The demo everyone shows is AI reading a medical scan. The deployment that's actually scaling inside health systems is far less glamorous: AI drafting clinical notes, filing prior authorizations, scheduling patients, triaging inboxes, and chasing denied claims.
This isn't an accident. Administrative work is where AI met healthcare's lowest barriers and its most acute pain at the same time. Ambient documentation tools don't diagnose anyone, so the regulatory bar is lower. And clinicians are drowning — documentation burden is one of the most-cited drivers of burnout. Give a physician back two hours a day and you don't need a clinical trial to prove the ROI; you can feel it.
Follow the money and the logic gets sharper. Administration consumes an enormous share of every healthcare dollar. Even single-digit efficiency gains there are worth staggering sums in absolute terms — far more, and far sooner, than the uncertain economics of AI-assisted diagnosis. The CFOs figured this out before the clinicians did.
But the interesting questions are second-order, and we're barely asking them.
First: who captures the savings? If AI cuts the cost of running a health system, do patients see lower bills, do providers see better margins, or do payers pocket the difference? Technology doesn't answer distributional questions; contracts and policy do.
Second: what happens to the administrative workforce? Healthcare employs armies of people doing scheduling, coding, billing, and authorizations. This is where the "AI and jobs" debate stops being abstract and gets a badge number. Retraining and transition planning should be starting now, not after the layoffs.
Third: accountability. An AI-drafted clinical note still gets signed by a human — but humans rubber-stamp things when they're tired, which is precisely when the stakes are highest. The liability frameworks here are unwritten.
My take: healthcare is the sector where AI's economic impact will be felt first and most concretely — not through robot doctors, but through the quiet automation of everything around the doctor. The policy conversation is years behind the deployment curve. It should catch up before the savings get captured by whoever moves fastest rather than whoever deserves them.
What do you think — where have you seen AI actually working in a hospital or clinic, versus just being piloted?