AI Prescribing in Utah and What Autonomous Healthcare Means for Dentistry
Healthcare is beginning to separate professional work into individual tasks and ask which of those tasks truly require a licensed human every time. Utah has become one of the clearest laboratories for that experiment, and dentists should pay attention because the same question is already reaching dentistry.
Utah first drew national attention with Doctronic, an artificial intelligence system designed to process renewals of medications that had already been prescribed by licensed clinicians. The distinction matters. Doctronic does not diagnose a new disease, start a new medication, change a dose, or prescribe controlled substances. Utah created a legal pathway that could eventually allow certain renewals to occur without prospective physician approval, but as of October 2026 the pilot remains in its first phase. A licensed medical practitioner still authorizes every request, and advancement to greater autonomy requires state approval. Utah has even tightened the requirements and removed medications from the approved list. The state is testing autonomy, not declaring that autonomous medicine has arrived.
A newer Utah pilot moves closer to the boundary. Nolla Health uses patient history and facial photographs to assess mild to moderate acne and recommend topical treatment. The state agreement permits the system to progress toward issuing first time topical prescriptions without prior physician review, but only through staged testing. In the first stage, two Utah licensed physicians review every prescription before it is sent. Later stages can shift to retrospective review and eventually sampling, but each step requires state approval. Severe acne, pregnancy, breastfeeding, immune compromise, and other risk factors stop the automated pathway. The important development is not that dermatologists have been replaced. They have not. It is that a state regulator has created a supervised path by which software may eventually move from recommending treatment to initiating it.
That distinction is easy to lose because artificial intelligence invites exaggerated language. A company may report that physicians agree with its recommendations 96 percent of the time, but agreement is not the same as improved outcomes, diagnostic accuracy across every population, or long term safety. Regulators want innovation to succeed. Technology companies want adoption. Clinicians may emphasize risks that protect patients, professional standards, and sometimes their own economic position. None of those incentives makes anyone dishonest, but they are reasons to ask harder questions. What was measured. Who measured it. What happened in the disagreements. How serious were the misses. Was the comparison against excellent clinical care or against the rushed, fragmented care patients actually receive.
That last comparison matters. The correct benchmark for artificial intelligence is rarely a perfect clinician practicing under perfect conditions. A routine prescription renewal may currently sit in an inbox until a tired doctor reviews it between patients. Dental documentation may be completed after hours from memory. Insurance narratives may be copied from templates. Radiographs may be interpreted while the hygienist is waiting and the next patient is already seated. If an artificial intelligence system can perform a narrow task more consistently, document every step, identify red flags, and escalate uncertainty, it does not need to become a doctor to change how doctor work is organized.
Dentistry has already seen an unusually relevant version of this experiment. Utah previously authorized a Dentacor pilot in which licensed dental hygienists could use an artificial intelligence radiograph tool for a very narrow set of conditions. When the hygienist and the software agreed, that agreement could stand in place of direct dentist supervision for specified diagnoses and procedures. When they disagreed, the case went to a dentist. The pilot was limited, temporary, and surrounded by consent and reporting requirements, but its significance was larger than its scope. Artificial intelligence was not merely helping a dentist read an image. It was participating in a regulatory structure that changed when a dentist had to be involved.
That is probably how automation will enter everyday dental practice. Not with a robot placing a molar implant while the dentist watches from the parking lot, but through hundreds of smaller decisions. Chart review. Radiographic screening. Periodontal risk assessment. Recall intervals. Medical history reconciliation. Referral triage. Emergency routing. Postoperative follow up. Insurance documentation. Treatment plan organization. Patient communication. These are the middle layers of practice, repetitive enough to standardize but important enough to consume enormous amounts of staff and doctor time.
The opportunity is substantial, but so is the danger of calling too many things routine. A toothache is routine until facial swelling, anticoagulants, renal disease, uncontrolled diabetes, pregnancy, or a complicated medical history enters the picture. A radiographic finding is routine until the lesion does not fit the pattern. A medication request is routine until the patient omitted something important. Good automation must therefore know when to stop. The most valuable system may not be the one that produces the most answers. It may be the one that recognizes uncertainty early and brings the dentist into the decision before a small problem becomes a clinical or legal one.
Liability will become one of the defining issues. If software recommends a diagnosis and the dentist follows it, responsibility does not disappear. If the dentist ignores an alert, that decision may later require an explanation. If the software misses something, courts, boards, insurers, manufacturers, and clinicians will eventually have to decide how responsibility is divided. Dentists evaluating these systems should look beyond accuracy claims and ask practical questions. What exactly can the software do. What can it not do. Who validated it. Who audits it. What happens when it is wrong. Does the patient know artificial intelligence is involved. Is there a clear human escalation pathway. Does the malpractice carrier understand how the product is being used.
The economic question may ultimately matter as much as the technology. Patients do not always choose the most human option. For a complicated diagnosis, surgery, or emotionally difficult decision, trust and personal judgment may dominate. For a simple problem, many patients may choose an automated service if it costs less, works immediately, and avoids travel or a three week wait. The same pressure can reach dental practices. When software makes a process faster, more consistent, easier to document, and cheaper to deliver, adoption does not require a revolution. It simply becomes workflow.
That is why the safest prediction is not that artificial intelligence will replace dentists, or that it never will. There is much stronger evidence that artificial intelligence will replace individual tasks than that it will replace dentistry as a profession. The harder question is what happens when enough tasks move. A dentist may eventually spend less time sorting information and producing documentation, and more time handling exceptions, explaining choices, managing risk, performing procedures, earning trust, and taking responsibility for outcomes. That could make the dentist more valuable. It could also allow fewer clinicians to manage more patients. Both possibilities can be true at the same time.
For a practicing dentist, the strategy is neither panic nor blind enthusiasm. Watch actual patient outcomes, not demonstrations. Watch reimbursement, malpractice experience, state board decisions, payer requirements, and whether regulators convert temporary pilots into permanent rules. Be skeptical of claims that autonomous doctors have already arrived, but equally skeptical of claims that nothing important is changing because a human still signs the chart. The real transformation is happening one task at a time, often inside ordinary software, where it attracts far less attention than a headline about an AI doctor.
Utah is showing healthcare the question that dentistry will increasingly have to answer. Not whether artificial intelligence can imitate a dentist, but which parts of dental care still require a dentist to make the call. What part of dentistry should AI be allowed to own first?
Join the Conversation!