The Rise of AI Agents in Healthcare | Omri Yoffe, Vi Labs

Manav: Today on the podcast we have Omri Yoffe. So happy to see you, Omri.

Omri Yoffe: Same here. Thanks for having us.

Manav: So I wanted to start the podcast on a very light note. What's wrong with the American healthcare industry?

Omri Yoffe: Healthcare is still, I would say, failing the people it's supposed to protect. Sometimes unintentionally, people are diagnosed but don't get the right treatment or the best next action fast enough. Drugs are very slow to market given some complexity. We can go into it later in terms of clinical trial completion. And grandmothers don't have access to affordable healthcare, although they can, but it's very hard to connect the dots for people to get access to the system.

Manav: So you're talking about a highly complex, sometimes misaligned incentive system. You've had a very multifaceted career going from an Air Force pilot to founding a biometrics/wearables company, and now you're in enterprise healthcare. Can you talk a little bit about that transition? I know healthcare is the common theme here.

Omri Yoffe: Sure. So I got into the health mission in a very abnormal, random way. As you've mentioned, I come originally from the aerospace domain and founded my first business as an R&D powerhouse called Clerka to basically deliver different types of avionics, command and control systems, and heads-up displays to the largest aerospace organizations out there in very high-stake environments.

I'm a bootstrap guy, so a big believer in trying to build things in a self-sustained way. In that case, Clerka's early years generated positive cash flow. I was looking for more impactful, scalable solutions using technology.

One of them was called LifeBam. We lost one of our best F-16 pilots in the Air Force. He basically pulled high Gs, passed out, and crashed. We built an application and solution designed to save lives of pilots, special forces, and astronauts during unique situations—fatigue, hypoxia, passing out, and different types of physiological weaknesses. We solved it by the end of 2016, and I fell in love with the problem of precision health and predictive health. I never looked back.

I founded V, did a material pivot in between. Same mission, but we went from large-scale direct-to-consumer to enterprise. I'm very obsessed and passionate about the health mission.

Manav: You were pretty early on so many things, right? We just saw Whoop get valued at $10 billion. There's Oura Ring, Eight Sleep. There are so many interesting companies in the wearable space. How do you think about that industry, and because you were so deep into it?

Omri Yoffe: You know, it's crazy. First of all, you're right. Timing—I don't want to say it's everything, but I think what I learned, a good lesson in humility, is the following.

From day one, we wanted to create an AI companion that would live within your earphones, wearables, and phone, and guide you to a healthier life journey. We did that. We got to $56 million in sales. We had the fastest and largest Kickstarter campaign out there. We delivered the full product with product-market fit.

Then we got a very important lesson from the Apples and Googles of the world when they went in with HealthKit, Google Fit, and so forth. The cost of acquisition of our solution went through the roof. I had two options: either bang our heads on the wall and fight a very unwise fight, or say, "We want to make people healthy. Let's use our edge and advantages and learnings from the wearable solution into the enterprise world."

I pivoted very aggressively. I had 53 people and needed to let go of 50. Some of them came back, by the way. We adapted pretty fast into the B2B world and never looked back.

Manav: I was watching one of your presentations on YouTube, and one thing I really took away was that you're a deeply technical person. Can you talk about that a little bit? You also have an MBA, but...

Omri Yoffe: I'm a very obsessed "deeper, not wider" guy. If there's something I want to be fluent with, I invest the time and surround myself with the right teachers and real-life environments to be fluent in it. I don't have an engineering degree, but you're right—anywhere from systems design, hardware, software, data science, to actual product design, I'm above average in terms of being in a room and having an educated discussion about it and pressure-testing questions versus just hiring a bunch of people and letting them fly.

I think in today's world it's a must if you want to have the basic intuition into what's feasible and what's not. You must be fluent with those data pieces. We'll talk AI native in a second. There are some new leverage coming in, but I think traditional self-learning will not change.

Manav: What was your early interaction with AI? Because AI became mainstream in 2022, and you were doing this in 2016.

Omri Yoffe: Correct. Back in 2016 and 2017, machine learning and deep learning started to be a thing from MITs to other institutions. For me, it was clear that I wanted to prove a thesis: can we mimic human practices into an AI companion?

Back then, you didn't have what we have now. We had basic text-to-speech and automatic speech recognition, and machine learning models that could predict human activity based on tagged data. To build a health coach, we needed to record extensive data. We worked on a human voice—we wanted it to be fully human and authentic versus synthetic. We recorded tens of thousands of voice snippets and interviewed many voice actresses.

We built a coach based on a voice actress named Crystal's voice. It was the only coach in the world that had predictive analytics and the ability to trigger the right commands and conversational parts, but also had stitched-in conversational sentences, numbers, names, and so forth. That was a non-scalable solution for an AI problem.

Today, it's a no-brainer that you can use LLMs or OpenAI to speak with our TTS voices. Second example: we didn't have large language models. Every model you needed to deploy required taking deterministic models, specific data, tagging it, training the model, and seeing the outcomes. In today's world, you have AI systems that can do all of it by themselves. There's a massive jump that people underestimate.

Manav: Here are some exciting numbers from today, 2026. V Labs has raised over $160 million and is currently valued at $1.64 billion. How does it feel to be a unicorn?

Omri Yoffe: To be completely direct, it doesn't feel different. We just had a town hall meeting with my team, and I ended it with a very simple slide called "Stay focused, stay humble." It's not an empty statement. I think if we were doing what we're doing for the sake of valuations or money, we'd probably sell the company by now or go to the beach with great restaurants, hotels, and movies—which I'm not against. I think it's important for people to enjoy their life.

But in V's case, I genuinely believe that it's day one, always. The type of people we hire and train and grow are people who are here to improve their version and do exceptional things, regardless if you have another zero to your valuation or not.

Practically speaking, valuation helps do two main things. First, you want to retain the world's top talent. You can use this as a way to get every single employee as a shareholder in the company and to think and drive enterprise value. Second, shareholders need some measurement and returns to recognize success in their minds.

But for me personally, my wife is joking that she never knows if we're up or down because I push the same level. I think it's very important for the rest of the team as well.

Manav: What's the elevator pitch of V Labs? I know some people have called you the Palantir for the healthcare industry. Can you explain in your own words?

Omri Yoffe: Happy to. In short, V is the AI execution layer, basically helping healthcare and life science enterprises turn their data into next best actions for patients and care teams.

Manav: These are hospitals.

Omri Yoffe: Exactly. Three quick examples: We help the most important clinical trials accelerate and complete faster. We help patients and physicians from surgery to appointments take their next best action, get second opinions, and move their health journey in a much more precise, predictive, and fast way. We help optimize supply chains and operations—staffing, scheduling, and different types of resource management.

The way you should look at us is we bring three main things that differentiate us from Palantir and other enterprise AI players. First is our data edge. We have arguably one of the largest world datasets for clinical, behavioral, and operational data—190 million lives and 1.2 petabytes of computed data. That's massive. For context, that's 1.2 million GB of data.

Manav: That's insane.

Omri Yoffe: It is. Second, we deploy and productize agents and models specifically specializing in the health domain—what we call vertical AI applications. Third, we have skin in the game. Our business model is very straightforward: we win only if our customers and their patients are winning, and we have a way to price and keep ourselves accountable for their outcomes.

Manav: So it's not a traditional SaaS model. How would you explain your business model?

Omri Yoffe: You're spot on, and it's actually a very critical point. For the last five-plus years, we're selling value. This isn't because of the current hype of selling ROI—we always believed that selling on a seat-based level is the wrong incentive. You want to be completely aligned between you, your customers, and anyone that is part of the valuation, including the patient itself.

In short, here's how it works: We go into a customer and underwrite and map out where we can help—reduce the cost of patient activation, accelerate clinical trials, increase usage of health services for a hospital or provider, or save tens of millions in operations. We bake it into our pricing model. We go into a three-month pilot with a control group, so you have full transparency with V and without V—control group versus action group. We prove value, and only when we prove value do we turn it into a flat monthly fee based on this ROI. There's full transparency for the client. If we don't deliver after six months or a year, we can stop and re-underwrite to make sure they do get the value they need.

Manav: That's a much better business model right now because SaaS is going through a bit of an apocalypse. Have you thought about being partners in these businesses and taking equity?

Omri Yoffe: We did. We definitely did. You want to make sure you're staying the Switzerland of the industry, if that makes sense. My job is to be as impactful and scalable as possible. If we want to create what we call "health abundance" in the next five to ten years, you can't do it if you're over-incentivized with a specific pharma company or healthcare company. So we make sure to get upside, but we never took equity.

You're asking the right questions. Many companies are saying, "You're an extension of our technology. You're an amplifier. We want you to think enterprise value." We usually turn it into cash normalized by ARR because you also don't want a $10 million one-off check. You want to create predictability, especially if the company at any point wants to go public—which is definitely not the case right now—or just wants to create predictability for investors or shareholders. You want to normalize it over two to three years ahead.

Manav: And I'm guessing you're not competing with the OpenAIs and the Clouds of the world?

Omri Yoffe: We work with them. They're one of our partners. V sits on top of the existing data stack. You keep your Databricks, you keep your Epic, you keep your Alteryx. We also leverage our own proprietary models, but third-party models like OpenAI and Anthropic, as well as more local, private, and open-source models. We use those models and productize and train them based on our data and expertise. They like it. Think about them as the OS. We build the actual application for the end client.

The good thing about OpenAI is they're infinitely scalable. How scalable is your company, and how many countries are you targeting?

Manav: How scalable is your company, and how many countries are you targeting?

Omri Yoffe: Again, we need to stay focused and humble, so I don't want to throw empty statements and numbers out there. But I can give you non-V-related numbers.

The healthcare and life sciences market by itself is 19% of US GDP and 12% of global GDP. We're talking about trillions of dollars compounded.

Second, the fact that V is "Switzerland"—I don't focus on specific clinical or population expertise, and I cover the full value chain. V activates new patients, engages existing patients, and operates for operations. I don't want to say limitless, but we have years of growth ahead of us.

Every time I'm asked by board members what we're doing now, we might be able to do it for mobility, governance, and defense. But I'm here because of the purpose itself, because of the mission. I think we're fortunate enough to have a big enough tent to focus on it for the coming years.

Manav: I saw in an interview that you use 20 LLMs. Can you talk a little bit about that?

Omri Yoffe: I don't want to treat it as a hard-coded number, but here's what you need to understand. We're cherry-picking deterministic models and probabilistic models—the LLMs of the world. Usually, it's 15 to 20 chosen models at a time. The reason is you want to pick novel models and specialize and productize them versus being a supermarket and reseller of models.

It's very nuanced. It's not like we change all models every quarter. It's incremental. If a new LLM comes out—like OPUS 4.7 is better for internal coding and agents—great. Maybe there's a different Mistral model we can deploy for operations. Let's check it. But it's a cherry-picking process versus a hard-coded list.

Manav: I still use a lot of agents in my businesses, and you still need a human in the loop.

Omri Yoffe: For sure. I think you're spot on. We're on two ends of the value chain. On the V side, we have QA people. It sounds obvious, but we believe we have a very flat, AI-native approach. I can explain how we do it if you're interested, but we basically have an IC culture from day one, and AI native is amplifying it. It's not that we have 10 humans per employee. We have small squads, very prominent and fluent AI people who usually have traditional software capabilities, but you also have QA testers. There's lots of hallucination and probabilistic potential mistakes. You need someone to work off the workflows.

On the patient side, you have moral judgment and compliance problems. So you need agents to do the heavy lifting. I'll give an example. If you went to a United Healthcare coverage or HCA hospital, were diagnosed with a kidney problem—God forbid—even if there's an agent that once you left the doctor, triggered the next best action and gathered and scheduled a second opinion appointment, got payment from your healthcare provider and employer, and sent you an update about your journey and expectations, great. You still need a human to intervene. It can be a nurse or engagement specialist.

First, to show empathy and for the person to understand that it's being managed by humans with the right values and care. Second, sometimes the AI will do stupid stuff. I can give examples. So we're moving into a world where you have massive heavy lifting by agents, but also human authenticity, judgment, and compliance.

Manav: One thing I noticed in the healthcare industry is too many parties involved—the insurance, the government, the patient, the doctor, the healthcare provider. How do you think about that? For someone who studies game theory, this is like 7D chess.

Omri Yoffe: I don't have a silver bullet answer here. This is an orchestrated political, regulated, capital-sensitive, and commercial battle. But I think there are two things that, if they happen, we have a massive opportunity to solve what you're saying.

I think the best future leaders, if they can run on the ticket of saying "We're going to create one source of truth and a connecting grid" so that people can sign a consent and all those parties you've mentioned will see their medical ID, clinical claims, and health journeys from dentist to, God forbid, leukemia—that's a massive opportunity. It's doable but requires leadership and execution.

Second, you need CEOs, CTOs, and COOs to turn their organizations to be AI-ready. You know, it's the Databricks, the Snowflakes of the world, the clean rooms of the AWS and Google Clouds. This is a must. Once those two things happen, health abundance could be achieved. Lots of work to do.

Manav: Kudos to you by the way for building in such a hard sector. A lot of people just avoid healthcare because of the bureaucracy in this industry. I want to touch a little bit on healthcare abundance. Let's make some crazy predictions. You know, we'll come back to it in five years and maybe laugh. Tell me some predictions because things are changing so much—like with full self-driving cars, we have so much abundance right now. Tell us about abundance in healthcare.

Omri Yoffe: I think there is a way, and I don't think it's crazy. I think we need to be very concrete and ambitious about how to measure health abundance. I'll use three main metrics.

First, you want to get to a precision level where you don't make stupid mistakes that kids, parents, and grandparents are dying or suffering for the rest of their lives just because there's a lack of precision. I don't know if it's 100%, but it's not 80, that's for sure. We want to get to the 80-90% precision level so your healthcare system is able to drive the next best action, the right second opinion, the right appointment, the right surgery, and so forth.

Second, you want to be predictive. Time is critical in health. I'm not talking 100%, but I imagine a world where a person, a minute after they left the doctor, or at least the same day, is getting a very clear GPS journey that is already predicting what are at least the avenues and junctions they need to take. I think it's very doable.

Third is affordability. It doesn't make sense that most people in the world, not only in the US, don't have access to affordable care. I think AI is the best cure and the best ingredient because it lowers the cost of opportunity. It lowers the heavy lifting of humans needed to do repetitive tasks in a very cost-non-effective way.

I think you're going to see better decisions, shortened time, and most importantly, I'm very confident that over the next five to seven years, most citizens in the western world using AI healthcare will not be the main burden for them as it is today. You'll have robots doing surgeries all day long, and you'll have abundance of access to data. So you don't need to think about it that way.

If I have, God forbid, a very nuanced clinical problem, I don't need the best Sloan or Mayo Clinic doctor. It's fully democratized, and each one of us across this room will have the exact same access to those decisions and the exact same access to the care at least executing or facilitating your career navigation, which is the key piece. It's not the drug or the science problem. The science is there. It's how you orchestrate and navigate the patient to take the right decisions.

Manav: I want to end on the note of what's next for V. What roles are you hiring for, or what's the next big vision for V?

Omri Yoffe: We want to build what we believe is going to be the most important and generational healthcare and life sciences enterprise solution in the coming years. There are two areas I'm most excited about.

First, I'll call it "context as data as a catalog." We're building a very powerful AI-native infrastructure for our teams to execute in a disproportional way. We serve today over 100 of the largest enterprises. I want to serve 300 with 250 people. So there's lots of investment being made behind the scenes. This is about building a team operating system where a data scientist, squad leader, and applied AI engineer can be disproportionately effective for a very large enterprise.

Second, human touch is critical in AI. So you're going to see V hiring. I actually want to use this opportunity to share with people who are watching or listening. We look for extremely competent deployed engineers. You still need people. If you work with a $50 billion healthcare or pharmaceutical organization, you want people who really live through their strategy and KPIs. It's not 20 people—it's one or two people per logo. But we're looking for people who have good technical and operational understanding of the customer but are also very fluent in AI and the technical stack.

Those are the two areas I think are going to unleash the next chapter at V.

Manav: If you're a smart engineer and want to be part of a unicorn, go to v.co. Where can people find you?

Omri Yoffe: We're at v.co. We're publishing most of those roles out there. I'm available on LinkedIn, so whoever wants to reach out, I'm very responsive to any question or anything we can be helpful with. You can follow us on LinkedIn, YouTube, and X.

Manav: One thing I want to ask you—what's one piece of advice for any founder or entrepreneur watching who wants to be the next Omri?

Omri Yoffe: Never give up. Don't be afraid of shame, humility, or optics. Focus. Make sure, as Einstein said, intelligence is the ability to change. So make sure you change over time, but never give up. Over compounded effort and time, you will prevail and win.

Manav: Thank you so much for coming on the show, Omri. I know how busy you are. Thank you.

Omri Yoffe: For sure. Thanks for having us.