Healthcare AI in Fall 2026: What Changed and Why It Matters

Every few months the healthcare-AI headlines reset. This episode is the two besties, Dr. Junaid Kalia and DR GPT™ Harvey Castro, pausing to ask what actually changed this fall, underneath the noise. No guest, just a deep dive, and it lands on two shifts that, from opposite ends, point to the same future.

“You don’t need language to please me. You need decisions to please the machine.” Dr. Junaid Kalia

Junaid’s shift is the one most people missed: we have started building AI models for machines rather than humans. Today’s large language models were trained with reinforcement learning from human feedback, optimized to produce text that pleases a person. But when AI is the back-end running software, you do not want paragraphs another model has to read. You want a calibrated decision and a probability. He describes a new class of System-1-style “decision” models that take a question and a set of choices and return an answer plus its likelihood, nothing else, which makes them astonishingly fast and cheap. The clinical payoff is concrete: feed a patient’s chart and ask “is this patient on warfarin? what was the last refill?” and get a decision with a probability in a fraction of a second. He claims you can run 600 such questions on a single chart in under a second, turning the whole EHR into a “decision machine,” at roughly $6,000 where a conventional model would cost $30,000. He also notes SaveLife.AI™ has brought its own GPUs online to build in this direction.

Harvey’s shift is human. Reactive medicine, he says, is personal: his father died of a heart attack in his early fifties, an emergency that had quietly been building for years with no symptoms. His mother is the opposite story, a cancer caught early by chance, and she is alive twenty-plus years later. That contrast is the case for proactive medicine, and his single favorite lever is VO2 max, the body’s ability to extract oxygen under exertion. The striking claim, backed by large cardiorespiratory-fitness data, is that improving low fitness reduces mortality risk more than resolving smoking or diabetes. He adds the un-glamorous, no-prescription basics: around 7,500 steps a day, sleep, and picking one habit you will actually keep. His grandfather, still strong enough to carry his own bags onto a cruise ship, is the north star. His line for the whole hour: the best emergency is the one that never comes.

Put the two together and the episode’s real argument appears. Faster, cheaper, decision-first AI is only worth building if we point it at the right goal, and the right goal, Harvey insists, is keeping the emergency from ever starting.

Connect with us

Dr. Junaid Kalia, Neurocritical Care Specialist, Founder of Savelife.AI™

DR GPT™ Harvey Castro, ER Physician, #DrGPT

Edward Marx, CEO, Advisor

Originally published on my Substack. Looking for a speaker on proactive medicine and AI? Explore my healthcare AI keynotes.