Six Conversations In: Where Healthcare AI Actually Is
Six conversations into Season 2, Steve and Leon sat down to ask what those conversations added up to. Both landed in the same place. Leon: “We’re a little bit past the dizzy and excitement of like, ‘Oh my God, it’s AI,’ and a little bit more deeply into implementation and applications.” Steve put it on the Gartner curve, past the peak and into “this dose of reality and what’s working, what’s not.”
Practices are buying capacity
Confido Health, where Raihan Faroqui leads partnerships (S2E01), puts AI agents on outpatient front desks to answer phones, book visits, and chase refills. Practices aren’t buying AI, Leon noted. They want more patients booked and a front office that isn’t underwater. The hard part turned out to be the sticky notes around every monitor at the front desk, the local rules nobody wrote down. Turning those into written procedure is most of the work, and some of it is teaching staff how to document for automation.
Accuracy takes longer
That led to the hypothesis Leon put to Steve: efficiency shows up first because it’s easier to measure and easier to automate, while “accuracy really depends on science.”
The autism episode (S2E06) tests it. Latha Valluripalli Soorya, who co-founded Argus Cognitive, and psychologist Erica Crawford, whose practice AT4K was a reimbursed site in the tool’s study, described smart glasses the clinician wears during the standard autism assessment. Asked whether it made things faster or better, they said better, on the kids who fool the adult: the shy child who gets coded as shy, the verbal kid who handles grown-ups well. The time their clinic saved came from a separate report-writing tool.
Travis Zack (S2E04) holds OpenEvidence to a similar bar: an answer has to be right for the right reasons. Steve, who picked apart clinical guidelines as a fellow, wasn’t sure the guidelines underneath were ever written to be computed.
The edge that averaging erases
Greg Raiz (S2E02), co-founder of the pre-seed fund Founders Edge (Leon is a limited partner, with no say in what it backs), argues that group vetting kills the deals worth making. Leon’s gloss was Francis Galton’s ox. A crowd’s average guess at its weight lands close because individual errors are random. If one person knew the ox was stuffed, averaging would bury the one estimate that mattered. The same logic applies to any committee that reviews AI.
A word that doesn’t hold
Two episodes ran into the word “AI” itself. Randi Seigel and Jared Augenstein of Manatt (S2E03) track a wave of state law, 280-plus healthcare AI bills this year and 29 signed by Manatt’s count, with very little precedent behind it. Leon’s worry: a consent rule written with chatbots in mind could also cover the machine learning radiology has used for years without asking patients.
With Sir Peter Donnelly of Genomics (S2E05), “is this statistics or AI?” turned out to be the wrong question. The right ones were “Is it good? Does it work?” As Leon put it, genetics gives the company something rare in health data, a causal arrow: when a variant changes a protein, you know which way the influence runs. The moat, tens of thousands of harmonized studies, is boring and very hard to copy.
Out of the clinic
Erica imagined a parent wearing the glasses while a child gets ready for school, or a teacher wearing them in class. Leon, setting the privacy questions aside, called it an amazing possibility. Steve compared it to sleep apnea testing, which moved from an overnight lab to a kit at home.
Is efficiency always first? Steve wasn’t sure he could answer that, and neither are we. It’s the question we’re carrying into the rest of the season.
Listen to the full conversation: https://practicalaiinhealthcare.com/episodes/#reflections-9
You Might Also Enjoy
- S2E02 with Greg Raiz — The full argument for why committees average away the edge, and Greg’s four-stage AI ladder.
- S2E06 with Latha Valluripalli Soorya and Erica Crawford — The faster-or-better question in detail, including what the published study does and doesn’t show.
- S2E01 with Raihan Faroqui — What it takes to turn a front desk’s sticky notes into something an agent can follow.