Robert Wachter has often discussed the productivity paradox: the frustrating delay between the introduction of a new technology and any actual gain in efficiency. In health care, we’ve lived this paradox through the electronic health record (EHR). We didn’t reinvent the medical record; we simply digitized the paper chart. We added billing artifacts, compliance language, and a deluge of alerts that serve the institution more than the patient.

Now, as generative AI enters the clinic, I worry we are about to make the same mistake again.

The questions I hear most often from colleagues and investors are: Will AI replace doctors? How should doctors use AI? When will we all have an AI doctor? These questions are anchored to a crumbling system. They assume the most important thing AI can do is make the “doctor-shaped” part of health care faster or cheaper.

This is health care’s “horseless carriage” moment. Early cars were defined by what they replaced. They pointed backward. The term “automobile,” meaning self-moving, pointed toward a new capability. The automobile didn’t just move people; it transformed passengers into drivers, and fundamentally reshaped the architecture of cities.

“AI Doctor” is a backward-looking metaphor. It tries to automate the physician instead of reimagining the work. The more interesting question is: What parts of health care should no longer require a doctor at all?

The reckoning of licensing and knowledge

We must acknowledge that the line of physician authority is not a law of nature. It is a social and legal arrangement built on the scarcity of knowledge. For decades, the education required to understand one’s own body was so out of reach that society gave up on giving citizens authority over their own health.

Fast forward to 2026. My own specialty, internal medicine, recently moved toward open-book online exams for maintenance of certification. The American Board of Internal Medicine (ABIM) is acknowledging that being an internist should not rest on the ability to retain biodegradable facts, but on the ability to problem-solve. The uncomfortable truth is that a non-clinician using ChatGPT can now ace these exams.

If knowledge is no longer a meaningful discriminator, we have to ask what the clinician is actually for. If we continue to license “AI Doctors” based on traditional metrics, we are ignoring the fact that our current system has reliably failed to measure what matters to patients: reliability, access, and transparency.

From patient-centered to patient-operated

I propose an alternate frame: patient-operated health care.

This is not about patients performing their own surgery with ChatGPT and a YouTube video. It means designing the system so that the work of health care can safely happen where life happens: at home. While the industry championed “patient-centeredness” for years, it often resulted in burdening patients with the responsibility of navigating a byzantine system without providing them the tools to do so.

Patient-operated health care asks what parts of the system need to be remade to enable patients and caregivers to operate it independently and with confidence. This is already happening. We see it in the rise of self-collected diagnostic swabs and direct-to-consumer lab platforms like Function Health. We see it in the expanded “Pharmacy First” models in the U.K. and Australia, where pharmacists prescribe for minor ailments.

The most significant signal is the FDA’s ACNU (Additional Condition for Nonprescription Use) rule. This marks the end of the “Label Era,” where safety was tethered to a static piece of cardboard. The ACNU pathway allows complex medications to move to the over-the-counter aisle, provided they are coupled with digital guardrails like diagnostic questionnaires. This is the technology filling the frame of medical licensure.

Making the invisible visible

This infrastructure leads to a radical possibility: AI may finally make patient and caregiver labor visible enough to pay for. Today, families perform an enormous amount of “ghost work,” including symptom monitoring and post-discharge surveillance. Once professional-grade tools allow AI to structure and audit this work, it moves from family duty to clinical output. Insurers will not just pay for an “AI Doctor”; they will pay for the structured patient work that keeps the patient out of the emergency room.

As a physician at UCSF and the founder of a health agent for medically complex patients, I believe our job is not to defend every inch of the old territory. It is to help train patients and caregivers to operate safely with these new tools.

We will still need clinicians for high-stakes procedures and embodied judgment under uncertainty. But much of medicine is not like flying a plane; it is more like riding a bike or driving a car. With the right system design, patients can steer their own care.

The old model was a historical compromise built on scarcity. Now, we can finally start designing beyond that compromise.

Michael Turken is an internal medicine physician.