An AI that writes acne prescriptions is now live in Utah under the tagline ‘skip the clinic’, and Australian clinics should take that personally.
On 5 October, a US start-up called Nolla Health switched on something no regulator had allowed before. Under a one-year pilot authorised by the Utah Office of Artificial Intelligence Policy, its Nolla Derm app can now assess a person's acne from a face scan and a short medical questionnaire and issue an initial prescription for a topical treatment, without a doctor approving that script before it reaches the pharmacy. AI has been refilling prescriptions in Utah for a little while (Doctronic renews close to 200 common chronic-condition medications there, and Legion Health runs a similar refill pilot for psychiatric medication), but this is the first time software has been trusted to make the first call on what a patient should be treated with.
The process takes ten to fifteen minutes from start to finish. A user verifies their identity, gives informed consent, answers an intake questionnaire and completes a five-angle face scan, and the app then chooses from a set of physician-approved topical treatment plans (eight medications, according to Bloomberg's reporting) before sending the script to the patient's home or a local pharmacy. It costs US$4.99 a month and is open to Utah residents aged 18 and over, and once someone is on treatment they scan their skin every day so the plan can be reviewed after every 28 scans. Severe acne and pregnancy both sit outside the pilot's scope, so those patients (along with anyone else the protocols don't cover) are handed to a licensed physician, and every user can message a doctor through the app at any time.
The boundaries of the AI recommendations have been designed to loosen as the evidence builds. For the first 100 patients, two physicians independently approve every AI prescription before it is filled, and up to 500 patients the app prescribes directly while a physician reviews every case afterwards, at least weekly. Beyond that point, physicians review at least 10% of prescriptions each month along with every case involving an escalation or a side effect. Nolla has to show 95% agreement with physicians and zero serious adverse events before the state will give written approval to move from one stage to the next, and the company says its clinicians already agree with the AI in 96% of real-world cases. There is also no free-text input anywhere in the prescribing flow, which means the system cannot improvise, and its overseeing physician, Dr Zaid Fadul, has been careful to describe it as selecting from protocols that doctors have already approved.
Nolla's argument for all of this is access. By the company's own figures, only around one in ten Utahns with acne ever sees a dermatologist, and the average wait for an appointment in the state is 61 days. Australians will recognise that story, because Cleanbill's most recent dermatologist listings put the average initial consult here at $230.22, and anyone living outside a capital city knows how long the wait for a skin specialist can stretch. An app that prescribes in fifteen minutes for less than the cost of a coffee each month is a pretty compelling answer for a client who has been sitting on a referral for weeks.
For the professional industry, the question we're asking is where does this cross the line? Skin analysis has been part of the clinic toolkit for years, and in most treatment rooms the scanner works as a consultation aid and, if we are honest, as a sales tool that helps a therapist show a client what is happening beneath the surface before recommending a treatment plan.
Nolla has taken that same scan and connected it directly to a prescription with a pharmacy waiting at the other end, which removes the consultation from the middle of the journey altogether. Once a client has experienced that, they walk into a clinic with very different expectations about how fast and how cheaply a skin concern can be handled.
It is unlikely to arrive in Australia in the same form any time soon. Prescription-only medicines here can only be prescribed by authorised practitioners under state and territory law, and the TGA's February 2026 guidance on AI made it clear that software intended to diagnose or recommend treatment is regulated as a medical device, with the marketing claims around a product counted as evidence of what it is intended to do. That second point matters for anyone in our industry selling skin analysis technology, because a device that starts suggesting treatments (even if it only does so in its marketing copy) can drift into regulated territory without anyone having planned for it. I would still treat the Utah pilot as a question of timing for Australia. Utah has built a model in which a regulator lets AI take one narrow, low-risk clinical decision, watches the data closely and widens the gate only when the evidence holds, and that model is far easier for other jurisdictions to copy than a wholesale change to who is allowed to prescribe.
For brands, the most interesting detail in the whole story is the formulary. Nolla's AI chooses from a short list of approved treatments, and whatever sits on that list is the only thing it will ever recommend. As AI takes on more of the first conversation a client has about their skin (whether that happens in a prescribing app or through an agent shopping on someone's behalf), the commercial question for brands moves from how to win a therapist's attention to how to earn a place on the shortlist a machine is working from. That shortlist gets built on clinical evidence and on protocol data that a model can read, which is a very different discipline from building brand love in the treatment room, and most brands in professional beauty have not started on it yet.
Nolla is pretty clear about who it sees as the competition, because the app’s own tagline reads ‘skip the clinic’. For clinics and derm-adjacent services, that makes the new competitor for mild to moderate acne an app that costs a few dollars a month and is selling itself on the promise that its customers never have to walk through your door. What it cannot do is everything that happens in the treatment room, from extractions and peels through to LED and the combination protocols that a good therapist builds around a prescription, and it cannot sit with a 19-year-old who has stopped going out because of their skin and help them feel like themselves again. The clinics that do well from here will be the ones that position themselves as the place people go when an app's protocols run out, or as the place that takes a prescription and builds the professional treatment journey around it, rather than competing on the triage step that software is now doing faster and more cheaply.
There are serious oversight questions that this pilot leaves open, and they are worth keeping front of mind. Who carries responsibility when an AI-issued prescription causes harm is still being worked out, and so is how well these models read darker skin tones, given that dermatology image datasets have historically underrepresented them. Beyond those, we still need to know what happens to years of daily facial scan data, and whether 96% agreement measured by a company's own clinicians is the right test of safety at all. The one-year term and the staged review exist precisely because nobody has those answers yet, and the data Utah publishes at the end of it will tell the rest of the world a great deal.
The shortlist is AI-built. The choice is human. The brands that get onto the shortlist and the clinics that own the human part of the choice will be fine, and the businesses whose value sits mostly in identifying a problem and naming a product are the ones that should be rethinking their offer this year rather than next. If you want the bigger picture of where agentic AI is taking professional beauty, our deep dive Agentic AI & The Future of Professional Beauty is waiting for you inside Intelligence.
Comments