Most clinical AI is boring.
It helps doctors type faster. It summarizes notes. It keeps the burnout at bay. Companies like Abridge and Cleo are building scribes. OpenEvidence helps with research. They are copilots. Useful, sure. But they stay in the passenger seat.
Then there are the outliers.
A tiny group of startups is ignoring the “assist” narrative entirely. They want the wheel. They want to replace the physician, period. Here are the three companies betting the house on that idea.
How Doctronic Got Legal Permission To Prescribe In Utah
Doctronic operates in New York. It raises money there. But it practices medicine in Utah.
The company describes itself as a free, 24-7 personal AI doctor. HIPAA-compliant, yes, but the real headline is autonomy. In March, they closed a $40 million Series B round led by Abstract and Lightspeed Venture Partners. Total funding sits at $65 million.
Utah allowed Doctronic to renew prescriptions for chronic conditions without a human doctor approving each one. This is the first time a U.S. state authorized an AI platform to handle routine prescribing alone.
The numbers from six months in are striking.
For 192 drugs treating hypertension, diabetes, and depression, the AI approved renewals without human involvement 72% of the time. The other 28%? The system escalated to a human doctor. Usually because it flagged a complication or needed labs. Independent reviewers agreed with those escalations 69% of the time.
When the AI did make the call alone, physicians agreed with it 91% of the time. But in 3% of cases, they disagreed entirely.
Utah’s Office of AI Policy reports no serious safety incidents. Yet the Utah Medical Licensing Board and the American Medical association are nervous. Researchers have flagged security vulnerabilities in the pilot. Regulators are still working through them.
Certuma’s Play For FDA Approval And Robotic Clinics
Martin Varsavsky founded Certuma. He is a serial entrepreneur. The Austin-based company targets the stuff that clogs urgent care waiting rooms. UTIs. Strep throat. Sinusitis.
They have raised $10 million in seed funding at a $60 million valuation, led by 8VC.
Certuma wants to be the first AI doctor to get formal FDA approval. They picked 25 low-risk, high-frequency conditions specifically because the clinical decision trees are rigid and well-established. Low variance. High predictability.
The strategy is two-track.
In Argentina, where Varsavsky is from and deregulation is friendly, they are already offering diagnosis and prescribing advice via a consumer chatbot. In the U.S., they are taking the slower, higher-bar path through the Food and Drug Administration.
The long-term vision involves autonomous robotic booths. Patients walk in. They get blood drawn. They get imaged. All without a nurse or a doctor.
Why Ada Health Calls Itself An AI Front Door
Ada Health is in Berlin. It has raised $242 million, including a $120 million Series B from investors like Inteligo Bank and Farallon Capital.
It doesn’t claim to be a doctor. Not yet.
It builds an AI health companion. You type your symptoms. It tells you what might be wrong. It suggests what to do next. Often, it suggests nothing. And you stay home.
Ada Health says it has 13 million global users who have completed over 32 million symptom assessments.
They point to a peer-reviewed BMJ study they led. The study compared eight symptom-checker apps. Ada’s tool came out on top for coverage, accuracy, and safety.
It is less “AI doctor” and more “AI gatekeeper.” But the outcome is the same. Fewer visits that require a human physician.
The Regulatory Loopholes Making This Possible
These companies cannot just decide to replace a doctor. The law gets in the way.
Two systems must bend for this to happen. Federal device approval. State medical licensing. Right now, different companies are testing different weaknesses in both.
The FDA has cleared over 1,000 AI-enabled medical devices. Most went through the 510(k pathway. This only requires the device to be “substantially equivalent” to something already on the market. It is not built for autonomous treatment decisions.
The De Novo pathway handles novel devices. It produced exactly one precedent for autonomy. LumineticsCore (formerly IDx-DD). Authorized in 2018. It detects diabetic retinopathy from a retinal Scan. No physician interpretation required.
That is the model Certuma is copying. As of now, no autonomous AI prescribing service has cleared any FDA pathway.
On the state side, Utah took a different route. Doctronic is not licensed to practice medicine. The state didn’t grant a license. Instead, the Utah Office of AI Policy signed a non-enforcement agreement. The state won’t prosecute the company under unlicensed-practice statutes. The catch? The company must follow specific safety and privacy guardrails. And it only covers renewing existing prescriptions.
Elsewhere, the doors are slamming shut.
Oregon and Delaware explicitly bar nonhuman entities from using licensed clinical titles. By mid-2026, about 37 states passed or introduced AI in healthcare laws. California, Texas, and Nebraska require a licensed human clinician to make the final call on medical necessity.
Utah’s sandbox is the exception. And it is already facing pushback.
Are Doctors Really Obsolete Yet?
The big players are playing it safe. Anthropic launched Claude for Healthcare at the J.P. Morgan Healthcare Conference in January 2026. Framed as a support tool. Large health systems want AI to handle documentation and administrative drudgery. They want to free up doctors for the judgment calls only humans can make.
But the line is blurring.
Revere Health made the first major AI-linked workforce cuts in U.S healthcare. Not doctors. Coders. Billers. Documentation staff. Roughly 200 jobs. About 7% of their workforce.
AI is hollowing out the roles around the physician first.
Doctors are uneasy. A recent Sermo poll found 58% believe AI will diminish their role or make them obsolete outright.
Companies like Doctronic argue that if an algorithm handles the routine 80%, humans can focus on the hard 20%. Patients get faster access.
But medicine is not just pattern matching.
A sore throat can be strep. Or it can be the first sign of leukemia. A good clinician catches the latter. An algorithm might miss it.
The Utah prescription deal. The FDA filings. The unanswered malpractice questions.
The safest bet is still that AI reshapesedicine. That it changes the work but keeps the worker. But for the first time, venture capital and state governments are actively testing what happens if that safest bet is wrong.
