马克·库班谈AI为何可能让独立创始人成为万亿富豪

We are not in an efficient healthcare market. The economics are broken.

One of the many ills of healthcare is that it completely lacks competitive dynamics. Just by the presence of Cost Plus Drugs, you’ve actually forced change in the incumbents.

UnitedHealth has 3,000-plus subsidiaries and $150 billion in intercompany transfers annually. There are often times when they’ll increase premiums, increase deductibles, and at the same time force you into a relationship with a company they already own—and you don’t know they own them. The ultimate mission is to disrupt healthcare so that it’s affordable again.

Is there some area where you’re like, “Why is no one building here?”

There’s no better time to be an entrepreneur than right now. The biggest challenge, and where AI, I think, is going to have the biggest impact, is…

Well, Mark, welcome to the podcast. Thanks so much for joining.

Thanks for having me on, guys.

Julie, what has struck you about Mark’s contributions in the space? What have you always found fascinating about what Mark’s been up to with Cost Plus Drugs, or just in general, and why you invited him onto the show?

Mark, first of all, just had the courage to enter into the morass that is healthcare in a way that made such a splash. Many, many people who are also in the privileged position Mark is in, with such a huge audience, have been intimidated from doing this for many decades.

I’ve loved watching the market react to what Mark has done in healthcare. Mark, you and I have talked about this in the past, but one of the many ills of healthcare is that it completely lacks competitive dynamics and competitive forces, free-market dynamics, all that kind of stuff. What’s been fascinating to watch is that just by the presence of Cost Plus Drugs, as one example—and we’ll talk about other things you’ve done, too—you’ve actually forced change in the incumbents. Even with policy and all sorts of other forces, they never changed over the course of many decades. That’s one aspect that I think has been interesting to watch.

I appreciate that. Thank you.

Yeah, it’s been interesting to watch.

I don’t get to see that stuff because I don’t talk to those guys.

Even absent a contract, I think people are recognizing from the demand side of the market that they need to show there’s some pathway for people to get your product. Perhaps even in the absence of a direct contract, I do hear people talking about—

That’s good. Yeah.

—your solution as part of their network—

But it truly does not surprise me when the big three or even the big six use our name as a way to counter the request in the discussion of pricing.

Yeah.

They know that our prices are public, that anybody can take our API, and big companies that have half a brain are pulling that data daily and saying, “You have to beat this price.” I can see that, but I’ve had at least one company, and probably more, say, “Oh yeah, one of the big three has told us that you’re part of their network, and we get your prices where you’re cheaper.” We don’t have an agreement at all with any of the big three.

Mm-hmm.

Mark, let’s zoom out and take stock of where you’re at with the business. You launched in 2022 with a few dozen drugs. In April this year, you said you were up to roughly 4,000. You’re getting into manufacturing, hospital contracts, potentially AI-based care. Why don’t you take stock a little bit of where the business is a few years in?

Sure. I’ll give you the rundown. We launched costplusdrugs.com on January 19, 2022, with 111 medications, purely generic, and we didn’t know what to expect at all. I went out and started talking about it, and it immediately blew up. Our volumes led the manufacturers to expand their portfolio with us, and we were able to add additional manufacturers. Now, as you said, we have 4,000 SKUs, give or take, and our business every week is up and to the right. Our biggest challenge is just keeping up with it. So that’s segment number one.

The obvious response to mail order is, “Well, I would need to pick things up today.” So we created what’s called teamcubancard.com. I did not name it. It’s basically a discount card that allows you to go to any of the 17,000 pharmacies that have affiliated with us and pick up most of our drugs locally. I’m a big believer in supporting independent pharmacies, so we pay our affiliate pharmacies a $12 fill fee. That means we’re going to be more expensive for the lowest-cost drugs. If you have a $5 statin plus $12, you’re going to get it cheaper somewhere else. But at the same time, if you really want to support your independent pharmacies, you can pick up the cheap drugs. On the more expensive drugs—the specialty generics and biosimilars—we’re actually much cheaper. For things like biosimilars and insulin, which require cold shipping, it’s actually cheaper to buy them and pick them up locally through our Team Cuban Card. So that’s the second stool in what we offer.

Third is Cost Plus Marketplace. Before I get to Cost Plus Marketplace, let me go back to our Dallas manufacturing. A big part of Dr. Alex Oshmyansky, my co-founder’s, original cold email to me was that he wanted to manufacture generic drugs that are on the FDA short supply list. There are pediatric cancer drugs and things widely used in hospitals, like Pitocin and epinephrine, and a long list of other drugs. We were able to use AI and robotics to shrink what would’ve been a football field or more into 5,000 square feet. Not only that, but we were able to turn drugs every four hours. We can make a six-month supply, turn the line, and then go to the next drug and continue doing it. We sell those products to hospitals and clinics.

An extension of that, because we started working on it five years ago, is that we’ve been able to take those same robotics and AI and shrink them into a manufacturing plant the size of a tractor trailer. We call those pods. Those pods can be used for anything and shipped anywhere. The government was our first customer. We shipped the first one recently, just got an order for a second, and they told us there’s more behind that. If there’s an ecological disaster, we can send it to FEMA, if FEMA’s around. If there’s a weather disaster or a military disaster, God forbid, we can ship them there and configure them to make any specific type of drug. We can configure a pod for sterile injectables. Multiple pods can be daisy-chained together so we can scale, or you can have one as a sterile room and another doing the tablet work for generics. I’m not the expert on all this; Alex can go into it in precise detail.

The thing I’m most proud of—and Stat News just wrote a story on this yesterday or the day before—is that we can configure these pods to do N-of-one therapies. We can ship a pod to a research hospital. These pods cost about $20 million. If the hospital specializes in therapies for people in the N-of-one category, they can make it and iterate very quickly there. Now we’re doing these ASO drugs. Don’t ask me what ASO stands for, because I can’t pronounce it. We’re making them at our first manufacturing facility in Dallas. So that’s part two.

From that manufacturing, we created costplusmarketplace.com—sorry, Cost Plus Marketplace. Cost Plus Marketplace takes the sterile injectables we make, along with others that we source from other providers or manufacturers, and sells them at cost plus 10 or cost plus 12 to clinics, hospitals, providers—whoever wants to buy them at the cheapest price. We can literally save hospitals millions of dollars a year. Our biggest challenge there is that we convince the CEOs, but we can’t always convince the purchasing people to do it.

Finally, because of my own experiences as a self-insured employer, and from going around and talking to hospital CEOs and CFOs and learning about their relationship with the big carriers—the big vertically integrated carriers—and all the things they like and dislike, mostly dislike, I went to them and said, “Look, I know you hate the delays, the denies, the pre-ops, the late pays, the clawbacks, and the fact that your plans—or the carrier plans—keep increasing deductibles, turning providers into subprime lenders because they have to collect the deductibles before they can offer care or before they can get access to the insurance money.”

Knowing all these things, I’d like to work out a direct contract, but I want 2 things. I’m going to tell you what I’m going to give you in exchange for those 2 things. First, I’m going to trust the doctor. Unless it’s something crazy, I’ll trust the doctor, so you’re not going to need a pre-op. There aren’t going to be denials.

We had somebody who had Munchausen syndrome that required a denial. But other than that, if it’s something crazy, I’m going to pay the price that we contract for, on time. I’m going to take responsibility for the patient’s out-of-pocket costs. Either I’ll pay them, or I’ll take the credit risk for the patient, so you’re not a subprime lender.

In exchange for all that, I want 2 things. I want a better price, because your associated administration and overhead and doctor time are much less. Your yield is going to be higher. And I want the ability to take the contract we signed—this direct contract between us—and publish it on costpluswellness.com so it can act as an open source for all of these contracts, so any employer can use them and any provider can post their contract. Even if you don’t want to post or use them, you can use them as templates in your own negotiations.

1. The Health Plan Power Shift

Well, I love the last piece, Mark, because I think that was the genesis of our back-and-forth when I published a piece a few weeks ago on what I’m calling the great replacement cycle of health plans. For so long, employer-sponsored health benefits have been the largest class of insurance in our country, if you think about it that way. Roughly 155 million Americans are covered by an employer-sponsored benefit, or—

More than that.

More?

Yeah.

Yeah, more. But in any case, it’s more than half the country—60% of the country.

Yeah.

Historically, it’s been completely monopolized by the BUCAs, these national carriers. It’s been an unaddressable market from a startup perspective. The relationships were super entrenched; you needed these national networks to be able to play in the game. The starting cost to enter the arena was too high for startups, and the relationships were just too entrenched. Now, the premium increases are just freaking ridiculous. This year, we’re seeing some BUCAs force 60% year-over-year increases in premiums, even among our startups. And it’s just—

It’s even—

Or the economics are—

It’s even worse than you think, Julie. Not to step in, but—

Yeah.

Because they’re vertically integrated. UnitedHealth has 3,000-plus subsidiaries and $150 billion in intercompany transfers annually. That would make them number 14 in GDP in the United States among states.

Mm-hmm.

That tells you how big they are, and CVS and Cigna are not far behind. Oftentimes, they’ll increase premiums, deductibles, and out-of-pocket costs, while at the same time redirecting you—or actually forcing you—into a relationship with a company they already own, and you don’t know they own them.

Right.

You talk about employer-based insurance, particularly those who self-insure. Sixty percent of people who get their healthcare from companies get it from self-insured employers.

Mm.

Those employers need someone to process all the medical claims. Put pharmacy aside; I’m talking about medical claims. Those are called third-party administrators, and if they’re owned by the big BUCAs, they’re called ASOs, or administrative services only. We hear all the time about the problems with PBMs and how they spread pricing. On the medical side, it’s a million times worse because—

It’s also a bigger chunk of the market relative to pharma.

What’s that?

It’s also the bigger chunk of the market relative to pharma.

Yeah, of course. The medical side is much bigger.

Yeah.

You know, 10% and growing versus the other 90%. The plans and the hospitals have their own contracts, and hospitals, unless they’re market dominant, hate the BUCAs in those contracts. More importantly, whatever that relationship is, the self-insured employer has no idea what its terms are, right? It’s possible to go to the hospital’s website and look up the TiC file—the Transparency in Coverage files for that relationship and that plan—but they’re not always correct.

What ends up happening is that the price the self-insured employer pays rarely matches the amount that was charged or the amount that was paid to the hospital. There’s spread pricing there. To make matters worse, those ASOs owned by the BUCAs make it impossible to reconcile the numbers, so you can see whether you’re paying more or less.

Right.

There are these things called 835 and 837 files, which contain the data sent to the hospital and the amount of money paid in response to those claims, and employers have no access to them. In many cases, they don’t even have access to their own claims. It is just so rigged. To your point, Julie, trying to remove them—to disintermediate them—is the key, the holy grail, to what’s happening in healthcare right now.

Yeah. We’re super excited to see an influx of startups that are now able to go after this market. A big reason, in addition to all these market forces—and again, the economics are getting to a breaking point—is that AI can play a significant role in enabling a full-stack health plan offering to be built in a way that reduces all the corruption you’re describing, in a very short period of time.

2. AI Meets Broken Contracts

But this all comes down to who takes the risk.

Mm-hmm.

I’m a huge AI fan. I live it, I love it. I think it’s important on the clinical side and on the operational side. I think it’s most important short term on the contractual side. But the reality is, in a broken financial system, even when you introduce AI into the economics, operations, and processes, you can’t fix it when it’s already broken.

You brought up the point right at the beginning, Julie, and you’re 100% right: We are not in an efficient healthcare market. The economics are broken. You see that because if you don’t have price discovery, that protects you from having an efficient market. But what’s even worse—and here’s where the market, the economics, really start to fall apart, and what makes it tough for AI in the short term operationally, and less so clinically—is that the contracts are so big and so long. The BUCAs, through their ASOs, get to be contractual terrorists.

By contractual terrorism, I mean the City of Dallas’s PBM contract—I haven’t been able to see their ASO contract yet—is 1,335 pages.

Mm-hmm.

There’s no group of humans with enough coffee on the planet to catch everything they try to squeeze by. As a result, no matter what AI you use, once you sign a contract—hopefully you use AI to prevent yourself from signing a bad contract, and I think that’s the low-hanging fruit for everybody—but once you sign that contract with terms you don’t fully understand and conditions you can’t even protect yourself against, they win.

Yeah.

There’s no AI on the planet that can protect you. When you try to compete with them on plans, that sounds good, and I’m all for it. What Sidecar Health and so many others are doing is really smart. But they’re having problems because, yes, the wraparound networks are needed, and yes, putting together a network is hard, but those BUCAs have so many subsidiaries, they mess with you somewhere and you don’t even realize it.

Yeah.

That’s the ultimate challenge, and that’s why I’m so big on the Break Up Big Medicine bill and going after these companies. I want this to be an open market for every entrepreneur—

Yeah.

—to use AI in every way possible.

You went exactly where I was going to go, Mark. One of the areas where AI is playing a role, for better or for worse, is that it’s shining a light on how broken the underlying system humans designed is. The whole prior-auth bot war situation happening right now between—

Yeah.

—payers and providers is one of those things people blame on AI. But actually, AI is just implementing the existing fee-for-service billing system to the full extent and showing us how broken that system is. I’m curious what your take is on the end state of that. Clearly, your direct-contracting methodology sounds like it can reduce a lot of the information asymmetry, improve contract clarity, and address things of that sort. Is that what it’s going to take?

The bot wars in terms of claims and upcoding and downcoding are short-term stuff. That’s just like, “Okay, I need to lose some weight. I’m not going to eat my Rice Krispies or my Frosted Flakes every day.” It’s nice on the periphery and it’s a good short-term play, but it doesn’t really change everything, because AI is a great equalizer.

Your AI, my AI, your OpenAI, my Anthropic—whatever it is—we’re going to battle each other out. You might win a couple, I’ll win a couple, but they’re going to net out at some level. So I don’t see dramatic change in the long term.

Maybe if you’re first in, you can pick up some things. You’ll see some lawsuits and challenges when one side has it and the other side doesn’t. But the revenue cycle management on the hospital side is pretty darn good.

The BUCAs have a big problem, in my opinion, because they grew through acquisitions. We mentioned these 3,000 subsidiaries. That’s a lot of siloed data. Even though you can try to find ways to get access to claims and specific things that allow you to upcode—and they’re doing that—it’s going to be hard for them to become AI-native companies on the BUCA side. I think over the long term, they’re going to be challenged.

But all of this starts—and I can’t say this enough—before those employers sign their contracts. If every one of your portfolio companies said, “You know what? I’m going to run my existing benefits contract—if I have a broker, a consultant, a TPA, or a PBM contract—through AI, and I’m going to ask a single question: ‘Where are they fucking me over, and what should I do about it?’” Within 30 seconds, you’re going to see: bam, bam, bam, bam, bam, bam, bam, bam, bam.

The things I’ve seen running through contracts—let me just tell you how much I love AI agents. I use them all. I use GrokBot. Erik and I are NBA fans, so I got the Luka Dončić plan because it’s $77.69 or something. I’m like, “This is brilliant. GrokBot, $77, the Luka plan. I got you.” Then I have the Gemini plan, OpenAI’s plan, et cetera, et cetera. I use the agents in Instinct and Dot, too.

I tell them, “Go out and find me every state, entity, organization, and enterprise that has an open-records opportunity. Find me every single contract, RFP, or whatever you can find from those entities. Write the letter to request it, put it in an email, put it in my draft, let me personalize it a little, tell me when it’s there, and I’ll hit send.” It’s amazing. Stuff that’s always been at the bottom of the list, that I always wanted to get to but never could, is now just coming in.

I put it up on Google Drive, and each of the different LLMs has access to it. Every time something new loads on the drive, I tell them, “Go through it, tell me the good news, the bad news, what should be done, and then give me the simple version so I can send it to the company, school, or whatever it is.” And bam, bam, bam, bam, bam. That’s where I get to see all these contracts that the big 3 on the PBM side and the BUCAs on the ASO side are signing. I get to see specifically how all these enterprises are getting ripped off.

Mm-hmm.

These school districts—oh my God. Cities from Denver to Dallas to Fort Worth. Unions. I mean, going back to your perfect point, Julie, if that 67% of employers representing 67% of the 160 million-plus Americans actually did one thing—ran their contracts through AI before they signed them and negotiated from that—the entire system for the big BUCAs, big PBMs, and all their subsidiaries would collapse. The cost of healthcare in this country would drop like a rock.

When I go talk to employers, investors, and their portfolio companies, they ask, “Do we need legislation? Do we need this, that?” I’m like, “No, you need to be smart and do your job. Bring in somebody I call a healthcare CEO or CFO whose only job is to manage the contracts and everything associated with them, and you’ll find that your healthcare costs will go down. Then, once we get you there, we can talk about carving out for direct contracts and doing all these other things.”

There are a couple of public companies I told to do this, and they came back to me and were like, “Oh my God, we had no idea how badly we were ripped off.” I’m begging them to mention it on their quarterly earnings conference calls.

Yeah.

They won’t do it because they still have to do business with those BUCAs, even if they left them for whatever reason.

Yeah.

If we just do that one thing—run the contracts through—literally, if you have more than 1,500 employees or members, the quickest way for that startup, that tech company, to save maybe millions or more, depending on how badly you’re ripped off, is to just do what I said. Run your contracts through AI.

Yeah.

Mark, maybe zooming out, can you share more about the grand vision for the business a few years out and how you think about sequencing to get there?

Sure. A couple things. The ultimate mission is to disrupt healthcare so that it’s affordable again. I’m a fan of universal healthcare, but you can’t get to universal healthcare until you have correct data and people can make decisions based on real data. That’s the bigger system. Whatever direction it ends up going, I just want to disrupt healthcare so that when people think about the economics, it becomes, “Okay, I know exactly what’s going on.”

How we get there is more about education than anything else, like I mentioned before. In terms of the business of Cost Plus Drugs, look, somebody could come along on the retail side and say, “We’re going to do cost plus 1%.” Okay. Trust is still our primary product. We don’t spend any money on marketing or advertising. When we first got started, all the people who came from pharmacy said, “Oh, you’re going to get killed.” I’m like, “No. People are going to tell others how much money they saved, and then they’re going to trust us.” That’s the one thing missing in all of healthcare economics: trust. Once we gain their trust by doing exactly what we said, being fully transparent, and having a set markup of 15%, we’ll garner people’s trust. That’ll protect us to some level.

The grandiose vision is those pods I told you about that sell for $20 million. There’s good margin in those, and we think we can sell a couple hundred a year within 2 years, and a couple thousand a year within 5 years. So, give me a thousand pods to sell at $20 million each, and we’re going to be doing really, really, really well. It’ll completely change healthcare because we can daisy-chain those things. We could literally manufacture tableted generic drugs in Dallas, Texas, using these, and make them for less than what we pay the manufacturers we partner with in India. We don’t go for the cheapest price; we go for the best manufacturer, so we pay a little bit more. But still, we can make them for less.

The only thing holding us back is that for each generic, we have to apply for something called an ANDA—Another New Drug Application—and each one costs $365,000, plus other fees. If you want to do 1,000 of them, that’s $365 million, plus all the cost of the pods. We’re trying to say, “Look, if we truly want to bring manufacturing onshore for medications, we’re here for it, and we can do it. You’re already working with us in other areas.”

We think we’ll get there as a big part of what we do. We’ve already planned so that on CostPlusDrugs.com, we’ll put a little American flag when something is manufactured by us here in the U.S. But that’s probably 2 years away, just because of all the work we have to do with the FDA. That’s the big picture. If we’re selling $20 billion worth of pods a year, watch out, Elon.

Is your view that, as we look out over the next few years, no matter how good the technology gets, it’s not going to disrupt healthcare on its own, and that other, more structural changes need to happen?

Yeah. There’s regulatory capture, and there’s scale capture. Right now, on the economic side, it’s all scale capture. When you listen to the quarterly earnings conference calls or read the transcripts for all those big vertically integrated companies, they talk about technology, but the one thing they never say is, “This is going to translate into lower costs for employers and patients.” They always say, “This is going to improve our earnings. We’re going to return more to shareholders. We’re going to buy back stock.”

Until I’m able to educate all these employers, they’re going to keep making it nearly impossible. That’s why, when Julie talked earlier about CVS, I mentioned we don’t deal with any of them directly. We have to work outside of them, because otherwise you fall into their trap, and they game you with their contracts.

Yeah.

Mark, you’re talking a lot about structural things, but there’s also the reality of what you described earlier: AI will change the nature of people’s jobs in this industry quite a bit. I’ve heard you talk a lot about how we future-proof our young pools of talent and what being career-ready looks like in this day and age. What do you think that looks like for healthcare? What does the healthcare workforce look like in the next 5 years?

I mean, it looks the same as any other industry.

You have to still hire entry-level people.

Mm-hmm.

Because no matter what the model is, the minute it’s released, it’s behind. There’s latency associated with every model, so you need people who are aware of everything that’s going on, create context, deal with the models, be the human in the loop, and understand the consequences. You need those people to grow with the organization over time so that, as all these things change, they’ve seen the history of the implementation and can bring value to the organization. That’s part one: You can’t just get rid of all entry-level positions.

Part two, you have to enable those people—and everybody else in white-collar roles—to use AI, because AI doesn’t replace jobs. People who know how to use AI replace the jobs of people who don’t know how to use AI. And then part three is the bigger-picture stuff. The first step is trying to educate the C-suite on how AI even works: teaching them how agents work, how to use a model, and the difference between open-source and closed models, along with the different things you have to do.

You have to help them understand the latency of a model. This brilliant conversation the three of us are having isn’t in any model anywhere, and won’t be until it’s published. If we say something that we can all go out and use, the models are all behind, and AI is all behind. That’s a given. And, of course, LLMs are text and some pictures; they don’t deal with all the real-world things that are going on in video. So, as a CEO and in the C-suite, you have to understand how all those things impact your company.

Once you get a good grasp on that, then you just start fooling around, because you can find new ways for things to happen that benefit your customers or create new opportunities. That’s the brilliance of AI. AI is not smart. It’s book smart, maybe, but it’s not street smart or entrepreneur smart. You’ve got to have somebody there who guides it and helps it, but until you have a C-suite that understands how it works, it’s going to be very difficult for that organization.

I literally sat with the head of a consumer food company, and he was talking to me about acquisitions months ago. He was like, “Yeah, because of GLP-1s,” yada yada yada. And I said, “Okay, well, let’s do this. Let’s take your number-one product, your cookie, and I’m going to ask ChatGPT to create a competitor that’s more GLP-1-friendly and priced lower. I want a full bill of materials, sources, everything, and what I have to do for manufacturing.” In about 3 minutes, we got it back, and he was like, “Oh, shit.” He had no idea.

Mm-hmm.

If you have no idea that’s a question you can ask an LLM, you have no idea how to enable your company to use AI. And that, to me, is the greater struggle for most organizations.

For entrepreneurs asking that question and trying to implement, no matter what it is, there’s no better time to be an entrepreneur than right now. There will be the trillionaire in the basement. There will be somebody who puts out their own agent swarm on an entrepreneurial basis.

There are still challenges, particularly for agents. Agents don’t see social media posts. They can’t review them, and it’s hard to get transcripts for any material amount of them. They can’t watch ads or see ads. So, if we’re moving to an agent-driven economy for individuals, there are going to have to be a lot of changes in how we communicate commerce to those agents.

I was fooling around with one of the guys in my company, and we created this thing called agentfeeds.dev. It’s not truly active. The whole idea is that you take the transcript of your post and put it into agentfeeds.dev. Then you copy and paste it for anybody you follow into their agents: “Hey, follow all the feeds for this.” It takes all the ads and gives you all the information you need, so your agent is more up to date on those categories or whatever. You’re going to see an explosion of those types of things.

But we still have to get through trust and all the things we know about AI and the issues we know about AI. In terms of opportunities, going back to healthcare, the biggest challenge—and where I think AI is going to have the biggest impact, other than clinical—is with patients, with consumers. That means being able to look at the things you’re given on the economic side, like the benefits bill, the EOB, the explanation of benefits.

Here’s where I think consumers will get the greatest benefit. Right now there’s terror in, I don’t know, half the country. Forty percent of people in the country can’t afford a $400 emergency bill. So what I always tell people is, before anything happens, go into your favorite AI model and say, “Get me a list of all the nonprofit hospitals.” Go to them and ask what their financial assistance services are, because all nonprofit hospitals have to have financial assistance programs, and some of them are really good.

For some people, you’re better off—and I don’t want to make the decisions for people—but you can ask the AI questions like, “Whatever my insurance plan premium was going to be, I’m going to put it in a money market fund and let it collect interest, because I’ve already discussed my situation with the local nonprofit hospital, and I know everything there is to know about their financial assistance plan.”

Those financial assistance plans are created, and the pricing is referenced to the federal poverty levels in most cases, but it can go up to 600% of the federal poverty level, which can be over $100,000 for some families in some states. Patients, consumers, and families understanding what their options are, and how to procure those options from nonprofit hospitals in particular, is going to be enormous.

If I were going to start a company in this space, I would go to every nonprofit hospital and set up a plan that hosts agents that I enable people to use. “Hey, if you’re concerned about how you’re going to be able to pay for your bills or your family’s bills, and you don’t have much money in the bank, or you’re making under X number of dollars, copy and paste this into your Muse or your Dot or your Claude Cowork, whatever it is, and it’ll know to come here to us and ask questions.”

When I talk to heads of hospitals, I’m like, “You don’t understand. You guys suck at marketing. You are dependent on the BUCAs and their plans and being part of their network to feed you patients. That’s ridiculous.”

Yeah.

If you make it easy for people to use your financial assistance programs, maybe go to a local bank, maybe show people how to use stop-loss for small or medium-sized companies—there are so many options for them to use it to make things better for patients, because that’s really where the big disconnect is.

You can offer better plans. You can use technology to reduce the price and become more efficient in how you create a network. We’re negotiating direct contracts. But until the end user truly embraces it and knows how to use it, there are going to be challenges.

All of us, at some level, learned to use spreadsheets back in the day when we thought they were too complicated. I’m old enough to remember when people thought spreadsheets were hard. You had to explain them to people, and they had to pay $500 for Lotus 1-2-3. The same thing is happening now as we go through this transition. There are a lot of people who are anti-AI, but there were a lot of people who were anti-calculators. People are anti-everything. That’s just a matter of time.

The way you get people to overcome AI, or their fear of AI, is to show them how it can really benefit and protect them, and reduce the inherent fear they have in the event that something goes really, really wrong.

There’s this famous Aaron Swartz quote from many years ago: “Information wants to be free.”

Yeah.

I know in healthcare, in a previous conversation, you’ve been a bit skeptical that that will happen because, as you mentioned, there’s a lot of siloed data, and there are strategic reasons why that data is siloed. That’s their advantage to some degree. So I ask you: What do you think the ultimate impact of the frontier labs—OpenAI and Anthropic have very big ambitions in this space, either them or others—will be in reshaping healthcare?

I’ll tell you exactly. I’m selling my own book here a little bit. The cost of designing a drug personalized to you, after reviewing your DNA and knowing everything that it knows, will start approaching zero as the capabilities of models go up.

We talk about GPUs and how they’ll get faster, the networks behind them, and the power required, but it’s the algorithms that are going to change everything and make them more efficient. We don’t even know what the speed of improvement is going to be for algorithms. At some point, whether it’s 5 years, 10 years, or 20 years, they’re going to take something from Erik, Julie, and Mark, and they’re going to say, “Starting with over-the-counter…”

Because if you think about it, you walk into a drugstore and get, excuse me, Advil, Tylenol, ibuprofen, and we all think it’s going to impact us the exact same way. We have to experiment to see how it impacts our bodies. It makes no sense whatsoever.

As the processing and analytics power of frontier models gets better, they’ll be able to truly personalize over-the-counter drugs without a lot of the mishegoss you have to go through to figure these things out. Now, all of a sudden, here’s my personal ibuprofen, my personal dose or mixture of these over-the-counter drugs. Bam, you start to change it. Then you get into making other drugs that are personalized for you.

Because, again, the fact that we all take Eliquis and decide between Eliquis, Xarelto, or warfarin makes no sense whatsoever in the big scheme of things. Being able to take those things and personalize that medicine—what I’ve told 2 of the frontier models is, “If you really want to be smart, take your $20 or your $200 subscription and include that service once it’s available.”

Your marginal cost for doing this at that point in time—look, the inference and everything that has to happen just to respond to a request today is expensive, right? It’s coming down very quickly, but the same thing is going to happen for this type of personalized medicine. Initially, it’s going to be enormous, but as we understand the human body, all the variables in our bodies and our brains, and the medications, and figure out how to do the N-of-1 for everybody, it will be part of a normal subscription, because there’s no reason it shouldn’t be.

How expensive is 23andMe? You pay $99 or less now to get your DNA looked at. Maybe I pay $99 up front and give them my DNA, if I’m willing to do that, obviously, for privacy concerns. They store it and continue to process it. Maybe you want to know what could be wrong based on your DNA, maybe you don’t. But if something does go wrong—

Yeah.

—they have access to all these drugs, all these opportunities, and the ability to create them in real time. In selling my book, we want our pods to be right there to be able to manufacture them, because we’re the only ones capable of doing it, and I don’t see that changing for a long time.

I think one of the things people probably don’t appreciate as much as they should in that context is that, as what you just described happens and drugs become better products for every individual in the country or in the world, that is actually one of the most disruptive forces for care delivery. How much—

Oh, for sure, right?

How much of healthcare is actually the doctor’s job of helping patients figure out why a drug doesn’t work for them, calibrating and doing med titration, and changing behavior to make your life work for the drug, as opposed to the other way around?

Absolutely.

And so that’s something that I think—

But—

—you know, people don’t—

But remember, because of that latency I mentioned earlier, you’re still going to need doctors. If Mark walks into the hospital and I just tried this new personalized medicine and I start puking my guts out—

Yeah. You still need calibration.

—you can have it on video, right? And you can have a robot there say, “How do you feel, Mark?” The Terminator might walk in and say, “Your time is up.” But I’d rather have a human there with all 5 senses, able to take a look, ask questions, see the crying mom or the crying son or child or whatever, and include that in the context. I don’t think we’ll ever replace doctors, and we shouldn’t even try. You guys know how adoption of technology is, particularly for this. You still have to know the questions to ask.

Mm-hmm.

You still have to know what the variables are that matter versus the ones that don’t. You still have to realize that even as processing costs drop toward zero, hopefully at some point in time, there’s still a cost associated with it, even if that cost is only time. You need somebody there who’s been educated in all those pieces.

I think, to your point, Julie, where things might change dramatically when it comes to the role of the doctor is how they’ll be trained in medical school. They won’t be taught the exact same things. They’ll have to learn how to use OpenEvans or whatever else is available out there, and how to extract information, even if it’s just that DNA information, understand what’s in there and what the model thinks about it, and be able to translate that to the person.

Effectively, in order for this to work perfectly, everybody has to have their own model, and that’s going to be difficult for everybody, even though there are going to be millions, if not hundreds of millions or billions, of models created and supported over the next years. Plus, IP becomes a totally different problem introduced into this equation and process. Up until recently, it’s been publish or perish. If you’re a doctor and you create something, you want everybody to have it. Maybe you can even license it and make some money off it.

But now, if you try to patent something, every 12-year-old is saying, “Hey, look at this patent, change it just a little bit, and make my own patent.” Or if it’s a new procedure in healthcare, you might say, “No, you have to come to my model, and you have to pay me $9,999 for every access,” or, “To use inference, you’re going to use this software that I require,” because nobody’s going to give away that IP for free. And that creates a lot more problems.

I think part of the underpinning that a lot of AI entrepreneurs have, going back to Erik, is that information wants to be free. The information may want to be free, but the person who owns the IP feels differently. That’s going to change how we interact with models. We can argue about front ends that choose which model is least expensive to answer the question, but we haven’t had the discussion about where we find the IP that’s needed to either train that model, post-train that model, RAG that model, whatever it may be, if that person isn’t making it available but it’s absolutely required. Does that make sense to you guys, or do you think I’m an idiot?

I think it’s really interesting because it also brings up the point of what the role of the doctor is in the future. Number one, I agree that I think we’re going to need more doctors in the future, because as care advice becomes more abundant, more people are going to demand healthcare, actual real-life healthcare services.

100%.

So we’re going to find we actually need more doctors. There’s something to be said about what you’re describing: the role of the doctor today is very rote. It’s kind of like, these are the tasks. You’ve got to do these quality measures. You barely need to use your brain, honestly, for a lot of the things doctors are asked to do in the traditional healthcare setting.

What you’re describing is, what if the paradigm moves toward physicians being more physician-scientists? Their job is to take the bleeding edge of information that we may not know at population scale yet, but on an individual basis, they’re at the tip of the spear, collecting novel data and generating—

Absolutely.

—novel insights about how people are reacting to things.

Novel data, by definition—

That’s a much more intellectually stimulating job description to me.

Yes. Yes. Because by definition, you’re exactly right. Novel data equals person.

Mm-hmm. Exactly. And with—

And novel data—

We’re going to have—

Yeah.

—longitudinal monitoring. Blood pressure is the one we always talk about. It’s the same principle as what you were saying. The fact that we take our blood pressure once a year is kind of ridiculous because it changes every day.

Not me. I got this shit right here, hon.

Yeah, you’re cutting edge. Soon we’re going to have wearables that are measuring it in real time, every single day, and you’ll have a much more precise understanding of how you’re responding to different therapeutic interventions and even diet and exercise and all that kind of stuff. That’s going to create—

But you bring up a good point, Julie, in terms of novel information, and that’s everybody in terms of finding jobs, right? You’re going to need AI translators for people—

Yes.

—who may not keep up with these things. Regular people, wherever they are in this country or in the world, who just want to live their abundant life, whatever that may be. You’re going to need people who are going to be the translators for what’s novel about this circumstance or this individual, and how you make it valid to AI and the output of the AI valid to the individual.

That’s going to be a job, and it’s going to be very difficult to extract everything from an individual that the AI is going to need. I think you’re going to need to have a person there to do it. Maybe somebody envisions a humanoid robot that’s going to ask you those questions, and somebody’s going to be comfortable with that. I’m not. I don’t think that’s going to happen. Yeah.

By the way, Mark, I don’t think I’ve ever heard you talk about another set of entrenched incumbents, which is the EHR companies.

Uh-huh.

What’s your theory of the case there? How does that market change or evolve—or not—in the age of AI, and what does that look like in the next few years?

Honestly, I don’t know. The EHR market is dominated by Epic, and it’s their world, you know? So I’ve just deferred to them and tried to stick where I can have an impact, because we partner with them. We think offering pricing, particularly pharmaceutical pricing, at the point of prescription, when the doctor is there, and giving them different options is a good start.

Right. You guys are part of that new announcement they just made about the—

Yeah, exactly. Exactly.

—kind of tie-up. Yeah.

With Epic and Amazon, et cetera.

Yeah.

So I think that’s a big plus, and I think it’s needed. But in terms of their actual business and where it’s disruptable, I’m not really the one qualified to speak to that.

Okay. My problem with that announcement, by the way, was that I love that they announced it, and then they said it’s not going to be available until May of next year. I was like, “What? How does it take so long?”

Yeah, I know.

Yeah.

I had nothing to do with that.

Okay.

But that’s typical in this industry, right?

Yeah.

Do you guys think that in 10 years healthcare is still going to be one-fifth or one-sixth of GDP, whatever it is now? Is there a kind of terminal state where there’s way more demand than we can supply, and even as it gets easier and more efficient, and GDP grows—even 5% a year, hopefully—our demand still just caps at one-fifth of the economy?

I do.

Yeah. Yeah.

But the value we get out of that spend should be—

Well, the value you get out of the people—

Yeah.

—will increase.

Yeah.

Yeah.

Because they’ll be healthier.

Yeah.

So there’s a difference between health and sick care, as a lot of people refer to it, versus keeping people healthy. They’re more productive, there’s less downtime, and they’re smarter because it’s easier to learn, et cetera. So I think we’ll get greater productivity that’ll more than account for it, which we don’t get now.

Zooming out a bit, Mark, we have a mutual friend who’s a very prominent pollster. I won’t out him here, but he’s done some work indicating that if you ever ran for office, he thinks you could win. Now, of course, you’ve said you’d never run.

No.

We’re speaking hypothetically now. If you were president, or advising the next president, I’m curious how you’d think about the leverage to most address our healthcare challenges. Then I want to hear your broader policy positions you’d be most excited about.

It’s really easy. It’s the same thing I tell states and that I’ve testified for. One: make all contracts with anybody who takes taxpayer money public. No redactions. The minute that happens, you start to have an efficient market, and the costs go down. Two: make it so that anybody who finds a better price for less than their current out-of-pocket cost under their plan can apply it to their deductible. Eight states do that at some level already.

Three: for nonprofit hospitals, require them to publish their entire general ledger, down to each detailed entry, and all their procurement receipts and requirements, so that everybody gets to see it. It’s our money. We’re the taxpayers. Four: instead of the current insurance plans, the way we have them now, 72% of those big carriers’ revenue comes from taxpayers. They don’t take risk anymore. They offload it. Everything’s an arbitrage for them.

Knowing that, we also have to consider that the one debt we never pay off till the day we die is our insurance premiums. Think about that. Whether it’s Medicare—unless you’re on Medicaid, right?—you are paying until the day you expire. And when you need it, you have to fight for it in the current system and probably foreseeable systems.

To me, if we were able to create a program where you deposited what you’d pay for a premium under an ACA Silver plan in your state, and we did a deal with the bank and said, “Okay, in Texas, for a family of five, it’s $2,100 at your income,” right? You just keep depositing $2,100 in escrow, and you own it, but it has to be deposited if you want to participate in this program.

Then, if you have a medical incident or a pharmaceutical incident, you can take that money. If it’s not enough, we’ve worked out a deal with the bank where 2 things happen. One, the bank will loan you money up to a point where a stop-loss kicks in, which means we’re going to take part of the money you deposit and use it for stop-loss coverage. Two, the federal government or the state is going to guarantee that loan to you so the provider isn’t out, and we’re still going to deduct it from your payroll so you pay it back.

If you’re lucky and you never have an incident, and you get to save all that money, when you turn 65 and go on Medicare, you just get all that money and all the accrued interest. Because if you think about it, the federal government will—

I mean, like, in a third-world country, this is—

—you know, cover or guarantee your student loans. They’ll guarantee SBA loans. They’ll guarantee home loans or give you a down payment, right?

This is a third-world country.

We’ll guarantee all these things.

Or what?

Even agricultural things. But the most important thing is covering our healthcare. Good luck. You’re in a bad accident, good luck to you, and we’ll maybe push it off to the hospitals that have to take you in.

So I would create a plan that says you have to deposit this money. The bank will have a guarantee, just like a mortgage or whatever, and will loan you this money. Then we’ll have federal stop-loss programs above some amount, $40,000, whatever it may be. Now all of a sudden, bye-bye, insurance companies. Now all of a sudden, you’re not fighting back because it’s your money to spend as you choose. We’ll have to set thresholds on where and how you can spend it, so you don’t just give it to your cousin Billy, who says he’s going to fix your broken toe. But within those constraints, I think we have a better system. So that’s what I would do if I were advising on healthcare.

What about one of your gifts, storytelling and marketing more generally? What about the AI industry? AI is uniquely unpopular in the US. If you were advising OpenAI or Anthropic on how to change those numbers, and then there’s the whole data center saga, I’m curious what your advice would be to the broader AI industry.

I’ve had these conversations with them, and I told them they’re idiots. What controls how pretty much everybody in this country thinks right now? Social media algorithms. I was literally having a conversation with a 23-year-old friend I work with. We were talking politics, discussing who we voted for yesterday and how and why. I asked, “What about your friends, the other 23- and 24-year-olds?” They said their friends, living in Texas, didn’t even know who Ken Paxton or James Talarico were—or are, right?

That’s insane in my mind, right? Insane in this person’s mind. But it reinforces the fact that you get more of what you seek via algorithms. If you really have the best AI, you should be able to not only reverse-engineer those algorithms, but also put together a marketing program. You tell everybody you can do marketing programs through your bots, your bot or Claude or whatever, and really start marketing there, because this applies to politics, AI, everything in this country—the divisiveness.

We all have our unique AI feed in the social media platforms we use, and those platforms don’t give a shit what they feed you within some minimal constraints. So everybody looks at the world differently. The days, particularly for younger people, of thinking, “Just buy more ads and ruin our football games and our basketball games, and everything’s fine,” are over. Maybe older people vote more, and that’s maybe why.

But it’s useless when you’re trying to change hearts and minds about AI specifically, even though the irony is that it’s the AI behind the algorithms that’s ruining everything. As long as Section 230 is there, we are where we are. And unless we take an Australia-type approach or a states-type approach for those 16 and under, what I told them specifically is: You need to use your AI to segment everything and figure out the best way to put together marketing, promotion, and however else you want to do it. Do it like the Russians do it. It doesn’t matter, right? Because that’s what’s at stake for you. But data center hate is just a proxy for not being able to understand AI—

Yeah.

—and not knowing all the pieces. I don’t blame people at all, right? Just like employers have been really dumb with their health care benefits, a lot of communities have been really dumb with the deals they’ve negotiated for data centers. They’ve fallen for contractual terrorism in a lot of cases.

To me, educating people about social media, and educating our legislators and whoever’s president about social media, is important because I don’t think they truly understand the impact. We don’t all read the same newspaper or magazines, look at the same websites, or watch the same news shows. We all have our own unique TikTok feed. Your feed is not like mine. The best way to find out about your kids is just to look at their social media feed. My 17-year-old son’s feed is basketball, basketball, basketball. Now I’m starting to see a little bit of politics pop up, but then hot girl, hot girl, hot girl, basketball, basketball, car, car, car. And I’m like, okay.

No, I’m okay.

Typical 17-year-old guy, right? But until the big frontier models understand completely that they have to figure out how to become part of everybody’s feed, they’re going to be putting out all these position papers that are useless.

Yeah, no, Mark, we agree with you that what you said earlier about this being the best time to be alive and an entrepreneur. So many things that were historically intractable are now all of a sudden addressable. But is there an area of the market—you meet a lot of entrepreneurs and see people building in all sorts of areas of the world—is there some area where you’re like, “Why is no one building here?” Is there a white-space opportunity for entrepreneurs that you think is still open?

All of them. Seriously. When I go talk to kids, I always tell them there are going to be 2 types of companies: those that are great at AI and those that have gone out of business. And there are 2 types of people in how they use AI: those who use AI to learn anything, and those who use AI so they don’t have to learn anything. If you’re one of those people who uses AI to learn anything, and you realize what types of companies and problems aren’t being solved by AI, don’t be shy about just talking to your AI.

I say this not just to kids, but to 50-, 60-, and 70-year-old entrepreneurs. The biggest challenge right now for entrepreneurs using AI is feeling comfortable talking to your AI, whether by voice or typing, because there’s still a lot of hesitancy. Just like back in the day, when people got their first spreadsheets, they still had their 10-key adding machine and would check it to make sure the spreadsheets were right. You do have to check the output, but you have the opportunity to take the output of Grok and put it into ChatGPT. When I go through all these contracts I was telling you about, I start with one and say, “Hey, one of your peers came out and said this about this contract. What do you think?”

Uh-huh.

And spend a couple of hours. That’s all, young entrepreneurs. So when you ask me if there’s one thing, I truly believe it’s everything. Everything will change. This is the greatest transitional technology we have ever seen, and I’ve been around for all of them. This blows them all away.

The hardest part is just going for it. That’s where Silicon Valley and kids from the Bay Area have a little bit of an advantage, because the culture there is go, go, go. But I think the rest of the country and the rest of the world, once they get that confidence level, are going to start seeing businesses we never imagined pop up. We’re not seeing that yet. We’re not seeing companies where it’s like, “Oh, what is that? Why are they doing that? What is the opportunity there?” And then they realize it before we did, right? It’s kind of like Friendster at first. Like, whoa, what was that?

I think the biggest thing to focus on, if you’re looking for something, is to ask anything. I do it all the time: “What do you think about this? What do you think about this? What do you think about this?” I’ve held phones up next to each other, taken the output, and had them have a conversation with each other so I can listen in. Don’t be shy about using AI in ways that feel really uncomfortable and weird, because that uncomfortableness is probably the path to something groundbreaking and game-changing.

Love it.

That’s a great note to wrap on. Mark, thank you so much for coming on the podcast.

It was fun.

Thanks, Mark.

Thanks, guys. Thanks, Julie. Thanks, Erik.