On October 5, Nolla Health said Utah’s AI sandbox has cleared its software to write first-time prescriptions. The company calls it the first U.S. AI to issue initial prescriptions rather than refill existing ones, according to its release. The scope is narrow: mild-to-moderate acne, topical treatments only, no isotretinoin and no oral drugs. Patients pay $4.99 a month.
How the three gates work
A patient spends 10 to 15 minutes on identity verification, consent, a questionnaire and a face scan from five angles, per Nolla’s blog. The AI can only choose from treatments physicians pre-approved. If nothing fits, it hands the patient to a licensed physician. The pilot is open to Utah residents 18 and older.
Oversight comes in three stages. For the first 100 patients, every prescription waits for a physician before it goes out. Utah’s pilot page says two Utah-licensed physicians do it. Through patient 500 the AI issues prescriptions itself and physicians review afterward. After that, physicians sample. Nolla’s blog says each step up requires 95% physician agreement and zero serious adverse events, and the state page adds that each advance needs written sign-off from Utah’s AI office.
The sources disagree on how thick the sample is. The company’s release has physicians reviewing daily in stage two and sampling weekly in stage three. Nolla’s blog and SiliconANGLE’s report put stage three at no less than 10% of prescriptions a month. Which one governs will show up once the pilot runs. SiliconANGLE also notes that the state office wants it understood that joining the sandbox is not a state endorsement.
The 96%
Nolla says licensed clinicians agree with the AI’s recommendations in more than 96% of real-world cases, and that most of the rest are small tweaks such as topical strength. Nobody audited that number, and the release does not say what the sample was.
The bar is the interesting part. The gate to advance is 95%. The company’s own figure clears it by one point. A few dozen disagreements in either direction could close the gap, and the pilot has not logged its first prescription under these rules.
Doctronic, a year in
This is not Utah’s first AI-and-prescriptions pilot. On January 6 the state announced Doctronic, an AI for renewing existing prescriptions. The pilot page dates its approval to October 24, 2025 and still shows it in phase one, with a licensed practitioner authorizing every request. It cannot start a new prescription.
The two gate structures differ. Doctronic advances by medication group, one application at a time. Nolla advances by patient count: 100, then 500. Nolla is attempting the harder act, the first prescription, and its first gate is a hundred patients wide.
The price and time math
Nolla cites Utah numbers for the gap it says it is filling. About 100,000 people live with active acne at any time, only 10% ever see a dermatologist, the average wait is 61 days, nearly double the national median, 11 counties have no dermatologist, and a visit runs up to $300. At the top of that range, one visit buys 60 months of the subscription. The $4.99 is the platform fee. The release does not say who pays for the medication.
The stages also have floors. Per Nolla’s blog, stage one lasts at least four weeks and stage two at least eight. If patients and agreement clear the bar as fast as possible, stage three starts in late December at the earliest. The 100 and the 500 are minimums, not dates.
The pilot is small, the funnel is not
Nolla says its Nolla Derm app has been downloaded more than 175,000 times across 44 states before this pilot began. The first two stages cover 600 patients. That is about 0.3% of the downloads. The gates are sized for caution, but the audience behind them is already large, and stage three is where the cap comes off. Each stage is also a staffing decision in disguise: stage one needs two physicians on every prescription, stage three needs them on a sample. The cost per patient of supervision is the number the company is really testing, and it has not published it.
What it means for the people in the room
Start with who is not on the list: dermatologists. By the company’s numbers nine in ten Utah acne patients never see one, so the AI is taking visits that mostly were not happening.
The exposed layer sits in the middle. That is the physicians, NPs and PAs who write first-visit prescriptions on teledermatology platforms, plus the intake and triage staff behind them. The mechanism is to shrink the first visit to a choice among pre-approved options and turn the physician into an auditor. In stage one, every prescription gets a human pass. In stage three, if the state’s 10% sample holds, the physician review per prescription falls by as much as 90%.
Set it beside last week’s Gemini 4 Argon result on Harvey’s legal benchmark, 19.6% in a vendor’s own table. Legal is still being scored. Medicine already holds a license and a patient count. Regulation takes a different path too. In our California coverage, SB 951 required 90 days’ notice before large displacement. Utah waives rules first and measures after.
What is not in the reporting
Four things are missing: who pays for the drugs, how large the 96% sample was and how agreement is scored, how long it takes to fill the first 100 slots, and what happens when the pilot ends. No dermatologist or medical society has responded on the record in the coverage we read.
Three things to watch. Whether the first 100 patients clear near the four-week minimum. Whether Utah’s office publishes an advance approval. And how the drug cost is billed. Those will say more about how far this model scales than the 96% does.
