Utah just became the first state to let an AI renew prescriptions without a doctor's review [1]. Not as a pilot program. Not with a human watching over its shoulder. Actual prescriptions, going to actual patients, approved by software.

The system costs $4 per renewal. It covers 191 medications. And a recent study suggests that when AI and human doctors disagree, the AI is right four times more often.

If that number makes you uncomfortable, good. It should. Not because it's wrong, but because it forces us to confront something most industries aren't ready to discuss.

The study that started a debate

Doctronic, the company behind Utah's AI prescription system, published a study comparing their technology against board-certified clinicians across 500 urgent care visits [2]. The headline finding: 99.2% alignment on treatment plans.

But that's not the number that matters.

In 97 cases, the AI and human doctor reached different diagnoses for the same patient. Independent expert reviewers evaluated each disagreement. The results:

  • AI was superior: 36.1%
  • Human was superior: 9.3%
  • Too close to call: 54.6%

When they disagreed, the AI was right roughly four times more often than the human.

I've spent 30 years working in technology and healthcare. I've seen plenty of vendor studies with impressive numbers that crumble under scrutiny. So my first instinct was skepticism. But the specifics here are worth examining.

Why the AI outperformed on disagreements

The AI showed stronger adherence to clinical guidelines, particularly for conditions with established protocols. It was more consistent when patients presented atypically—exactly the situations where human pattern recognition sometimes fails.

Here's what I found most interesting: a third of the "disagreements" weren't real disagreements at all. The human simply documented with less precision. A doctor wrote "viral infection" while the AI specified "viral pharyngitis." Same diagnosis, different specificity.

The caveats that actually matter

Before anyone declares victory for the machines, let me be direct about the limitations.

The study is a preprint. Not yet peer-reviewed [2]. All authors hold equity in Doctronic, a significant conflict of interest. The human doctors saw the AI's notes before their own evaluation, which may have biased them toward agreement.

And perhaps most importantly: urgent care telehealth is routine medicine. We're talking about prescription renewals and common conditions, not complex diagnostics or rare diseases.

Utah's Office of AI Policy built in safeguards. The first 250 prescriptions get reviewed by a human doctor. Licensed physicians remain on call. If the AI is uncertain, it refers patients to a human [1].

These are exactly the kinds of guardrails I recommend when clients ask about AI implementation. Start narrow. Monitor closely. Expand only with evidence.

The pattern you should recognize

What's happening in healthcare has already happened in legal document review. In financial analysis. In code review. In every domain where routine expertise commands premium pricing.

The question is no longer "can AI do this work?" The technology has answered that. The real question is harder: what happens to the economics of industries built on human expertise when AI matches or exceeds that expertise for routine tasks?

I've watched executives cycle through predictable reactions. First denial: "our work is different." Then fear: "we'll be replaced." Then, for the thoughtful ones, strategic clarity: the value shifts from routine execution to judgment, oversight, and handling the cases AI can't.

What this means for you

Don't extrapolate too far. A $4 prescription renewal is not the same as a complex diagnosis. The AI succeeded in a narrow, protocol-driven domain. Know what "narrow" means for your industry before drawing conclusions.

Watch for this pattern in your sector. If your business involves routine expertise—reviewing contracts, processing claims, analyzing reports—the economics are about to shift. Not because AI is perfect, but because it's consistent and cheap.

Focus on what AI reveals, not just what it replaces. The study showed that a third of "disagreements" were actually documentation differences. AI often exposes process inconsistencies we've learned to ignore. That's valuable intelligence for operations leaders.

Build evaluation frameworks now. Utah required expert review of AI-human disagreements. That's the model. You need a way to assess when AI is right, when humans are right, and when the question itself is wrong.

The uncomfortable conclusion

The Utah experiment isn't about whether AI should practice medicine. It's about what happens when AI demonstrably performs routine tasks as well as—or better than—the professionals we've always trusted to do them.

The professionals who thrive won't be the ones who fight this transition. They'll be the ones who understand where human judgment still matters and where it doesn't.

For business leaders, the question isn't whether this pattern will reach your industry. It already has, or it will soon. The question is whether you'll lead the transition thoughtfully or be forced into it reactively.

The $4 prescription renewal is just the beginning.

References

[1] Utah Office of AI Policy, https://ai.utah.gov/

[2] Doctronic preprint study on AI diagnostic accuracy, https://www.doctronic.ai/research