Hims & Hers CTO: Mo Elshenawy

I'm very excited to have this discussion together today. I've been a long admirer of what you built before and what you're building now. So I'm excited to dive deep into your background and what's next with Hims. But why don't we start at the real beginnings? We spoke a bit about Egypt. Should we go there? How did you grow up until today?

Excited to be here. Thanks for having me. As I said, it's a long journey. It started as a kid. Dating myself, I got an electronic chess board that changed my life. For the listeners who don't know what that is, it's a raw electronic chess board where you click on a piece and move it and it gives you the coordinates. You're basically playing a computer at chess. And this was way before deep networks, so it was basically brute search algorithms, and I was fascinated by it. That drove my curiosity to read a lot about tech. And ironically, it was a time of the second AI winter. So there was a lot of skepticism about what AI will never be able to do. Imagine reading about that in the early 90s or late 80s and then seeing every step of the way, seeing that journey shatter, from Deep Blue to AlphaGo to AlexNet and all the progress that happened. It was quite a long journey, but it started from something as simple as that. That led into bachelor degrees in computer engineering and electrical engineering, and later an MBA. I remained an individual contributor for a long period of my career, very intentionally. I love building, and then I moved to management. I worked in small and big tech, Amazon and later at Cruise, where I was president and CTO and helped deploy the first driverless ride share, the first commercial ride share in San Francisco. And along the journey, an MBA to help me put technology where customer and business needs it and help take things from zero to one. I had some startups on the side, and all that led me to Hims and Hers.

You said you love building, and it goes without question what building is today given all of that AI. How much did it change, maybe over the past five years?

Oh, that's a good question. That's fascinating. As I mentioned, even on the side I've had these startups. I had three startups with successful exits, and the only reason I had them on the side was I couldn't really get away from coding and building myself. Now, as I reflect on this, there were a lot of nights and weekends because I always had them next to a full-time job. Looking back at all these zero to one startups and the effort, lots of long nights and weekends, if I had access to Claude at the time, I think I would have achieved the same results in a much, much shorter time. And the learnings remain the same. There are a lot of things you can't learn just through reading your way through it. You have to do it. You have to go venture with building things from zero to one and understand the market and the customers you're after. That all remained relevant and the same, but the actual coding piece, of course, changed significantly.

When was your move into Silicon Valley?

I moved in with Cruise back in 2018. I was at Amazon headquarters in Seattle, and when Cruise approached me, I had a couple of conversations with Kyle. This was before Cruise was on the radar at all and self-driving was not as popular or talked about as it is now. It intrigued me a lot, and I wanted to really be part of the first generation that makes driverless a reality. And we did.

What's the hardest problem between the two? Is it healthcare or is it self-driving cars? Because with one you have a lot of regulatory hurdles, and the other one is incredibly difficult operationally.

Both also have some regulatory similarities, where there isn't a playbook on the regulatory side and you have to kind of write it as you go, you have to build relationships. There are a lot of similarities. I think the similarity in both is that both are closed loop systems in my view. And I'm talking about what we really aspire to do in healthcare, not just simple feature innovations. Those are not what I'm talking about. If you're talking about reinventing healthcare, that is a deep technical problem and it's truly a closed loop system. There are a lot of similarities in the data scale, in how you build eval loops and harnesses and safety cases, and in working with regulators. But it's a bit harder in a sense because there are no counterfactuals. In self-driving it's really signals to control, and you can really see applying the steering signal and brake signals, what happened to the EV. You can argue, yes, there isn't necessarily a playbook for perfect driving, but there are ways to measure that more easily. In medicine, the outcome sometimes takes years. The counterfactuals are what make it harder. How do you know that this treatment, while achieving the desired output, was the most optimal treatment to get you there? Maybe there is another dose that would get you there without side effects or in a shorter time frame. Without the counterfactuals, it's really hard to figure out what that optimal path looks like.

There is a lot of depth we can go into. Why is healthcare not first principles control looped, and instead starts by being statistical? Did you have any thoughts around it?

This is a great question. Honestly, that's one of the reasons I joined Hims and Hers. I have to admit, initially it seemed like an odd reach. But the more I learned, the more I talked with Andrew and the team, it was almost a trifecta that made it very straightforward for me to join. On one hand, healthcare touches everyone's life. This is not an optional thing. You can't choose not to worry about healthcare. Everyone does. And at the same time, you look at the state of healthcare today and it's unequivocally broken, big time. We spend twice as much as other developed nations and have a life expectancy about 3.7 years shorter. One person dies worldwide every two seconds from something that took 10 to 20 years to develop, something very easy to find out and treat. It truly is a completely broken status. This is not a case of oh, can we do better. It's absolutely, we can do better, we should do better, there's no question. It's a massive opportunity. So that's just one piece of the trifecta. On the other side, I've come from places where I've been an AI practitioner for the past dozen years, pushing the edge of the technology to reach what I wanted to do. And we're now in a state where AI advancement and technological advancement is at a pace that we've never seen before. It can truly reason and help patients and customers to a great extent. It's not about replacing doctors at all. It's about elevating doctors. So technology is advancing, the status of healthcare is completely broken, which is a massive business opportunity. And then you look at Hims and Hers and you have this incredible company with a great brand, 2.6 plus million customers, tens of millions of telehealth visits. And they control the full closed loop system, from the minute the patient shows up with intention, to diagnosis, to treatment, to titrations, to side effects, to outcomes. They have affiliated providers, 1,500 plus certified providers on the platform. They have their own pharmacy, a million square feet of pharmacies in the US, and they truly control the end-to-end loop. So the ingredients are all there and it's ripe for a change. That really made it very clear, especially after talking to Andrew and seeing his vision, that this is the right fit to make that change happen. So I haven't reflected as much on your question of why it's the case that that's not happening right now, but I believe it's ripe for a change.

When you joined Cruise, how far... I remember the cars in San Francisco, Cruise was everywhere. In your view at the time, how far away was it that there would be self-driving cars, and that it was a solvable problem? Because right now you are in the streets, and Waymo is everywhere, and the Cybertruck is everywhere. But at the time, how did it look?

In 2018, this looked like a very out of reach problem. Waymo operated drivers at that time in the suburbs of Phoenix. It was not crowded streets, but we were after real commercial service, and we were after deploying hundreds of EVs in San Francisco, no strings attached. There is a lot of work that comes into making that a reality. I had the conviction we could make it happen, and it did. At the time, there was a lot of skepticism, I would say even more so than there is on healthcare transformation now, because the technology leaps at the time were not as great as they are today.

The keyword is conviction. How did you build that conviction?

I'm very analytical. I take my time reading, and as I mentioned, maybe that's part of growing up reading a lot about AI, being an AI practitioner before the hype. It gives you a very different perspective. Right now, everyone is talking about AI all the time, and they have a very different perspective than if you've been in AI for the past dozen years, because they just see it from one view. I think that helps you build the right convictions of the capabilities. Even growing up following the history of what shallow nets can or can't do, and why AI is a complete bust, and then seeing the waves of deep networks come to mind, the transformer papers come to mind, understanding the underlying technology and its capabilities early on gives you a lot of perspective and a lot of conviction of what can be done. It was indeed a very hard problem, a massive scale problem. I think we've built at the time one of the largest, if not the largest pipelines in the world, ingesting exabytes of data every month. To make that happen, and it worked, it takes a lot of talent, a big operations, a large team here in the Bay Area, and great founders. It took a lot of work, but it happened.

It even outlasted the founders after the crisis in the company. What are the learnings you took from that experience?

During my last year and a half at Cruise or so as president and CTO, there are tons of learnings. The main one is the rest of the ecosystem that comes around the technology. Technology is never the only important piece of the puzzle. It's not the full solution ever. You can't build a technically right stack without getting the proper infrastructure in place, the proper regulatory relations in place, even the public readiness in place. All these pieces have to work together for you to have a smooth operation, especially as you go through transformations. I think it took a decade or so, like people having an operator stand in an elevator and press buttons for folks, so you can imagine.

There are places in Greece where this still happens, right?

It's an interesting job for sure. So there's a lot of that. A big part of it is, again, the conviction and the relationship you build with the team. Cruise at one point of time had around 3,000 engineers across every discipline of engineering. These folks were very, very mission focused. I have a ton of respect for them. I couldn't just turn my back and walk away. We were all convicted in the mission, and I wanted to stay and make it happen. I got the privilege to deploy driverless twice, once before that and once afterwards, and there are different frameworks, different regulatory frameworks. It was great learnings all around.

How does the Hims opportunity come to you? Did you speak to someone from the team? Did you hear about it?

I had lunch with Andrew. I have to admit I was almost embarrassed going home. I usually try to optimize my time, and I rarely drive. I usually either take an Uber or a service just to utilize my time working. That time I was driving, and I usually listen to something going in or out, because I can't stand just doing nothing. I remember on my way home after my first lunch with Andrew, I didn't listen to anything. I just kept replaying the conversation over and over, and I almost felt embarrassed of how ignorant I was related to the state of where healthcare is. We're all very fortunate to be in the one percentile, or maybe two percentile, of getting excellent healthcare, getting access to incredible doctors, incredible systems, so you don't see the opportunities, you don't see how broken the system is. It was after he talked to me, a founder with such passion and such clear vision. I went home and did more research on my own and discovered more and more of the opportunities, and started to see that trifecta, as I mentioned: the state of healthcare is broken, the technology that can fix it already exists, and then Hims and Hers is very uniquely positioned because they control that closed loop. So it was very clear to me that this could be a fit.

How do you define closed-loop, and maybe what was a closed-loop at the time?

Well, maybe the easiest thing is to define what open loop is, which is the current system. First of all, I don't think what we have today is health care. I think it's sick care. You wait till something is broken, then you go try to fix it. You walk into a doctor's office, you take an appointment, maybe you go there a few weeks later, you fill out a form on a clipboard, and you go talk to a doctor for maybe five minutes if you're lucky, maybe less. And then you walk out with a treatment, and that's about it. Maybe you follow up with another doctor later on, or maybe you go have another exam. Things are all fragmented, they're disconnected. There isn't one system that controls everything about you, not just in this particular category of health, but across all categories, longitudinally. From the minute you walk into a platform, what's your intent? How are you feeling about it? What treatment was prescribed? What kind of side effects did you get? How did you titrate this medicine up or down? Did you reach your outcome or not, the almost optimal clinical outcome or not? And what else is going on? Maybe this happened five years ago, but then you took another lab test and your readings changed, and then you started treatment in a completely different category of care. That's what closed loop is, because all this data is in one place, tied to you as an individual, as a customer, as a patient for Hims and Hers, so you have that holistic view that's very hard to replicate somewhere else.

You always have this example in mind, which is that in the 70s and 80s you're sending rockets to the moon and you have more sensing on those rockets than you have on the human body. It's crazy when you think about it. So you have this conversation with Andrew, and what happens next?

What happens next is I talk to the rest of the team and verify that there are great folks in place that they really would love to work with. On this level, you really have to double click on that aspect, and the conviction develops. Once the conviction develops, for me it's clear that this is a place I want to join and be part of the change they seek. I've been there for about a year and some change right now. Lots of folks from prior lives joined me, from Cruise or Amazon or other places, and we're just really doing some good work.

What's your advice on onboarding? When an executive joins a company, it's always very difficult to onboard that person and ensure that they're productive. It takes a lot of time for them to be able to be productive. What's your advice to other executives that go through that, and maybe teams that are onboarding executives?

I always try to double check myself before giving advice, just to be clear. In my own situation, similar to how you onboard other executives, what do you do for them? You write an onboarding doc. At an executive level, you have to manage through outcomes, not through the details of the hows, but really the what and the why. What is expected to be changed and by when, and what kind of resources or autonomy this person has. It's not much different from that, except that you have to do it yourself, create your own onboarding doc. What is expected of you to change in the first 30, 60, 90 days and in one year, and iterate on it with the founder, with the CEO, to see if these are aligned incentives or not as far as where you want to be a year from now, and then execute accordingly. There is no playbook. It's very easy to give you the regular abstract playbooks, like you go in and you meet with the team and you understand the business. All this stuff is a waste of your listeners' time. Everyone knows that. But the thing that I would focus on is start with the outcomes and iterate. Write your own onboarding, define what success looks like, and then the rest is easy.

What are you trying to do today? What's the product evolution like?

It's been a lot. We took a dual-pronged approach as far as the product evolution. First is how we deliver care itself, and then the second prong is all around AI and how we integrate AI into care. The first prong is the unsexy prong, and this is the most important one. It's how to truly build a healthcare platform, because AI without a platform is not going to get you far. It's just going to be demo-ware. And we focused on that equally.

How do you define the platform there?

The foundation where your product and AI sits on, whether it's headless or not. For example, how do you launch a new category of care? How do you launch a new SKU inside that category of care? What are the clinical outcomes and guidelines around launching a new category? What does that intake form look like? The component that makes care possible in the first place, but doing it right. It's like an e-commerce engine that allows you to be in multiple categories of care at the same time, to decouple pricing from SKUs, from how you fulfill your products, to the state rules around the product, the pharmacy rules and so on, the content management to give the business teams the full power to self-serve how that experience looks like for the customer, how to run promos or bundling or offerings like memberships, all these abilities, how to communicate with your customers by emails or text and so on. What are the quiet hours and data privacy issues? All that stuff needs to be in one place. You don't have to replicate that every time you're building a mobile app or a web app or an AI app. They all have to go through the same platform, and you need to fix that platform and think of it as building Lego blocks that would allow the business to go in whichever direction in the future, modularized, component-oriented pieces that you can put together to get to the business agility, to be able to launch everything that you want to launch very fast. That's one prong, and we've worked a lot on that. Since that time, we've launched a lot of new categories for our customers: hormonal health, testosterone, labs, which was a gateway to a multi-categorical care, membership models we've offered, and lots of new SKUs with partners on the weight loss journey. That's one dimension. And then another dimension is AI, which I'm sure is a whole different conversation. Our approach for how you deliver care in a different way is basically to embed intelligence into every step in the customer journey, and reimagine care using AI as a tool, as a means to an end, not a goal by itself. That was the different prong, and we've had a lot of success in a lot of product launches there too that I'm happy to get into details on.

On that point, given that there are all of the AI companies like OpenAI and the fabrics of the world and their moves into healthcare, with many people having similar models, what becomes uniquely true and what's the unique advantage for Hims over the years?

First of all, it's great for the consumer. It's also a great thing for the models to be out there and help educate customers on even what questions to ask when they meet their doctors. The main differentiator is you can have a conversation with these models and it's going to remind you multiple times they're not a doctor, and you can't take it for granted. It's non-actionable. In our case, that's not the case. You have a doctor at the center of everything. We have providers affiliated and everything is well integrated so you can act on it as well. You don't have to have a conversation and port it somewhere, hoping to remember that you're going to ask your doctor the question you got advice to ask, and maybe have half the time to ask it and not the others. In our case, everything is integrated in one thread. The doctors already know what kind of conversation you've been having with your AI in whichever fields. The AI knows a complete view of the customers: their intake, their side effects, their lab results, everything they have interacted with on the system, the outcomes from prior treatments, and their doctor is right there. You can actually turn that into an action. You can have a prescription shipped from our pharmacies to be at your doorstep in a couple of days, and you can see results, and you can follow up with another lab test to confirm that these results are actually happening, or step on the scale and see that your weight loss is actually happening. All that is interconnected. There's a big difference there.

When I spoke with the team, you have a new product coming up and new product announcements. What is it? How does it work?

It's called AI in weight loss, and it's really not about AI at all. As I mentioned before, AI is never the goal here. As a matter of fact, the best AI is invisible AI. This is almost a new clinical care experience where your doctors, your AI coaches, your care team, your care coaches, your customer service are all in one thread. You know exactly who you're talking to, but they're always there for you 24-7. This is akin to, back to the privilege we've all enjoyed without knowing. I had the privilege of having amazing doctors and some doctor friends, and I was able to text them anytime. If I wanted to ask them about something, they knew me very well, they would respond compassionately about what I asked. And that's very similar to that. There's a doctor at the center of the conversation. One of the biggest features for AI built right in health care is to know its boundaries, to be clinically grounded, not rely on the open internet to give you advice, and then most importantly to know when to get a doctor into the conversation at the right moments, and not just tell you okay, you're on your own, go talk to a doctor about XYZ. That's part of what's happening. So not only are they 24-7, not only is it interactive, it's also proactive. Meaning it's not going to wait for you to reach out with a question. It also knows when to reach out to you at the right moment, which would be different from one person to another, and how to help you at the end of the day reach the most optimal clinical outcome you're after.

What does proactive really mean in this case? In what cases would it reach out to me?

There are a few cases. As I mentioned, it would eventually be different from one customer to another, but imagine we're launching this for a weight loss journey, because it's one of the most involved, longitudinal, hard journeys for a lot of patients. We've seen a lot of patients with weight loss either drop off because of side effects or because of hitting plateaus and so on. So there are clear activation points where someone reaching out at a certain moment can make a big difference for that patient's journey. It could be the first dose, to take you through it. It could be the first time you step on a connected scale where we already know your weight, and every time you step on that scale we get to know your weight and how your progression has been through your journey. It could be a time expected when you start to really have some side effects from the journey, at points where you need to titrate up or down based on your progress. So all those could be an activation point where it reaches out. In a nutshell, the beauty about AI here is it can be objective oriented, and I stress this on the team very much. There's a huge difference between being in health care and consumer tech. We're not in consumer tech. I'm not after engagement metrics. I don't care about that. I only care about clinical outcome. And that makes it easy, because then you reach out only when reaching out is going to make a difference. Other than that, silence is a feature. We take the consumer attention very seriously. The more you reach out for no reason, the more it would dilute the important reach out points. We're very excited about AI in weight loss and how that can change the journey for many customers.

Can we walk exactly into the user journey of this product?

Yeah, sure. The user journey, as I mentioned, AI is a means to an end. We are reinventing health care in a way that is reimagined based on today's capabilities, or in fact even tomorrow's capabilities. But it's not about bolting a chatbot on the edges and saying now we're AI integrated. It's not that at all. This is about the same customers we have. We understand the journey very well, we have a lot of data, we have a lot of counterfactuals of customers having different treatments and different iterations, and we're using all that plus AI plus our amazing provider networks plus our amazing care coaches, customer service, and converging all this in one seamless integration in a new app we're launching and a new interface with the AI, so that it's always there for you as a customer. It's as simple as that. In one thread, one conversation, you don't have to worry about whether you need to talk to a doctor or a care coach or customer service or the AI. The AI is going to figure all that out for you, and you're going to know who you're talking to at every point in time, but it's going to be very seamless from your point of view.

So the AI understands by itself what's the right moment to bring a human in the loop, and it brings a practitioner to answer the right questions. Isn't this a difficult problem to solve by its nature, knowing exactly when a doctor is necessary?

It's a very difficult problem to solve. However, you can make it more difficult on yourself depending on what goals you set on precision and recall. But if you're optimizing for recall and you want to make sure that you never under-escalate, you always will get a human in the loop when needed, it becomes solvable. The advantage we have is that we have incredible provider networks that are already in the loop that we can bring into this conversation, which could be very difficult for others to replicate. There's a classifier that would help orient the conversation in the first place. The AI itself is designed in a way that has to be grounded to clinical input, which we work with our providers very closely from day one to make happen. This is basically clinical guidelines as code, and any output coming out of the AI has to be grounded to that.

When you say clinical guidelines as code, this is a very interesting topic. Can we dig into that? What does it mean?

It means that for every treatment, for every category, there's tons of tacit knowledge and clinical guidelines that we have. And we are now putting those in as guardrails. Not as suggestions, as actual guardrails for the AI. They're managed by our doctors, by our clinical ops team. They're checked in as code, similar to how you evolved into configuration as code in other areas. And that's what the AI gets grounded to. Very similar to RAG, how to get the best of both worlds in reasoning but also be grounded to our own clinical guidelines of what you can say, where you stay within bounds, and where to escalate to a doctor. If you design the system, as I said, with a classifier that helps route things in the first place, whether to a human or to the AI system, and then another layer of defense is the AI system itself using these guidelines checked in as code to make sure that every sentence and every answer is grounded in these clinical guidelines, followed by another LLM judge in real time to double-check that and score it before any customer can see it, and all that happens in real time. Then you have multiple layers of defense to ensure that escalations happen properly, there are no hallucinations, and everything we're saying we can stand behind.

At the scale of teams, you mentioned 25 million users, and what is the deployment like? Which customers do you deploy first, and how does this work at this scale?

We're starting this for the Hers weight loss journey, which is a big segment of what we have. We chose Hers weight loss because that is a complicated segment. We wanted to start with a high touch journey with a lot of customers. Of course we have all the proper engineering guardrails as we roll this out from 1% to 5% to 10% to 30% and so on, with rollback guidelines. We have drift monitors. We have QA sampling with real-time monitoring, human monitoring as well. And at the end of the day, we are really after one thing only, which is an improved clinical outcome and patient satisfaction. That's super important, because there's a lot of benchmark theater right now with AI. You can go in and release a few headlines about how we've had a chatbot that does XYZ or whatever. But at the end of the day, for us as a business that cares deeply about its customers, I put customers first and then everything else falls into place. When you satisfy the customer in their journey, when they meet their outcome, when they are happy and satisfied and they retain better, you see engagement go up as a second or first derivative. You see your retention go up, you see conversion go up, you see new customers coming to the platform. All that happens after you get to the most optimal clinical outcome.

What was actually built in this case versus what existed in the past? Are you using specific models that are available generally? What are the new technologies that you had to build?

I'll answer your question twofold. What we're building is everything I've mentioned, because we didn't have an AI department before. We've built all that up from scratch with our 40 AI engineers, and that team is growing quickly.

Can we dig a bit into that? What does this mean? When you build a new segment inside of Hims, which is a very big company today, how do you decide on the structure of that? And have you decided if you want to build an AI organization within a company, for example?

In this case it was clear to me the kind of innovation we need is frankly almost reinventing. I understand exactly what it takes to build something as high quality and high rigor as what we wanted to build. Bear in mind, just to be clear, in my view AI models right now are almost a commodity, and they're commoditizing faster than anyone can ever imagine. So that's not the advantage we're after. We're not after inventing yet another LLM. Other people are doing a great job doing that for us, and the economics are actually in our favor. The advantage is really in the workflow and the data. Data and workflow compound while models commoditize. And to build that closed loop, to turn it from that theoretical abstract I've talked about to an actual machine learning infrastructure where you can build models, build eval loops, continuously improve these models, that's the skill set that was not necessarily there in health care before, and this is where we're building our AI organization around it.

But Mo, when one builds a new team, you either go top down or bottom up. I'm wondering about this. And second, given that it's a new innovation and it doesn't really exist, you don't have this pocket of talent that exists in a specific company that you can work with, how do you build a new team that knows it's going to work?

In this case, I've had the privilege to know a lot of great talent in the Bay Area from previous jobs. I know exactly the kind of talent we need here. The mission-oriented mindset, the conviction, the desire to really have a big impact, not just be in technology for technology's sake, is a key ingredient. I was fortunate to have a lot of great talent follow me from previous places, and that made it easy to hire this level of talent and hire it quickly.

So it was mostly that you knew them from previous experience and the qualities you'd worked with in the past. I see. So I had this conversation with Vinod Khosla yesterday. I think he's much more extreme in saying that clinicians are obsolete and we need to be using AI for everything. So I wanted to ask you about a comparison between AI and clinicians. Who makes more mistakes today, and how do you solve this type of problem?

That's an impossible question. First of all, I love Vinod, and I'm going to disagree with him on this one. So Vinod, please don't hate me for this. But no, I genuinely respect him a ton. Let me take a step back. First of all, AI is not here to replace doctors. That's not my view at all, and that's not Hims and Hers's view. On the contrary, it is to elevate doctors. You can talk to economists about this, and you can look at the current supply and demand curves and conclude something. In my view, that's completely the wrong approach. There is a massive induced demand that you can generate if you truly fix the health care experience. Right now, health care is broken. That's why you see these supply and demand curves, and despite that, you see more demand than there is supply. But now imagine if you reinvent health care. We can talk about what that looks like in the future, where it's actually accessible to every human that ever lives, and frictionless and affordable and easy to understand. That's going to generate an induced demand where there's just tremendous demand for human judgment, and doctors, and the empathy the doctor can provide, and actually having a human conversation. If you look at how doctors spend their time right now, there's a lot of tax on admin tasks and documentation, and even chasing your own information and context gathering, and understanding who you are as a human, and switching context between one and the other. AI can help with all that. So it's a platform to truly elevate doctors and get them to focus on what they enjoy the most and what they joined medicine in the first place to do, and to be that accountable human in the loop at the end of the day. So I don't see it, at least for the time being, as something that's going to replace all doctors. Now, to your second impossible question, whether it makes more mistakes, this is something you have to answer via studies, and it depends on what AI you're referring to, what harnesses you've built around it, what guardrails, what kind of specialty this AI is focused on. There's a lot of disconnect between what you read in the news and the reality of the implementations. Similar to how I'm sure you're going to read about tens of driverless car companies, but the reality is Waymo is the driverless car company right now. There are a lot of things attached with, can it deploy a car in a limited domain, can it change lanes, can it turn left? That's not really driverless. There's a lot of AI in medicine right now that can prescribe, but you look in the fine details and it's like, can it really, or does it just renew a prescription where two doctors are reviewing it? That's a very different story. What I would say is, back to the counterfactual, it's very hard to even run such a study at a massive scale and just lump some doctors in one place and see, here's the doctors' performance. At Hims and Hers, we are very focused on making sure we're not falling into the trap of imitation learning. It's very easy to look at back data and build an AI and a deep learning model that you do backtesting on, and it looks incredible on the backtested data because it imitated exactly what the previous doctors have prescribed. Then you fall into a lot of traps with that. So there are ways to make sure that you're isolating the confounding factors and looking at other measures to make sure that you have a golden data set, not just a back data set, and you understand what the optimal treatment looks like from multiple views of doctors. That's why we're lucky to be partnered with our doctors and our medical directors from day zero. This is something I really love about Hims and Hers: we don't have an AI team and a medical team, there's one team across the board. The medical directors work with our head of AI from day zero, from inception, before even the design of anything. It's a true partnership of, where is this balance? How do you build the eval loops? How do you escalate? What are the escalation guidelines? What are the QA guidelines? How do you red team this? How do you do regression testing? That's all built from scratch, and built with doctors' partnership from day zero at Hers.

On that point, if we project a bit forward, what this means is if you are a clinician, you have a number of agents that act for you and remove from the process many things that are admin, for example. What do you see this becoming? I have this example in mind that includes coding. You can have so many agents, and then the PR review expands so much given that so much more work is being done. There are so many companies working in the area of removing the PR reviews and so on. But I think, yes, it empowers a coder to have so much more output. Do we have some sort of equivalent here in medicine?

Let me pause on this for a second, because I'm very familiar with the PR space you're talking about, and I think we have a superset of that. At least at Hims and Hers, what we're building for our doctors is not as simple as this, because there is bias in accepting automation. I'm sure you see that in engineers. Forget medicine for a second.

Yes, there's too much bias into safety.

In my view, AI with engineering is a magnifier. It's a tool that either makes a great engineer the greatest, or makes a mediocre engineer even more so, and makes a bad engineer really, really bad, because he's going to be chucking out a lot of PRs that are all wrong. If you try to mimic that in medicine, that's not a good thing. One of the biggest problems in AI is to also show how certain or uncertain you are about your own capabilities, to prevent that false assertion when you're wrong. That's something you can build for from day one. AI, in my view, should be invisible. Doctors shouldn't care about how many agents are working in the background. What do you care about as a doctor? You go in, there's a patient, whether they went through an intake and want to get a certain treatment, or they've been already on a journey and are sending you a message. As a doctor, I would imagine the best thing for you is to know everything about this patient in the most efficient way possible. It shouldn't overload you cognitively with irrelevant information about this person related to the question or the prescription you're about to approve or disapprove, because at the end of the day it's your call as a doctor. But make that extremely easy and frictionless for you, to know everything about this patient: what other treatments they have, what other conversations they've been having in the past, what side effects they had from other treatments, what dose titration they've been on or off. Then we have a system that's been out for a while called MedMatch that can recommend to the doctor, based on this and based on tens of millions of other treatments we've seen globally for similar patients, that this could be a good starting point. It shows all the details for the doctor and has the doctor make the right and final call at the end of the day. They should see all that in the most seamless way. This is why it's an all-integrated system. AI shouldn't be the focus at the end of the day. It's the right information at the right time, the right level of certainty, the right color coding. That's what helps the doctor be as efficient as they can without bias.

We spoke about proactive health, we spoke about AI and weight loss. What I'm wondering is, if we project this a bit further, if every person starts to achieve their goals for their weight, you can further help them afterwards and say, can I achieve a different type of goal, which is, I want to achieve greatness in the Olympics, for example, or I'm a student and I want to become a better student. Are we at any point going to arrive at this situation where the person decides that I want to have this goal, and it's not only about me being sick or me having a weight issue?

So it's true health care.

Exactly, like us saying, hey, I want to win at this Olympic game, or I want to be the best student, and the system allows me to get there.

I love this question. Yeah. So this goes back to, we have sick care today, and you're asking, is there a world that would truly be health-optimal care? You're asking this as an engineer. You want to optimize your own health, and it's going to be different from one person to another.

Plus, we might need to change the word health care into something.

Something better. Yeah, well-being, an operating system of living well. I don't know how many can relate to this, but we're standing in line in the Blockbuster franchise store to rent a DVD, and every conversation about health care innovation is about how can we arrange the DVDs on the shelf, or change the lighting in the store, or make it more appealing for customers. It's completely the wrong approach. It's like, you should be asking, how can this be Netflix? It's insane that we've all gotten acclimated to accepting health care in its current status, which you wouldn't accept in your music feed. I open my podcast app and I see yours and others that are actually oriented towards me as a listener. Imagine if everyone in the universe opened their music app and they saw the same music feed, which is the current health care system, or even worse. So yes, it is a complete reinventing. And layer on top of that, look at the status of AI today. There's a lot of noise, a lot of hype, a lot of drama, Hollywood writing, but the reality is it is moving at a very fast pace. What that means for drug discovery, in my view, is that it's also going to shrink this cycle for drug discovery greatly, which means the selection that you have to optimize your health is going to get so complicated, almost Amazon-like complicated. Five years from now you can have so many different choices to optimize your weight, or your cognitive ability, or longevity, or hair loss, or sexual health. There are going to be so many choices for consumers, and every customer is different. This is where you really need to rethink how you reinvent that experience altogether, to be centered around the customer or the patient, and be personalized at its core to help you really achieve the outcome you desire. So yes.

Absolutely. I think that's the direction that inspires all of us to be in a company like this one, and AI can help a lot with that. You can just imagine a very different future layered on top of that: integration with wearables and connected scales and many other methods and genetic sequences, and then the sky is the limit. Because then we know a lot more about the customer, we know that information in real time. You can imagine waking up someday in 2030 and your app is saying, hey, I've been noticing this trend from your watch or your ring, and we're worried about you. We'll send you a lab kit in the mail. It should be in your inbox. You go in and you put it on your shoulder, you click a button, which we do now anyway, and you send it over. And by the end of the day you get the results back, and the results don't need to be deciphered. You can understand them. It's in plain English. It's in the same conversation where your doctor is, and it comes with an action, with an option for you to do something about it. And the prescription can be at your doorstep a couple hours later. That is not out of reach. We've achieved that in many other categories. Why not in the most important thing that affects every single one of us? All the ingredients are there. I don't know what would stop us from getting there. So yes, I'm very excited about getting us there.

Mo, if we walk five years into the future, what does Hims look like?

I don't know how many people are able to predict five years into the future with accuracy, so I'd like to stay humble and say this is as much as I can predict. As I mentioned before, there are dimensions where I feel that AI is really changing the face of drug discovery in the first place. As you know, today already 50% of web traffic is automated. It's not even human. So there's a big move towards headless software. Take that blockbuster approach: I think Hims and Hers looks very different than today. If you're really reinventing care, there is no reason to stick to the bounds that we're in today. Even in our case, you look at how patients go through online forms to get to a certain treatment, and it's not much different than what you fill out at the doctor's office, except that one is online and one is on a clipboard. All that changes, because at the end of the day, this is also a problem of how you can extract the most relevant signal from every person, which is very different, in the least amount of friction. You can do that with AI conversationally, via text, via many different forms, and get to the same outcome in a much better way. You can get to a world where your longitudinal treatments are very seamless and they're managed all in one interface. This interconnected system of AI agents and doctors and care coaches, truly what's called today concierge healthcare, that costs hundreds of thousands of dollars, or tens of thousands of dollars if you're lucky to get one, the economics of that collapses completely. That should be affordable and in the pocket for everyone. Everyone can have that same level of experience, frankly even better, that is obsessed and personalized for you, knows everything about you, and is there for you 24-7 with a doctor in the loop. There are a lot of opportunities for us to truly reinvent that space, not just to renovate it.

I think it makes a lot of sense to ask you this question, given that we are both Mediterranean sitting in the center of San Francisco. A lot of things are going wrong in the majority of Mediterranean countries, but I want to end with an optimistic angle. What would your advice be, or where would you want to see some changes? So countries in the Mediterranean, maybe Egypt, maybe Cyprus, to improve so they have better healthcare and better outcomes for their people.

I'm not a politician, obviously.

That's why I'm asking you. If I asked a politician, they would give a bad answer.

Right, they'd give you an approved answer. Here's what I genuinely believe. As I mentioned before, I think raw intelligence is equally distributed, and raw talent is also equally distributed. But opportunity is not. We are at the edge of Western civilization. I love this country. I love all the values, because it gives everyone opportunities to reach their maximum potential. That's not necessarily the case in many other places in the world, which is very unfortunate. The good news is I do believe AI is going to be the biggest equalizer we've ever seen in our life, and it's going to give people a lot of chances, it's going to level a lot of playing fields, and it could be an opportunity for many countries, for many small companies and small startups to emerge in different places and build an ecosystem. That can lead into prosperity, which I do believe at the end of the day is one of the key ingredients toward peace. This is a bigger topic, but I truly believe that. I think a man with nothing to lose is the most dangerous thing for society, and there are a lot of people with nothing to lose in many areas of the world. You can see here that tech prosperity is all around us, and it's my hope that you see this wave of AI inducing a lot of tech prosperity in many regions of the world worldwide, that truly improves everyday life and creates an economic cycle that turns things to the better for everyone else, including here in the States.

Is your advice to people to work and iterate around AI?

The barriers to entry to get something off the ground were much higher even two years ago than they are today. So my advice is that you can have a lot of solo entrepreneurs now coming from various areas of the world with amazing ideas and big hits, and that could be the start of something big. A big part of that is to shatter some glass ceilings, some beliefs that people have about their own capabilities, or how bounded they are, or what conspiracy theories they live in. As I mentioned, today AI is the greatest equalizer, and frankly, if you have access to the cloud and the latest models, no matter where you are in the world, I don't know if you have much excuse at this point to not be your best self and reach something great. It's been a fantastic discussion. Thanks, Kyriakos. It's been a lot of fun. Thanks for having me.