Utah's AI prescribing pilot shows how a profession hands over a task in three phases

TechSpot reported on 6 October that Utah will become the first US state to let an AI examine patients and prescribe medication without direct human oversight. The healthcare start-up Nolla Health runs the pilot, which Bloomberg first reported, according to TechSpot. It lasts one year and covers only patients with mild-to-moderate acne. A patient pays $4.99 a month for the app, scans their face and fills in a medical history. The AI then produces an acne severity score and recommends medication from a list of eight approved treatments. All eight are topical.

The headline says a doctor is being replaced. The detail is more useful than that: a published schedule for how a profession hands over a task.

The three-phase handover

TechSpot sets out the sequence. Doctors review the first 100 prescriptions before they reach the pharmacy. For the next 400 patients, the AI submits prescriptions itself and a physician reviews them each week, after they have gone out. In the final phase, a doctor reviews at least 10% of prescriptions, once a month.

Read that as a job description changing in real time. In phase one the physician approves every decision. In phase two the physician checks decisions already made. In phase three the physician checks a sample. The title of doctor lasts through all three phases. Prescribing for this condition does not.

This is the pattern The War Room keeps pointing to. Tasks are absorbed first, then the role is hollowed out, and the title goes one restructuring cycle later. TechSpot itself describes doctors as one of the jobs expected to be least at risk from AI replacement. Inside a protected profession, the narrow, well-bounded tasks where an error does little harm go first.

Nolla Health's chief executive, Luis Wenus, said the side effects of the eight medications are limited to localised skin irritation or dryness, according to TechSpot. That is the selection test: errors are cheap, the outputs come from a fixed list and the input can be measured.

Where the value moves

After phase three, the physician keeps the part the system cannot carry: answering for the sample, judging the edge case and deciding whether the programme should widen. TechSpot reports why that role matters. It cites the case of a 60-year-old man who followed ChatGPT's advice, replaced the salt in his diet with sodium bromide and developed bromism. The audit role exists because that kind of failure exists. Whoever owns the audit owns the risk, and whoever owns the risk holds the leverage.

AI does the routine work. You do the thinking. The Utah pilot writes that sentence into a regulatory timeline and puts a number on it: review drops from every prescription to at least one in ten.

The usual mistake is to ask whether your profession is safe. A better question is which of your tasks fit the Nolla profile, and who writes the review schedule when those tasks move.

Your Next Move

• Run the Nolla test on your own week. Mark every task with a measurable input, a fixed list of outputs and a low cost of error. Those are the tasks most likely to be handed over in phases first. The method for sorting a week of work by how much of it AI can absorb takes an afternoon.
• Take the review seat before it is assigned. If your team is piloting an AI tool, offer to define the checks. Decide what gets reviewed before release, what gets audited afterwards and what sample size is enough. The person who writes the oversight schedule keeps the judgement layer.
• Log your catches. Keep a running record of decisions where the standard answer was wrong and you caught it. That record is your evidence for being the person who does the audit, and the audit is the part of the job that stays with a person.