How AI Will Finally Make Healthcare Deflationary | Eric Larsen

If I look at healthcare, out of our $5.3 trillion sector, 2.9 trillion is in labor. We have 23.8 million people employed, the only industrial vertical to see negative productivity growth. You know, healthcare is great in this country if you're rich, educated, white and urban. It's not great if you're not in one of those privileged categories. And so we have 250 billion in medical debt. The leading cost of bankruptcy in this country is medical debt. So the deflation has to happen in healthcare. And I predict we're going to see five to 700 basis points taken out of US GDP allocated to healthcare over the next several years. If we can penetrate the regulatory walls and if we begin to see the labor substitution. And I'm not, I don't want to be reductionist in this because right now there are 1.8 million unfilled jobs in healthcare. If you lower the cost of something, there's something that you and I are familiar with called the Jevin's Paradox. The cheaper a commodity becomes the more people use it. The cheaper healthcare becomes the more people are going to use it. And if we add a couple decades to human longevity, we're going to need more healthcare. Today, on the show, Steve speaks with Eric Larson, the president of Towerbrook Advisors and Adventure Partner at Thrive Capital. Welcome to The Heart of Healthcare Podcast. I'm Halle Teco. I'm Michael Eskadell. And I'm Steve Krause. And every Monday, we bring you the latest in healthcare innovation as we sit down with entrepreneurs and industry experts. So buckle up and join us as we figure out how to improve healthcare for all. Hello listeners and welcome back to The Heart of Healthcare. I'm your co-host Steve Krause. And today, I am so pleased to have a repeat guest, a Webby award-winning guest, may I say. Mr. Eric Larson, who for our listeners, our loyal listeners, will remember we had on the pod about a year ago where we discussed his tone of thought piece called The Gen AI Juggernaut. Healthcare is not prepared. And we're going to do a look back on many of the predictions and the discussion that we had a year ago, that obviously was a very popular episode. We thank all of those who voted for that in the Webby Awards. We're very proud. And thank you, Eric, for being such a great guest. We're going to do that in today's episode. And also, as a preview, we are going to have Eric back on because he has a new, very lengthy thought piece. If you have a question for Eric that we can discuss on our next episode, which we'll probably have in about a month, drop it in the link in the show notes. And we will compile them and we're going to bring Eric back on to answer them. But Eric, welcome back to the Heart of Healthcare. Steve, it's awesome to be here. I really listened to our pod for April last year this morning and it was super fun. And the Webby was nice, but we've got a big bar at a super seed today. I know. I mean, hopefully we can run it back, you know, and we can probably do this for the next decade. Hopefully we do. Hey, let's dive into it on a scorecard. You called this the most important moment in US history and said that the US Healthcare industry has the greatest surface area exposure for Janet. I to disrupt now that we're you're into it. Did your predictions hold up the macro one that start with that? Do you feel vindicated? Or does a year of watching the actual deployment make you want to walk back on anything that you said? I think we were pretty right. And that's not, I don't say that in a self-aggrandizing way. I say it in more a totally humble, observational way. I think if anything, we're getting, I think the intuition is getting super corroborated, right? Like most powerful technology in history, multiplication of intelligence. I think 12 months later, almost 13 months to the day, I think we know a lot more about almost the neurology of this intelligence. I think we can describe it with more precision than a year ago. And maybe if you'll indulge me, I'll take a pass at that in a minute. But I think in terms of healthcare's susceptibility to disruption, I think we're super vindicated for better, for good or for bad. I mean, last year, I kind of articulated the reasons I thought, healthcare, when you strip away the mesmerizing, generative elements of this, it's about brute force productivity augmentation. And I kind of postulated that US healthcare as the greatest labor addiction, labor intensity, 3.8 million people. And since then, Steve, all we've done is add humans to healthcare. Yeah. Yeah. See, that's generative to your point, because we talked about how it's, we're gonna have to reshape the labor force. And yet, I think of anything, this has been inflationary, not deflationary so far, and that we've added the AI, the cost of the AI, and granted, there are some outcomes that probably have been reduced and proved. But I think total cost is going up over the last year when you add the AI plus humans, because I don't see a lot of humans being removed. No, in fact, quite the opposite. And we've added 750,000 people to the labor ranks. Right. And if you think about it, like US healthcare is atlasing the entire hiring economy. Oh, yeah. I think it drives 50% of net new hires, I think, in the labor statistics if I'm correct. You're right. If you take away the 750, the labor force lost 200,000 employees, and can you imagine the convulsions that the equity markets would go into? You'd have emergency Fed proclamations, probably a 150 basis point cut to the Fed's fund, right? It'd be catastrophic. And so healthcare has been buttressing up the entire labor economy. And I think it's just a matter of time before you start to see the labor substitution. And when we were together a year ago, I kind of made the observation that the early applications ambient listening, summarization, documentation, coding were pretty unimaginative. Yeah, yeah, I think you called them a little boring. I did. I was super uncharitable. And some buddies in the industry like Shiv Rao, who we really love from a bridge, was like dude, that was super uncharitable. But as we know, Shiv has big ambitions and has done great things in the interim. But I would say that that is not true anymore. I think there are two milestones, one in December and one on February 5th. And we saw the emergence of a real nonlinearity in the models. We saw the emergence of Anthropics, Claude code in December. And then you had a lot of the engineering class and the programming class like basically disappear away from their families during the holidays and just like vibe code. And you had people like luminaries like Andre Carpathi come out and say, this is a massive magnification of capability. And then you saw another step function jump on February 5th of this past year, where you had this almost like, cosmically coordinated release between open AI and Anthropic, chat GPT, five, three codex. And you had Anthropic release Claude 4.6. And that was just another massive, not just evolutionary, but a really revolutionary jump and capabilities. And so when I think about the first chapter in the deployment of the application layer, it was inflationary why? Because the incumbents weaponized the existing model of fee for service and coding optimizations. And it was really just an entrenchment of the old payment hydraulics, right? And you saw it sort of asymmetrically benefit the providers more than the payers. So HCA hit over 115 billion dollars in market capitalization. You saw, you know, going from a trillion dollars market capitalization for the seven publicly traded managed care companies down to about half of that. You saw almost a half trillion incinerated. You saw two out of three blues plans last year lose money, right? So if trend for the country was 9%, trend for the blues was 13%. And a lot of CEOs in unguarded moments would say totally explicitly this is about coding optimizations. And then I'm not going to weigh in on that debate. But I will say that I think the first chapter was inflationary because it was used to reinforce the old adversarial model. The next chapter is going to be deflationary. And the way I think about it is, if you're using the technology to amplify your reimbursement mechanism, your touchment mechanism. The second chapter in that is going to be automating the humans that do that, right? So we spend $5.3 trillion on our industry every year, $2.9 trillion is in labor, right? And I imagine we'll get to it, but we talked last year about automating, augmenting, or eliminating. And in the new paper, I've tried to decompose the 23.8 million jobs in US health care. Getting at the 7.2 million jobs for hospitals and health systems, the 611,000 jobs for payers, the additional 209,000 jobs for PBMs and TPAs. And really ask the question, how much is augmentable and how much is substitutable? But it kind of, it passes the intuition, smell test, if you begin to automate some of those functions, especially in areas like functional verifiability. And I might take a step back, Steve, because one of the things that's happened in the past 12 months is I think. the two most important words in AI are functional verifiability. And what's I talk a little bit about more about this, 'cause I actually haven't heard this yet, and I wanna learn more. - Yeah, it's a little bit of my rubric. I mean, it's Andre Cupathy again, this is the second time I've quoted him, 'cause he's such a, he's sort of a demigod in the programming in AI world. He said, in the previous version of software, software 1.0, if you could specify the function you could automate it. In software 2.0, which is AI, if you can verify the function, you can automate it. And it's one of the reasons we've seen this absolutely parabolic takeoff in coding. It's why cursor's getting acquired by Elon for $60 billion. It's why Anthropic has gone from $9 billion in revenue in December to a projected 80 to $100 billion at the end of this year, right? - Yeah, it's an issue. - Yeah, and you know, we at Towerbrook are investors in Anthropic, you at Bestmer are at earlier investors in Anthropic, but the functional verifiability comment matters because anything with math or coding or a right or wrong answer or provability is automatable. And that's one of the reasons we've seen such step function jumps in revenue cycle management. And we talked a little bit last year, Towerbrook and CDNAR own R1 and our CEO, Joe Flanagan, I think is one of the most prescient technologists, CEOs in deploying the technology into areas of functional verifiability where you get massive augmentation. And so, but I do think this next chapter is going to be deflationary because it's a little bit reduxio added at absurdum. Like my bots are going to fight your bots, right? The providers were quicker off the draw than the payers, but the payers are smart and they are industrializing their own agentate capabilities really quickly. Yeah, eventually it's just going to cancel each other out. I want to come back to it, but let's just stick on the application layer because I think that was where you were the least charitable, as you shall say. And it seems like the application layer is so far, at least in healthcare, while there's been massive, massive companies built, as you said, in Anthropic others in the foundational model layer. And let's come back to it as it replies to healthcare. The application layer is where we've seen the most explosion and growth of these companies. And so I want to ask you a question as it relates to those companies that I think a lot about is if AI, you're able to build software really quickly, right? And you constantly have to stay ahead of the game. Like long-term, if you were the CEO of an application layer, or founder, CEO of an application layer company, I think maybe you can opine on what you think would be protectable versus not, but like what are the moats in this new world, especially in our industry? In other words, are we going to have another sassacre for vertical AI application layer companies? Or do you think there's a way in building those companies that you can create real defensible moats versus an epic incumbent who might use AI or versus an Anthropic who presumably can do almost anything. And Dario clearly has this site set on almost anything. Like how do you think about that? - Yeah, I mean, I'll be a little provocative. I think software is largely uninvestable at this point. And I just look at traditional software. - Well, I'm actually going to be a little bit more inclusive in that, and I think this is a real sort of like admonition to the startups. The way I think about it is there's just this inexorability around the God models, right? And their generalized capabilities are spilling over into every verticalized domain. And if your moat before was the difficulty of the programming of your software, that's obliterated. There is no moat. I'm a technophile, but I have zero coding capabilities, and yet I am vibe coding all the time with cloud code these days. And the barriers to entry and the activation energy is just really low. To get that escape velocity and just begin going, you can do it in natural language. And so if the moat before was the complexity of your tech stack, that doesn't exist anymore. And so if we're talking about healthcare, yes. I think that's one of the domains where you do have some defensibility, largely because of the regulatory enclosure and the hyper litigiousness of our sector. And so I think the regulatory moat is our friend in the short term, and it's how do you get to the network effects of where each incremental user adds to the relevancy of your platform, right? The conventional definition around network effects. How do you get such subatomic expertise on your vertical like unquestioned domain expertise that you can encode that into your platform? How do you create a UX UI around the platform that it really gets that stickiness and the frontline users are gonna clamor for it? That's a really high bar. And I look at, you know, I'm a big fan of hypocratic and what Manjala's built. I'm a big fan of Shiv and what he's built. I'm a big fan of a lot of Wallach and what he's built at ADOC. There are gonna be great companies that emerge from this, but there was 202 billion invested into startups last year. I think it was, and you and I should check our math, but I think it was probably 250 in total, but disproportionately it was an AI. But the other kind of part that we got to disambiguate was a lot of that went to the God models, right? So that consumed a lot of it. - Yes. - But the point is, I think you're gonna see a lot of capital incineration in this vintage, a lot of malinvestment. And it's a little bit of a San Hill road sort of conundrum right now, which, and you guys are the intelligentsia and Boston, so you guys are removed from the chaos of San Hill road. - It's everything safe here. Nothing's different here. - Everything here, yeah. But, you know, what's investible? Like, everybody's throwing money into the God models, right? For obvious reasons. And you start to look at companies like, you know, Harvey is a really interesting example. Obviously not in healthcare, but you know. - Yeah, we're in La Dora, for instance, which is a Harvey equivalent. - Yeah, yeah, yeah, yeah. - So the question is, how much is defensible? They, you know, they had a $12 billion valuation, they built their own foundation model, and it was super like, it was super tuned exclusively to this very esoteric legal domain. - And great, I'll take for both companies, right? 'Cause we're talking this, yeah, yeah. - But then, but then Sacha goes ahead and just throws it in co-pilot, you know? And so now, I think they're very wisely pivoting more toward adapting on top of the foundation models. And so, you know, I listen to Jensen who I have incredible respect for. And he talks about the five layer, hey, right? Power and then the GPU stack and then the God model stack and then the application layer, et cetera. And where does the monetary value accrue? - Well, it's pretty unambiguous. When we were together a year ago, Jensen had a $3 trillion valuation. Today, he's got a $5 trillion valuation. So check that box. You are seeing a massive industrialization in CAPEX. Like, last year when we talked, you know, the hyperscalers had invest $370 billion in CAPEX the year before, and actually it turned out to be about 420. This year, we're projecting $800 billion in CAPEX. Next year, we're projecting 1.2 trillion. And next year at 1.2 trillion, the hyperscalers CAPEX will be 3.3% of US GDP. - That's crazy. - The value is accruing at the God model layer. It's accruing at the hyperscaler layer. I mean, the mag 7 is now worth 22 trillion. I don't see it accruing as much on the application layer. And so healthcare is, you know, it's not gonna have the same mass extinction rate that we saw peripendemic with 13,000 diagnostic therapeutic, care, augmentation, clinical, nonsensical workflow companies with a 95% attrition. But it's hard to be a VC right now and no, you know, I don't need to tell you, brother. Let's set your domain. - Thanks for showing me a little sympathy, yeah. It's actually, it's, I agree in where you place your bets is it's hard to see. It's so dynamic right now is what I always say, which makes it also really fun. So while it's hard like strategically, it's also like just, I mean, if we all can't do well in this end, you know, all of us entrepreneurs, you know, our industry investors can't do well. I don't know when we will, 'cause this is just a crazy, crazy, crazy fun moment. But hey, that's super helpful. Let's tackle another subject. Then I might jump in with a few questions, but last year we talked about the incumbents versus the insurgents. - Yeah. - You know, when you sort of sat on the fence last year, I'd be honest, you weren't really clear, you know, you said actually it had to be a marriage, but you were skeptical about the incumbent's track record as sort of co-development partners. And what's your take? - I actually don't think I was hedging last year as much as I was making sort of a declaration that it has to be incumbents and insurgents together. And I still believe that for the short term. You know, and I think I said that, you know, in currency is not some invariant law of nature. It's a head start. And I still believe that. I mean, look, you know, US health care is just as, it's even more oligopolistic than it was a year ago. The top 10 health systems now represent $455 billion out of the $1.6 trillion sector. The top 100 health systems represent $1.2 trillion out of that $1.6 trillion. It is as all agopalistic as ever. I still think that incumbents in our industry, they have to co-develop and, you know, they've sort of abdicated every tech phase shift to the past generation. And with each turn of the crank, you know, in the year since we were together, Steve, you've seen memory, so reasoning had emerged toward the end of 24. And then with the advent of '03 in December of '24, it really took a step function. Across 25, you saw real augmentation in memory. Then you saw tool control. Then you said, kind of went from the scaling hypothesis, much more to test time compute and sort of inference time compute. And the models got smarter the longer they thought. And then you got to agentification, right? Proto-agentification. And with each turn of the crank in the models, from reasoning to memory to tool control agentification, the incumbents lose their agency. They lose their jurisdiction. They lose their self-determination. And so what I've seen over the last year is a real stratification. A few pair and provider in life sciences and med tech CEOs have decided to be autocratic. This isn't a democratized thing. One of the lessons the last year is that this isn't, you can't make this a, and you can't make this a really gentle enfranchisement persuasion thing. It's got to be unambiguous from the CEO. Yeah, you got to repill it. You got to repill it. You got to go all in. And one of the lessons from history, you know, we did not talk about this last year, is that ever since the first industrial revolution, we've seen this absolute like flourishing of humanity. And the three industrial revolutions that we kind of looked at, the mechanization, electrification, computerization, now agentification, I think people have internalized all the wrong lessons from history. They kind of looked at that and said, hey, it's the inventor of the technology that wins. You know, if you invented the steam engine, you know, in Great Britain, that's why they got all of the military and political and economic and sociological benefits. In the second industrial revolution, Germany sprinted forward in chemical capabilities and the US sprinted forward because it defused electrification. And the third industrial revolution, the US beat Japan because we invented more on the computer side. It's not invention, it's diffusion. The country, the company, the CEO that spreads in and beds the technology, most horizontally, that does the installation the fastest wins. And the same is true for the 150, what's really funny, just a little as little as I, you my friend last year characterized the 150, you said the F-150. I've never said that in my life. And it was so hilarious, is that now everybody refers to it as the F-150 now. Now I have to use the F-150. So, so I thank you and I, I think it's back a little bit to the history of the, there you go, the Ford motor company. Yeah, that's true. But the, so your, the, the inside here is that the CEOs in the payers, providers, med tech and life sciences companies that have basically come up with a blueprint on diffusion. I've got some ideas and some observations and some recommendations. They're the ones that are not abdicating their co-development responsibility. They're the ones that are shaping the technology because this generalized technology from the base models is, healthcare is too sacred and it's too idiosyncratic just to have some 20-something genius techno-solutionists design. It's an installation for healthcare. Let me ask you a question. I totally get you. Who do you think's doing it best in healthcare right now? If you had to point to one CEO, one company, who's the leader? Warner Thomas at Sutter is, I think, sprinting ahead. And how has that shown up in Sutter's business or is it too early to tell? Well, I think they asked Warner and he would tell you that he is, he's been super unambiguous with the board, with the senior leadership team that, you know, Sutter's going to be an AI learning organization. And so they're diffusing a base model. They're really empowering the, the, the frontline staff and the tens of thousands of Sutter associates to become a facile with the tech. They are, you know, doing sort of a tight-loose tight regime where tight, they're sort of unambiguously. This is what we're going to do, but we're going to be hip-a-compliant, we're going to make sure we have all data sovereignty, no data exfiltration. So that's the tight part. The loose part is how do you get that creativity that is happening at the individual level? Yeah, right? This is incredibly democratizing technology and in a way, it's really empowered and enfranchised the individual, but the productivity benefits haven't accrued to the enterprise. They've accrued to the individual. And so it's a little bit of like, how do you capture that creativity from folks that are going to be ingenious and going to be inventive and come up with productivity improvements? But how do those diffuse into the enterprise? And then that's the loose part. The tight part at the end is absolute rigorous measurement, codifying the winds, putting it into monetary terms. You know, what is our reduction in S-GNA? You know, it's not about reducing labor per se. It's about mitigating the need to hire more. I really admire what Lily is doing. And they're actually building, I think, the largest supercomputer ever in life sciences with Jensen and really focusing on AI engineered biomolecules and biologics and then also simplifying all the machinery around drug discovery and development. One company you didn't mention, who might be the largest, or probably is the largest software company's epic? And man, are they great at creating fun in the market? I'm curious, where do you think they are in their journey? How would you grade them and related maybe to our application layer question? They are software, for sure. Like, is there ever a world where, you know, my colleague Sophia calls it the octopus strategy, where some of these other application agentic layers come around and sort of strangle epic, which by the way, I think entrepreneurs across America would like raise a glass too. And all of a sudden we might have, you know, so you said software is no longer that special. I understand Epic has a lot of other advantages. Do we ever see Epic go away? Maybe, maybe not under Judy's regime, but somebody else's. Like, where do you think Epic stands? Is one of the largest incumbents? >> Yeah, I mean, I have cognitive dissonance on this one because on the one hand, I'm sympathetic to the entrepreneurial community and the, you know, the sort of the startup ecosystem that is dealing with a juggernaut in Epic. And on the other hand, I think very highly obviously of Judy, but also Sumerana, their president and Seth Hain, who's just a brilliant guy. And so I admire what Epic has done from a sort of mobilization of a platform. And I think Comet, I think Cosmos has something on the order of 300 million longitudinal patient records. So it's the biggest structured and unstructured data repository in the industry. I admired when they built their own foundation model. And at the first they called it Comet. I think Judy came up with a much better better name for it afterward. And in terms of an incumbent mobilizing, I think Epic has done a very fantastic job. But I do think that you're going to see an arm the rebels strategy. And, and you, you've seen this proliferation of a lot of brilliant founders that are focusing on not just the administrative simplification domains, but the clinical AI domains, both in, you know, differential diagnosis and imaging and treatment and care protocols and then very specialty specific emergence. It's an unnatural act for an incumbent to survive each different tech paradigm shift. We've never really seen it across history. And so I would bet that insurgents are going to displace the incumbents outside of domains that have three things. You know, if you look across history, how have incumbents preserved their advantage? There's three time-pested strategies. One is regulatory enclosure. Two is narrative warfare. And three is bringing up the drawbridge on interoperability. And, and you've seen this across history. I mean, you know, when railroads threatened to supersede canals, they did the same playbook. When, when internal combustion threatened to supersede stagecoach, they did the same thing. When the PC threatened to supersede the mini computer, it did the same thing. You're seeing the same playbook from monopolists today. And, you know, health care has got a little bit of time because of those three properties, regulatory enclosure, narrative warfare. It's unsafe. It's too fast. We're moving imprudently and then drawing up the bring up the drawbridge on interoperability. So I'm a little agnostic. I think the future is not written for Epic. I think it, I think you're going to see creative destruction in every sector of the economy, I personally agree with Venosa. So I think half the fortune 500 is going to turn over in the next few years. Right now, you've got this just ascendant group of 10 companies that have over a $30 trillion market capitalization, and then you've got a long tail of innovation. Healthcare is a little bit artificially preserved for the moment, but I think that the regulatory walls are going to be breached. And part of the counterfactual here, Steve, is what's happening in the GCC, in the King of Saudi Arabia and in the Emirates, and what's happening in the CCP and what's happening in China. And just one of the disagreeable facts of history is that autocracies and monarchies are much better at the implementation phase than messy Western pluralistic democracies like ours. I think especially in clinical AI, where you can only go at the speed of liability. You can only go at the speed of blame allocation. You are going to see the diffusion of clinical AI, much faster and broader in China and in the Gulf Coast states than you are in the United States. And we're going to be looking to those regions and having to reverse import some of the longevity, expanding things, some of the deflationary things, and that's something I worry about. Let's stick on that for a second, because I think there's been a lot of money spent on operational, administrative, patient communication layer of healthcare. And I do believe that ultimately that should, some of that should be deflationary. It's not right now as we discussed, but I think we'll get there. But I think that much more interesting layer is the clinical layer, where we're spending a lot of time, I'm sure you are. And I talked a lot of, you know, health system CEOs, you talked a lot more. And when you talked about clinical AI and taken to its fullest extent, yes, we're going to need like my friend who's a world leading neurosurgeon to go in and operate on an AVM. I don't think, well, I'd be curious in your mind when robots can do that, because it's going to happen. Like robotic world models operating on AVMs at some point in history that's going to happen. But at the same time we have, when I talk to CEOs, they kind of look at me like on this weird, futuristic, you know, and I don't even wear hoodie. I look respectful and I talk to them, but they kind of look at me like, you, what's this guy talking about? Some of them roll their eyes. Some of them get it. And I said, hey, guys, if you don't do this and get ahead of this, it's kind of similar to you. If you a CEO don't red pill this, your point, I didn't think about other nations are going to do this, right? And then what's going to happen is consumers are going to demand this. And you already see this. You see consumers using AI for, like I use cloth sometimes for diagnosis, right? And recommendations, people use cloth as a mental health companion, right? So the consumers are going to drive our system there. Like how does this play out? Because I think you'd agree or I'd be curious, like five years from now, maybe even sooner, like we're going to be there where you can use these technologies actually to get the right diagnosis. We're probably already there. And then it's a question of how does our system regulatory reimbursement, the actual systems themselves adopt that? And that's all the friction and some of those gates that you talked about. But like, how does that, how do those gates get taken down? And how do we get to, when do we get to a place where we have a lot more clinical AI driving the care of patients in our country? >> Yeah, I'm a huge evangelist for clinical AI and not in a sequential way, meaning I don't think we need to start with administrative simplification. And then very kind of circumspecly inch toward clinical AI. I actually think that's a moral and ethical failure. Because that reticence, that almost professionalized hesitation, comes from a, I think a pretty big fallacy. And the fallacy is that the current system is great. The current system is not great. The current system in a lot of ways is intolerable. I mean, we have 377,000 deaths attributable to medical error per year. Another 441,000 injuries that happen. And so, if you start from the premise, Steve, that I have reverence for US healthcare and the practitioners. And I would not want to be treated in any other country on Earth. And so I start from the premise of deep gratitude. But I also think that this preservationist instinct that we can't deploy this too fast because it's too risky. I actually think this is a bigger societal question. I think we're really evensing characteristics of a late stage declining empire, where we're just trying to be protective. I mean, less than 30% of Americans are optimistic about AI. 80% of Chinese are optimistic about AI. And I think this is a really interesting juxtaposition. And so I think that there is a broader hyper litigious sense in this country. I mean, China is an engineering state. The United States is a lawyerly society. What do lawyers do? And I don't mean to denigrate lawyers, but we've got 1.3 million of them. And the purpose of their profession is to be proceduralists and to be a source of the eternal know. I think we've sort of over corrected in this regard because the promise of clinical AI is not just deflation. It is going to be, I subscribe to Dario's view that we're going to add decades to human longevity. And just today it was announced, I'm a venture partner at Thrive Capital. We announced that we just led the 2.1 billion dollar round into Isomorphic. And Isomorphic won the Nobel Prize for chemistry for what I would assert is the greatest scientific achievement in the last 50 years, which is alpha-fold. And if you think about diffusing clinical AI and the fact that the God models right now are already superhuman from a differential diagnosis and care treatment and protocol pathway point of view, it's almost unethical. Not to defuse them. How does that play out? And how does that play out practically though? So I totally agree with your, what are you just postulated? But like, how do you, someone who thinks a lot about this and talks, I mean, you talk to so many different CEOs in our industry, you've talked to Dario and your podcasts like literally practically, how do you see this playing out? I think this will go at the speed of blame allocation and whoever underwrites the medical email practice and the product liability will win 100 to zero. And the analogy I use is B.Y.D. in China. B.Y.D. is the largest EVAV company in the world. It's called Bilger Dreams, they are a hundred billion dollar company and I think they've got something on the board. They've got more autonomous vehicles and electric vehicles on the road in the world than Tesla. They have level four autonomous arcing and driving in Shanghai and Beijing and the company assumed the product liability and the medical malpractice for use of their autonomous driving capability. And I want to go back a second, Steve, because this is so important. I think in this country, we hold technology to a hypocritical standard of infallibility and perfection. Yeah. Not human equivalence or superiority. Yeah, I agree with that. And if you think about autonomous driving as an example, we have 45,000 fatalities in this country every year. The top three causes are drug driving, distracted driving and texting, human, human, human. And yet you get a single error from an autonomous vehicle and you can bankrupt the company. Crews a couple of years ago had a really unfortunate incident in San Francisco, bankrupted the company. So my belief is that whoever underwrites their product from a medical malpractice and product liability point of view will win. And so I think about it, there are precedents for this. There's a company called IDX and I think they changed their name to Digital Diagnostics for Diabetic Retinopathy. Yes. And it was the first FDA approved autonomous diagnostician. They underwrote the product liability in the medical malpractice. And I think this is a clarion call to the startups. If you have absolute conviction in your product and you assume the liability, this is a clarion call to the hospitals, to the payers, who is willing to underwrite the liability? By the way, it sounds like a great opportunity for like a med mal AI insurer who, because like the risk to your point is probably better than the way better than humans. And probably given the risk off environment of like this is a new area that generally is happening, there's probably money to be made in that malpractice world too to underwrite these things. And then I think you're exactly right. Now I think this is a re-emergent moment for the hyperscalers in healthcare. You look at Google, which is about to be the most valuable company in the world. It's going to crest in Vidyan just probably the next three weeks. And there, you know, under Demis, who I think his Nobel Prize was in a coronation. I think it's a, I think it's a starting gun. I mean, there are a couple of people across history of one, two Nobel prizes, including Marie Curie. And I think Demis is going to win multiple Nobel prizes. And I think what he's doing to mobilize Google across so many different vectors is the vector that they've not explored yet is assuming the liability. If they're a $5 trillion company with 150 billion in free cash flow, and obviously their catpacks is off the charts, but and they underwrite the liability for clinical AI. This is a challenge to, whomever has the highest conviction. Now, there's a ton of ambiguity in-- It's fascinating, by the way, just to interrupt, because I've always thought Google, like Pre-AI had kind of had the right to win the medical search. this is Gen 1, let's call it of like consumers going to Google the search on health care. But I like they never exploited that. And maybe you know, I've talked about this in the past, but I think it was frankly because they were so worried about the negative Wall Street Journal headline or maybe the liability and their cash cow historically, as you know, has been an amazing business that had business they have. Now with the changing of GEO and some of the ways that agents are interacting with agents like I used that could, you know, I'm sure Google's way ahead on that front. That's not my expertise. But the dynamics are changing. So maybe that cash cow versus like actually being the super intelligent medical source to your point. Maybe that maybe the tables as have turned there where they're willing to risk it, risk the core business to assume the liability on the health care side. I think the live and I don't want to be glib about this. But you know, with reinsurance and other mechanisms, I don't know that this is about the company strategy. I think this is how do you insinuate yourself into the 18.3 percent of US GDP that is health care in a really central way? And you know, I think you're going to see like the God models have decisively entered health care. So chat GBT has 950 million weekly active users. They have 40 million health care users per day. And we talked about this last year, the propensity of humans to lie to other humans to avoid stigmatization or judgment or discrimination. But they'll tell the truth confessionally to chat GBT. They'll tell the truth confessionally to cloud. They'll tell the truth confessionally to Gemini. And I think with GPT Rosalind, with Google's co-clinician, with anthropic buying co-efficient bio, you know, you are going to see the God models, you know, step into this space. I think Dario has been the most astute at recognizing that health care in this country is not a consumer business. I'm sorry to say it. And I'm sorry to puncture the rhetoric of the past decade. But health care is not mediated through the consumer. Now eventually, it may be, but as long as the reimbursement hydraulic reinforces the entrenchment of the 150, you got to go through the 150. I think Dario has been very astute and we talked a little bit about how do you align with the decision makers on the establishment side in health care. But I think the field is wide open and the liability question is going to be determinative of this. Now the other thing here, Steve, is we need to win. And what I mean is that we're in a race between souring public perception, the regulatory and capture of the incumbents and then the exponentials in the tech. You know, I said alpha fold was the greatest scientific achievement in the last 50 years. It's too esoteric to make its way to kitchen table conversations. We need to let the labs rip. And I know when we do our follow-up podcast this, if you'll have me in a few weeks, I guess we're going to talk about my new paper. But there's a whole chapter on the center of gravity is going away from the gerontocracy, the 60 and 70 year olds that lead the peer reviewed journals and the IV colored academics on the east coast, including in your beloved Ferris City of Boston. And you know, from this institutionally mediated, you know, sort of oligopoly. It's done. We're not the creators are not looking over the Atlantic Ocean. They're looking over the Pacific Ocean. It's Silicon Valley 20 something insurrectionists that are tech facile. They have the GPUs, they've got the algorithms. And I think the center of gravity on scientific advancement is going from east to west, from the gerontocracy to the young Turks. And hopefully we can talk about that because the very reasons that we have these institutions that mediate healthcare comes from scarcity, scarcity of cognition, scarcity of credentialing, scarcity of expertise. What do these multimodal LLMs do? They democratize expertise. They're super abundance now of data and intelligence. And so I think the center of gravity on biomedical advancement is totally changing geographic. The nucleus is shifting geographically. And so I expect we'll talk a little bit about that. But Dario is going to get Dario is aggressively getting into this, Sam is aggressively getting into this, Demis is probably the furthest along and you and I love Dario's essay on machines of loving grace. You know who he dedicated that to? Demis? The Demis. He dedicated it to Demis. So I think the 150 bringing us full circle have limited time to get in the game and defuse this and co-create and a good number of them have. And the last year I've probably brought another 75 of the top 150 with me to Silicon Valley to do these immersive days with Dario and with others to get really smart and become facile with the exponentials. And I don't think you've got too many of the 150 that aren't red pill at this point. You've got a small stratified number that have gone from situational awareness to urgent mobilization. A couple last things curious your take on regulatory because our industry is so governed by regulatory. You know on the one hand this administration is pretty hands off and you know maybe that helps to your point of the insurgents all flowers bloom the creators the east to west shift and it sounds like you're you like that. On the other hand you you stated some stats that are pretty scaly hyperscalers 3.3% of the GDP like oh my god I don't know like that's a lot of power rested in very few hands and that can go right or that could go wrong. How do you think about what this administration's view they've had a few within a health care specifically a few regulations they promulgated like where would you put us as a country in being prepared from that perspective for this greatest moment in our history. Yeah. I mean look there's a lot in this administration that I don't love. You know least like the 17th century mercantilist tariff nonsense you know I don't love everything about our foreign policy. I love some of that. I don't there's others. I don't but the two areas that I do really appreciate about this administration are in AI and health care and I think David Sachs entry arm Christian have done a fabulous job although I do want to talk about labor dislocation before we end if we can because I think Sachs in particular is perpetuating a view that this isn't going to have labor dislocation that I think is very short-sighted and I think it's going to come back to bite them. But the other domain that I that I respect is the health care team I think specifically Dr. Oz Chris clump Amy Gleason Abe Sutton Steph Carlton I just think this is a that's an all-star team and I think that they're not technocratic in the way that their predecessor CMS HHS sounds like we may get a new leader at FDA here we'll see but they're not as technocratic and I appreciate that and what I mean by that is they're not they're not weaponizing the instrumentality of government to overengineer I think that they're wisely lazy fair but I think they've got a very enlightened view of how do you stimulate deployment even in clinical AI not just administrative simplification so I think they're they're a little bit hamstrung with the Chevron ruling which limits what they can do unilaterally as departments I also think that they're limited in the crazy quilt of absolute insanity of 50 states and endless municipalities that have their own regulatory patchwork there are 895,000 laws and and regulations and guidances of thou shalt or shalt not across the country at the state local and municipal level 895,000 and so I think this leadership and especially with with Chris clump as just an amazingly articulate and thoughtful leader I think that they are going to create a very conducive environment to stimulating this but the reality Steve is that most of the innovation should come outside of an irrespective of what happens in Washington and that's why I think whoever assumes the liability it's going to be a real tailwind but you know we also have one trillion dollars of administrative spend that is automatable augmentable or eliminable right and I look at the 600 billion that is due to the adversarial payer provider system and I think that there's a lot of de escalation that's going to happen again the first chapter was inflationary because people just weaponized the existing system this next chapter is going to be deflationary and I mentioned I think that there is this sort of Orwellian conspiracy of silence on job dislocation in this country yeah why is that yeah I think it's well I understandable like we don't I mean there's a reason why ice has a higher approval rating than AI in this country that's crazy it's crazy but I'm also sympathetic to that because people are alarmed about job dislocation in an absolutely compassionate way. about that. But I look at health care out of our $5.3 trillion sector, $2.9 trillion is in labor. And, you know, health care is greatness country if you're rich, educated, white, and urban. And so we have 250 billion in medical debt, the leading cost to bankruptcy in this country is medical debt. And so the deflation has to happen in health care. And I predict we're going to see five to 700 basis points taken out of US GDP allocated to health care over the next several years. And I'm not, I don't want to be reductionist in this because right now there are 1.8 million unfilled jobs in health care. The cheaper health care becomes the more people are going to use it. And if we add a couple decades to human longevity, we're going to need more health care. And if we finally begin to synchronize medical management with pharmacologic data and behavioral insights and STOH, we're going to need more, we're going to need more of all of this. But we are going to see job dislocation. We're already seeing it in areas of automation, starting with areas of functional verifiability. And then from functional verifiability, you're going to go to verticals that have codifiability or are rules based. Anything you can verify, you can automate, anything you can decontextualize or take out of the workflow you can automate. I think this goes to BPO, all of the offshoring and business process outsourcing that is definitionally decontextualized, right? You can take the process out and put it in India or the Philippines. I think we're going to repatriate all those jobs, but not give them to humans. We're going to give them to agents. We're going to give them to agents, right? And so I think it's really short-sighted not to talk about potential job disintermediation, because if you can't acknowledge it, you can't begin to come up with a solution for it. And I think the solution is going to be some combination of Elon talks about universal basic income, but universal high income, right? I mean, Elon's projecting we're going to 10x global GDP in a decade, which is kind of remarkable. If our current global GDP is $127 trillion now, he's talking about 1.2 quadrillion dollars in a decade. To the point where GDP and the economy definition sort of become obsolete, right? And so then it becomes not universal high income, but universal provision. We're just going to provide goods and services, because AI is going to be massively deflationary. Why? Well, if you reduce the labor dependency, 60 trillion out of the global 127 trillion GDP is labor. So we've got to find you can only go at the speed of your bottlenecks, right? So if you can partially automate a process, the bottlenecks are where you're going to need to double down on human labor. And while you're overall SGNA, while you're overall salary wages and benefits, SWB as well are going to go down, you're going to have superstar employees in those bottlenecks that you're going to have to pay more to. Automation is not some monolithic thing. Like if you automate the low skilled tasks, then compensation goes up and employment goes down because you need higher qualifications to do the job. If you automate the high skilled tasks, employment goes up, but compensation goes down because lesser skilled, lesser trained employees can do the job. If you automate both of them, you enter the period we're talking about now. And so how do we think about the things that are automatable, augmentable, or eliminable, and then have an enlightened conversation about the bottlenecks? How do we retrain humans to do those irreducibly human things? And the fault that I find with Jensen and David Sachs and others that refuse to acknowledge the possibility of job dislocation is that it's preventing us from having an open conversation about reskilling and upskilling into those bottlenecks. I agree. I agree with that. This is going to be a big task. It's doing the disservice to every technology has goods and beds. It's doing a disservice technology where the beds are and then dealing with it. Frankly, we as a country have not been great on reskilling our workforce. So that's a conversation we'll have for another time and how we do it in the AI era. Let me close with one last question on this year chapter of looking back. We're here a year from now, which I hope we will be. I hope we're here for another hundred years with more longevity. But what's your one prediction that you think will be different a year from now than today as it relates to healthcare? And in the field of AI. I think we're going to enter the first deflationary period in healthcare. But it's not it's going to be a jagged frontier. Because right now the intelligence is a jagged frontier. You spike in certain areas of objective function as we talked about. But then where there's squishier subjectivity or irreducibly human stuff or data scarcity, it's not as good. But I think a year from now we are going to and we can subdivide it. You know, if you'll give me 60 seconds here, I'll kind of subdivide it. I think for health systems, you are going to see the biggest market share shift in a generation. And you are going to see a resurgence of non-contiguous M&A that goes across geographies. And the biggest determinant of who's the aggregator versus the aggregated is AI diffusion. With a liberalizing FTC and DOJ, you are going to see a lot of non-contiguous 50 billion dollar health system mergers. And the top 10 that control 455 billion, it's going to go up. I think it's going to be deflationary because just for hospitals, we spend a trillion dollars on labor. 270 billion is on the administrative side. The 730 billion is on the clinical side. So I think you're going to see a real stratification in performance and not anemic margins. The 1% operating margin for the AI adept is going to go out. For payers, I think you're going to see the realization that we are entering into a period of structural, secular decline for the payers. That is not about V28 or anemic rate notices or star ratings or Wall Street Journal investigative reporting on insured driven coding. I think you are going to see the creation of an AI-enabled sector with the margins of a utility. And a real stratification again here for which payers move most swiftly to automate SGNA and use this emergent medical superintelligence to predict when patients are going to decompensate, focus on longitudinal pair, close gaps in care. Hopefully when we get back together for our next conversation, we could talk about this sort of bigger abstracting out battle between payers and providers for who owns the patient. I think for life sciences companies, you are now going to see AI-engineered biologics and biomolecules enter stage two and stage three of clinical trials. And we might actually see our first FDA approval and commercialization. I think instead of the 2.6 billion that we spend on a molecule in 10-year Odyssey with a 90% attrition rate, I think we may see it in 120th of the time at 120th of the cost. I can't wait to talk about that. Yeah, for sure. All right, well, those are the three big sub-sectors. You heard to hear first. A year from now, we're going to come back and see if Eric was right, whether there has been deflation. It sounds like on those who have been red-pilled in particular, payers, providers, and life science companies, and whether we see it there, those will be the early green shoots of actually the hope of this being deflationary for the most expensive industry in the world that is healthcare. Well, Eric, as always, a masterclass. I love our conversations. It was great to look back. Thank you for joining us. And for the listeners, please again, drop any questions you have in the show notes for Eric. He'll be back on in a couple episodes and we'll talk about his new piece. And also, we answered some listener Q&A, which will be a lot of fun to Eric. Thanks again for joining us. All right, my friend. Thanks, Steve. Thanks for listening to The Heart of Healthcare. If you enjoyed this episode and you'd like to support the podcast, please leave a rating and review. And don't forget to subscribe. The Heart of Healthcare is produced by Halle Teco and hosted by Michael Eskabel, Steve Krauss, and Halle Teco. The show is engineered, edited, and mixed by Kyle Moore. Visit our website, heartofhealthcarepodcast.com for show notes and details.