The New Provider: Doctor, A.I.

A recent industry report making the rounds — published by a venture capital firm that invests in healthcare technology — lays out a vision it calls "multi-agent care." The pitch is straightforward: AI agents will become the next tier of "medical provider," the same way physician assistants and nurse practitioners did in 1965. A primary care physician could theoretically manage a panel of 12,000 patients instead of 1,800, cosigning AI-drafted encounters and supervising concurrent AI virtual visits. A psychiatrist could cosign 100 AI-delivered therapy sessions a day. A telemetry nurse could monitor 40 rooms instead of four.

I read that report as a physician who has spent his career in interventional pain and spine medicine, who has trained residents, and who built a practice from the ground up specifically so that patients could have a real relationship with their doctor. I want to talk about what's true in that vision, what's exaggerated, and what every patient and every physician — regardless of specialty — needs to understand about where this is headed.

The Basics: What's Actually Happening

This isn't science fiction, and it isn't far off. The American Medical Association's CPT Editorial Panel is actively developing a new billing framework — tentatively called CMAA, for Clinically Meaningful Algorithmic Analyses — designed to let AI systems bill for clinical work performed with no physician involved at the point of care. This builds on real precedent: the AMA created a classification system for AI back in 2021, and in 2022 an existing billing code for an autonomous diabetic retinal exam device had the word "automated" replaced with "autonomous." The comment period on the newest proposal closes this year, with CMS observing the process closely.

So the honest answer to "is this coming" is yes. The honest answer to "is it already fully deployed at the scale these reports describe" is no — not yet. But the direction of travel is unmistakable, and every large health system, every academic medical center, and every insurance-driven payer model has a financial incentive to move toward it quickly once the billing infrastructure exists.

Key Distinction: Efficiency Tool vs. Replacement Provider

There is a meaningful difference between AI that removes administrative burden from a physician and AI that is positioned as a billable, autonomous "provider" in its own right. The first is something I welcome. The second is where I get concerned, and it's where patients should be paying attention too.

Since the widespread implementation of electronic medical records, physicians have absorbed an enormous amount of clerical work that used to belong to support staff. Documentation, inbox management, prior authorization phone calls, coding — in many respects, physicians have become their own secretaries, spending hours after clinic finishing notes they will likely never read again. If ambient scribing and AI-assisted documentation genuinely give that time back to physicians, that is a real and welcome improvement. I hope it materializes. There is a light at the end of that particular tunnel.

But that is a very different thing from an AI agent independently delivering a therapy session, triaging a patient's symptoms, or managing a monitored hospital bed with a human physician functioning as a distant, overextended cosigner. When the AMA and the venture capital world start talking about AI as a "billable unit of clinical work" rather than a tool that supports the clinician, the incentive structure shifts. Once something can be billed, it will be scaled — and it will be scaled by the organizations with the leverage to do it: large hospital systems, private equity-owned groups, and academic centers under margin pressure.

Why This Matters: A Slippery Slope, Not a Single Step

If you have ever been frustrated navigating an automated phone system when you just needed to speak to a person, you already understand the trajectory here. It rarely starts with the phone tree replacing the person entirely. It starts as a convenience, an efficiency, a way to route simple requests faster. Then the menu gets longer. Then reaching an actual human becomes the exception rather than the rule.

Healthcare is on the same slope, and the stakes are considerably higher than a customer service call. Whether you receive care at a large academic institution, a big hospital system, or a corporate-owned outpatient group, you will increasingly be pushed toward AI-mediated touchpoints — not because the evidence definitively shows better outcomes, but because the math is attractive to the organization. More encounters per clinician, or fewer clinicians per encounter, both improve the numbers on a spreadsheet. That tension is spelled out plainly in the report itself: AI adoption creates value on either the revenue side or the cost side of a health system's income statement. Patients are not a line item in either version of that story, even though the story is being written about them.

The honest concern is not that AI will be used. It's that healthcare, delivered this way, will keep drifting from a relationship into a service — something transacted rather than something built. And a service optimized for volume and margin does not reliably produce better outcomes. It often produces more expensive care with worse outcomes, because the thing that actually drives good medicine — a clinician who knows your history, examines your whole body, and has enough time to think — gets systematically squeezed out.

Where AI Genuinely Belongs

I want to be fair to the technology, because there absolutely are clear-cut cases where automation is the right answer. Autonomous AI reading a diabetic retinal screening image is a legitimate, FDA-cleared use case — it is a narrow, well-defined task with a clear right answer, and it expands access to a screening that would otherwise go undone. Ambient documentation that drafts a note while I'm actually looking at my patient instead of typing is a genuine improvement to my day and, more importantly, to the quality of attention my patient receives during the visit. Administrative automation — prior authorization routing, inbox triage, scheduling — is overdue and welcome.

The distinction I keep coming back to is this: automating a discrete, well-bounded task is not the same as automating judgment. A patient in front of me is not a checklist. They are a whole person — a specific spine, a specific nervous system, a specific set of comorbidities, fears, and goals — and the exam, the conversation, and the years of pattern recognition that go into a clinical decision do not reduce cleanly to an algorithm, no matter how sophisticated. An algorithm optimized across a population will always, by definition, be built for the average patient. Medicine that actually works is built for the one in front of you.

What I Recommend: Protect the Relationship While You Can

For patients: seek out primary care physicians and specialists in independent, private practice when you have the option. Corporate medicine — whether a large hospital system or a private equity-owned group — operates under productivity and margin pressures that are only going to intensify as this technology matures. A physician in private practice answers to their patients first. That structural difference is not a small thing. It is often the entire difference between a doctor who has ten minutes and a doctor who has the time to actually know you.

For physicians: this is not a conversation to have passively, after the billing codes and the corporate protocols are already in place. Our specialty societies need to be discussing, now, how autonomous AI should and shouldn't be implemented within our specific specialties and the patient populations we serve. Radiology, psychiatry, primary care, and interventional pain medicine all carry different risk profiles, different degrees of diagnostic nuance, and different tolerances for delegating judgment to an algorithm. That conversation should be led by the physicians who understand those nuances — not exclusively by the venture capital firms and hospital administrators who stand to benefit financially from scale.

Perspective: Medicine Is a Relationship, Not a Transaction

I did not go into medicine, and I did not build my practice, to manage a panel of 12,000 patients through a dashboard of AI agents. I went into it to look at the whole patient — to understand not just the pain generator on an MRI, but the person living with it. That is not nostalgia. It is the actual mechanism by which good outcomes happen. The physician-patient relationship is not a sentimental extra layered on top of clinical care. It is a diagnostic tool in its own right, and it is the thing most at risk as corporate medicine reaches for the efficiency this technology offers.

I hope AI fulfills its more modest, more genuinely useful promise — giving physicians back the hours we've lost to documentation and administrative burden so we can spend more time, not less, actually practicing medicine with our patients. That version of this future is worth welcoming. The version where AI becomes a billable substitute for the relationship at the center of care is worth resisting, and it is worth having a very clear conversation about — as patients, as physicians, and as the specialty societies responsible for setting the standard — before the decision gets made for us.

Disclosure

This article reflects the author's clinical perspective and interpretation of a publicly available industry report on clinical AI, along with publicly reported information on AMA CPT Editorial Panel proceedings. It is intended for educational and discussion purposes and is not a comprehensive analysis of AI policy, reimbursement law, or regulatory developments, which continue to evolve.

About the Author

Dr. Mahajer is double board-certified in Physical Medicine and Rehabilitation and Sports Medicine, fellowship-trained in Interventional Pain and Sports Medicine at the Icahn School of Medicine at Mount Sinai Hospital. He is the Founding Physiatrist of Osso Health in South Florida, with a research focus in regenerative and biologic therapies. He serves as Past President of the American Osteopathic College of Physical Medicine and Rehabilitation and as Assistant Professor of Neuroscience at Florida International University Herbert Wertheim College of Medicine. He holds medical licenses in Florida, New York, and California. A published author and book chapter contributor, his work appears in peer-reviewed journals and texts from Oxford University Press, Human Kinetics, and Springer. He has been featured in Vogue, US News & World Report, PBS, and Healio, and has been recognized as a Top Physiatrist and Top Doctor in Florida and New York, a New York Times Rising Star, and one of America's Best Doctors.