To some, this may look like a scene from another country, Baugh said. But to him, it’s just Thursday.

Baugh uses artificial intelligence, in the form of an ambient scribe, to assist in navigating the busy evening as he attends patients desperately waiting for help. Still, the technology is far from the panacea that developers claim it can be for what ails American health care.

The CEOs of some of the biggest AI companies have predicted the technology will “eliminate most cancer” and “improve treatment of most other ailments,” or perhaps “cure all disease” within the next 10 years. But nowhere better illustrates the challenge facing the technology than an emergency department. Brigham and Women’s own evaluation of the benefits of AI in the emergency room shows that it falls far short of the techno-optimist vision of AI acting as an efficiency-generating machine that can make abundant health care available for marginal cost.

While AI documentation may help the physician focus on the patient and figure out what’s wrong, scribes are doing “almost nothing” to improve the quagmire of challenges of delivering health care, said Graham Walker, an emergency physician in San Francisco and founder of health tech startups MDCalc and Offcall. He listed hallway patients, boarding, and the lack of skilled nursing facility beds as some of the ED challenges AI can’t do anything about.

AI could potentially help patients better understand what to do when they leave the hospital, act as translators, triage care, or act as scribes for patients. But those applications are currently all academic, Mass General Brigham ED executives told STAT, or face opposition from health systems.

“The idea that AI is just going to magically wave its wand and fix all of the problems of health care?” Walker said. “I would argue that there are probably more problems totally unrelated to AI in health care than there are problems that AI can fix in the next five years.”

The emergency room challenge

The emergency department is where all health care’s ills find a place to land. Patients without insurance, rising out-of-pocket costs, primary care shortages, homelessness, and an aging population with multiple chronic illnesses often mean more people showing up at ERs, urgently in need of help, where it’s almost always illegal to turn someone away.

That increased number of people increases wait times, which are long because there is no room in the emergency room. There’s no room in the emergency room because patients are “boarding,” i.e., stuck there because there are no beds in other parts of the hospital. There are no beds in the ICU, because there aren’t enough nurses to staff them. There aren’t enough nurses, because hospitals insist they don’t have the resources to hire that staff. And so on.

One of the most widely adopted uses of AI in medicine is the ambient AI scribe. Baugh is one of the early adopters of the technology, which the Brigham and Women’s emergency department started piloting in April 2024.

Emergency departments pose special problems for AI scribes. Unlike in outpatient settings such as primary care or orthopedics, the workflow in the emergency room is messy and fragmented — a complicated dance of residents, physician assistants, and nurses all working on the same patient at different times, ordering tests, and deducing what’s wrong.

It’s also noisy. Everyone spoke in muted tones that September night, but the emergency bay at Brigham and Women’s was still loud, punctuated with high-pitched and low-pitched beeps from monitors, the clanking metal of nurses popping bed guardrails up and down and opening and closing drawers — all overlapping with the voices of doctors, residents, and caregivers huddled around a patient in a bed in the middle of the hallway.

Just a few years ago, very few AI scribes were tooled for use in the emergency room. Now, the technology can handle multiple speakers giving the patient’s history, the high ambient noise level, and has been tuned to produce the kinds of clinical notes that are most useful in the emergency department, said Reid Conant, an emergency physician who works for Abridge, a leading vendor in the AI scribes market.

Baugh, who uses Abridge’s scribe, described it as a “wellness” tool. Thanks to the scribe, he gets his notes done within 24 hours of seeing a patient, rather than the typical 72 hours. The technology also lets him “untether” and be more available to staff during his shift, allowing him to do more teaching and interacting with patients — parts of his job he enjoys more than clinical documentation. He usually hits the button to make all of his patient recordings into notes as he’s walking out the door at 11 p.m.

The alternative, he said, is that “I’ll go home and try to do this at 1 a.m., or over the weekend.”

Though the AI scribe may be helping Baugh, a recent study in Mass General Brigham ERs suggests that the technology is not changing the status quo in the emergency room.

While the AI scribes saved doctors 1.6 minutes per note, and human scribes saved double that number, neither type of scribe increased how many patients a clinician can see in a shift, or how much money the hospital was able to collect for each patient. Though Baugh was part of a pilot with two different scribe products, MGB at the beginning of 2026 switched to using Microsoft’s AI scribe, a MGB spokesperson said.

Mass General Brigham’s community emergency medicine division chief Melisa Lai-Becker told STAT that there’s “something priceless” to relieving the cognitive load for an emergency physician, especially when it can decrease interruptions and let physicians focus on a task during a shift when they might have dozens of patients to care for. But it’s unclear where the actual time saved with the AI scribe is going. Those few minutes, the only measurable benefit of the technology, “are unfortunately being erased by all of the other roadblocks and bottlenecks of the system,” she said.

Does the AI benefit patients?

At Brigham and Women’s, Baugh has seen 15 patients in the first four hours of his shift, and he thinks that might go up to 25 or 35 by the time he goes home. He’s the only attending physician on this bay. When the residents or physician assistants lean across the central row of computers to present a new patient’s case to him, he slides his phone, recording on Abridge, between them.

The residents and PAs working with Baugh acknowledged the benefits AI scribe companies typically tout: Without the pressure to document during his shift, he’s less distracted, they said; he can be more thoughtful and listen more carefully when consulting on what to do with a patient.

But it’s unclear where the benefits accrue outside of the person using the AI scribe, or those working with them. The patients STAT talked to at Brigham and Women’s viewed the tool as more of a positive than a negative, but it didn’t assuage their larger concerns.

Isaac Cohen, whose bare feet stuck out from under the blanket on his hallway bed, came in to the emergency department because of dizziness, headache, and chest pains, at the urging of his physical therapist. Baugh had ordered an X-ray and a CT scan, and Cohen was anxiously waiting to be told the results.

He had no problem with the use of AI in his care but said it didn’t feel any different from any other visit he’d had to the emergency room before. The AI didn’t feel like it was helping him at all. He wondered if his blood sugar was part of what caused his symptoms, but he didn’t have any answers. “That’s my concern. I don’t care about anything else,” he said. “I want to know what happened to me.”

Similarly, Kiki Wenzel, an ovarian cancer patient who came in because she was out of breath, groggy, and twitching after her sixth chemo infusion, didn’t think the technology addressed her bigger worries. Someone told her she was getting admitted, but she was still stuck in a busy corner of the hallway. “Why am I still sitting down here in the emergency room?” she asked.

Her oncologist had called ahead to the emergency room, which put her at the front of the line — but Wenzel and her husband still waited for four hours. She worried about those she’d left behind, still waiting to be seen. “There’s still people out there that have been outside in the waiting room for like three hours before we got here,” she said.

Sayon Dutta, an emergency room physician and clinical informaticist at Mass General Brigham who co-authored the study on scribe use in the ED at MGB, said that it’s an open question as to what physicians do with the time — perhaps up to an hour per shift — they’re saving on documentation, since the study indicated they are not seeing more patients. It’s also a tough question to study.

“Are they talking to residents? Are they teaching? Are they more at the bedside? Are they looking at other parts of the chart and reviewing stuff? Are they spending time doing other things?” he said. “Or are they just checking their email?”

At EDs, the wait for a fix continues

STAT talked to several emergency physicians at different institutions who use AI scribes. They concurred: The AI used in emergency rooms today isn’t going to substantially change wait times, how long it takes to get results back, or how fast rooms become available.

An AI scribe “is one of the lowest hanging fruits” for AI, said Josh Lesko, who works for a Virginia group called the Emergency Physicians of Tidewater that staffs the Sentara health system. “It’s a much easier and simpler fix than ED boarding or patient discharge rates, which have plagued the hospital system.”

MGB didn’t even expect the AI tool to affect patient throughput because of how full of bottlenecks the ED is, Dutta said. “If you’ve got a lot of boarders, for example, that can’t get a bed upstairs, you may not be able to impact how many patients you see because you’re constrained by these other things that are outside of your control, to some extent,” he said.

Other generative AI tools in the pipeline for the emergency department — such as creating discharge instructions in simpler or other languages, triaging patients, summarizing a patient’s complex medical history for ED physicians, suggesting diagnoses and next steps, or bringing the most urgent radiology scans to the front of the line — are far behind AI scribes.

For example, Lai-Becker told STAT that MGB hasn’t yet found an AI triage vendor whose product “looks like it’s quite ready to really launch into a pilot.” Dutta said that AI for translating patient documents to other languages first needs to demonstrate safety and “is still within the academic realm.” And Lai-Becker said that she wasn’t aware of any non-emergency-medicine tools, such as radiology AI, currently in use that would impact the flow in the health system’s emergency department.

Walker imagines a world where there’s an AI tool that can better help his patients decide whether they need to go to the emergency room or not — to encourage people who come in too late to come in sooner, and to re-route would-be ER patients who would be better served at an urgent care. But even that AI system would be throttled by access in other parts of the health care system: “‘Now the system is telling me my only options are, I can see a specialist in seven months or I could go to the ER and get my workup started,’” he said.

There are “dozens of issues in health care that need addressing, arguably maybe even more than improving and fixing the rev[enue] cycle and prior auth[orization],” said Walker, referencing the billing and insurance applications of AI that are also currently among the most popular uses of AI in medicine. He listed infant mortality, rural health, maternal mortality, and under-insurance as bigger issues, just to start.

“But those problems are so perpendicular to what AI can do right now,” he said.