The CEO behind the AI Doctor ($41m raised)

I had a conversation with Vinod Khosla. He made the argument that doctors are over. AI is just so much better.

It might be considered medical malpractice not to use AI to treat a patient.

KJ Dhaliwal is the founder and CEO of Lotus AI. He's building an AI doctor and has raised $41 million from investors including OpenAI, Microsoft, and Stanford Medicine.

You've heard that joke, right? In the future, every plane will have one pilot and one dog, and the dog is there so the pilot doesn't touch the controls.

KJ is a serial entrepreneur who exited his previous company at a $50 million valuation before turning his focus to innovating the healthcare industry.

Healthcare investors are the worst. They're used to investing in an asset class that is inherently not innovative. Go fucking take some risks, man. There are so many dead bodies in healthcare. Every person I would talk to would say, "Oh yeah, healthcare is really hard. Don't do it." And when Sam Altman wakes up in the morning, the last thing he wants to worry about is running a medical practice.

You mentioned that before Lotus, you did another company, which you sold, and before that you were a trader. So let's go from the beginning. Walk me through: what was your first job and how did you start?

My first job was Domino's Pizza. When I was about 15 years old on the East Coast. I grew up partly in New York, then Pennsylvania. I've always kind of been an entrepreneur. I was building things, doing things. We had a snow shoveling business in my neighborhood. I had six different kids. This is when I was 15, 16. We would shovel snow and get people to pay us, basically a subscription during the winter. I was a power seller of sneakers on eBay, just always thinking about...

Early days of eBay?

Early days, yeah. I guess you could say early days. But my first real, professional job was when I was in university. In parallel, I was studying to get my Series 56, which is an options trading license. I was able to pass that pretty quickly because I started investing when I was pretty young, just a little bit of money here and there in the stock market, and learned how it worked. I was always fascinated with that. So to your point, I was an equity options trader for about a year while I was in school.

Which school were you at?

I went to school in Philadelphia.

Yeah, yeah. I went to Lafayette for a little bit, took some classes there, but then ultimately went to Drexel University, and took some classes at Penn as well. I was always just trying to learn, had this thirst for knowledge, I guess you could say, figuring out ways to make money and build businesses, right? But when I was in Philly, we started a few companies as well. One of them was a QR code scanning app. This is before QR codes were popular, right? There was a period about 10 years ago where QR codes kind of got popular and then they fell off again, and then with COVID they got really popular again. But we built this app called Eyeball, which was a scanning app for QR codes.

So generic QR codes, not a specific use case.

The idea was we would generate our own QR codes and then give them to businesses, and then allow their customers or potential customers to scan them to do things. We actually had a good use case. The Susan G. Komen Cancer Foundation worked with us. They have a marathon in Philly every year, and they basically put the QR code on the marathon runners' sleeves. So people could scan the QR code and donate to the Cancer Foundation. And they ended up raising a few million dollars through that, and that was incredible because it was something that no one had really done before, and QR codes were just a new thing. But that company taught me a lesson, which was that we were too early. The market wasn't ready for QR codes, and people didn't really know what they were.

What's the context of phones at the time? What phones existed?

This is 2013, you know? So it's not that long ago. Phones were iPhones, right? And we ended up shutting it down, but then a few years later, this company Scan came out with a very similar idea, and Snapchat acquired them for several hundred million dollars, and we were like, "Shit, we were early." And that was my first lesson on timing in life.

They acquired them? Why?

Snapchat was doing really well at the time. I think they were trying to build their own QR code scanner, and they were becoming this camera company, a lens company, and I think they probably liked the team too. It was one of those things where it wasn't obvious when we started that it was too early.

And then, I guess to answer your question, I moved to San Francisco after university. I was just like, I'm not gonna stay on the East Coast because it was a horrible place to build a business, right? Even today, I love New York. I've thought about moving back a lot ever since I moved here. But I would say the best companies in the world are still built in San Francisco when it comes to technology companies.

For what reason, from your take?

I mean, the obvious ones. I think it's just the combination of talent in the software engineering space and then risky capital, right? Venture capitalists that are willing to take risks because they've made money taking risks before. At the time in Philly, for example, First Round Capital was there, but they weren't really taking that many risks. Philly wasn't a great city for tech companies. Even New York wasn't either, and I would argue it still isn't. New York is a great place to build a brand and build a community, and do things when it comes to CPG. Even for our marketing, for example, we're hiring in New York, right? But when it comes to building the actual core product and the engineering, I found that San Francisco just has a DNA that doesn't exist in the rest of the world. Because it also attracts the best entrepreneurs from the rest of the world to come here to build something.

I mean, we sit here. That's Anthropic. And then that's Strava. And then Oura is in the corner. Everything is just five minutes away.

Yeah, we got LinkedIn right there. Every major tech company is here, and that's...

And you can literally see, not from this window, but if we go higher in this building, you can see OpenAI.

OpenAI's office. They just moved most of their team from Mission to that office. I mean, why are you here, right? Same thing, right?

Yeah, it is the same thing.

You're not from here, right?

Exactly. For me, it's also the same thing, maybe reflected in a different way. It's serendipity, right? It's the fact that you are five minutes off South Park, and all of the investors are there. Opposite this office are all of the people. I go for a walk in the park, and I see the Strava folks every day and the Anthropic folks every day and so on. You do an event, and everybody joins. And everybody's receptive to joining. When I was in London, I realized that in Europe value comes from scarcity, and here value comes from exposure. I spent ages in London, and I was thinking, "Oh, it's a startup place," because they were advertising it as a startup place. And you couldn't meet anyone, nobody meets. And then I go to Boston one day, and I'm at MIT, and we're doing this hackathon. And then I see Jared Friedman, the partner from YC, and he's there doing the same thing. I'm like, "Man, first of all, the US is different." And then you come here, and everybody tries to get exposure, so you get to meet everyone. It's such a small place, and everybody meets. You go for a walk, and you find everyone. And the other thing is, there was this conversation about Oura going public maybe next month or something. It was on the news yesterday.

Yeah. I mean, everyone knows they're going public, yeah.

Yeah, but I was in their office three months ago, speaking to them, and they told me. You kind of know all of these things that are happening before they happen.

Yeah. Another reason why San Francisco is a productive place is because at 9:00 p.m. on a Wednesday, there's nothing going on. Versus New York, 6:00 p.m. on a Monday, whatever time, there's always something going on. You can have fun, go out, whatever. So that distracts a lot of people as well. And the thing here is you have all the talent, and then you have all the big tech companies that spend a lot of money attracting the best talent from around the world and around the country to come here. Then those people get bored. They realize the problems with corporate America and how slow these big companies move, and then they wanna go start their own companies, or they wanna join startups. So you have this constant flow of talent, right? And then you have all this capital, and then you have nothing to do. So it's like, might as well build shit. And try to change the world and work with people that are equally passionate about problems that you're interested in.

How did you hear about San Francisco in the first place?

Actually, it's a good question. I remember this. When I was in Philly, I was at this investor event. I think it was a Boston-based VC fund. I forgot the name of the fund, it's Flybridge or something like that. One of the partners was there, and he was talking about Boston and San Francisco and all this, and I had already been thinking, "I wanna go." I've always known I wanted to build things, I wanted to be an entrepreneur, and I wanted to build a tech company. When I was younger, I always thought I'm gonna move to San Francisco. But I didn't have enough context, proof points you could say. Like, to your point, how did I know? So when I went to this event, I went up to him afterwards. I was like, "Hey, here's the deal. I'm doing really well. I have this Wall Street thing going. I'm graduating. I'm doing all these other things out here. I'm building companies. I have a good network. But I'm thinking about dropping everything and moving to San Francisco. I don't really know anyone out there, but I feel in my gut that's the place for me to be. Should I do it?" And without any hesitation, he was just like, "Absolutely. Just go. You'll figure it out. Just go there." And honestly, that was the reason. I was like, "This guy seems smart." He obviously has built a great fund and invested in a lot of great companies. He knows what he's talking about. I was like, "I'm gonna do it." And I told my parents. My parents were devastated because they were just like, "Wait, you have all this stuff going on here." I had offers from Goldman and JP and all of them, and I just dropped it.

It seems like everyone has an offer from Goldman, and they just drop Goldman all the time. Do you see that?

I mean, I don't know anymore. At least back then, it was an accomplishment. The way Goldman and these investment banks and trading firms worked back in the day, and they probably still do, is they would scout different schools for talent early, give them summer internships, and then hire them full-time. If you got an internship, you were very likely to get the job. So the idea at that time was that you kind of made it. And I think now it's almost a flex to say, "I dropped out of college," you know? I meet people all the time who are like, "Oh yeah, I was a freshman. I got into the college just so I could prove to myself I could get in, and then I dropped out." And I was like, "That kind of makes sense." It's the same thing: you get a job offer from Goldman just to prove to yourself that you can do it, and then you're like, "You know what? I'm actually better than that. I'm gonna go do something else." And I'm really happy, because I see a lot of my friends that are still doing that, and they're miserable in New York, right?

Everyone is miserable, aren't they? To be honest, I always saw this as a fake scheme. I was at Imperial, which is considered a very prestigious and very good school in London. And I remember the day McKinsey came to give a speech in one of the first weeks of university. They brought this one guy who was working at McKinsey and used to be at Imperial. That guy comes there, he sits, and there are 500 people in the class. You cannot get in. It's impossible for everyone to attend. And it's like the guy is God. Right? So they create this perception that it's so incredible for you to be there, that it's such a smart guy who went there. And then they also reject everyone. Everyone applies, and then one person makes it to McKinsey, and everybody's like, "Oh my God, this guy got a job at McKinsey." But then you see it from the outside and you start thinking, "Hmm." Most of my friends that went to McKinsey were so disappointed. And it's the same with Goldman and so on. So at the end of the day, it gets to the point of: where do you want to be five years from now? And to touch on your own experience, so many people come with this question, "Should I be an entrepreneur?" Whereas in your shoes, you're 15 years old and actually doing the thing. So the answer to that is: do the thing.

And that's all I knew, right? People ask me this all the time: "Oh, when did you figure out you wanted to be a founder?" As far back as I can remember, I always wanted to be a founder.

I was a founder. There are many ways to describe a founder. One way is that they're crazy, right? Travis from Uber recently described it as someone that can endure a lot of pain and continue to push forward. But I think another way is someone that feels the status quo and the safe path in life is not for them, and they want to create things, do things differently, and change the world around them. That was something I always believed. Making money at an early age through these different ventures, you could call them, was just a way to itch that need and satisfy it. So going back to your original question, when I moved to San Francisco, I was on the early team at a company called Union Metrics, where I quickly learned. My goal was to learn. It always is and still is. How did that company work? How does a startup work? How do you raise money out here? We actually built some really cool things. We built the first Instagram enterprise analytics tool, and the first Tumblr analytics tool for large enterprises like Nike and the NBA. The company actually did quite well and got acquired. But I had left, and I started a dating app called Dil Mil, which ended up becoming the largest South Asian dating app. And again, I just realized...

Wait, wait, there's a lot to that. How was it the biggest? How did you acquire customers? What was the product?

Yeah. This is 2013-ish, 2014. 2013 is when I had the idea.

That's pretty early. What other dating apps existed at the time?

This is early Tinder, Hinge, and Bumble. Still very, very early. For the South Asian community, the only alternative at the time was a website called shaadi.com and a few others like matrimony.com. They were kind of '90s websites, just catalogs of people, and half of their users were people's parents. So I thought, okay, mobile is obviously a thing. Dating apps are a thing now. I need to apply this technology to the problem of South Asians wanting to find someone to date and marry, which was a really big problem because a lot of South Asians tend to date and marry within their community. And even within the South Asian community, there are subcommunities, different religions, different languages, et cetera. So there were a lot of nuanced needs in that community that weren't being serviced by the mass market, the Tinders of the world. And even the word dating doesn't actually exist in Punjabi or Hindi. So we kind of had to...

What's the word?

There is no word.

There is no word?

Yeah. There's no concept. The concept of dating doesn't exist. You just get married, right? Well, now we like to say we changed that. But traditionally in South Asian cultures, it's all arranged. Your families find somebody for you, and that's probably the first person you're dating, and then you get married.

So there's basically no trial period.

There never was, traditionally. And this is common in a lot of cultures, by the way, if you go back far enough. So we not only had to create a great product and do all the software and technology, but we had to create a brand and change culture, which was really hard. Looking back at how we got distribution, it was through creating a brand and finding early adopters who resonated with it. That happened to be on the East Coast, in New York, where a lot of my friends grew up. Their friends started using it, then their friends, and eventually it got to the point of viral growth. To your point, all great consumer companies start with a small community of people that deeply care about that problem or opportunity.

What would you say was the one thing you did absolute best? What were you the best in the world at? Was it brand building?

At Dil Mil? It's funny you bring that up. I would say we had features inside the app that were arguably better than Tinder and the mass market apps at the time. Not just in design or conceptually, but even performance-wise, because we spent so much time trying to fix certain problems. How do you actually match somebody within a community that has ethnic tastes, variability, and features that Tinder or Bumble may not care about, at least at the time? Now they have a lot more filters. You can filter by language, religion, and all these other things. But very early, Tinder was just age and miles away, and that doesn't really work for the South Asian community. So we were building things much deeper than that, thinking about matching algorithms and how we could build really good recommendation systems. So I would say our product was just a lot better. But we were also really good at building brand and creating content that would resonate with our community and create trust. Because dating apps are all about trust. This is actually a good topic, because one of the questions I would often get when I was first thinking about Lotus and starting to raise money was, "What did you learn at Dil Mil that you can apply to Lotus?" They're very different: one's a dating company and now it's a healthcare company. Beyond being able to build a great product, hire great people, and be good at consumer, it was this concept of building trust with a community, and I think that's really important in healthcare as well. It's really important in dating apps because you have to convince a lot of women that this is a good product, that it's going to find them somebody good, that they can trust it, and that it's a cool place to be. If you can't do that, your product will never work. Once you get the women on the platform, the men will always follow. That's the biggest challenge for most dating apps: attracting enough women to create the critical mass for the product to take off, and very few dating apps have been able to do that. There are a lot of dating apps, by the way. When I sold the company, I was at the parent company, on the M&A team as well, and there was a new dating app almost every other day. And most of them never succeeded. I think there are some opportunities now with AI dating apps, but we'd have to get into that. In general, I would say we were really good at building brand and creating trust with our user base, but that was done through doing the right thing for them and building the right product. So they knew they were actually going to find somebody valuable and quality, versus somebody five miles away who just wants to hook up, which was the vibe on Tinder.

Yeah. How do you go from that to raising... How much did you guys raise, $40 million, $45 million?

$41 million.

$41 million. Was this the first raise you've done for Lotus?

For Lotus? No, it was a combination of a seed and Series A.

Okay. So how did this happen? If a founder is listening to you now, what was your process? How can they learn from it?

It's going back to what I was just talking about. Firstly, it's actually building a product that people want, a good product that works. It's the obvious thing people say, but I don't think people fully appreciate it. What I'm seeing in Silicon Valley right now is that you can now use AI to build stuff. So how do you really know if this product actually works at scale? It's very easy to just vibe code a product now. Building a good product takes a lot of effort, not just coding it, but thinking about the use cases for it, what kind of brand you want to build around it, what kind of community you're going to build around it. We spent a lot of time on that early. Then when investors would meet us and ask questions, I would be able to answer, because I was predominantly the only one that fundraised. My co-founder is more technical. He doesn't really like investors that much, to be honest, so he never goes to these meetings with me. I would be able to answer the questions so well that it was apparent to them that I deeply understood not only what I was trying to do, but the space, and that I had thought about it for a really long time. I was an expert in this space in their eyes, and I think that is the biggest thing.

So how did you become the expert in this space before you were...

I just spent a lot of time learning. I started Lotus in 2024, but I was probably thinking about it for a major part of my life, the last 10 years, and especially the previous four years. From 2020 to 2024, after I sold Dill Mill, I was really thinking about building something in healthcare. Even when I was at the parent company that acquired Dill Mill, I was leading strategy there as the chief strategy officer. Every single night I was reading, learning, and thinking about how to build something in healthcare. How does healthcare work? It's a very complex space, as you know. By the time I started the company, I had formulated a really good understanding of what I wanted to do, but also how all of healthcare worked, and the problems and challenges inside of it. By the way, I don't think that's necessary in every domain. It's necessary in very complex domains where, if you don't know what you're doing, you're either going to get in trouble or you're not going to build a good product. So I don't want the advice to founders to be, "Don't start a company until you're an expert in the space," because I actually think generally that's not true. You should just get started, and you'll figure it out on the way. But for healthcare, every person I would talk to would say, "Healthcare is really hard. Don't do it. There are so many dead bodies in healthcare." And I was like, "Huh. I wonder why. I want to learn why." One of the things I learned, and this led to me having even more conviction in Lotus, was that most healthcare innovation is either for providers, like hospital systems, or for payers, like insurance companies, because that's really where you can make most of the money. Healthcare is this weird thing where the end user, the patient, actually isn't used to paying for it. They're being forced to pay a premium, their insurance premium on a monthly basis, either through their employer or the government or Obamacare, whatever you want to call it. But normally when you go to the doctor, outside of your copay, you always feel that your insurance company is going to cover the bill. So you're not paying for this thing. That's not true in almost every other part of our life. Whenever we consume something, we generally are willing to pay for it. Because of that, all of the innovation in healthcare has gone to the payers and the providers. So one of the big distinctions we made very early on was that we're not going to bill insurance, and we're not going to try to integrate into the existing healthcare system. We're just going to go directly to the patient and figure out what their needs are and what we can do for them really, really well, and we'll figure out how to make money later. I think that's been the single most important thing we've done really well. Because of that, we were able to build something that patients really need and care about and love. That was reflected in our early retention numbers and our early engagement numbers. And going back to your question, when investors saw that, and then saw how much we knew about the space and what we were doing, they were lining up to invest. So we had a pretty competitive Series A. I think we're still very early. I don't actually think a lot of investors really understand what we do yet, and we kind of like to keep it that way. Some of them knew, some of them we talked to here and there, but I get an inbound at least every other day now from top VCs wanting to meet. It's not that I don't want to meet them. It's just that I don't have anything to talk to them about right now. We're really busy building. Maybe when we think about fundraising again, we'll talk to them. But an interesting piece of advice I've gone back and forth on, for founders thinking about fundraising, is that you don't actually raise money when you really need or want to raise money. You raise money when you're just doing the right thing, and you happen to now need to raise money or be thinking about it, and then the money kind of comes to you.

Just on the point you're mentioning about investors, I had this conversation yesterday. The first time, you're excited to meet that investor, and you're like, "Oh, my God." You put them in some sort of God position. And then over the years, what ends up happening is they reach out to you and tell you the same things. They tell you, "I'm going to help you with introductions," and so on. It's the same story every single time. It just becomes this boring, mundane thing that keeps happening.

Yeah. Investors are interesting people. If you think about it, they're all selling a commodity, and they're trying to put some branding on it to convince founders that they're doing something different. They actually have the hardest job in the world: how do you differentiate and build a brand when everyone's selling a commodity? Capital is a commodity. So then it comes down to the people, the actual partners at these funds that you deeply get to know and actually like. Many people have this saying: "Most investors aren't going to add any value." There's maybe 5% that would add value in certain edge cases. But the bigger problem is that most investors will distract you and maybe remove value, or hurt you, because they waste your time on something. Investors are glorified, and now they're becoming influencers. I've always thought of investors as really smart people who don't have the courage to go start a company. They'd rather take the safer route: "I'm going to invest other people's money into a bunch of companies and hope one of them 100Xs and returns the portfolio," because they don't have the balls to go do it themselves. But they want to find founders that can do it. Look, I have nothing against investors. A lot of people listening are probably my friends who are investors. I love you guys. But at the end of the day, go fucking take some risk, man. Life is short. I think a life worth living is a life where you've tried to do things differently and change the things around you that you feel are wrong. What are you really going to do in venture capital? How are you going to improve venture capital? Inherently, venture capital is a game of greed. If you're a VC, you're basically helping rich people get richer.

Yeah, it's interesting, investors telling you to take risks, but if you think of their position, how much risk do they take? It's this modernity problem which stems from doing the thing you speak about. So yeah, I think we could spend an hour on that. If we get to healthcare, what's AI replacing and what's AI keeping?

Oh, by the way, just to close out the investor thing, now that you bring up healthcare: healthcare investors are the worst. Healthcare investors are used to investing in an asset class that is inherently not innovative, right? And so they have learned to look for the wrong things. Even as a healthcare company, you kind of never want to raise money from a healthcare investor, unless that investor is the owner of five hospitals and they're going to give you a bunch of contracts right away and you're going to make a lot of money. Sure, go for it. But ultimately, you want to raise money from the best generalist funds because they're investing in you for the right reasons. I think healthcare investors are so risk-averse that they almost shouldn't be in VC. It's kind of funny to say that, right? Because we're a healthcare company. You would think we raise from healthcare funds, but we've raised from Kleiner Perkins, which is now starting to do more healthcare but is by no means a healthcare fund. Or CRV, which is by no means a healthcare fund, right? And I think it's because we're also consumer, so it's slightly different. We are trying to raise from the best consumer investors. Going to your point though about what AI is doing in healthcare, your question is basically what AI is going to kill in healthcare?

Yeah, maybe two to five years from now, what do you expect is going to be the same, and what do you expect to be different?

Well, first of all, broadly speaking, I think AI is pretty disruptive in every field. Because healthcare has a lot of complexity, regulatory capture, risk and liability, AI can't completely kill it or automate it, right? Even with what we do, we have real doctors that give treatment to patients. We are willing to be accountable and responsible and hold the liability for what we do. AI is never really going to be able to do that, at least in the current legal framework. So I think healthcare is shielded from AI much more than most other industries. Any regulated industry is somewhat shielded, I think.

Is this a good thing?

I don't think it's necessarily a good thing. If you put the best group of doctors in a room today and give them the entire context of a patient and ask them to come up with the diagnosis or treat that patient at that single point of care, they might actually outperform the AI. But if you ask them to sit in that room for an entire year, monitor everything that patient's doing, and guide them throughout that year in optimizing their health or solving their problems, AI will outperform them, because the AI's context is much larger. It can talk to the patient for hours and hours without getting bored or tired. It can pull in the latest clinical research and evidence at its fingertips to understand what's going on with this patient, and then compare it across different patients. It's just obvious that a human is not going to be able to compete with AI in terms of cognitive performance, right? Or just knowledge in general. So to your question, is it a good thing for society if AI is regulated in spaces like healthcare to the point that you are defaulting to humans just because of some old regulation? I do think doctors are really important, and we love doctors. We think they're heroes. We work with a lot of doctors. But even they are starting to tell us about Lotus Cortex, which is our physician-facing tool. Just for clarity, we have Lotus AI, which is our patient-facing consumer app, and then we have Lotus Cortex, which is our physician-facing tool. It's sort of like a Copilot or Cursor for doctors that we built. Routinely, our physicians are starting to tell me, "Hey, this Cortex thing is actually really good. It was better than me in this one situation. I didn't know what to do, and I was able to figure it out because of it." So in the way that AI is de-skilling people in a lot of industries, AI is starting to upskill doctors in healthcare. Meaning now they're comfortable practicing in areas of medicine that traditionally they weren't. Maybe they learned about something in medical school but then forgot about it. Now with the power of Lotus Cortex, or even something like OpenEvidence, for example, they can start to learn about those areas of medicine again and be confident practicing things they might not be confident in otherwise. With Lotus Cortex in particular, because we pull in the entire context of the patient as well, we're able to make them really confident for this particular clinical encounter or point of care that the patient is seeking treatment for. So in that respect, at some point it might be considered medical malpractice not to use AI to treat a patient. Because you're basically depriving a human of the best care just because of some law, right? So I do think the laws will change. I think it will take a long time, because the practice of medicine is regulated at the state level, right? Every state has jurisdiction over this, and some have their own laws and regulations around what doctors can do, what AI can do, and how medicine is practiced. So unfortunately that will take a really long time to change. But you're seeing it in other industries. Think about Waymo, for example. I always use Waymo as an example. Waymo is actually safer than human drivers, right?

It pisses me off every day that it goes slow or just stops you somewhere.

But I will say, if you're crossing the road and you're using your phone, you feel way safer if it's a Waymo at the red light versus a human at the red light. Right? I almost don't even have to look up if it's a Waymo. Even if it's not a red light, I know I can just cross the road. At that point, yes, it's slow, but it's also really reliable and really safe, right? And in things like healthcare, you'd rather have that even if it was slower, which by the way it's not, it's faster. Healthcare traditionally is super slow, right? Lotus is really fast. For example, if you signed up to Lotus right now and had some issue, fully onboarded and became a patient of ours, and you needed a prescription for something, you could theoretically get that prescription sent to your pharmacy down the street within five minutes of downloading the app. That means a physician would be looped in, they would review everything, and they would decide if it's clinically appropriate to prescribe you this medication. We would send that medication to the pharmacy, and you would go down the street and pick it up.

But wait a second. So the AI makes the recommendation, then the clinician looks at the recommendation and approves it, and then the patient receives it?

You could say that, yeah. It's a clinical decision support tool, yeah.

And what is the percentage right now that the physician rejects the proposal of the AI?

Yeah. It was in the tens of percentage points. Now it's in the single digits, and it's getting better and better.

So where this is going is the physician is only necessary for maybe the approval, because the law states so.

Yeah, and we do think, mostly from a patient safety perspective and a regulatory perspective, yes, that's why the physician is there. But there is also an element of trust and humanity to it, which is interesting. Our patients know that there's a doctor giving them the prescription. It's not just AI, right? And they're more likely to engage and...

But what's the use case of the clinician in that context? Because I had a conversation with Vinod Khosla two weeks ago. He made the argument, and he published in the Journal of...

Journal of Medicine AI.

Yeah. He published, it became super viral and so on, and he was saying that doctors are over. AI is just so much better. Then I also had the Hims & Hers CTO last time, and he said, "Look, this is not going to happen. Clinicians are going to be powered by AI and they're going to have even more work." But it sounds from what you say that we are partially using the clinician because we trust them more, but not necessarily for the deduction.

I think another analogy is: would you get on a plane that didn't have a pilot?

I would think of it like this. If you look at the rate of progress of aviation over the past fifty years, it comes from context on accidents. As an accident happens, the system improves, because all of the airplanes improve themselves. And then aviation reduces the error rate to an extreme level. It's so much signal over time that makes the airplane do better. You have someone taking the risk. But to be honest, it seems to me that if you left it as it was fifty years ago, with all of the decisions left to the pilot, the error rate would be so much higher.

To be clear, planes today, the pilot really doesn't do much. It flies itself. So my point is, you've heard that joke, right? In the future, every plane will have one pilot and one dog in the cockpit. Yeah. And the dog is there so the pilot doesn't touch the controls. And so basically, it's proven in aviation that the machine is better at flying the plane than the pilot is, but people still won't ever get on a plane today without a pilot in it, right?

Why? I would.

I would too, yeah. I'm just saying it's a cultural thing, right? And that's where we are in healthcare and medicine today. Minus, again, the people that live in San Francisco, and maybe New York and other major areas where they're really deep in tech, who would probably do it and are using it. Some Lotus patients don't even care that there's a real doctor involved, right? But when you look at the majority of the world today, or even the majority of the US, they are culturally still used to having a human in the loop when it comes to taking a new medication, for example, right? And I think...

To be honest, I'm the exact opposite. Every time I go to a doctor, I get an opinion, and that opinion is always different across doctors. So you don't get the diagnosis based on the disease, you don't get the facts. You just get an opinion. So...

No, I'm with you. I do think it's coming. You're asking why the Hims & Hers CTO and Vinod have such opposing views, and I think Vinod is right. Obviously, we're building an AI doctor, if you will, right? We believe at some point in the future the laws will evolve, physicians will no longer be needed to do certain things, and AI can autonomously do them. And by the way, we have all of the data too. What Lotus is doing is actually very interesting in that sense: every single interaction where a patient is escalated to clinical care and our physicians disagree with our AI systems, our care intelligence, on something is a learning opportunity for our system. And the reason that's really special compared to a benchmark that ChatGPT or Claude hits on health questions is that the point of care data that Lotus and Lotus Cortex capture is in the context of the very complex longitudinal care that we're delivering to a patient, and what other health systems are delivering to that patient. So it's not just us. Say we give you a prescription through Lotus, then three weeks go by and you develop some disease, you go to the ER, you go to a cardiologist, and a bunch of other things happen, and by the way, we ingest all that data. Then you come back to Lotus and modify a prescription that cardiologist gave you, and let's say our AI suggests something and our physicians disagree with it. That learning data point is very unique and very proprietary to us. It doesn't exist in the world today, because healthcare data doesn't get tracked like that. One of the foundational things we did really well early on was figuring out how to build a system that captures all those data points and continuously reinforces and learns on them to improve the quality of the recommendations of our care intelligence system.

It's even more important because, if you think about it, clinicians do not share their mistakes with other clinicians, because if they do, they admit a specific error they can be sued for. So across the whole industry, it's very difficult to share your learnings.

Let's put it this way. What's the thing that leaves first? What's the sequence of things that get replaced with AI?

Again, anything that's unregulated and that doesn't require a lot of longitudinal context and memory is very easily replaced. And even within the regulated space, anything that's deterministic is very easily replaced. You brought up an interesting point earlier, which was that if you go to different doctors, they'll give you different opinions. The reason for that is there's no standard playbook in healthcare. Most of the practice of medicine is probabilistic. That's why it's called the differential diagnosis. When a doctor comes up with a differential diagnosis, it's literally a list of things that could be wrong with you, ranked in order, with the thing at the top being what's most likely wrong with you. And by the way, that's never 100%. So when you think about how a doctor works, a doctor is actually probabilistic. So is an LLM, so is an AI model. But my point is those things are harder to evaluate in the real world unless you're evaluating the entire system. I was in a meeting a couple weeks ago where HHS and CMS invited us to comment on how to evaluate an AI doctor, for example. The example I gave was that you can't just evaluate based on a benchmark. You have to evaluate it in the real world within a system. You actually have to evaluate the entire medical practice. And that just turns into how high quality of a practice you are, the same way you would evaluate a real-world medical practice today. If you were to evaluate a real-world medical practice, it's not just going to be the decisions the doctor is making. It's going to be how the information is organized, what the patient's experience is, what the protocol is when something goes wrong, all these things. So to answer your question, that's why I think the parts that are probabilistic in regulated spaces are going to take a while to be fully replaced. Regulation exists for those reasons, because of the complexity of the system. It's not just that somebody decided it. I mean, there are industries like that, where someone decided a person needed a license, like taxi drivers, like with Uber. Travis talks about that, right? The Taxi and Limousine Commission in New York is like a cartel. They gave out an arbitrary number of licenses in the form of medallions, and now people basically sell those and created a marketplace, and then Uber came in and disrupted that. But there's no reason somebody needed a taxi license other than training.

On that, you have to see what happened in Greece. When Uber went to Greece, the taxi drivers were getting on the Uber cars and smashing them. So then Uber left Greece.

Yeah. It makes sense.

There's this argument around a study, I believe it was at Stanford University, where they had the AI evaluate a patient and the diseases, and it was 88% correct. Then they had humans do it, and they were 73% correct. And then they had humans supervise the AI, and it went to 76%. So basically, the humans degraded the AI. And obviously you have a massive error rate in the actual thing. But if you guys are learning from the approval of the physician, how do you know their diagnosis is correct? Even if it's legal, even if they take the responsibility and so on, how do you evaluate whether it's actually correct?

That's a great question. We don't just do it on one doctor. We have a lot of doctors treating a lot of patients, so we're learning across that entire universe. It's not just one doctor's opinion anymore. And that's the other problem in traditional medicine. That's why people get second opinions, because some people don't trust doctors, and they're like, "Hey, I don't know if this doctor is right. I'm going to get a second or third opinion." So again, you have to evaluate the whole system, not just what the doctor is saying. And sometimes the doctors are wrong. We've found that too, and the doctors have admitted it to us. Doctors within our medical practice will come to us and say, "Hey, I was actually wrong about this thing." A lot of times it'll be doctors coming onto Lotus Cortex who are a little hesitant, not sure how we operate or what we do, until we educate them. Then they get really confident, and they're like, "Wow, this thing is actually really good." And eventually they'll say, "You know what? I initially didn't believe this thing was good or even decent, and now I believe it is definitely better than me." I've heard this multiple times now. So I think the best doctors in the world are starting to realize that in order to succeed in the future, they need to learn how to be more human in their interactions with patients, because technology is going to automate a lot of the busywork they spend their time on. When you think about what a doctor does, the actual clinical decision doesn't take a lot of time. In traditional healthcare, most of the time is spent on busywork that's unnecessary, because of how complex the system is, whether it's documenting ICD-10 codes so you can bill CPT codes to insurance companies, or making sure your practices abide by certain laws, or what have you. A physician, given the appropriate context on a patient, can decide in under a minute what's going on and what's wrong with you, and that's actually the most valuable part of a doctor. When they take their USMLEs, for example, they're spending about two minutes on a question. Those questions are typically a clinical vignette: "Here's what's going on, these things are wrong with the patient, what's the most likely diagnosis?" They're fundamentally trained to be really fast at making that decision. Basically, they're pattern matching. And we all know AI is incredible at pattern matching. That's why AI has blown the USMLE benchmarks out of the water compared to a human doctor. So the way we work is we give the doctor the best and most appropriate context, but we allow the doctor to manually verify it if they need to. A lot of times they do, but they can still make that clinical decision in a couple of minutes. That's the most powerful part of what we're doing. And ultimately, what you're describing that study found is that the humans were slowing the AI down and potentially introducing their bias. So if you do it on a per-doctor basis, it's going to be wrong. But if you can aggregate across thousands of clinical encounters for similar patients, you can start to paint a pretty good picture of what the doctors collectively are saying about something that typically happens. Then you can actually improve the accuracy of your models, compared to a single point in a single interaction. For sure, the doctor could just be tired that day, or they could be biased. And they might be wrong. But we believe the more data we have, the more accurate we can be in pretty much every respect.

I have two more questions. The first one is, what is a doctor doing five years from now?

Five years from now? Yeah, I always use the Waymo analogy here. Today, Lotus at least is building an Uber kind of model, where we have our own patients and our own doctors. The doctors are the drivers, and the patients are the passengers. In the future, and we're already getting ready for this and we have the data on it, we'll be building a Waymo, where the patients will have agency over their health and be directing the car. But the doctors are that support person. When you don't put your seatbelt on, the support person comes on, stops the car, and says, "Hey, please put your seatbelt on." So I think the doctors will be there to make sure the car's not going off the rails, and also, when the passengers, or the patients in our case, don't trust the AI, they will be there to help. I think that might actually happen a lot more than we think. There are going to be a lot of scenarios where the human just wants to talk to a human. Look at call centers today. When you call customer support, a lot of times the automated system is actually more efficient than talking to a human. But most of the time, people are trying to skip it as fast as possible: "Just connect me to a human. I don't want to..."

I always skip that shit.

Yeah. Everyone does, right? And that's my point. Even though it's been proven that the system is more efficient, because it's going to ask you the appropriate questions and get you to the solution, while a human might be wrong, or biased, or have a language barrier and not understand what the hell you're saying, people still choose to talk to a human. So my point is, I think doctors will become that: when a human needs to talk to a human. And in healthcare, that happens a lot. Especially with mental health issues, there are a lot of scenarios where a human in the loop is necessary, and I don't think that will go away. So five years from now, maybe even ten years from now, physicians will still be required in that loop, but in those special situations. Who knows what's going to happen 20 years from now, right? Some people believe that eventually no one will have jobs, and we'll all be sitting on a beach writing poems while AI does everything for us. But I don't think that future is that close yet in healthcare, or at least in the practice of medicine.

Given that Anthropic is here and OpenAI is down the street, they've both invested heavily in healthcare. What's your take? What are they going to do, and will they manage it?

I always think it's funny that people classify it that way. Healthcare is almost a quarter of the US economy now. So investing in healthcare is no longer a distinct decision. In order to operate as a company, you have to think about healthcare because it's such a big part of GDP. You could argue Apple has invested in healthcare, and Google has invested in healthcare for a really long time. But healthcare is still a massive market, and there are hundreds of billions of dollars in market value created by venture-backed companies building in healthcare. Generally speaking, when it comes to general intelligence, I think it's a good thing that these companies are taking healthcare seriously, because they're starting to evaluate their models against important real-world situations and bringing in physicians to improve the clinical accuracy of the models. So net-net, it's really good for society. I don't think OpenAI or Anthropic, and we're pretty close to them, I was just talking to OpenAI today, are going to hire doctors to give people medical treatment. It just doesn't make sense for them. Their mission is AGI. If they get busy with the mess of the healthcare industry from a medical practice perspective, they risk missing the target on AGI as a company. When Sam Altman wakes up in the morning, he has a lot on his mind, whether it's the government doing something to him, Anthropic releasing a model, or Gemini beating them on monthly active users. The last thing he wants to worry about is running a medical practice, giving people care, and being liable for it. So in what we're doing, we work with the frontier labs to improve our intelligence stack. But effectively, like I said, we're kind of like an Uber: we're taking the average doctor and upskilling them to be the best doctor. And we're bringing care closer to the home. We're giving people the ability to get world-class care at scale, really fast, from the comfort of their homes, and allowing doctors to practice from the comfort of their homes too. So we're building a parallel healthcare system outside the traditional one. The way we look at it, the big labs doing stuff in healthcare is net-net good for us. Because I'm pretty sure hospitals doing stuff in healthcare isn't necessarily going to add value for us. It's just them improving their margins, right?

Cool. It's been a good chat, man.

Cool, man. Thanks for having me.