.png)
On October 5, a Utah company called Nolla Health announced that adults in the state can now get a first-time acne prescription from an AI system. You chat with an app, upload a photo of your face, and the software decides whether you qualify for one of a short list of creams and gels. The pilot price is $4.99 a month, and the company says it is the first time a US state has let an AI issue an initial prescription.
It is tempting to read that as the day AI became a doctor. The details say something more interesting, and they say it about the whole field. This first article in the series looks at where AI is already doing real work in healthcare. The surprise is that the clearest wins are not in diagnosis at all.
What the Utah pilot actually is
Under an agreement the state signed in late September, the app can issue first-time prescriptions and renewals for topical acne treatments only, for adults with mild to moderate acne. Oral medication, isotretinoin and severe acne are all excluded.
The rollout is staged. For the first 100 patients, two Utah-licensed physicians approve every prescription before it goes out. In the second stage, the AI sends prescriptions first and doctors review them weekly afterward. Any further loosening depends on how the data looks and on the state signing off. It is a one-year pilot, and Utah has said plainly that the agreement is not an endorsement of the product.
The company's case is about access. Its filing points out that Utah dermatology wait times average around 61 days against a national median of about 32, and that eleven Utah counties have no dermatologist. Acne is common, low risk and treated with fairly standard protocols, so it is a reasonable first test.
Not everyone is convinced. Earlier this year Utah ran a separate pilot, by a company called Doctronic, that let AI handle certain prescription renewals. In April the state's medical licensing board urged officials to suspend it immediately. The skepticism from physicians is already on the record.
What stands out to me is the staging. This is graduated autonomy: the system earns trust in steps, against data, with a regulator who can say stop. A lot of automation in the wild does the opposite. It is switched on for everyone on day one, and the safety review is bolted on later, if at all. Whatever you think of AI prescribing, the structure of this pilot is a template worth copying.
The boring stuff is winning
If you ask where AI has made a measurable difference in clinics, the answer is paperwork.
Ambient scribes, the tools that listen to a visit and draft the clinical note, are now among the most widely adopted clinical AI in the United States. A Doximity survey found the share of US physicians regularly using voice-based documentation tools rose from about 20 percent in April 2025 to 29 percent by January 2026.
The evidence has also become more sober, which is a good sign. In April, JAMA published the largest controlled study of these tools so far, led by researchers at UCSF and Mass General Brigham. It compared more than 1,800 clinicians using AI scribes with roughly 6,800 comparison clinicians across five hospitals over two years. The result was real but modest: less time spent on notes and in the electronic record, and a small but statistically significant revenue bump of about $167 a month per clinician. Time spent in the record outside working hours did not change significantly. The clinicians who got the most out of the tools used them for more than half of their visits.
On burnout the picture is better. A separate JAMA Network Open study of 263 clinicians across six health systems found self-reported burnout fell from 51.9 percent to 38.8 percent after 30 days with a scribe. That is a survey, not a controlled trial, so it deserves caution. But it matches what clinicians keep saying: it gives them their evenings back.
Money is following. On October 13, the scribe company Suki announced a $90 million Series E and said it plans to move beyond notes into revenue work and clinical reasoning. Scribes are the foot in the door. The real ambition is everything that happens after the note is written.
Scribes are workflow automation in its purest form: a repetitive, text-heavy task, a model that produces a draft, and a human who reviews and signs. They work because the human checkpoint is part of the design, not an afterthought.
The front door is becoming an AI
Patient-facing automation is scaling just as quietly. UnitedHealthcare told Healthcare Brew on October 8 that its generative AI assistant, Avery, now reaches 21 million commercial, Medicare and Medicaid members, past its original goal of 20.5 million. It is not glamorous, but it changes how millions of people deal with their insurer.
Mass General Brigham's Care Connect program, launched last fall, has a patient chat with an AI agent for about ten minutes, then sends a summary to a physician who can see them by video, often within a day or two. Boston Globe reporting counted more than 14,000 appointments booked this way, while noting that some doctors see the tool as a distraction. Patients in a region with multi-year primary care waits clearly value it. Critics worry it is a cheaper substitute for the primary care system that actually needs fixing.
And people don't wait for health systems to deploy anything. In March, KFF found that one in six US adults had used an AI chatbot for information or advice about their mental health or emotional wellbeing in the past year. Among adults aged 18 to 29 it was 28 percent.
Put together, the pattern is clear. The tools succeeding today are narrow, supervised and built around a person who can catch mistakes. The next article asks what happens when AI moves from the paperwork into the diagnosis and the drug pipeline, where the stakes, and the evidence standards, are much higher.