The shift no model can cover

One in thirteen American emergency departments has no attending physician on site around the clock. The work those departments are missing is hands-on, time-critical, and physically present, and it is the work on which every index of AI exposure, whichever lab built it, bottoms out. The virtual hospitalist companies stop at procedures. If you have been thinking about airway, access, or rural coverage, the market is telling you to do it. The advice is easy to give; following it means twelve to twenty-four months of courses, case logs, and committee calendars, and the credentialing is where people give up. What follows is the case for doing it and the path, documented as far as the providers' own prices, rules, and dates allow.

This is written for the family physician, internist, pediatrician, or hospitalist who trained for a version of medicine that a model is now quite good at, and who wants to add the part that a model is not. The safest ground left in clinical medicine is work that needs a physician in the building with trained hands. The rural emergency department is where that work is scarcest and where hospitals pay a premium for it, and a primary-care physician can be credentialed to do it without a second residency. What the data say, what the rules are, what the courses cost, and what the paperwork looks like follow, in that order.

Why the hands will outlast the screen

Start with the exposure data. Microsoft's occupational study, built from 200,000 real Copilot conversations mapped to the government's task database, gives healthcare diagnosing and treating practitioners an AI applicability score of 0.13, against 0.38 for media and communication workers, and puts home health aides and nursing assistants at 0.04, in the same band as forest and conservation workers. The authors warn that a high score does not mean job loss and a low score does not mean safety; the score says only how much of a job overlaps with what people are using AI for. The warning is fair, and by that measure the parts of a physician's day spent talking, documenting, looking things up, and pattern-matching on text overlap a great deal, while the parts spent with hands on a patient barely register.

Anthropic's economic index arrives at the same place from the other direction. Its January report singles out radiologists: the model does well on their two biggest tasks, reading images and writing reports, and yet "these occupations have low task coverage because AI can't do the hands-on or administrative work in their job profiles". OpenAI's benchmark for economically valuable work, GDPval, admits only occupations in which at least 60 percent of tasks involve no physical work, which is why it includes nurses and medical secretaries and not a single physician who touches patients. The three documents agree. The labs measure themselves against work that fits on a screen, and none of them is measuring against a chest tube.

None of that is permanent. Last September researchers published in Science Translational Medicine a soft robotic device that guides its own way into the trachea; on cadavers, in the hands of prehospital providers given five minutes of training, it beat video laryngoscopy on first-pass success, 87 percent to 63. But it has been tested on manikins and cadavers, not on a living patient, and a 2025 review of 28 airway robots found most still in validation, a handful commercialized, and almost no human data. Autonomous airway management remains a research program. On that evidence, the physician who can secure an airway, place a line, put in a chest tube, and run a resuscitation in a building with no backup will be scarce for at least another decade, a longer horizon than I would give any purely cognitive part of the job.

Two more measurements point the same way. The virtual hospitalist programs that have now logged a million visits covering nights and surges will admit, round, and manage a ventilator by video, and they say plainly that they stop at central lines and intubations, because a camera cannot do them. When Mass General Brigham studied 198,000 emergency visits, AI scribes saved emergency physicians 1.6 minutes of documentation per note and changed nothing about the work RVUs they generated per hour. The tools speed up the screen work and leave the hands-on work where it was.

The market for a doctor in the building

The 2022 national inventory of emergency departments found that 344 of 4,621 responding EDs, 7.4 percent, had no attending physician on site around the clock. Eighty-nine percent of those were critical access hospitals and 72 percent were rural. In North Dakota the figure was 58 percent of emergency departments, in South Dakota 56 percent, in Montana 46 percent. KFF Health News went to see how those departments run: a physician assistant or nurse practitioner alone on a twelve-hour night, a physician reachable by phone. The AMA and ACEP both want laws requiring a doctor in the building, and Indiana, Virginia, and South Carolina have passed them. Those laws will need doctors to fill the shifts.

The supply side is the strange part. In 2021 the specialty's own workforce study projected a surplus of 7,845 emergency physicians by 2030, and the profession has spent five years worrying about having too many emergency physicians. The surplus is urban. ACEP's own rural task force reported that only 8 percent of emergency physicians work in rural EDs and that many rural departments are staffed by physicians without EM board certification and by PAs and NPs with no EM physician oversight at all. The family medicine journal's count is that primary-care-trained physicians are about a fifth of the rural emergency workforce.

The rules were written for this reality. The Medicare conditions for a critical access hospital require a physician, PA, NP, or clinical nurse specialist "with training or experience in emergency care" to be on call and on site within 30 minutes, or 60 in frontier counties. The newer Rural Emergency Hospital rules require someone "competent in the skills needed to address emergency medical care" in the department around the clock, with a physician available by phone. Neither rule requires a board-certified emergency physician.

There are 1,388 critical access hospitals as of July, every one of them obliged to run a 24-hour emergency service, and 55 hospitals now operating as Rural Emergency Hospitals, a designation that did not exist before 2023 and that is, in effect, an emergency department with a few observation beds and no inpatient unit. Meanwhile 197 rural hospitals have closed or converted since 2005, eight of them last year by the Sheps Center's count. The money arriving to stop the bleeding is real.

The $50 billion Rural Health Transformation Program pays $10 billion a year through 2030, every state got an award in December, and the fifth approved use is recruiting and retaining clinicians who commit to rural communities for at least five years. Montana got $233 million for the first year, Colorado $200 million, the Dakotas about $190 million to $199 million each. The administration's pitch is that AI will make rural care affordable. Rural hospital leaders told STAT they do not believe it, and no amount of it staffs an emergency department at two in the morning.

The locum market puts a price on that scarcity. Locum emergency physicians are earning $250 to $300 an hour this year, against $120 to $135 for locum family physicians and $120 to $130 for internists. Postings for critical access hospital emergency shifts this week run $267 to $404 an hour. Employed emergency physician pay averaged $421,000, up about 8 percent in a year. The spread between those lines is not automatically yours: the rate follows the privileges you hold and the shifts you can safely cover, and the shifts are hard. But the spread is a direct measurement of what hospitals pay for a physician who can be present and capable at the bedside, and none of the models in the indices above narrows it.

The objections, and what they concede

ACEP's position, stated for years, is that emergency care is best provided by physicians who are residency trained in emergency medicine and that there is no substitute for residency training. I agree. The same document concedes that EM-trained physicians may never fully meet the demand, and the numbers above show what that concession means at night: the choice a critical access hospital faces is between a primary-care physician with the right training and a physician assistant or nurse practitioner alone with a phone, because the residency-trained emergency physician it would prefer is part of an urban surplus. Take this path and you will be the least trained person in the room for this kind of medicine and, on some nights, the only physician within sixty miles. The objection sets the terms of the path rather than closing it, and ACEP's own task force has written the terms: for rural non-EM physicians it proposed a defined knowledge base, supervised experience, and mentorship from an EM board-certified physician by telemedicine. That is the sensible response; arrange it for yourself before you need it.

The second objection is about your own hands. Internal medicine residents now report 30 percent confidence in procedures, with 9 percent confident in lumbar puncture and 20 percent in thoracentesis, and three quarters of the academic hospitals surveyed have moved bedside procedures onto a dedicated procedure service. The training pipeline has been producing less of this skill for a generation. A weekend course is the start of getting it back, and a proctored year is the rest. The deskilling evidence cuts both ways here too: endoscopists lost adenoma detection after three months of AI assistance, and a procedural skill you do not use decays the same way. Whatever you build, plan to keep using it.

From here to a credentialed shift

The sequence below carries costs and dates as the providers state them today. Two paths run through it. One is procedural competence where you already work, which for a hospitalist means never calling the ED for a line again. The other is rural emergency coverage. They share the first four steps. Budget twelve to twenty-four months and, by my arithmetic from the prices below, north of $10,000 in course fees before the first paid shift, less if your hospital pays for courses, which many will if you ask.

Step 1. Decide which of the two you are building, and write down the date. Procedural competence alone can be built inside a hospitalist job in a year. Rural ED coverage needs the same procedures plus trauma, pediatrics, obstetrics, airway under pressure, and the certifications that every medical staff office asks for before it will look at you. If you want the second, aim for the first rural shift eighteen months out and work backward. Tell your department chief and your medical staff office what you are doing at the start, because the privileging steps below take months of calendar time that you cannot compress.

Step 2. Get the three cards that gate everything, then the two that make you useful in a small building. Nearly every rural bylaws committee will want three cards in hand before it looks at you, and the primary-care board route to EM certification requires all three to be current. They are ACLS and PALS from the American Heart Association, and ATLS from the American College of Surgeons, now in its eleventh edition with 26 online modules and a 2 to 2.5 day course, or 1.5 days in the hybrid format; completion is valid four years. Then two courses built for exactly this situation. Comprehensive Advanced Life Support is the rural all-hazards course: twelve hours online and two days of simulation covering airway, trauma, cardiac, stroke, pediatric, obstetric, and neonatal emergencies and sepsis, up to 23.5 credits, $1,697 for physicians in Minnesota, with the next open course November 11 and 12 in Aitkin. Fundamental Critical Care Support from the Society of Critical Care Medicine is written for the non-intensivist who has to manage a critically ill patient for the first 24 hours until transfer; the self-study version is $275 and hosted courses with skill stations run about $800. If the department you are aiming at delivers babies, or catches them in the parking lot, add ALSO from the AAFP, one day after an online course, about $595 at an AAFP-run site.

Step 3. Learn the airway from people who do it for a living, then practice it in an operating room. Privileging committees count cases, and a course certificate is not a case, so this step has two halves: the courses, then the operating room. Take two courses, and I would space them a year apart. The Difficult Airway Course: Emergency is three days, 21 credits, $1,950, with simulation stations and code-airway drills; the September Atlanta course is sold out and 2027 dates open soon. The two-day cadaver course that Rich Levitan and Sal Rezaie run in Texas spends the entire second day in a lab with a dozen prepared cadavers, 16 credits, 36 seats, with dates remaining this year on September 21, October 26, November 16, and December 3; the last posted tuition was $2,250, and there is a June version at Jackson Lake if you want the mountains with it. Then the part no course can sell you: ask an anesthesiologist at your own hospital for supervised intubations in the operating room, one morning a week for a few months, and log every one.

Step 4. Lines, taps, tubes, and the ultrasound that makes them safe. The ultrasound is what makes these procedures safe, so learn them together and hold a real ultrasound credential rather than a course certificate. The Hospital Procedures Consultants weekend course is the one built for hospitalists: 23 online modules, then two ten-hour days at a six-to-one ratio covering ultrasound-guided central and arterial lines, thoracentesis, paracentesis, lumbar puncture, chest tubes and pigtails, pericardiocentesis, intubation, and cricothyroidotomy, 20 credits, with a simulation certificate at the end; remaining 2026 dates are Seattle September 19, New Orleans October 24, and San Antonio November 14. The SHM and CHEST point-of-care ultrasound certificate requires online modules, a live course, a regional course, a portfolio of 20 cardiac, five lung, five abdominal, five DVT, and three soft-tissue studies, and a skills and knowledge assessment, all within three years; the assessments carry a fee and retakes are $300. If you want the emergency flavor, the Ohio ACEP two-day emergency ultrasound course is $999 to $1,325 with the next one November 4 and 5, and Gulfcoast runs a three-day course in St. Petersburg in February and August at a three-to-one ratio.

Step 5. Turn the courses into privileges. This is where the calendar time goes, and where people give up. Ask your medical staff office for the delineation-of-privileges form for each procedure and read the numbers on it: how many supervised cases for initial privileges, how many a year to keep them. The Society of Hospital Medicine has published a position statement on credentialing hospitalists in ultrasound-guided procedures that most medical staff offices will accept as the framework if yours has none. Joint Commission standards put every newly granted privilege under a focused professional practice evaluation, meaning a proctor and a defined number of observed cases, and the clock starts the day the privilege is granted; after that, ongoing evaluation cannot be more than twelve months apart. So you need a proctor. Interventional radiology, the ICU, and the emergency department all have people who will sign for you if you show up and ask; the procedure service, if your hospital has one, is the natural home. Keep a log from the first case: date, indication, ultrasound or landmark, attempts, complications. The log is what a committee actually reads; the courses only get you in the door.

Step 6. Choose your rural emergency credential path. There are three paths, and only two are open to a practicing physician. The first is experience: get ED privileges at a critical access hospital, whose bylaws, following the AAFP's position that privileges should rest on training, experience, and demonstrated competence, will usually ask for the three cards, references, and a supervised start; begin with overlap shifts beside the incumbent physician, then nights with a phone, then alone. The second is board certification through the American Board of Physician Specialties, the only board that certifies primary-care-trained physicians in emergency medicine. Its practice route is five years of full-time EM practice with at least 7,000 hours and 1,400 hours in every twelve-month period, plus current ACLS, ATLS, and PALS, three letters, ten case reports, and a 325-question written exam followed by an oral exam, good for eight years; California adds a sixth year and 300 EM-specific CME hours. Its fellowship route is a 12- or 24-month program from a list of fifteen approved programs in nine states. Marshfield Clinic in Rice Lake, Wisconsin takes two fellows a year, requires 1,400 ED hours and rotations through trauma, pediatric EM, the ICU, and anesthesia, and is accepting applications for 2027. Tamarack Health in Ashland runs twelve to fourteen ED shifts a month for three fellows and accepts applications on a rolling basis. Texas Tech in Odessa trains in a Level 2 trauma center with 60,000 visits a year under ABEM-certified faculty. The third path is closed to you: ABEM certifies only graduates of an accredited EM residency or of a five-year combined EM and family medicine residency, and there is no practice track. Know that before a recruiter tells you otherwise.

Step 7. Build the backup before the first night. This is where the AI and telemedicine story turns in your favor. Avel eCare puts board-certified emergency physicians on video in 137 rural hospitals and claims a 21-minute reduction in door-to-physician time and 700 transfers avoided in a year; those are the company's numbers, but the service has run since 2009 and it is what the ACEP task force asked for. A telehospitalist triage line studied across four community EDs sent 19 percent of admit-likely patients home and kept 58 percent local instead of transferring them. Before you take a shift, know which of these your hospital has, know the transfer agreements and the flight times, and know the rule you are working under: 30 minutes to the bedside at a critical access hospital, 60 in frontier counties, a physician always reachable by phone at a Rural Emergency Hospital. The backup is there so that you can be the physician in the building without being the only brain on the case.

Step 8. Handle the money and the paperwork early. The dull work is what separates a plan from a shift. Ask the rate first. Critical access hospital emergency shifts are posting at $267 to $404 an hour through locum agencies, and a hospital that hires you directly will pay less per hour and offer more in return: malpractice, a schedule you control, and sometimes loan repayment. The National Health Service Corps pays up to $75,000 for two years of full-time primary care at an approved site, and a critical access or rural emergency hospital qualifies through its affiliated outpatient clinic, not through the emergency department itself; the 2026 cycle has closed, so plan for 2027. Ask your state office of rural health how it is spending its Rural Health Transformation award on clinician recruitment, because that money exists and comes with a five-year commitment attached. Call your malpractice carrier before the first shift and confirm in writing that emergency department scope, procedural sedation, and the procedures you have added are covered. Get the license in the state you will actually work in months ahead of time.

Step 9. Keep the skill you paid for. A skill you do not use decays, as the endoscopists found. Set a floor: a minimum number of each procedure a year, written down, with the log to prove it. Recertify ATLS on the four-year cycle and take the airway course again when it lapses. Keep deliberate AI-off intervals for the cognitive work too, and use the workflow the Stanford data support: commit to your own assessment, then compare. What lasts in a physician is being there, deciding under uncertainty, answering for the result, and doing the physical work. The physical work fades faster than the rest when you stop doing it, and right now nobody has built a machine to replace it.

If you follow this and end up at a rural bedside at two in the morning with an airway to secure and no one else to do it, you will be doing the one job in medicine I am confident a model will not take from you, in one of the few places where the shortage is simply a missing person. Hospitals have been advertising that shortage for years. This is how to answer the ad.