Advertise on podcast: Practical AI in Healthcare

Six conversations into Season 2, Steve Labkoff and Leon Rozenblit look back at what they added up to. They revisit Raihan Faroqui on AI agents at the front desk, Greg Raiz on data over gut feel in pre-seed investing, Randi Seigel and Jared Augenstein on the state-by-state map of AI law, Travis Zack on an evidence engine that grades its own sources, Sir Peter Donnelly on genetics as nature's drug trial, and Latha Valluripalli Soorya and Erica Crawford on smart glasses in the autism evaluation. Their read: the field is moving from hype into implementation, efficiency keeps arriving before accuracy, and "AI" may be too broad a word to regulate or even to argue about.

Families who suspect autism can wait up to two years for an evaluation. Dr. Latha Valluripalli Soorya, a clinical psychologist at Rush University Medical Center and co-founder of Argus Cognitive, and Dr. Erica Crawford, Clinical Director at Autism Testing 4 Kids in Dallas–Fort Worth, join Steve Labkoff and Leon Rozenblit to discuss Neurora: smart glasses a clinician wears during a standard autism assessment, with software that measures a child's gaze, facial movement and speech. AT4K was a reimbursed site in its FDA study. They cover what the clinician gets back, the hard cases where the data helps, what the published study does and doesn't show, and why the biggest time savings so far came from report writing.

Ninety-five percent of drug targets fail, and Professor Sir Peter Donnelly thinks the reason is simple: we can't run experiments on humans. Genetics is the next best thing. People who carry natural variants in a gene show what happens when that gene is dialed up or down over a lifetime. Peter, CEO and co-founder of Genomics, joins Steve Labkoff and Leon Rozenblit to explain how his company harmonized tens of thousands of genetic studies into one resource, what genetics revealed about "good" cholesterol, why the AI agent sits on top rather than at the core, and what genetics still can't tell a drug developer.

Doctors hallucinate too. That's the provocation Travis Zack, MD, PhD, chief medical officer of OpenEvidence, brings to this conversation with Steve Labkoff and Leon Rozenblit. A physician with 20 patients and no time to look things up goes with their gut, and can get it wrong in the same ways a language model does. Travis explains why OpenEvidence was built on retrieving evidence rather than generating answers, what makes one set of references better than another, how it grades the strength of evidence behind an answer in real time, why health systems and EHR vendors act as gatekeepers, and why the last mile of a clinical decision still belongs to a human.

By Manatt’s count, more than 280 bills to regulate AI in healthcare were introduced in 2026, and almost none of the resulting law has been tested in court. Randi Seigel, a partner at Manatt who leads its health care regulatory practice, and Jared Augenstein, a senior managing director at Manatt Health, track every one of them. They join Steve Labkoff and Leon Rozenblit to map where the states are headed, explain why consent and disclosure rules often miss the mark, describe how regulation can penalize companies that tried to do the right thing, and offer practical advice on AI governance for hospitals with one-person committees and academic systems with too many.

Early-stage investing is one of the last places where big decisions get made on gut feel. Greg Raiz, co-founder and general partner of the pre-seed fund Founders Edge, is trying to change that with data: a survey of more than 3,000 entrepreneurs and a review of some 42 published studies on what predicts founder success. In a solo conversation with Leon Rozenblit, Greg explains why venture capital may be backing the wrong age group, why angel groups are structured to kill their best deals, who actually gets the leverage from AI coding tools, and why electronic medical records are ripe for disruption. Leon is a limited partner in Founders Edge and discloses it at the top of the episode.

Raihan Faroqui trained in internal medicine and then went to work on the front desk. As VP of partnerships at Confido Health, he puts AI agents on the phone lines of what he says are about 1,500 outpatient practice sites, handling appointments, refills, insurance checks, and billing questions in twenty languages. His framing is that healthcare never had a software problem so much as a system of record problem, and agents are the first tool that does not demand a migration. In this conversation, we get into the economics of the unanswered phone call, why the deployment model looks more like consulting than software, how they define and measure agent failure, and the question the episode keeps circling: these agents do not tell patients they are agents.

Every technology healthcare has adopted moved through the same curve: a bleeding-edge few went first and documented what broke, and the majority followed. Large language models skipped that entirely. Jeremy Harper, a biomedical informatician who has worked at Epic, Ohio State, and Regenstrief, and who wrote Large Language Models (LLMs) for Healthcare, explains what we gave up by going all at once. Ambient scribes are everywhere, and because most vendors discard the audio as soon as they transcribe it, nobody can say how often the notes are wrong. He offers a fix borrowed from de-identification, a one-question test for any AI vendor, and a lament for the reusable knowledge we keep paying for and deleting.

Steve and Leon close Season 1 by looking back across all fifty-two episodes. They start with what surprised them (how fast conservative institutions adopted, and how slowly AI literacy is moving), then work through five conclusions a year of guests kept reaching independently: the models are no longer the hard part, almost nobody monitors these systems after deployment, an interface built for a novice can degrade an expert, "compared to what" is the question everyone skips, and money is usually the real gate. They also notice the show has started talking to itself, with guests answering each other across dozens of episodes. Several questions are left open on purpose and handed to Season 2.

In their seventh Reflections episode, and their fifty-first overall, Steve and Leon look back across six conversations: Mika Newton on interoperability that finally started working, Peter Embi on monitoring clinical AI after deployment, Vimla Patel on how clinicians actually reason, Renee Deehan on an engine built so it can't hallucinate, Christine Dymek on AI literacy, and Jeremy Harper on an adoption curve that broke. They decide against forcing a single grand lesson and find three threads anyway: keeping a human in the loop as a deliberate design decision, the national clearinghouse for AI errors that still doesn't exist, and whether literacy is even the right word for what healthcare workers need.

Chris Dymek came to healthcare informatics by way of philosophy, and she still thinks like a philosopher: the first question is what we actually mean. As Director of Digital Healthcare Research at AHRQ, she funded AI research, helped launch work on AI and patient safety, and wrote a request for information asking healthcare organizations how they thought about AI literacy. It was never published. She left in May 2025 and carried the work into the DCI network instead. In this conversation, she lays out the distinction at the center of her framework, knowing-that versus knowing-how, the three constituencies who each need something different, and why a literate staff and a literate patient population are what let a health system move forward without breaking things.

The internet is full of wellness advice built on a single cherry-picked study. Renee Deehan, a molecular and cell biologist who leads science and AI at InsideTracker, spent two decades building the opposite. In this episode, she explains why the core of their recommendation engine is symbolic AI, knowledge representation and reasoning, rather than a large language model: it's deterministic, fully auditable, and by design cannot hallucinate. The LLMs are fenced off to chat and summaries, while humans still write and review every recommendation against convergent clinical evidence. She and the hosts dig into a 20,000-user outcomes study, the discipline of refusing to claim causality, the MCT-oil case where the system decides not to recommend, and how a data-science team grew its own AI literacy instead of hiring it.

Dr. Vimla Patel has spent four decades studying how physicians actually reason, and what happens when technology ignores it. In this episode, she explains the difference between forward reasoning (the fast, pattern-driven hallmark of expertise) and backward reasoning (the slower, hypothesis-testing mode of novices), and why most clinical AI is built for the wrong one. Through two vivid cases, a textbook expert diagnosis and a near-fatal potassium overdose driven by a flawed order-entry system, she shows how good design preserves clinical judgment and bad design erodes it. She closes with a warning about confident AI and the quiet loss of independent thinking.

Peter Embi has spent his career at the intersection of medicine and informatics: coining the term "algorithmovigilance," serving as the nation's first Chief Research Information Officer, and leading AI work at Vanderbilt. He is also a patient. A rare adrenal tumor took nearly 15 years and a self-diagnosis to catch, and it nearly killed him. In this conversation, Embi connects that diagnostic odyssey to the case for monitoring clinical AI the way we monitor drugs: continuously, in the real world, across a network of institutions. The discussion runs from zebras and missed diagnoses to VAMOS, the "air traffic control tower" his team built to keep deployed models honest.

Interoperability has been healthcare's twenty-year broken promise, but something genuinely changed. Mika Newton, CEO of xCures, explains how provider data exchange jumped from roughly 30% to 85–90% in just two years, and why that's only half the story. Moving records, it turns out, was never the hard part. As Newton puts it, "records travel, but they don't translate" — most of a record arrives as duplicated notes and scanned images that still have to be made usable. Steve and Leon dig into the honest limits of AI parsing, the card-network model behind nationwide exchange, and why the patient is becoming the real access point to their own data.