Utah lets Nolla Health's AI write first acne prescriptions as physician review tapers to a 10% sample

Patient number 101 is the first person in Nolla Health's Utah pilot whose acne prescription goes out before any doctor has looked at it. The first 100 get two physicians signing off in advance ^[3]. From there to patient 500, physicians check the work after the drug is prescribed ^[4]. That means 400 people will get a treatment chosen by software and reviewed later ^[20].

The company's pitch is about independence. "Unlike telehealth platforms that use AI only to collect information before handing patients off to clinicians, Nolla's AI system is designed to independently issue initial prescriptions within tightly defined, physician-approved treatment protocols," Nolla wrote ^[16]. What a user actually does is smaller. They fill in a 15-minute questionnaire and scan their face ^[9], and they get one of the eight skin treatments the system can currently prescribe, according to Bloomberg as cited by The Verge ^[11]. The app's pitch is wider than what it is cleared to prescribe. Digital Trends reports Nolla describing skincare plans for redness, fine lines, wrinkles, dark spots and under-eye bags as well as acne ^[15].

The product is for an adult with ordinary acne who would rather pay $4.99 a month, or $59.88 a year, than book a clinic visit or sit on a waitlist ^[2][22][19]. Nolla names a second beneficiary. "By handling routine, low-acuity cases, Nolla Derm frees dermatologists to focus on complex, high-risk patients," the company wrote in its press release ^[17].

The 97% figure measures how often a doctor agreed with the AI's pick ^[12]. It does not tell you whether the skin cleared or whether patients kept taking the drug, and it was reported without a case count. The app already lets patients take regular face scans to track how they respond ^[18], so outcome data is being gathered alongside the approvals.

The two published accounts of the oversight differ. Digital Trends describes a single healthcare expert approving each of the first 100 prescriptions, with review later easing to weekly checks ^[14]. The Verge quotes Nolla describing two physicians up front and a monthly sampling regime later ^[3][5].

Utah began allowing AI systems to renew certain medications earlier this year, and several companies are testing AI prescribing in the state ^[6]. Nolla chose acne because it is relatively low-risk and has been studied in clinical settings for decades, according to Digital Trends ^[13]. With two steps on the record, the evidence supports a pattern of Utah widening AI autonomy one narrow use at a time: renewals first, then first prescriptions for a single well-studied condition ^[6][7].

A team planning its own autonomous pilot can sort every human check on two axes. One is whether the check happens before or after the patient gets the drug. The other is whether it covers every case or only some. Nolla's first 100 patients sit in before-and-every, and patients 101 to 500 sit in after-and-every ^[3][4]. After 500, the plan moves to after-and-some ^[5]. The before-and-some cell holds two checks: the model handing off when it is unsure, and the patient choosing to contact a licensed doctor ^[10][18].

In my view that last cell needs the most testing in any copy of this design, because after patient 500 it is the only route to a doctor before a prescription ships ^[5]. I'd track two rates from day one: how often the model hands off, and how often a sampled case shows it should have. The tradeoff is the wait. Every handoff sends the patient back toward the clinic visit and the waitlist the app was pitched to remove ^[19].