In Part 2 of our conversation with Matt Truppo, Global Head of Research Platforms and Computational R&D at Sanofi, we move from discovery to development, where the real stakes begin. Matt unpacks the promise and limitations of “digital patient twins,” a concept often described as the holy grail of drug development. With nearly 90% of drugs failing in clinical trials, even modest gains in predicting efficacy or patient response could transform the industry. Through real-world examples, including Dupixent and rare disease therapies, Matt shows how quantitative systems pharmacology (QSP) and AI-driven simulations are already shortening timelines, reducing patient burden, and, in some cases, eliminating the need for entire trials.

But the story doesn’t stop at modeling. We explore how AI is reshaping clinical operations, from Sanofi’s “clinical control tower” that integrates trial data across 4,000 users, to generative AI tools that are cutting regulatory document creation time by more than a third. Matt also shares a personal experiment, building a network of AI agents modeled on his own workflow, reclaiming 30% of his time and offering a glimpse into a more “agentic” future of work. The throughline is clear: AI is not replacing human expertise, but amplifying it, helping the industry finally bend the cost and time curve of drug development.

Ted Shortliffe built MYCIN at Stanford in the 1970s, one of the first medical AI systems ever deployed in a clinical setting. Five decades later, he joins Steve and Leon to examine what has persisted in clinical decision support — above all, the demand for explainability — what has changed (computational power finally caught up to the ideas), and what the field may have lost along the way. The conversation includes a direct response to Bob Wachter's claim from S1E24 that AI in healthcare decision support was "too hard a problem to start with," and a case for why structured knowledge representation deserves a second look in the age of LLMs. For anyone tracing the arc of medical AI history, this episode is a rare primary source.

AI in drug discovery has been long on promise and short on delivery. Matt Truppo, Global Head of Research Platforms and Computational R&D at Sanofi, presents a different picture. His team used AI to identify 10+ novel drug targets in 12 months, screen 30 million target combinations in days, and produce AI-designed compounds with 75% synthesizability. But Truppo is equally candid about the gaps: data integration, explainability, and change management remain real barriers. In Part 1 of this two-part conversation, hosts Steve Labkoff and Leon Rozenblit explore what happens when AI moves past pilot projects into core pharmaceutical science.

Every five episodes, Steve and Leon step back to examine what picture forms when you put their guest conversations side by side. This time, five guests from completely different healthcare domains -- data quality, clinical trials, medical translation, patient data, participatory medicine -- independently converged on the same conclusion: the AI works; the infrastructure around it doesn't yet. From Charlie Harp's data quality metrics to Adam Blum's 60-to-90% scaffolding story to Amy Price's reframing of healthcare AI as "unfinished, not broken," Block 4 reveals what industry maturation actually looks like -- not a breakthrough, but a quiet shift in what the conversation is about.

Amy Price survived a car accident that left her with a broken neck, severe brain injury, and $4 million in medical bills. She was told she'd need to be institutionalized. Instead, she earned a DPhil at Oxford and became Editor-in-Chief of the Journal of Participatory Medicine. In this episode, Amy sits down with Leon to discuss why patients belong inside the AI design process, what it really means to have a "knowledgeable human who cares" in the loop, and why healthcare AI is an unfinished system worth building on, not a broken one worth scrapping. She also shares how she uses AI tools for her own health decisions and what she's learned about closing the patient AI literacy gap.

Shashi Shankar spent nearly a decade at Genentech before a family cancer journey and a broken data landscape pushed him to build something different. His company Novellia works directly with patients — not data brokers — to collect and consolidate health records across multiple providers using SMART on FHIR. The result: longitudinal, patient-authorized real-world data that fills the gaps left by claims databases, single-site EMRs, and health information exchanges. We explore why previous PHR companies failed, how AI catches clinical data errors that humans miss, and whether Big Tech should be trusted with patient data.

When Adam Blum was diagnosed with follicular lymphoma, he tried over a dozen commercial trial matchers. None returned actual matches. So the serial AI entrepreneur built CancerBot, a free precision-matching service that assesses 100% of eligibility criteria — not the five surface-level attributes most matchers use. On this episode, Blum explains the Prompt Workbench (where biomedical experts refine extraction prompts to above 90% accuracy), how conjunctive normal form makes complex eligibility logic tractable, and why "best trial" means something different for every patient. A masterclass in AI scaffolding for healthcare.

For 37 years, Charlie Harp heard the same thing from healthcare organizations: "Our data quality is fine." They were right — for billing and scheduling. But AI changed the equation. Harp, founder of Clinical Architecture, built the PIQI framework to measure patient data quality across four dimensions: availability, accuracy, conformance, and plausibility. His PIQXL Gateway scores data on a 1-100 scale before it enters your systems — not after. Early deployments reveal uncomfortable truths: lab data averages 70% quality against USCDI standards, and one facility coded every blood test to a single LOINC code. The framework is now going through HL7 balloting as an open national standard.