Zocdoc CEO Predicts: Dr. Google to be Replaced by Dr. AI 🤖🩺

Today’s Decoder episode is a special one: I’m talking to Zocdoc CEO Oliver Kharraz, and we chatted live onstage at the TechFutures conference here in New York City.

You’re almost certainly familiar with Zocdoc — it’s a platform that helps people find and book appointments with doctors. It’s a classic of the early app economy, right alongside Uber, Airbnb, DoorDash, and others — it’s a friendly mobile app that efficiently matches supply and demand in a way that ultimately reshapes the market.

The big difference is that Zocdoc plugs into the United States healthcare system, which is a huge mess. And that means Zocdoc has a pretty big moat — it’s hard to make a database of all the doctors, and all the insurances they take, and understand healthcare privacy laws, and get a bunch of verified reviews from patients that comply with those laws, and on and on.

So, Zocdoc has a very different relationship to big platforms like Google and new AI tools like ChatGPT, which promise to just take commands and do things like book doctor appointments for you. They all sort of need Zocdoc’s infrastructure to run in the background, and you’ll hear Oliver talk about that pretty directly here. It’s a very different relationship than the one between AI companies and DoorDash, Airbnb, TaskRabbit, and others that we’ve talked about here on Decoder in the past.

You’ll also hear us go back and forth here on the shift from “Dr. Google” to “Dr. ChatGPT” — my entire family is full of doctors, and they tell me that people are increasingly asking AI chatbots for medical advice that runs the range from really useful to outright dangerous. You’ll hear Oliver say Zocdoc will use AI for mundane takes — the company has an assistant called Zo that can help with booking — but he’s drawn a hard line at giving medical advice. There’s a lot in this conversation, and Oliver is very direct. I really enjoyed it.

Just a quick note before we start: the TechFutures stage was on a beautiful rooftop in downtown Manhattan overlooking the Brooklyn Bridge, so while we certainly felt charmed sitting there and talking, you might pick up on a little wind noise and even the occasional helicopter. After all, it’s a live production.

Okay, Zocdoc CEO Oliver Kharraz — here we go.

This interview has been lightly edited for length and clarity.

Oliver Kharraz, you are the cofounder and CEO of Zocdoc. Welcome to Decoder.

I am very excited to talk to you. There’s a lot going on in how apps are built, how people experience services on devices, in healthcare in America. AI is tied up in a lot of that. I think there’s a lot of that to unpack with you that I’m excited to get into.

But let’s start at the beginning. I think people understand one version of what Zocdoc is. You need a doctor; if you open this app, maybe you’ll find one. But it’s a lot more than that now. Explain what you think Zocdoc is.

Zocdoc is really a platform that connects patients and doctors wherever they are. Obviously, as you point out, the marketplace and the app are really well-known, where people can just do that self-directed. But we are making sure that wherever you are as a patient, you can get access to care.

We have a partnership with some health insurance companies, like Blue Shield of California, for example. When you go to their website, you can get access to care. We help veterans get care. We have other services that are very annoying, like the phone, which seems weird for us to do, given that we started out to eliminate the phone from the healthcare process. But we’ve recently released a product that allows you to call your doctor and schedule an appointment with an AI agent completely autonomously. Our current trajectory is really about how we make getting access to care easy for any patient anywhere.

So Zocdoc was founded, I would say, in the era of smartphone apps: “we’re going to move everything into a screen on a phone and we’re going to have marketplaces, especially these two-sided marketplaces.” So, Uber for doctors.

There was a way of talking about apps and services at that time, which I think was very powerful and led to a lot of investment and to a lot of great companies. That’s changing now. Do you still think of yourself in that model? Or do you think Zocdoc is going to have to be something else in the future?

I think we’re definitely an app model, and we have figured out how to do access to care better than anyone else in the US. When you pick up the phone and you start dialing for doctors, it takes you, on average, 30 days till you can actually see one. Zocdoc, the plurality of all appointments happened within 24 hours. Nearly all of them happened within 72 hours. So that’s an experience that’s an order of magnitude better than what you get through the phone and the old modalities.

But we’re not trying to take the platform captive. We are opening it up for others as well, some of the health insurance players that I mentioned before, but we are generally thinking of ourselves as something that can be useful in meeting patients where they are and allowing them to see their doctor.

That expansion into telehealth is not just “I’m just going to book a doctor appointment and go to an office.” If someone books a doctor appointment, the doctor will show up here. There’s a lot of competition in that space. Zoom just sort of accidentally started a telehealth business in the pandemic, just by nature of existing. Other providers, insurance companies, want to be in that business. Is that a future growth area for you? Or is that just a continuation of the services you have now?

We offer telehealth, but if we’re being totally honest, and this was visible early on, patients just don’t really want it. We offer telehealth options, and we offer in-person options. For everything except mental health, about 95 percent of all appointments are in-person. Here’s the interesting thing: even doctors who offer both telehealth and in-person visits get more bookings than doctors who only offer one or the other.

But the bookings are all for the in-person visits, so the patient really only values the option of, “Okay, maybe in the future I want to see that doctor in a telehealth visit, but right now I have a body. They want to look at my mouth, they want to listen to my heart, they want to poke my abdomen.” One of the things about somatic medicine is that telehealth is a little bit like telepizza. It’s great, except you can only eat the pizza when you’re in the same room with it.

Now, mental health is very different. In mental health, the picture is exactly reversed. Nearly all of it is happening remotely, and it just has tremendous advantages for both parties to do that. So I think it’s a very nuanced picture, and one blanket statement isn’t going to do it complete justice. We offer that as we offer all other modalities. We offer urgent care and primary care, and 250 specialties, all the way to cardiac surgeons and oncologists. So you can find really any type of care on Zocdoc.

I think one of the interesting things about Zoom, for example, or other telehealth services, is the notion that you will end up speaking to an AI. I interviewed the CEO of Zoom, one of the strangest episodes of Decoder in history, and he said that the future of Zoom is that he will make an avatar of you, and then your robot avatar will go to your Zoom meetings for you, and you will go to the beach instead. And I said to him, “At the end of this, all the avatars will be having meetings, and I don’t know what we’ll be doing.” And he said, “That’s interesting.”

That might be fine for a number of corporations. It’s very different for a doctor or a healthcare organization, where you’ve outsourced the decision-making process or the patient relationship to an AI, or an agent, or an avatar. It feels dicey. It also feels like something consumers will increasingly demand. How do you think about that for your platform?

Yeah, so I have some skepticism about that future, mostly because I do think there will be more self-medication. Dr. Google is going to be replaced by Dr. AI, and the patient will develop their own judgment where they think that an AI is good enough to give them guidance, and where they actually want human judgment. I think it would be maybe misleading to blur the line and say, “Oh, you’re talking to an AI, but I make it look like you’re speaking to a human,” because the patient’s self-selected into, “I want human eyes on that because I think the potential for an error is too great and the change in outcome is too significant.” So this is where I think we just need to be honest with ourselves — not everything that is possible is actually useful.

So you have an AI part of the platform now called Zo. It’s an assistant. As you said, it helps with scheduling and customer service. That’s expressed, you described it, as on the phone. You can call and talk to a voice; it will talk back to you. Do you feel the same tensions there that people have self-selected into an AI, or are they just calling the phone and getting it?

Yeah, obviously, they know it’s an AI, and they can opt out of that experience. Frequently playing Tetris on the phone with another human isn’t actually that fun, particularly when you have to wait 20 minutes to actually talk to that person, and people are okay with that. But one of the big misunderstandings about how AI solutions work is that “Oh, we’re just automating the work of the receptionist or the call center agent.” I think if you aim for that, you’re aiming too low as an AI enablement company. Because what you need to think about is, “Hey, now that I have this AI and I have essentially unlimited bandwidth, how would I design this job from scratch?”

So, for example, for us, it’s not “Okay, how does our AI compare to human agents?” But it’s actually measuring the effectiveness of all the human agents, knowing the effectiveness of the AI for every type of patient, and then connecting the patient to the right resource. If you call in for a routine thing, you just want to confirm the office location or you want to reschedule an appointment you’ve already made, well, do that with an AI because it’s so straightforward. You’ll get faster service, and it will be super friendly.

But if you have a complex question, well, let’s connect you to the human who is best informed about that in the practice. And the AI can know that, and it can dynamically triage these patients to come in and give you a much better experience than you had before. So you should really rethink your call center, not as how do I reduce my expenses in a cost center, but how do I actually turn this into a profit center where I now lose fewer patients and have less leakage on the front-end, and make sure that patients have a great experience when they call me?

Let me push on this a little bit. So, the idea that I need to reschedule an appointment, I feel like that has been conclusively solved by smartphones. I don’t necessarily need to talk to a robot. I actually want to use the visual interface of my smartphone and hit the button. And maybe I’m actually taking the action, and maybe I’m just sending a note to another back office, or whatever it is.

But it feels like I’m actually doing it, and that problem feels solved. But “I have a complex medical question and I need to dive through a series of screening questions to find the right provider and schedule that” — that does feel like a natural language processing task that AI might be good for. But then that’s also a little bit diagnostic. It’s a little bit that you need some insight there. How much insight are you willing to let your AI have in that process?

So it’s actually very interesting, because what you say makes absolute sense, minus the fact that as a patient, your experience is actually that you have hundreds of different logins to all these different doctor systems. Obviously, I hope everyone uses Zocdoc so that you have only one login. But in reality, some patients still use the phone to make an appointment, and they don’t think about the app as an alternative. So you’d be surprised what percentage of calls that come in are actually simple things like scheduling that clog up the pipes for the patients that are coming in and calling about complex issues. So there is probably a transitory period until everyone uses Zocdoc, where these reschedules still happen over the phone.

But then, in terms of the insight, what we see is actually that humans don’t perform equally on all complex issues either. We can measure the successful conversion rate for a call that comes in, to the average human, to Zo, to other AI solutions, and to the best humans. And when you look at this — and there’s been an independent study that has been done on that recently — but they found Zocdoc, among the AI solutions, is actually the best. It has a conversion rate of roughly 52 percent, where everyone else was below 40 percent. The average human, typically, is in the high 40s, so comparable to the AI.

The best humans are 65 percent, so they are dramatically better. But are they at 65 percent for everything, and should you use them for everything? No, you should make sure that whatever they are doing, you teach all the other people who are answering your phone, so you up-level in general. But then also, you want to make sure that you route the patient that actually has this problem that this call center rep is an expert in, that patient and that expert need to talk to each other, not some other random person on either end of that.

To ask that question in a slightly different way, that feels like it requires some expertise, some insight into what the patient is saying, into what services are available. There has to be a limit on how much thinking you want the AI to do, how much judgment you want the AI to do. That feels like the problem writ large for our industry. Where are we going to stop the AI and say it’s time to talk to a person?

Well, the AI needs to be self-conscious in that way, and that’s why you can’t just leave it to the AI. I think anyone who uses LLMs finds that they are too confident when they shouldn’t be, and they’re not curious enough when more questions would actually be adequate to get to the correct solution. So, we have solved this in a completely different way, where we have a deterministic orchestration layer that then uses LLMs selectively to make sure we parse the answers from the patient correctly.

But we have a master plan, and we know when a conversation goes outside the bounds of the master plan and should be transferred over to a human, and therefore, we can take accountability for that. This is very different from just dumping everything in the context window of an LLM and praying for the best.

Okay, I want you to hold onto that, and I’ll come back to it because I think the entire industry is restructuring itself around that problem, and that’s one very important solution. But I do want to ask the Decoder questions and understand Zocdoc as a company. How is Zocdoc structured right now? How many employees do you have, and how are they organized?

We’re a little bit over 1,000 employees, and we are still functionally structured. We have a head of sales, a head of marketing, a head of government relations, and what have you. And the reason why that works for a company of our size and why I think it’s going to work is because of our quite unique history.

We didn’t have a straight lineup. We’ve been around for a long time. We went through a major business model transition, a turnaround you could call it, and it has created a kind of cohesion that a one Zocdoc philosophy still works. Everyone in leadership is oriented toward the same number, and it’s a number for Zocdoc in its totality, and this is why we can bring functional teams together, and we don’t get the typical corporate politics that make this not work.

What’s the number? When you say there’s one number to go for, what’s the number?

It’s a revenue number, it’s a profitability number, and we fuse that together into one score.

The business model change you’re talking about was that you went from flat fees for doctors to per-patient referrals. You’ve given a lot of interviews about how that unlocked growth, and now you’re profitable. The doctors didn’t love it. And the idea that you are now the market maker for doctors, some of them have decided to find their own customers. Doctors being on Instagram to find their own customers is a whole situation over there. Is that putting pressure on your model?

No. So obviously, some doctors didn’t like it, and some doctors liked it a lot. The interesting thing about marketplaces in general is that the utilization follows a power curve. As you may imagine, if you have one flat fee, the people who are on the top end of the power curve are getting value for free. Obviously, the people who are on the low end of that distribution don’t get enough value.

So everyone who was to the left of that distribution of our new price loves this model. And a lot more, like orders of magnitude more, doctors are on Zocdoc today than when we started that. Obviously, some doctors had to pay more. If you were getting 10,000 patients from us a year and we had a $3,000 fee, on a cents-per-patient basis, there’s no way you’re getting that anywhere, including on Instagram. But also, obviously, now that we ask you to pay a fee per patient, it’s going to be a lot more. So clearly, there was some adjustment.

What is super interesting is that despite the fact that we had to have conversations like, “Oh, your price is going up 100x,” which, if you ever had the conversation like that, it’s not fun. But all of these doctors, all the big spenders, actually came back to Zocdoc, except for one. And they came back and said, “The quality of the patients I’m getting, the volume I’m getting, the predictability for my business, is such that there is just no alternative.”

So when you think about that patient matching, again, I look broadly at the industry and I think, “Okay, well, Meta’s thesis is that AI will help us target ads better. Google’s thesis, they’re less loud about it, but their thesis is that the AI will help them target ads better.” That’s fundamentally what you’re doing: you’re matching customers and providers in a real way. Are you employing AI there as well?

Yes. For the matching process, absolutely, yes, we do that.

What are the parameters there?

We understand a lot about the patients, and obviously, they also answer questions for us. And we understand a lot about the doctors. There are, in some ways, layers of information that are not broadly documented. Really, these are things that we know between the doctors and Zocdoc, between the patients and Zocdoc, and that’s the information we can use to make that match as efficiently as possible.

There’s a lot of public information that you also need to take into account for that. Which doctor accepts your insurance card? Which doctor actually accepts new patients? What type of patients does this doctor see? How long does a doctor typically take for a patient with your chief complaint? Do they see them in the morning? Do they see them in the afternoon? How many of those can they see consecutively?

These are all meta information that we have about the doctor, and we have the direct connection to their schedules to see, “Okay, given that those are all the rules, which slots are even potentially available for you?” And then obviously there are clinical fit questions, which we tackle and actually is, I think, a very, very interesting area of growth for us.

The reason I ask these questions this way is because that’s the heart of Zocdoc, right? Every one of these referrals, now that you’ve made the business model change, is revenue for you. And especially if the patient shows up, everyone’s very happy. You have to make an investment in making that matching process better, and the investment here is an investment into AI, which is in its early stages.

We were talking before about the return on these investments being somewhat unknown. How did you decide, “Okay, I’m going to make the forward investment to put AI into our functional teams on the thesis that the matches will become correct, that the doctors will be happier, and the patients will be happier?”

Yeah, so first of all, we are not making referrals; the patients are using us to book with their doctors. But within the scope of that, from day one, the challenge was about how we make this match better. For anyone who is doing business in the actual physical world, understanding all the outliers and all the ways in which this can be off are critical pieces. Because if you apply the 80/20 rule, you’re going to piss off 20 percent of your customers, and you cannot do that very often. So you constantly need to zoom in and say, “Okay, great, what are the remaining edge cases where this doesn’t truly work?”

This is a problem that’s a little bit like the coastline of England. If you look at it from a map, it seems like, oh, I can just trace this and I can measure that. But as you zoom in and you say, “Oh, but here’s a little bay, like it’s really going in there. And in the bay is a rock, and so there’s another surface. And in the rock, there’s a crack, and then I go into the crack, and there are microcracks.” And the smaller you go in and measure, the more you realize, “Oh God, I will never be done with that. There’s just too much to do.” Now, AI is great because it can accelerate the kinds of problems that we can solve to make this an even more seamless experience for the patient and for the doctor.

But you had to make an investment, right? You have a functional team. You’re building one product together against one number to say, “Okay, we’re going to make this investment into AI.” Presumably, you had some goals here. I know you’re not calling them referrals, but the goal was for more patients to book with more doctors. How did you decide that it was worth it?

We had a team on that since day one, except that obviously, back in 2007, they were not using AI, but we were using machine learning and other techniques to improve the quality of the match. We have a belief, actually, that the quality of the match is a huge determinant. We are not trying to optimize the number of bookings in any given moment; we’re trying to optimize the experience that the user has because we believe that’s a determinant of where they come back and use us again. Do they have a preference for Zocdoc, because that’s the tool that just works?

Have you seen it pay off? Have you seen the return on the investment?

18 years later, we’re still here.

[Laughs] Well, on AI specifically. On Zocdoc, yes, but on AI specifically?

Yes, absolutely. I think there too, we’re thinking about ways to use AI to not just make what we have already been doing or what has already been done more efficient, but what new things are now possible because AI exists that were just not possible before. And so there are interesting things coming out in the future, and I’m happy to chat when we’re ready to announce them.

Let me ask you the other Decoder question, and I want to ask you about some of these interesting things. How do you make decisions? What’s your framework?

I am not in founder mode, if that’s the question. I actually think I only make three types of decisions. The first one is, who are the people that I trust and I bring on the bus? So what’s the senior leadership team, and who do I think can actually help us get to that next milestone? Once I have these people in place, if I choose them well, they should know their area better than I ever could. If I hire an enterprise sales executive, and I have to teach them how to do their job, I have mishired. So this needs to be on autopilot, and the only way that can happen is if I don’t get into their hair.

The second type of decision is where risk is involved. I think organizations tend to drive people to not take enough risk, and that is something that, as a founder, you’re uniquely positioned to say, “You know what? I’m going to absorb all the blame if this doesn’t go right. You could say I instructed you to do that. And if it does go right, it’s all yours. You came up with it, go forward.” So when I see that there are areas where we should be taking a risk, I get involved and I make sure that everyone knows that there is an absolute license to take the risk if it’s a smart one. We are not trying to jump off buildings, but there’s a lot of opportunity there.

The third type of decision is when it comes to where the puck’s going. This is a thing where you need to integrate a lot of different inputs, so there’s obviously what’s technically feasible. I also talk a lot to our customers. I understand how they’re thinking about the world where they sort of have pebbles in their shoe. And then I spent a lot of time in Washington, DC, to understand, “Okay, what does the regulator want?” And then you need to triangulate all these things and say, “Okay, great, given that, what do we need to do? What new capabilities do we need to bring in-house to be able to manage that next challenge?” I’m a believer that companies can evolve and develop new capabilities. I don’t think core capabilities are boxing you in in any way, but you need to know what you want and what you need; otherwise, you can’t build it with confidence.

Let me put some stress on where the puck is going. So Zocdoc is a service provider, again, of a generation of apps where consumers open the phone, and they take some control of what you might think of as back-office functions. I’m going to book a car, and I’m going to find a doctor. Those service providers all expanded in different ways, vertically and horizontally. You have businesses.

Yesterday, OpenAI had DevDay. Anthropic was just on stage to introduce [Model Context Protocol]. The idea that the AIs are going to disintermediate service providers feels very real. I call this the DoorDash problem. If I say, “Alexa, order me a sandwich,” and it goes and clicks around on the DoorDash website, and the sandwich shows up, DoorDash might be out of business.

Because all of the revenue that’s associated with me actually using DoorDash will go away, and they will become a commodity of sandwiches, which is not a great business to be in. That might happen to you. I might say, “Alexa, find me a doctor,” and it might traverse the Zocdoc back-end and take you out of it, and all these new capabilities you want to build might be disintermediated. Are you thinking about that? Are you thinking that you want to integrate with these new kinds of agents, or are you going to try to build them yourself?

We’ll integrate with these agents, and the reason is that I think that fear, the DoorDash fear, might be slightly flawed thinking. Here’s why I think that. Here are the questions you should ask yourself. Question number one: Are these agents simply going to completely displace you? Anyone who’s running a business that interacts with the real world knows that that’s not going to be the case, because of that learning curve, because of all the edge cases, and all these things. Even if the AIs were to start learning about them, we’re so much further ahead that we can always deliver a better experience. So this is the coast of England problem. Our cartographers have been at this for 20 years; there’s no way that anyone would catch up to us anytime soon. So they’re not going to put us out of business.

Now, the second question: Are they going to drain the profit pools for these things? You could say, “Well, there’s a world where you could imagine this happening, where consumers pay a subscription fee to people who built these agents, and then the agents find the optimal price for you.” That flies in the face of the entire monetization model of the internet. If you look at it, everything has been monetized through advertising, and so you’d have to believe that there’s going to be an anthropological change where people suddenly say, “Yeah, I’m actually happy to pay upfront and then maybe collect rewards over time where this is potentially giving me better deals.” But if that were true, everyone would be eating healthy, working out, taking all preventative tests, etc. So I just think that that is not how humans actually work.

So, the third thing is, okay fine, the profit pools will not be completely drained, but are they going to take most of my profits away? I think we are all anchored in these last 20-plus years where Google was a monopolist and could ask for these tolls. I think the tables have actually turned very much. There are five major LLMs or AI companies that are competing to be your agent. Imagine you had the one that doesn’t let you order a sandwich, that doesn’t let you book an Airbnb, that doesn’t let you call an Uber, that doesn’t let you book a doctor. Would you use that one? No. And so the providers of these services actually have a lot of leverage right now to negotiate the kinds of relationships with these AI agents that they never had with Google, because Google was already the monopolist when they came up.

Well, okay, there’s a lot in that answer, but I actually want to focus on that last piece, about where the leverage comes from, for one second. I think there’s a lot of leverage if everyone agrees that MCP is the way this is going to work. And then you can say, “My MCP server is open to Amazon and Google, but closed to Microsoft,” or however this plays out. And then now we’re just negotiating. We’re just negotiating API access with a different set of vocabulary.

I look at some of these companies, and they say, “Well, screw it. We’re just going to go click around on your website. We’re just going to open a browser, and we’re going to click the buttons for the user, and we’ll do that in the background.” And you might never know. You might never know that this happened. Perplexity is going to do this with its browser. Knowing Perplexity, that is probably how its agent will work. That destroys your leverage. You have to detect their agent and say, “You can’t do automated browsing.” And there’s no framework. There’s no negotiation framework for that.

While they do that, they’re not making any money, and I make money as I used to. So that’s actually cool. Give me free traffic.

But you don’t get your advertising money.

Well, how do you know? Because I might know which agent is coming to my website.

[Laughs] I agree that internet advertising is rife with automated fraud. That’s not the right answer.

Let’s look at Uber. Uber is making money from the drivers. That wasn’t the model. Uber would be getting all that free traffic from Perplexity. I’m sure they love that, and I’m sure Airbnb would, too. If you book through Perplexity and no money flows to Perplexity, I’m sure Airbnb would love that. Oh, you order through my DoorDash app, and I don’t have to pay you for traffic? Great. Why wouldn’t people want that?

This is the other outcome. There’s “let’s negotiate MCP access on the front-end and have revenue share,” and then there’s the bet that automated browsing will bring so much traffic or money, and there won’t be negotiations, but it’ll all work out. That’s the split I see right now. There’s more heat in browser coverage as a tech journalist than there’s been in over a decade, because people want to build new kinds of browsers that take action for the user. And then there’s a lot of heat on MCP.

Yeah, but if you look at the companies that create the most value, they’re not trying to do this through pure advertising. Obviously, advertising is a part of everyone’s revenue, but they are taking transaction fees. If you order that sandwich, you pay a service fee to DoorDash. When you book this Airbnb, they’re taking a cut of the booking fee from you. But yeah, use the website. That is a totally fine mechanism. Airbnb doesn’t even have advertising, but if less money comes in through advertising, you will take that right back in other ways.

So I don’t think there’s really a threat there. And if they are going to negotiate, if they do want to have some of that money, I think these companies that are the Ubers, the Airbnbs, the DoorDash of this world, are in a unique position to dictate their terms in a way that they could never do with Google.

Well, Google’s a really interesting case, and Google also owns a browser. It seems like Chrome is going to be automated in a lot of ways. Google is also the search engine of record. Do you feel yourself in a position to negotiate with Google differently than every other kind of vertical search engine has in the past, right now?

Look, I think we are always looking to help patients wherever they are in whatever way they want to interact with us. We even work with health insurance companies where Zocdoc is completely hidden. You log in with your health insurance company login, and you see the doctors that are in-network with your health insurance. You book one. You use the Zocdoc pipes, but as the patient, as the member of that insurance company, you don’t need to go to-

Let me ask this slightly differently. If you went to Google and said, “Look, people are going to talk to Gemini instead of the Google Search box. When they look for a doctor, just have Gemini use our pipes and pay us for it,” a year or two years ago, the door wouldn’t have even been opened. You would’ve just been at the door of Mountain View, saying, “Use our pipes, pay us money,” and they would’ve not paid any attention to you. Do you have the leverage to open that door today?

I think these doors are more open than ever. That’s exactly right. And I think as Gemini is trying to be your AI agent — and ChatGPT, Grok, Perplexity, and Claude to some degree — well, do you want to be the chat agent that uniquely doesn’t have the capability to use Uber’s pipes, or DoorDash’s or Zocdoc’s pipes? That would put you at a competitive disadvantage, and I think that is a reality that all these companies have to grapple with, no one more than Google, which has historically enjoyed this monopoly.

Who is Zocdoc’s biggest competitor?

So there’s obviously still a lot of inertia–

No, no, when you’re like, “We got to beat those guys,” who is it?

In terms of our core marketplace, it is such a difficult business that competitive waves have come and gone. Right now, there aren’t necessarily-

But this is why you’re special, right? I asked that for a reason. If Google, ChatGPT, or Perplexity wants to get a doctor for you, they have to come talk to you. In a very direct way, you are the database of record for that thing.

If you’re DoorDash, well, Uber Eats exists. There are many other ways to do this. I’m wondering if you see the opportunity for one of your tangential or orthogonal competitors to say, “Actually, we have a database of doctors too. We just never built the front-end to let patients book directly, but your agent can come use our database and do it for them.” And now this is a new kind of threat for you.

I think, again, the cartography problem, the coast of England problem, is the reason why there are no other ships sailing in our direction, because you need to be very patient. Literally, we did not leave New York for four years just to make sure that we got to a base level of this functioning, because there is the technology problem of integrating with all these [electronic health record] systems.

But then there’s an anthropology problem on top of this: how do these practice managers and front office folks, how do they actually use these EHRs? What’s the hidden information that you cannot extract from electronic systems? We’ve gone through all of that, and we have learned it the hard way over many years, and we’ve continued to learn it for two decades. So could you start a Zocdoc competitor today? Of course, you could. Would it be a dramatically worse experience than using Zocdoc? Yeah, it would be. So this is why I think that these AI agents will want to work with someone like us who can deliver a great experience for their users.

I would say at least in the case of OpenAI, what ChatGPT has proven is like, “Oh, we’ll take anything. This robot will tell me I’m in love with it, and that might be better than a real relationship.” That kind of disruption is real here. It will do the job slightly worse, but it’s doing the job in this interface, and that’s the kind of disruption I think not just Zocdoc, but also the whole industry is facing.

I think that is going to be great until you’re trying to catch your flight and the Uber doesn’t show up that you’d gotten through ChatGPT. Or you are hungry, all the restaurants are now closed, and it turns out your DoorDash order didn’t go through. You’re arriving in Miami, and your Airbnb is occupied by someone else. How often can you do that? It’s very different from telling you, “Oh, I love you.” That works, it’s probably true, but even if it wasn’t true, we have fewer expectations about how these communication challenges resolve, versus things that happen in the real world. This is where I think the experience head start that all these operators in the real world have compared to ChatGPT is going to be a sustainable advantage.

I do feel like we should spend the last 20 minutes here talking about the stakes of saying, “I love you,” versus the stakes of booking a flight.

The idea that the stakes of saying I love you are lower than missing a flight, I do feel like we need more than 20 minutes, but that there’s a lot to say about the AI conversation in that idea. There’s one more platform I want to talk about, and then I want to talk about some other things, specifically about healthcare.

Apple announced Siri with App Intents, which was going to be this high-powered assistant. I think a lot of people assumed that they would have a huge head start because all the apps are already on the phone. There are already some hooks for automating apps on the phone in various ways. That seemed like a bit of a false start.

Apple recently made some noises about MCP, which is kind of wild for Apple, as the owner of iOS, to say that MCP might be the way they go. Would you allow Siri on the phone to use your app in an automated way?

Because that also seems like a disintermediation.

For the same reason that I allow agents at the Veterans Administration or care coordinators at Blue Shield of California to use the app in an unbranded way, I would absolutely allow Siri to do that.

Would you expect it to actually open your app and click around, or would you just expose the database and the service of your app to Siri?

We’d obviously have to explore what consumers really want, but I’m very open to finding a path that is optimal for the patient. That’s why we ultimately exist. And that’s a completely orthogonal topic to what the relationship between Siri and Zocdoc is going to be.

App developers have had a, I would say, bumpy relationship with Apple over the past few years. In the same way you’re describing the doors are open at Google, do you feel like the doors are open to have different kinds of relationships with Apple now?

We are really into win-wins, and that’s why we’ve always had great relationships with everyone. I can’t remember being at war with any of those guys. And we were very focused on the things that we really want to do and want to do really, really well, and sometimes that overlaps with what someone else wants. And then you can say, “I love you,” and sometimes it doesn’t, and then we both stay friends and go our own ways. I think that those conversations will be ongoing, and I think it’s a very quickly evolving space where even folks like Apple will have to rethink how they are approaching the optimal solution for their users.

Are you making the same bet on MCP as everyone else, or are you more agnostic about how these agents will work?

Look, I think you should just try out a bunch of things. It’s not well-known at this point how these agents will be structured in a way that really gives the patient confidence, or the user confidence, rather, and leads to using the tools correctly. Now, I will say that sometimes complex information, we’ve played around with it, and sometimes you want visual feedback because you can just convey a lot more of it in one glance than talking you through all your options, etc.

So I think it’s going to be evolving paradigms for simple things where I can just tell you, “Hey, order me toothpaste” versus, “Oh, give me my options to do X, Y, Z, and now the options need to be arranged in a way that I can take that information in quickly,” because the narrative of it will be maybe too much for me. And so I think this will evolve, but we are there for it, and we are happy to partner with anyone who’s interested in making this better.

One of the reasons I wanted to ask you that specifically is that the criticism of MCP is that it has an enormous number of security issues with it. It’s going to expose a lot of data. You have just API access to databases in non-deterministic ways. You don’t really know how both sides of the transaction will work. In healthcare, you have an obligation to the patient, to the government, and to the provider to keep so much information private. Do you think MCP is compatible with your business?

Look, I won’t opine on the technical constraints that you have to put in. All I can say is that we use AI in some arenas where it’s critical that you get to the right results, and that what you do is unit testable. And we have managed to put frameworks in place that give us complete confidence that we’re not hallucinating, that we’re not going out of bounds of what is allowable. And this is that hybrid framework between deterministic parts of the application and LLM-based ones. And I think we’ll have to figure out how that actually works in the future, to make sure that we continue to put that safety and the safety of the data first, and we don’t create unforeseen results for the end users. But I just take this as a given, and I think that’s something that we can invent around, and we still come to good results.

Well, see, you mentioned this hybrid approach to development at the topic of conversation, and I want to spend a minute on it here. The bet, all the money in AI, is that the AI will eat everything. This is the way computers are going to work. This is the way we’re going to write applications. This is the way that programs will talk to each other. This is the way that services interact. And all this will happen in the context of AI, specifically LLMs and MCP, and that’s the future of everything. That’s a bet that is supporting a lot of investment right now, that everything will eventually operate in this framework.

You are describing a very different framework. You’re saying, “I need to surround these models with traditional deterministic algorithms and systems that guarantee the results I need, and this is actually the future for our business.” That’s not the prevailing bet; that’s not how the investment will pay off for all the massive investment. But having talked to you about it, you seem very confident in that way of working. Do you think there’s a path for the AI systems as they’re being built now to actually do the job as well as the hybrid model that you’re describing?

Not today. Not today. And is there a path for it to get there over time? People smarter than me are investing hundreds of billions of dollars into that.

Are they smarter than you?

For sure. That is the one sure thing to say. But they’re investing a lot of money in that, and I think there is probably a belief that would justify that money that we can get to AGI, and maybe that will happen tomorrow. I think as an observer of the scene, I would say that’s probably less likely. We just had the release of Sora. If you were expecting AGI in the near term, would you really invest in a video editing tool? No, you’d be working towards AGI. So I think we’re probably many, many years away from reaching this point in actuality, which gives us enough time to learn which elements of that are useful in which situation.

In life, the answer is nearly always, “It depends.” And for some tasks, obviously, the LLMs as they come out of the box today are just wonderful. For some tasks, you can’t trust them enough, and you need to put them into an orchestration layer, and I think we’ll see how that evolves. But I cannot imagine a world where everything is one thing, because as we talked about earlier, we’re still making [Intel’s] 8086 chips, and they were in when I was a kid 40 years ago.

Now the United States government is in the business of making 8086 chips, which is a real mind-bender. Let’s actually go there. To wrap it up, healthcare is a deeply regulated space. Healthcare in America is under threat. We’re talking in the middle of a government shutdown. That shutdown hinges on the future of the Affordable Care Act, for example, and how those payments might work.

Zocdoc exists because people have to go to the doctor, and in many cases, because they have an insurance provider, and that first filter is just finding a doctor who’ll take my insurance. Obviously, the market is under enormous amounts of pressure and stress right now. What are you seeing as the maker of the market in response to that?

Yeah, so the secret behind Zocdoc, the contrarian insight, is actually that doctors are not as busy as it seems. Doctors have roughly 30 percent spare capacity that comes from last-minute cancellations, no-shows, and rescheduling. As doctors are put under pressure because of the current budget disputes and reallocation of funds, it becomes more and more pressing for them to actually utilize the last 30 percent.

So they can tend to use Zocdoc more than they maybe did before. Obviously, we are in the business of helping patients and doctors connect, and so we’re happy to fill in the bridge here for the doctors and make sure they stay viable businesses. Broadly, our ambition is to realize the full potential of our marketplace, which means you can improve access, quality, and cost. We started with the access because it was the most broken thing, and it was also our way to get to enough scale to focus on these other problems in the future. But these are very much near and dear to our hearts, and we want to be a true market maker that helps patients find cost-efficient care of high quality that they can actually use.

So, cost efficiency is the thing that’s under pressure right now. Will the ACA subsidies across the country survive in various ways? Obviously, that’s deeply political, but one potential outcome here is that the subsidies go away and costs skyrocket, and some providers have to go out of business.

Is that something that you’re prepared for, that customers are going to open Zocdoc and look for providers that aren’t there? Or you might have to find cheaper providers for them?

I don’t think it’s going to happen in that way. Simply look back at the times before the ACA was around; there were more uninsured patients, and ultimately, we still treated them. We still treat them, but it was uncompensated care. The doctors made up for that by charging the patients who had commercial insurance more money. And so as we migrated uncompensated care into the ACA, the overall increase in rates may have slowed down a little bit versus what it would’ve done. Hard to say because there’s no counterfactual here, but that is one way to look at it.

We’re not actually lowering the total expenditure of care. The only way you could do that is by saying, “No, not only are we locking people out of Medicaid or the ACA, we’re also preventing them from receiving treatment.” I haven’t really heard anyone say that yet, because that has very dramatic implications on how we understand ourselves as a society that has solidarity with other citizens of this country that are not as fortunate as we are, either from a health perspective or from an affluence perspective. So that’s a completely separate political debate that hasn’t even been had yet.

I would say broadly, a criticism of the entire healthcare system in America, ACA or not, is that it has become commercialized. It is more market-driven than idealistically-driven, as you’re describing. My whole family is doctors. They have a lot of thoughts about this.

But the idea that there’s not actually price transparency in this very commercialized healthcare system, that prices are often locked away or pre-negotiated, and you get a lot of bills, doesn’t make any sense. All that is very true for people. It’s very frustrating. As the market maker, if the system becomes even more commercialized, if we start to move these numbers around because the regulatory framework has changed, would you put price transparency into Zocdoc and say, “This is how much these doctors cost?”

Yeah, so at the right time, the answer is yes. The way that we understand ourselves is actually, in some ways, as a union of all the patients that are using Zocdoc, and we are using their collective purchasing power to start affecting change in the system. We have seen providers being quite responsive. We say, “Oh, patients really would like to see you early in the morning or later in the evening, and they need insight into certain elements of what you’re doing and what you might be charging.”

So this is where the existence of Zocdoc as a marketplace that’s bundling decisions of millions and millions and millions of patients is actually a catalyst to the type of change that we want to see. And I think it’s very different from how the government is trying to effect this change, because we have regulation in place that says that payers and hospitals need to publish their prices. But that regulation is punitive. “If you don’t, I’m going to find you.”

Whenever you do that, you have all the smartest people in these organizations trying to figure out how to obey the letter of the law, but circumvent the spirit. Whereas Zocdoc can actually reward you for the right behavior. “Hey, if you do give the patient more information, well, maybe you’re listed in a more prominent spot on the marketplace.” And therefore, now they have all the smartest people working on, “Well, how can we give Zocdoc the information they need to make this better for the patient?” And so this is, I think, the internal optimist in me, thinking that, yes, we can build a better system. It’s not going to be instantaneous. It’s unfortunately not a fiat by the government, but it is something that we can build from the bottom up.

I like that you described it as a union of consumers. That is just another way of saying you have a lot of demand, and you can apply it to the market in focused ways. That said, I would not say most healthcare consumers in America are thrilled. They don’t seem all that happy. No one seems happy with the system as it’s currently designed.

When you think about the leverage Zocdoc has with the aggregate demand that you have on your platform, where are the most effective places for you to apply that pressure to make change, such that people are actually happier?

We are already doing that today. We’re working with the Veterans Administration. It used to be many, many weeks for a veteran to get access to a provider. We have cut that down to just a few days. The same is true with Blue Shield of California, where we have given people access much more quickly to more specific doctors who are better suited for their actual conditions. We are starting to grind away at this. We are firm believers that you can come into healthcare and say, “F the system. We’re tearing it all down and we’re building new.”

There are multi-trillion-dollar worth of deployed assets in healthcare. You have to improve it from the bottom up and work with the institutions that are really doing their best in many ways to try and help patients. But they just don’t have the technology layer necessarily, and they can’t overcome the collective action problem on their own, and they need a facilitator like Zocdoc to get there.

You’re describing the Veterans Administration and the state of California. Those are large government entities, some of the largest that exist. Is the government more responsive to tech solutions lately because of AI? I listened to this administration, and it was basically, “The AI will do it.” The promise of DOGE was, “AI will do everything.” I don’t think that was true. I don’t think that worked out. But there’s a different attitude that I hear from so many people in tech about this administration, their willingness to adopt new tools, or at least their faith that the new tools can lower costs in some way. Has that borne out for you?

Look, as an entrepreneur, I obviously love interacting with optimists, so anyone who thinks that the world can change and can be better, I love dealing with. But as Zocdoc, we have worked with five administrations over the years. We have always had good bipartisan relationships. We are really on the side of the patient more than anyone else, and we’ll work with anyone who is trying to come up with better solutions for Americans.

When you think about the biggest request from that patient base that you have on the platform right now, what’s the number one thing that they want that you can’t quite give them yet?

We are still cartographing. The reality is that healthcare is incredibly complex, so we’ll forever be busy making just the simple things that we do today even better, and making sure that we meet you with more doctors to choose from who are more specialized for what you do. But I think the journey that we’re on right now is to make sure that you don’t have to come to Zocdoc to experience that.

Wherever you are, we will meet you there, and we’ll start making this better for you with the same convenience that you’re experiencing on Zocdoc. And then to the extent that you have to take these steps offline, like calling the doctor’s office, we want to make that experience better for you as well. So we are really trying to be an all-around system for you as the patient, which makes every interaction with the US healthcare system better for you, whether you know that Zocdoc is inside or not.

When you think about that overall experience, I think it’s kind of where we started, and it’s where I want to wrap up. The idea that you could expand into the actual provision of healthcare is right in front of you, where you have a patient, you know their specialists, and you know their doctors. They might tell you some symptoms. You might know who’s available. And then they might ask you for that last twist of advice, “My knee hurts, what can I do for my knee?” And right now, Zocdoc won’t do that, but ChatGPT certainly will. It’ll just give you medical advice. It’ll say it shouldn’t sometimes, but mostly it’ll just do it. Is that a threat, that last turn, or is that something you want to expand into?

I think Dr. Google has been around since before Zocdoc was launched, and there’s obviously going to be some comfort level that patients have to ask ChatGPT or Dr. Google for advice.

Can I just make the distinction a little more sharply? My family hates Dr. Google — again, they’re all doctors — but at least Dr. Google is dropping you on the Cleveland Clinic website, and it’s like, “Here’s some stuff from this reputable organization,” and it’s all bracketed with, “Talk to doctors.” ChatGPT is like, “Here’s some answers. Go get this drug from your doctor.” It’s a very different set of authorities, symbols, and experiences. That’s going to change something. Is that a threat?

I don’t think we’ve really seen the full cycle of that. I think people will do that, and sometimes people will have great experiences, and sometimes they will have not-so-great experiences. And then over time, norms will develop when you actually let ChatGPT stand in for Dr. Google and when you actually want to talk to a human being. I don’t know that we know the surface area right now. And obviously, look, ultimately it’s a free country. We’re all adults. I have my own judgment where I would let LLMs inform me.

I think there are a lot of things that you can get extremely well out of LLMs today, that can help you actually structure your conversation with the doctor in a way that you get everything out of that that you could. So I think there’s definitely a lot of upside. Where the exact boundaries are, I think experience will show. And it’s a little bit like when you go to college, how much should you drink? You’ll figure it out over the course of four years.

Where’s the boundary on Zocdoc today?

We don’t give medical advice.

And that’s going to stay firm until something else changes?

What would make you change it?

We’d really have to define buckets where we know that the LLM or the AI knows what it does know, and it knows when it has a curiosity gap, and the stakes of the advice are low enough. These are two-way doors, okay? Worst case, your headache takes another three hours. Great. Maybe that’s a risk you could take. Whether you should take a medication that has far-reaching and long-term effects, I think I’d be very, very hesitant to do that outside of a human-in-the-loop at this point.

Obviously, you could stipulate, “Okay, AGI is going to solve all of that. I think that’s a totally different discussion altogether when we say, “Okay, humans are going to be broadly obsolete.” I happen to think that will happen in medicine as one of the last passions. Because we have all the physicality of our body that needs to be examined, and we have so many degrees of freedom in how we live our lives that bring surprising twists to the body of knowledge, I think doctors have a pretty safe future.

Yeah, I just think the other side of that is deepfake Sam Altman saying, “Take drugs,” and I don’t know how that’s going to play out.

Last question, and then we’ll wrap it up. It’s an easy one. Do you think this is a bubble?

If I knew that, I could make a lot more money on the stock market than sitting here. I think there’s always a risk. I think it’s a big bet, and as bets go, they can go in two directions. I think this is also one of those that could go in either direction. I think more and more people have questioned more recently whether this is going in the right direction.

I think in either scenario, AI is a useful technology that will endure. Whether we’re paying the right prices for certain assets right now, who am I to judge?

Well, this has been a great conversation. We’ve got to catch up again soon. Thank you for being on Decoder.

_{Questions or comments about this episode? Hit us up at decoder@theverge.com. We really do read every email!}