Running Through Walls

The episode contains a handful of genuinely substantive moments - the GLP-1 market structure analogy, the LLM-vs-OpenEvidence Nature paper, and the AI durability problem - but roughly half the runtime is consumed by personal anecdotes, insider jokes, and banter that adds nothing actionable for an operator.

“my bet is is that the profit margin in the GLP-1 market gets squeezed pretty hard on both ends of the spectrum. Both the generic side - you know, to pennies on the dollar - and the rebates on the high-end side”

“you used to have your first-horizon business last five to seven years and during that time, you can figure out what you're going to then build on top of that. And now you have less than a year”

The GLP-1 market comparison to the statin/ED convergence is a genuinely useful and non-obvious framing, and the point about frontier models collapsing the value of specialized healthcare AI products is sharp and timely; most other takes are competent but unsurprising VC commentary.

“the GLP-1 market is going to end up looking as though the erectile dysfunction and cholesterol-lowering markets had a baby”

“we are seeing time and again that the hot company from a year ago is the product roadmap for the large frontier models this year”

Bryan Roberts and Bob Kocher are legitimate, senior healthcare VCs at Venrock with real portfolio exposure to the companies and dynamics they discuss, giving their predictions genuine grounding; however, this is an internal team roundtable rather than an interview with an operator or founder who has executed at scale.

“many of our portfolio companies have already discovered that because they're delivering care and AI is enabling it”

“The most shocking M&A transaction that I've ever heard of where they didn't recut the price when the revenues were going to fall by 15%”

The episode anchors discussion in their proprietary survey data with named percentages, a specific price threshold prediction ($50 - 75/month in two years), the Nature paper as evidence, and a named deal (Aetna/Oak Street); the predictions are directional rather than rigorously evidenced, but the specificity level is above average for a short-form VC pod.

“54% of people believe that revenue cycle management and AI-powered administrative tools will have promising exits in the next 12 months”

“all of the top three LLMs - so, Gemini, ChatGPT, and Claude - performed better than OpenEvidence, significantly more so on clinical intelligence metrics”

The format is a casual internal team chat with minimal probing; Bryan occasionally pushes back ('How do you know what before was?') and asks a useful follow-up on payor coverage restrictions, but most exchanges devolve into affirmation loops or jokes rather than productive challenge or depth-seeking.

“How do you know what before was? You don't know what before was. It wasn't in the question”

“Even the coverage restrictions, right? Like what's been going on in the first half of this year. You think those dissipate?”