other words, “hospital management has created a staffing crisis.” Staffing costs money that hospital administrators would rather not spend on nurses, even though a recent analysis of New York Medicare patients found that appropriate nurse staffing would have saved $720 million over two years if most nurses had no more than four patients to care for, rather than the six, or even 10, that are typical. Such short staffing can result in working conditions that are not only untenable but fatal. Multiple studies have shown that when nurses are overworked, patients die who would not have otherwise. You’d think that hospitals would hang on to these hardworking nurses for dear life. Not so. Unlike physicians, nurses’ services are not billed to insurance companies (but are included in patients’ room charges). Because nurses don’t directly earn money for hospitals, eliminating their positions saves hospitals money. As a 2022 article in the Journal of Nursing clarifies, “Hospital systems have an economic incentive to keep their nursing staff as small as possible.” Into this vexed clinical environment comes the most inhuman solution of all, A.I., with its promise of increased efficiency and decreased labor costs. Analysts from the consulting firm McKinsey & Company estimate that A.I. adoption in U.S. healthcare could lead to savings of “roughly $200 billion to $360 billion annually.” It’s not unreasonable to assume that at least some of those savings would come from eliminating jobs. Whispers of a wider “A.I. job apocalypse” have even been hinted at by tech leaders themselves. Anthropic CEO Dario Amodei predicted that the tech could nuke half of white-collar jobs and push unemployment to 20 percent. (A.I. has already decimated tens of thousands of jobs at Target, Amazon, and other large companies.) As of now, it’s still unclear how much of a threat this is to nursing, which is not technically a white-collar profession. But some nurses have already lost their jobs to A.I. In July, 12 nurses at Montefiore hospital, in the Bronx, were laid off and replaced with A.I.-powered software. The nurses were responsible for utilization review, the process of evaluating patient records to make sure hospital admissions are appropriate and that patients’ bills get paid. I spoke with Montefiore nurse Shaiju Kalathil, a member of National Nurses United, who works in the same department as the fired nurses. Kalathil’s take on the use of A.I. to do nursing jobs surprised me. I thought he would protest it, but instead he insisted that the problem with bringing this tech into healthcare is that it’s simply too much of an unknown. “We don’t know what it can and cannot do,” he told me. “We cannot use our patients as guinea pigs.” In a Guardian article, a hospital spokesperson disagreed with the nurses’ view that they had been laid off in favor of A.I., but they also indirectly affirmed Kalathil’s concern. The hospital is “always investing in new technology,” the spokesperson asserted. New technology sounds great, but that doesn’t necessarily mean it’s been rigorously tested and well validated. A 2026 analysis of 1,357 A.I. medical devices cleared by the Food and Drug Administration found that only three had been tested on patient outcomes and just 34 were registered for prospective trials. “If AI in medicine is to be life-saving,” the authors wrote, “it must first be life-tested.” Julia Truelove, a nurse in D.C., works at a facility where this kind of spaghetti-at-the-wall experimentation left a bad taste in her mouth. As she explained to me, her hospital reportedly adopted a new A.I. sepsis-detection tool without consulting the nurses who respond to sepsis alerts. The tool alerted too often and on patients who were not septic, said Truelove. Similarly, in 2025, Adam Hart, a nurse at Dignity Health, in Henderson, Nevada, was told by A.I. to rapidly give intravenous fluids to a septic emergency department patient. That tends to be good clinical advice, except that the patient in question was on dialysis as a result of kidney failure, and flooding them with fluids could cause pulmonary edema (fluid in the lungs). The nurse got a physician to stop the order for IV fluids, but a less vigilant nurse, or one who was already overreliant on A.I., would probably not have caught the mistake. In fact, when Hart reported his concerns to his direct supervisor, he was told to follow the A.I. protocol. A hospital spokesperson supported Hart’s decision, stipulating that the technology is a tool “that supports, not supersedes, the expertise and judgment of our care teams.” Recognizing the expertise of human clinicians is important. A recent independent analysis in JAMA of an A.I. sepsis tool found that it performed worse than standard clinical practice, and more poorly than the company had claimed. The authors described an “underbelly” of company-generated claims about product precision “that may not accurately reflect real-world model performance.” Truelove reminded me that every medication and protocol used in healthcare is subjected to rigorous testing, but neither testing nor safety standards are currently required for healthcare A.I.—a concerning trend that will likely continue as the Trump administration makes it harder for states to regulate this new tech. In general, healthcare A.I. inhabits “a liminal regulatory space,” meaning that it falls outside existing laws that govern medical privacy, research, and clinical products, explains another recent JAMA article. “The sparseness of formal regulation contrasts starkly with AI’s potential to cause harm,” the authors wrote. At the very least, it’s creating confusion. Pa Vue, a triage nurse for a large California hospital system and a member of the California Nurses Association, told me that although her hospital officially “paused” the use of A.I. in call centers, it’s still listening in on employees and rating their empathy via a “score for sentiment.” It is beyond ironic to have A.I. evaluating human empathy, but Vue further explained that call center nurses are expected to be empathetic and fast. When I asked her how often those two mandates come into conflict, she answered, “All the time.” In addition to monitoring the nurses’ calls for empathy, the A.I. identifies when patients’ conversations cut into the nurses’ efficiency by “wandering,” Vue said. I’ve done many triage calls as a nurse, and I can confirm that the sicker patients are, the more rambling their stories tend to be. Illness, and the fear and stresses it brings, leads people to become poor communicators about what ails them. A nurse who stops a patient from “wandering” will not be empathetic, and an empathetic nurse will have to give patients time to get their stories out. In real life, empathy and efficiency tend to be mutually exclusive, even if in the world of A.I. they are not. It’s doubtful that an A.I. could ever replace the value of what nurses call “eyes on”: observing patients using clinical judgment that has been honed over time. I remember a patient, young and cognitively intact, to whom I gave medication. It’s a routine task, except that he raised his medicine cup full of pills up to his ear instead of his mouth. An A.I. would likely understand this action to be an error, but would it grasp just how serious an error this was? That the patient, who had been mentally clear, made a mistake no one of sound mind ever would? And could it communicate that concern as forcefully as I did, due to my own human worry for the patient? None of the urgent needs that I described at the start of this article could have been addressed by A.I., or at least would not have been addressed particularly well. Still, all of the nurses I spoke with made it clear that they support the use of technology—they just want it to be tested for safety and effectiveness before being deployed, and they want transparency as it’s rolled out. Truelove insisted that an even better option for guaranteeing the safety of A.I. would be to involve nurses in the process of evaluating this new technology. Hospitals could better reach their patient safety goals for sepsis and other issues simply by “asking people who do the work what is important and what might keep people safer,” she told me. A recent systematic review of A.I. helping nurses better manage chronic diseases found improved health outcomes for patients when nurses were clinical leaders. “Nurses should implement and lead the application of AI in clinical practice to improve patient-centered care,” it reads. Ultimately, A.I. that is simple and practical may prove to be the most valuable to nurses—and one of the things that make their jobs tolerable enough to keep. According to Becker’s Hospital Review, some hospitals are already using A.I. to better manage their supply chains, ensuring that essential equipment is always where it ought to be on hospital floors. Other A.I. is being used to get staff needed information faster. Any nurse who has desperately scoured a hospital ward for an oxygen connector or struggled to make sense of clinical protocols for high-risk medications like chemotherapy will appreciate A.I. helping make those aspects of the job easier. And, really, a slightly easier job—one that allows more time to listen to, monitor, and care for patients, that allows them to be more human—is all nurses want. Maybe then they’d stick around. Artificial Intelligence Economy Health Medicine Workplace Jobs