从5年荒漠期走到产品市场契合与53亿美元估值：Abridge创始人Shiv Rao

When we created our company values, one of them is that you have to taste good things to have good taste.

Oh, that's a good line.

Now, joining me in the hot seat today, we have Shiv Rao, founder and CEO of Abacus. What I love about Abacus is, yes, it is a $5.3 billion company today. They have the best investors in the world in the company today, but they went through a 5- to 6-year wilderness period. They were founded in 2018, but now they are one of the hottest companies in AI.

Being one of the first in vertical AI is critical. When you saw the announcements from OpenAI and Anthropic about forward-deployed engineers and these partnerships with big private equity groups to go help their portfolios, if that wasn't a sign that there is absolutely an incredible opportunity for the foreseeable future for vertical AI, I don't know what is. We're in for a tsunami of healthcare needs, and the system is not prepared to deliver it.

Ready to go?

Shiv, I am so excited for this. People don't know, but we actually met because I used you as a reference on a company, and you were phenomenally helpful—to the extent that my partner was like, “Can we just invest in him instead?”

Which really would have been a much better idea, and I should have gone, “Yes, 100%,” and broken all the rules for it. Thank you so much for joining me.

1. Five Years in the Wilderness: Lessons on Market Timing

Oh, it's such a privilege. All right, you're a legend. Dude, I so appreciate that.

Let's start with one that I often think about. Great CEOs are motivated by one of 2 things, I've found: either the fear of losing or the thrill of winning. There's no right or wrong, but if you were to say which one you are, which would you say you are?

Oh, it's the thrill of winning, 100%. The high-stakes games, going for broke, trying to do something really big, trying to build a new product, and trying to bend a new market or create a new one.

You said “bend a new market.” Again, I sent you this beforehand, you write it on the fly, and now I'm just kind of going off script.

I think you started the company in 2018. Now it's 2026. You've been hot for 2 or 3 years. That's a 5-year wilderness walk. What are your biggest lessons on market timing as a result of reflecting on that?

I think that certain companies, certainly ours, have to have a true north. You live every day in anticipation that the sky is going to open up because you have this thesis about the market. You might not necessarily know exactly when it's going to come true, but you know it's going to come true.

When you feel that in your bones, then I think you can draw a lot of resilience. You can figure out that you just need to stay standing. You just need to not die. You just need to be there when it's happening. For us, it was absolutely that way.

It was this thesis, this idea, this secret hiding in plain sight: that this data set—these conversations—it's the most human signal, and it really powers healthcare. You can build a whole new set of rails on top of this signal. The thinking was, well, at some point in time, the technology is going to be right, but also the market is going to be open.

Would you have died on this hill? What I mean by that is, sometimes you can have a thesis, you can even be right, but the market doesn't move fast enough.

I would have died on this hill. I was willing to pivot on the specific product and what specific order we'd put a feature out. I was willing to pivot on go-to-market. Certainly, we learned so much. Healthcare is incredibly complicated. Figuring out go-to-market is a really big part of how you win.

But I wasn't willing to move on the thesis. For us, that thesis—that healthcare is about people and they're having conversations—I think was core to the identity, and it would just have to be an entirely new thing. We would just shut it down and start something new if there was something else we could get really passionate about.

You said before about the references that I take. I do. I stalked the shit out of you before this, which should make you feel very comfortable sitting in this dark room alone with me. End of the day in London, why not?

I spoke to Andy Weissman, who led your seed round. I think market timing's one thing founders often worry about. Fundraising's another really hard thing. Was it an easy pre-seed, seed, early fundraising environment for you?

I wouldn't say it was easy, but it wasn't hard at the same time. This was 2018, 2019. Our seed round was like—we raised $5 million on a pre of $15 million. So, not the world we live in right now, but it made sense then. We felt like this was the market, and we did a really good job.

I think that there is founder-market fit, but there's also founder-partner fit. When you just have chemistry with a person at a firm, you know you're going to find a way to work together.

With USV and Andy, I stalked Union Square Ventures for years ahead of the first meeting that I had with them and with him. I would read all of their tweets back when it was Twitter. They had this ritual where they would talk about music, and I was always really impressed that they seemed to pattern-match across totally different genres: country western on Monday, indie hip-hop on Wednesday, and Swedish death metal on Friday.

To me, that's how I enjoy music as well. My thesis was that if they could extract at that level, they could be the right tech investor to think about healthcare in a new, different way. So I found a way to get to them through an angel, through a friend at MIT. Then we went in and pitched, and I knew at the end of that meeting that we were going to work together.

So, would you not approve of my liking Celine Dion and Taylor Swift? Would that go against me?

Speaking of Andy, Andy says this thing. He's the biggest coffee snob in the world. I once asked him about exactly what his rituals are—the temperature, the bean, and all that stuff—and he was just like, “Coffee's coffee, man. I'll have any coffee.”

When you ask him about music, he'll say, “I like all the sounds.” So maybe on some level, you can see the patterns and enjoy anything.

I know it's very random, but what you said about coffee there and Andy made me think about taste. Everyone talks about taste today. It's the difference between those that are truly special and what separates humans and AI. How do you reflect on taste being the differentiator?

When we created our company values—this was before this whole taste cycle—one of them is that you have to taste good things to have good taste.

This idea of taste makes a ton of sense to us. It's not just judgment, but being able to see patterns and put things together in interesting ways that hit different, that feel different, that feel authentic, and that really represent a person or people.

I think that's where the truly magical companies are these days: where you can feel the human behind the product and feel the decisions that they made, the things that they said no to, and the strong opinions held tightly that they died on the hill for.

Taste—I love that. “Taste good things to have good taste.” Does that mean you should taste everything?

No, but I do think it means you've got to be discerning. If you're not exposing yourself to certain flavors, then you're just not going to be aware of them.

I think the way we translate that in the company, for example, is that we should be reading the latest arXiv papers about some new type of machine learning model that maybe we can leverage. We should be thinking about the latest UI/UX patterns out there and where we could take these primitives in our own specific space.

We should be living at the edge of culture if we also want to create it. I think the best companies, in some way, shape, or form, are creating culture.

Can I ask—you said “strong opinions loosely held.” What was the strongest opinion that you've had that's loosely held?

Yeah. The loosely held opinions—there are the tightly held ones, the ones that we would die on the hill for. The loosely held ones are the ones that I've shifted on pretty profoundly over time.

I've had loosely held opinions about where the company's limits are—how far we could expand and how quickly we could expand. I think a lot of our priors, all of our priors, are getting updated.

How do you feel about growing as quickly as you have done?

I'm incredibly grateful and feel incredibly privileged. But in some way, I feel like we're on this journey that we were meant to be on. We're doing this thing that we're supposed to do.

By “we,” I mean it's not just the company, but all of our health system partners. It feels like we're part of a movement that we're trying to create right now. So I can't say that I ever woke up being like, “What is going on?”

But I can say that I've gone to bed at night feeling like this is awesome, this is great, and we have to think about making it even greater. You wake up in the morning and you're not at all satisfied. But at night, there's one thing that one mentor, a cardiology mentor, once told me.

He's like, “Try to go to bed with a little bit of perspective and a little bit of gratitude, at least, knowing full well you're going to wake up with that fire in your belly again.”

As a founder, do you think you can ever be satisfied?

Probably not. You always want something more. You always want to build something bigger, and your ambition always gets bigger. Maybe it doesn't get bigger; maybe you just reconcile yourself to a change in scope over time.

2. Healthcare Adoption of AI Is Now Table Stakes

You said that about the healthcare partners you work with today. I'm an investor in Legora. By no means similar in terms of space, but it's a large space that's traditional and now seeing a tidal wave of adoption, similar to healthcare. Every single law firm right now is going, “AI—I need AI to stay relevant.” Is that the same with healthcare and healthcare partners, or is the downside fear factor making it slower?

No, 100%. It's every single health system right now. There are doctors right now we've heard who won't sign up to join a health system, a hospital, if they don't have a bridge. It's one of those moments where, all of a sudden, almost overnight, over the course of 3 to 4 years, something just became table stakes. It's part and parcel with how care is delivered, especially in the outpatient space.

3. The GTM Trap Most Healthcare Founders Fall Into

Mhm. What did you not expect in GTM that surprised you most? This is where strong ideas held loosely.

When you first raise capital, maybe you hear an investor give you advice about starting down-market and swimming upstream, disrupting over time. Get that PMF, find those fast feedback loops. It's true to an extent, but you just have to be really careful and mindful because healthcare, specifically in the United States, is not one $5.3 trillion market. It's a bunch of different markets, and depending on who you're trying to serve, you have to be really careful about how you segment.

If you're trying to serve clinicians—doctors—there are about 1 million doctors in the country. Maybe 800,000 of them are actually practicing. The vast majority of them are concentrated in large care delivery systems. They're called integrated delivery networks or payer-providers. They also have payer arms, insurance arms, or they're academic medical centers like the Emorys, the Yales, the UCSFs, and beyond.

When they're concentrated there, you have to think about getting there as fast as you possibly can. Obviously, you've got huge ambition. You want to create as much impact as you need to. I think the trap that a lot of healthcare founders fall into is that they stay down-market. They don't time their YOLO shot to go up-market at the right moment.

For us, things really worked out. We're incredibly lucky and incredibly privileged, but things worked out. A lot of different things happened in the marketplace at the same time. On the other side of the pandemic, I'd say 40%–50% of the doctors in this country were saying that they were burned out. There was a study about nurses recently that was published in JAMA that suggested that 30% of nurses in the country didn't want to be nurses in the next 12 months.

There's all this burnout and fatigue. There are incredible financial pressures that have only continued since then, coming down from, for example, Medicare cuts. At the same time, there's this new technology. There's AI. We were in that moment where, without even knowing it, we were pre-selling the market and preparing the market for what we could bring. Then, in 2023, when the sky opened, we ran right in.

I want to unpack a load of things. How important is being first?

I think it is critical in vertical AI. I think being one of the first—being early—I think being late doesn't work. But there's that refrain that being early is also being wrong. I think we're a good case study in being resilient. You can overcome that issue of being perhaps too early.

There are maybe 3 variants of an AI-native company. In my head, there's a post-Transformer-paper, pre-LLM company. There's a post-LLM, pre-agent company. There's a post-agent company. Depending on your vintage, you have to make sure that you become the latest variant as fast as you possibly can.

That might mean that your product evolves pretty significantly. It might also mean that the way you organize and operate your company evolves pretty significantly, as we're seeing in this new era with agents. We were a post-Transformer, pre-LLM company. As soon as the LLM moment happened, we became that as fast as we possibly could in all the different ways. We've been doing the same thing in this new world with agents.

Totally understand that. Can I ask you how you think about it? There are a couple of different ways that you can break down competition.

Yeah.

3 come to mind for me, and you can tell me I'm an idiot. Very likely, as a venture investor, one of them is the age-old question. I see Matt from Legora pushing back against this now: Anthropic can release a legal product. The foundation models could very easily do an Abridge and build out a product suite. How do you fight back against foundation models that will build healthcare applications and replace you?

4. How Being First in Vertical AI Creates an Unbeatable Moat

If you are fighting against them, you've already lost. If you haven't figured out how you're going to win with them, are you going to just coexist but actually find ways to collaborate, potentially? If the tailwinds that they create are not yours to leverage, then you're screwed.

5. If Foundation Models Are Your Enemy, You've Already Lost

What does that mean for you, then, if we take—

Yeah. Healthcare—we say this is a vertical AI company, but think about it: healthcare is a $5.3 trillion market. It's about 18%–19% of US GDP. It's a big part of the US economy. In a sense, I don't know if it's a vertical AI company. We are an AI company, in my head, serving one of the biggest opportunities that is out there.

In this market, you can go millions of miles deep. Especially on the enterprise side of the spectrum of healthcare, you can go millions of miles deep in a regulated industry with proprietary datasets that you can build into very, very specific workflows in a way that's really, really hard to replicate.

Being fast and getting scale very, very quickly is crucial because scale enables you to not just build more products and get more surface area, but we do a lot of mid-training and post-training in the company. On the post-training side, it's being able to learn from all the user edits on a daily basis. It's being able to be useful for all the different types of doctors, in all the different settings where they deliver care, and all the different spoken languages they might use with their patients.

People in 2023 thought that was the last mile. That's actually most of it. It's that part. The model piece is much less. But even on the model side, I think the vertical AI companies that have the most leverage and the most upside are the ones that can reach farther down into the stack and own their destiny in a differentiated way, that can control their P&L.

How much of the quality of your product is derived from the quality of the model that you sit on top of?

About 40% of our model outputs inside our product are generated by in-house models.

Huh. What do you expect that to be in 2 to 3 years?

That's a great question. I think this is where nobody knows the answer, but you just have to be principled in how you approach this. There are any number of different problems we solve as a company with our product.

Some of those problems you can imagine ringing the bell relatively easily. If we help the nurse with this piece of data getting into this discrete field at this time in their workflow, we won the game. There are no bells and whistles around this. Once you do it, you've done it. It's binary.

Those sorts of tasks—being able to crush it with an in-house model—make total sense for a lot of different reasons, but most importantly for the user. Chances are it's going to be faster. You can optimize that model over time from not just a latency-per-token perspective, but also from a cost standpoint.

Once you've done it, too, you can set and forget to some extent. “Okay, let's move on to the next challenge inside this new product that we're delivering for nurses.”

But then there are certain challenges in your product suite that you're never going to be perfect on. Every week, every month, you want to be able to look back and say you're less imperfect than you were before, and it's going to matter. It needs to matter to the market.

If the market cares about you always improving on something that you don't think you'll ever be perfect on—and healthcare has a lot of use cases like this—it makes more sense to ride the frontier wave.

Why did you decide to build your own models when Caresso did, and TBD on whether that was actually effective, when you have the historical precedent to say that underlying models move faster and more efficiently than you could create them yourself?

Well, number 1, it starts with the end user.

The milliseconds matter when you're in workflow. These are doctors who have no patience for new technology. I think our refrain inside the company is, “Reduce, reduce, reduce,” from a product and user experience standpoint. We want to be like good air conditioning, where, when it's set right, we're in the background and you're just focusing on everything that's more important.

Oh, that's a good line.

And to get to that, though, you really need to think about being in workflow. The Thinking Machines stuff that just came out is really fascinating. It's like an agent that's in the moment with you, right? But that kind of experience needs insane performance and latency, and that's something that we weren't able to achieve in an easy, elegant way with frontier models.

For certain use cases, there was really no choice. It started with the end user. We were like, “Okay, to deliver this kind of value, this specific type of value in workflow, so that before the patient leaves the room, the doctor is okaying an order or putting in the right visit diagnosis that ends up informing billing and their own reimbursement”—these are high-stakes workflows. You need to be fast, and so that's where things started.

But quality is a part of it, too. A crucial part of it.

Yeah. What does the relationship look like for you between you and frontier models, and you and open-source models? Has your reliance on one changed over time?

It's dynamic. Inside the company, we always remind ourselves that the only thing that matters is the end user. We just have to have the best product, whatever it's going to take. There cannot be any pride of ownership here. It's just about what gives us the best user experience.

That has ended up serving us really, really well, because especially over the last weeks, there's been a lot of angst about perhaps costs going up in the coming weeks and months and what's that going to do?

How cost-conscious are you? If you look at open source, obviously, it's much cheaper to run versus being reliant on a frontier model. Do you care if the output is different?

I don't care now, but I also know that my future self might care more. That attitude is, again, I think, just putting the user first. In our specific space, given our product and where it sits in workflow, that has allowed us to build at a layer of the stack that gives us agency on the P&L that we wouldn't have otherwise.

That's awesome, but that's not what we're optimizing for right now, obviously. It's nice that we have those little levers.

I have this conversation with a lot of founders, and it's the same as this: “Bloody hell, we've got a lot of funding. We don't need to optimize for costs now.” Absolutely the right answer. When do you?

When it gets in the way of our ability to create more impact—to raise more money, to impact more users. I don't think it necessarily has to do with anything around IPOing.

One of our investors is Henry Kravis.

Yeah, he's the dude who created 2 and 20. Amazing.

Yeah, I'm a student of venture.

That's like saying you created a continent.

Yeah, yeah, yeah. It's so incredible, but he's such an approachable person despite everything he's accomplished. I remember sitting down with him in the first meeting, and one of his questions was, “Do you want to go public?”

I remember starting to clumsily navigate an answer on the fly, and he interrupted me and said, “No, you don't. You don't need to. Why are you even thinking about it?” The right answer was, “No, I don't need to. I have a mission, and to serve that mission, I will do what it takes. If I can do it in the private markets, I'll raise money there. If I need to go public to do that, then I will go public.”

I think one of the biggest barriers to AI progression within large enterprises in the U.S. will be data cleanliness and data structure within large enterprises. Is it a mess in large healthcare organizations in terms of data preparation, data structure, and data cleanliness?

Yeah, it is. When you saw the announcements from OpenAI and Anthropic about forward-deployed engineers and these partnerships with big private equity groups to go help their portfolios, if that wasn't a sign that there is an incredible opportunity for the foreseeable future for vertical AI, I don't know what is.

Help me understand that. Why is that?

It's not easy to go into one of these enterprises and figure stuff out: to get access to the data, to clean the data, to organize the data, to be able to integrate into specific workflows, to do it in a compliant way, and to check off all the boxes. It's not SOC 2; it's HITRUST. It's all the cybersecurity stuff.

It's being able to know that behavioral health information from a psychological encounter, for example, should be treated very differently than data that's coming from an internal medicine primary care physician's encounter. There's just so much to it, and it takes a lot of effort to build a machine that's scalable.

Do you have to have an FDA process to be successful selling to enterprises with AI tools?

No, you don't. Not in our case. You've got to get to that zero to one. You've got to get to that first product in all the ways: people in the health systems, in the clinics, listening to the doctors and the nurses, figuring it out, and building a product that fits.

But that's not a forward-deployed motion. That's just figuring out how to build a product that fits. If your first product is something that is everywhere, if you've attacked a workflow that can scale, then you're set.

It's like the classic motion: we picked a first product that we knew was everywhere in healthcare. Practically speaking, every single doctor has to speak to a patient. Great, that's our signal. We're going to attach ourselves to that spoken signal and create value.

What's our first bit of value? It's a note. Got it. Do they hate notes? Yep, they hate notes. We can automate them. Great, we created notes. Now let's go scale notes.

Then it was like, “What's next?” They hate placing those orders. They have to hit all kinds of drop-down menus and figure out what the right order is. Cool, let's do orders.

They hate billing. No doctor went to billing school, accounting school, or revenue cycle school. When I was a corporate VC at a large health system, we used to have these lunch-and-learns as a doctor where we tried to teach doctors about revenue cycle with pizza and PowerPoints. Every single doctor had a thousand-yard stare. Nobody wanted to be there.

So, can we automate that for them? Absolutely. Now we've gotten to this point over the course of a few years where every single clinician has this team of assistants that we're deploying in the background to go do all the jobs. Many of these are jobs that nobody wanted to do in the first place.

I spoke to one of your investors, who we'll remain nameless for this one, and they said, “Amazing products and amazing everything that you said, but if you are to fulfill the enterprise value that you have today and that you will have moving forward, you have to move closer to the flow of money.” Is that fair?

I think it's not only fair; it's also what we want to do. When you think about it, in the United States—but really, this is true in many parts of the world—clinicians and doctors aren't compensated for the care that they deliver. They're compensated for the care that they documented that they deliver.

When we first realized that we were going after this spoken signal, and the first thing that we were going to build with it, before, during, and after we captured it, was a note, we knew we were billing. We were creating bills, too.

You've got to take that responsibility very seriously, because the enterprise motion in healthcare is complicated. You have a lot of different stakeholders. You have your end users, but then you have the decision-makers. There's a CMIO, a chief medical information officer. There is a CIO, a chief information officer. And then there is a CFO. They all have a different lens.

Being able to thread the needle and resonate with all 3 is really how you win the day. Being able to recognize from the get-go that these notes are actually going to end up being bills means we need to approach this note-generation architecture, that workflow, and what's now become a much more agent-driven process very deliberately, so that we have the immediate extension afterward into the CFO's world.

How do you compete when Epic does the same product? They're the product that no one likes, but it's so entrenched—so entrenched. How do you compete in that world?

We're not competing with Epic.

Do they not have a competitive product to you?

They have a product or a feature. They have something out there now that's been out for a beat that helps with notes. But what I was explaining to you earlier is really, really critical. For us, it was never about the note. It was always about the signal—the conversation that we were using as the wedge.

That conversation, that spoken signal, allows us to build into any number of workflows, even beyond notes.

And when you're building notes with, for example, all of the revenue cycle that comes next, you build them differently. And so it's a moment where, kind of like the foundation model company, we don't ever want to be an electronic medical record.

By the way, I grew up with Epic, and I was at a health system that had multiple medical records. I was always the happiest when I was using Epic compared to any others out there. So we build on top of them, but the layer of the stack that we are building is the intelligence layer.

The wedge that we've chosen is the conversation because it's that sacrosanct moment in health care where the actual value is getting exchanged. It's where the doctor and the patient, or the nurse and the patient, are talking about the care plans. Being able to capture those conversations in a way that's not just clinically useful but compliant is a way to build out that layer over time, and you can extend very easily.

I think lessons are often learned from reflecting on what you did that you shouldn't have done. What did you do that, with the benefit of hindsight, you're like, “Mhm, it would have been better to choose an alternative path”?

During the pandemic, I think everybody was just figuring it out. I think perhaps we could have doubled down on research and just hibernated for a beat in a research cave.

We started our company 3 months after the Transformer paper, and a lot of the tech that we leveraged at first was fine-tuning BERT models: BERT, BioBERT, Longformer, Pegasus, and T5. We published a paper in 2021 around how you could actually do some of the jobs we do today at scale, but do it with models that predated LLMs.

There's a world where we could have done more R&D, but the decision that we made pretty early on, even before the pandemic, was that the barrier to entry on the doctor side was so high: let's go build for patients. So we built a direct-to-consumer app that would allow anyone of us to ask our doctor, “Hey, can I record?” and capture the conversation and create a summary for themselves.

I remember when we first pitched USV for that seed round, one of our slides had a doctor on one side, a patient on the other, and there was this mobile phone in the middle. We were like, “We can help both.” Both sides need help, both sides want agency, both sides want to own and control the story or have that story, and we can do that. We can serve them with AI.

We started on one side. I think there's a part of me that thinks that we could have perhaps built differently there on the patient side. It's super core to who we are, and we'll get back to that. Actually, we're getting back to the patient side of that slide now again.

But it was a strategic mistake then to build that patient side. I think that, relatively speaking, maybe we could have spent a couple more cycles on R&D as opposed to—I mean, we were really banging our heads against the wall trying to figure out business models that could work. Maybe that's the mistake. It's not actually that we built in that space; I really worked hard to figure out if we had a company there, and we didn't.

Because the companies that you can build on top of just the consumer in health care are often the kind of companies that paint themselves into a corner and can never be trustworthy enough to go enterprise. They're the kind of companies where the business models would require you to take multiple showers every single day because you're selling really private, sensitive information to who knows what entity. We didn't want to be that, but we realized that too late.

I think the pivot that I should have made is probably, okay, we built something really valuable here, but let's not lose sight of where the actual revenues are going to be. Back to your original question, I think I feel very strongly about what the thesis of the company was, but we pivoted a lot—not just on go-to-market, but also on business model.

You said about the kind of business model pivot that I think a lot of people are talking about: “Oh, the rise of consumption-based pricing. This is an incredibly innovative thing.”

Yeah.

Which is not. Twilio's been around for a while.

But AI-type nerds love that.

And CFOs—you mentioned different stakeholders. CFOs want to see pricing. They want to know that, like, “I've got $10, so I've got 10 bills.” What are your lessons on the pricing mechanism of the future for many of these frontier businesses?

Well, it's interesting. I think we're still learning. We started with a keep-it-simple kind of idea, a sensibility: per seat, and enterprise licensing the entire system.

The first big competitor for us was Microsoft. They had bought a company called Nuance for $20-some billion some years ago, and they had a product in our category. We still compete with them in the marketplace. But I think we've established ourselves as a completely different type of company with a completely different kind of offering, and at this point in time, we don't approach them the same way we used to.

So you don't buy bundling as a real threat? I remember—I'm completely forgetting Nabler. Nabler's a competitor of yours, I think.

Yeah.

Candidly, I looked at them and I was really concerned that bundling would be a very prominent threat to their ability to dominate the market. Was that a wrong concern of mine?

6. When Doctors Use "Abridge" as a Verb

We don't see either of those companies much anymore. I'm not sure if it has anything to do with bundling. I think it has to do with us being able to carve off a category—create a category that I think has resonated with the marketplace.

I was 2 days ago in Atlanta at Emory and was on some panels over there. One of the doctors on the panel used a bridge as a verb. Then another doctor later used it as a verb. I asked one over dinner, “What do you mean when you said that?”

When they described what they mean by that, they mean something that's a lot more than, “Oh, it created a note for me.” It's like, “Oh, it did all these different jobs for me, and it unburdened me and allowed me to just focus on the patient in front of me.” I think that's what's allowed us to win.

Back to your pricing question, I think there's a part of me that thinks that there are a lot of complex models out there, and especially the Microsofts of the world just obfuscate things so much and make things so complicated. We've won so far by, I think, even competitively positioning against that.

But there's a part of me that thinks that we need to continue to take notes here, because I see some companies in other verticals with incredibly complex structures, find that it's working, and wonder if there's something to it.

I always want my main course to be someone else's dessert because I'm going to care much more than they do. You care much more than Microsoft does about this market, by the very nature of the fact that it's all you do.

Any advice to founders listening? We have a lot of companies where it's like, well, they kind of could do, and they kind of do have, a competitive product. Should we just run our own race? Should we be unwaveringly focused on them also? Any advice to them if you're sitting down with them?

I mean, you want to compete against them. You obviously want to kill the competition. But the way you can kill the competition is mostly by focusing on your product and keeping in mind what they're doing so that you can competitively position, but also counter-position.

Big companies oftentimes have a lot of soft spots that you can counter-position against, where they just have to watch you.

When you say counter-position in that way, I know because I love Hamilton Helmer's 7 Powers. Can you explain what that means and how you think about using it to win?

When you're building in a way where the competitor couldn't build because it would impact their current business, it would hurt them, it really pushes them to think hard and probably prohibits them from directly competing against you.

What could you make a large amount of money on today, but for some reason you do not, and why is that?

We don't make any money selling data. We don't because trust is everything in health care. The industry moves at the speed of trust. All the platitudes—it really does. It takes so long to build up.

7. Why Abridge Will Never Sell Its Data

We've been able to build it up, I think, in record time, relatively speaking, in this industry. We've done in 4 years what a lot of companies take 15 to 20 years to be able to do, and that's all thanks to the people inside the company, the relationships that they build, and at the core, the product that we've been able to deliver and the value that we've been able to create.

Can I ask you a question? Is that a moral question for the greater good?

Yeah.

What I mean by that is, if you were to create a market for health care data, you would actually be empowering frontier models to create better models for the world to benefit global health care services. Is there not a moral—

But how do you do it, though? How do you do it? Who do you partner with to do it? And who's involved?

Obviously, Sam Altman.

Duh. But who's a stakeholder? For us, when we build new features that leverage data, we have a refrain in the company, which is: earn the right.

So we always go back to our health system partners, and oftentimes they know about our roadmap in advance. They’ve already blessed it, and we’ve already papered this into the contract: that we can build XYZ with the insights that we get so that we can improve the product or build other adjacent products that create whatever value.

But building that foundation model is, like we talked about before, where we do reach lower into the stack. We do train our own models, and to some extent, these models are the real workhorse for us. So could we build a model that ends up serving that kind of function for the greater good? Absolutely. We would just want to do that with everybody’s eyes wide open, all of our partners saying, “Let’s go. We want that too.”

In this industry, it works. People kind of get it. You can have access to Option A—frontier models before anyone else for 6 months—or Option B, a talent pool of the best researchers and engineers for 6 months. Which would you rather?

B.

B over A?

Yeah, no question about it. No question about it. Because ultimately, it’s always about the people, and if you’ve got incredible people, you’re going to build your own models, your own primitives. And yes, the frontier models you’re going to be able to ride, but that arbitrage, that period of arbitrage, you can—you know, that’s not the be-all and end-all for us.

Is the talent marketplace as hard as everyone suggests it is?

Yeah, absolutely. I think we have the benefit of being able to release a good amount of oxytocin for candidates. We’re a purpose company. We’re a meaning company. You can be post-money but still want to put your best years into this company because we’re at scale and we’re trying to do 3 things.

We’re trying to save time for the people who matter most in health care. We’re trying to save money for the system—we need deflationary economics in health care. We talked about the portion of GDP it represents, but we also want to save lives. You talked about the greater good. We want to help clinicians feel like superheroes.

8. AI Won't Replace Doctors

We have a feature, for example, that we just released, where a doctor goes in and we give them cues on what questions they should ask or diagnoses they should consider. And we’re doing it in a totally differentiated way, unlike a lot of clinical decision support products out there. We’re using context; we’re engineering the context about who this patient is. That context comes from all those different systems of record and from the conversation.

With the greatest respect, dude, if you’re giving them prompts for questions and diagnoses, for goodness’ sake, just replace them.

It depends. Every clinician—if a bot can do their job, I think almost every clinician will say, “What are you talking about?” There was an article in the Journal of General Internal Medicine that was published a few years ago that suggested doctors need 30 hours a day to get all of their work done.

So, understandably, we’ve looked at the paper, we’ve parsed all the different tasks that lead to 30 hours, and we’re picking them off. We’re creating these teams of assistants, of agents, that can go and get all the pre-charting done for the doctor before they walk in, tell them what questions to ask, help them create their note, do their billing—all the other things that we can do, we absolutely will try to do.

At the same time, though, I think, to your question, there’s high-frequency, low-stakes care that absolutely needs to be automated. Right now in Utah, for example, there’s a bunch of controversy around companies being able to—or starting to—automate medication refills. Can AI automate prescription refills for a patient? That’s, I think, as low-stakes and high-frequency as it gets.

But it’s those types of workflows that will progressively get automated until we get to some approximation of maybe the most boring variant of primary care that nobody wants to practice. But as soon as you get into the hard stuff—you talked about taste and judgment earlier—that’s where I think things get very real. And that’s where, invoke Jevons paradox or whatever, but we’re in for a tsunami of health care needs, and the system is not prepared to deliver it.

I unwaveringly don’t worry about, like, “Oh, we’re going to fire doctors because we don’t need them.”

Yeah, yeah, yeah. You know, there are people out there who think that. Have you ever been to, like, an emergency room?

There are people out there.

There are people out there.

Yeah, yeah, I know. They need to go to an emergency room on Friday night.

Goldman Sachs recently—I was reading an article by Goldman Sachs. Weirdly, I have bedtime reading that I queue up every night. I need to get out more. I used to party; now I have bedtime reading where I save articles during the day, and I was reading one last night.

It said that they estimate agents, within a 5-year period, will increase token consumption by 24x.

Very specific. 24x. 25 would be exaggerated. Do you think that is overplayed, underplayed, or about right?

I don’t know about the specific number, but directionally, I think so. I think that you just can’t get enough of this technology once you start to use it. Even as inference costs come down, you just end up using more and more and more of it.

I think for certain jobs, especially the jobs that you’ll never be good enough on, those are bottomless pits. But I think, now more than ever before, even reflected in the way people are organizing their companies these days, you can cover a lot more surface area. So you just do that much more.

9. The Hardest Role to Hire: High-Judgment Executives

What’s the hardest role to hire for today?

I just think getting really amazing executives—really high-judgment executives—into a company like ours is critical. And part of it is because this moment is pretty unprecedented.

I think being able to have folks who can serve as guides, who have a lot of patterns that they can match against—priors—but who can also go against those priors and reflect upon them, is important. The time between a decision and an action is getting compressed.

Obviously, product development—we do things differently now than we did a year ago, even. Every layer of the company is getting compressed. And I think getting high-judgment people—incredible executives—is still a thing that I’m working on.

How do you think about that? Brian Chesky obviously speaks about founder mode and the importance of getting as close to the bare metal as possible. How do you think about that, on the one hand, with the need for exact pattern-matching playbooks that you mentioned there?

I don’t think they go together. I think founder mode, to me, is about tours of duty. Depending on what’s on fire inside of the company, or what’s the most important thing and what you’re best at, you just go crush it. You just have to go kill whatever that challenge is.

10. What Jensen Huang Told Shiv at Midnight

It doesn’t mean getting into the details and micromanaging everybody in the company. It means doing the tour of duty when it makes sense.

And NVIDIA is one of our investors, and Jensen told me this. He once called me at midnight, and it was on his way home—a cold call—and I knew it was him. I picked it up, and it was awesome.

Were you in bed?

I was. And I got up.

All right, it’s, “Hi, darling. It’s Jensen.”

Yeah, exactly. You’re going to apply this even to your wife, aren’t you?

Yeah, that’s actually what I said. “My buddy Jensen.” That’s actually what I said. She was like, “Whatever.” I was like, “Hello, Jensen.” I’m whispering to him.

What did he say?

He was calling me back because of an email I had sent him. I’d sent him a 2-line email earlier in the day, so his SLAs are insane. He responds that fast; it’s really wild.

He wanted to unpack a challenge that I was experiencing, but one of the lessons for me that day was just: your job is to fall in love with whatever the job is. And that is something you can do. You can convince yourself.

You have to find a way to bend your DNA, to CRISPR new DNA in that loves living on airplanes. I live on airplanes now, and I think if you caught me 5 years ago, I’d have been like, “Never. We’re not going to do that. We’re going to find other ways to grow and scale.”

But this is what the job requires, and I enjoy it.

What have you not CRISPRed your way into yet that you need to?

I think, yeah, it’s a good question. I’d say people’s strengths are their weaknesses, too. There’s a shadow to what you’re very, very good at.

I think I have HR reporting to me. That kind of stuff is not the kind of stuff where I think I’m operating at the top of my license. That comes to mind.

I said the other day that no great CEO ever liked HR.

Yeah, I’ve never gotten so much hate back. I mean, it was just surprising. I thought it was wild. We all joke about Karen in HR, you know? But everyone’s like, “You’re a dick.”

I’m surprised by that.

Oh my God. Oh, savage. And then I said, “Listen, if you want to work from home on Friday, that’s okay. But just accept that, bluntly, you’re not going to work as hard. You’re going to go to Pilates at 10:00, and working from home Fridays is effectively 4 days a week.”

Are you 5 days a week in the office?

We’re 3 days in the office.

Why?

We just have certain folks who have to travel long distances in order to get into the office.

And so, we're still adjusting what exactly the sweet spot is.

Is that okay? Would you not rather have 5 days a week?

I love being around the people. I love being at the whiteboards. But I'm also living on airplanes. I also just know how much, especially EPD, benefits everybody in the company. At this moment, things are moving so quickly that the more time you have in person, the better. There's no question about it. We're not a top-down 996 company, obviously, but we don't have that kind of culture.

I think there are a lot of people who are token-maxing, and it's Goodhart's law, you know? Do you worry that, at your size, you might meander into slowness, process, and policy? How many people do you have?

450.

450. Yeah. Is it possible to only have A players at 450?

Honestly, it's a great question. I think so.

You do?

Yeah, I think so. I think it takes everybody in the company continually looking at what their bar is and reassessing.

When did you stop meeting everyone who joined?

It's been a long time.

It's been a long time. At 100?

Even probably a little bit before that, honestly. It was around then.

Is that right? How would you advise other founders on how to think about that?

I don't think there's any playbook, honestly. I think it depends on who's around you, too—what executives around you can do a lot of that work as well. I think culture is critical, and no matter what the founder CEO thinks, they play a huge part in setting it.

There are a lot of different ways you can convey what you believe is going to be the culture that's going to help you win and achieve your mission. Meeting every single candidate isn't necessarily, in my estimation, one of them. If you really trust your hiring managers and you really trust your executives as carriers of that culture, you can work it out.

What is the most non-obvious but core cultural element that isn't standard? Everyone's like, you want ambitious people, people that care, mission-oriented, all that. I want a weird one. For me, it's called the Titanic rule. I mentioned Celine Dion earlier—we love Titanic here. Great movie. It's 3 hours long. Everyone should be able to go to a cinema with their families. But if you're asked a question on WhatsApp, you need to respond within 3 hours. It drives efficiency and urgency.

Yeah. Titanic rule.

Anything that guides your principles that's less obvious, special, or countercultural?

11. Chips on Shoulders Make Chips in Pockets: The Hiring Philosophy

You know, Josh Wolfe from Lux—I remember meeting him years ago—and I think he says this a lot: “Chips on shoulders makes chips in pockets.” I think that's a hard thing to be able to assess in a person, but there are just certain people who have insane slope, and that slope is coming from something very, very deep and core to who they are.

They're just on a mission, and you see them and you just want to invest in them. You want stock in them, and you sort of feel it. I don't think it's that you're looking for—I don't know. It's like there's a bunch of VCs out there looking for broken people.

We were talking about psychological assessments and things before we started. Some VCs will do that, and I had one before we took money from one of our investors. But I think it's just looking for a level of fire that's going to translate into not only those SLAs you talked about, but resilience and grind. It's always wartime, and it's a different kind of war now than ever before. Not everybody's a warrior.

Are you a wartime or a peacetime CEO?

Oh, come on.

Well, come on. You're a wartime CEO? You can't be.

Yeah, there's no other way to be. You're always competing. You're in a market that is moving so fast. Who's a peacetime CEO? Even the Costco CEO, who I think is awesome and undercelebrated—is he a peacetime CEO? Maybe he is. Maybe he's a good peacetime CEO.

If the founder transitions out of the CEO role, I'm out. Do you agree with that?

Yeah, I definitely believe in those types of companies. Those are the types of companies I want.

What stage of Abridge's development have you struggled with most?

You know, I always think product-market fit is a series of chapters that we always get wrong. It's like, “Oh, I had product-market fit,” and then, boom, you're constantly finding new product-market fits.

What chapter of company development did you find the most challenging to grapple with, and what do you know now?

Honestly, the chapter of the company that sucks the most is the one where you don't feel pressure. We have a saying in the company: pressure makes diamonds. When it somehow feels like you're on cruise control, you're on autopilot, that kind of sucks.

I think you want dynamism. If you're not figuring out how to 10x your ambitions, if you're not figuring out how to think outside the box or find new pockets to pull from, new value to create, and you don't feel that pressure for some reason—if you're coasting—I think that doesn't feel good at all.

What was the largest round you've raised?

12. Raising $300M: How to Stop Spending From Getting Loose

Our last one was $300 million, about a year ago.

When you raise $300 million, how do you prevent expenses and budgets from becoming too loose? It's a lot of money, and I see it with a lot of companies where suddenly spending just becomes loose—events, travel, hotels, everything. How do you prevent that?

You hire really principled finance leaders who can work together with you to unpack what the goals of the company are, and then recognize that against certain goals, you don't want to blink. You raised this capital because it's a blunt instrument that's going to allow you to win. Against certain things, you just don't want to even debate. If you are, you're doing it wrong.

Having the right business-minded folks and finance leaders means that you should probably also have the discipline to know that you didn't need what I just ordered for our New York City office, like some fancy speakers. But sometimes that stuff creates culture, too.

What was the hardest round to raise?

The hardest round was a Series A1. You know venture capital; you obviously know it was a weak round. We were in a weak spot because there's a 1 at the end of it. A1 is terrible, and people are like, “Oh my God, what the fuck is this? Obviously, things are tough over there.”

But those first years—we talked about how the consumer-product business model wasn't feeling right, wasn't feeling big enough. We got a handful of no's on that round, for sure.

Does USV not just guarantee your next round? They did the seed and the A.

USV was our business model initially. We had other great investors like Bessemer and Pillar, and plenty of others, but we were pre-inflection. We knew it. We could feel it.

What was happening in that A1?

We found folks who believed. Whittington Ventures, for example—we did a round at a 2x. Folks believed that it was coming soon. We got it done.

$8 million on $40 million?

It was at $100 million.

Oh, wow. Yeah, that's pretty good. Oh, man. With Whittington Ventures?

Yeah. They did it alongside USV and Bessemer and some others.

Do you ever worry about becoming the jewel in a VC's portfolio?

How so? What does that mean?

When a VC realizes that you're the kingmaker, all the pressure goes on you. If I've got Abridge in my seed fund, you are my multiple fund returner. You're the only one that matters now. My attention, pressure, and focus concentrate on you. It's kind of like the spotlight shining on the one that works.

No, I don't think about it at all.

You don't think about it?

No. I think we have a job to do. We're doing it. They're part of it.

Do you notice the power transition changing from VC to founder, where suddenly you're invited to the AGM, you're invited to speak at their event, and you're invited to do their reference calls?

Yeah, I think in 2023 we did 3 LP meetings. That's when I knew things were going well. I don't think about it.

13. Being a Great CEO and a Great Parent at the Same Time

You have 3 children as well. I'm always contemplating family. I think it's one of the most important things. Do you have any lessons for me on how to be a great CEO and how to be a great parent?

I think there are trade-offs and sacrifices, and the folks who say you can have everything are lying. You have to be eyes wide open on what you're giving up. You also have to be eyes wide open on the high-level perspective.

I have a mentor who said, “Just go to sleep at night thinking about PPG: perspective, purpose, and gratitude.” The perspective I have is that this is what powers me. I have to do this work, and I also love my family and they come first, but sometimes there are trade-offs and I can't be at the kid event.

I live on the road. I'm traveling 5 days a week. Basically, I'm in San Francisco Monday, Tuesday, and Wednesday, and then I'm probably seeing customers. Saturday and Sunday I'm with my wife and kids. I'm missing a lot, but it's just one of those things that you have to come to peace with and recognize, too, that there are times in life—and this is one of those.

Do you know who Luis von Ahn is?

Yeah, of course.

Do you remember I interviewed him? I remember in his S-1, he's from Pittsburgh as well.

Duolingo was also Union Square Ventures-funded, and I'm sure that had something to do with USV being more comfortable with me, because I certainly didn't have a track record ahead of that seed round. But he has that one line in the S-1 about how the only thing you need to know is that I'm dedicating my life to this. I think, on some level, my family and I are all dedicating a portion of our lives to this. We're all eyes wide open on the sacrifices that we're making.

What are you not willing to miss?

For me, I walk a marathon every week with my mother on Saturdays. I'm not willing to miss that. I'll get the super, super-late flight back Friday night. I'll not go—I don't care. I'm not missing Saturday morning.

What's your “I don't care, I'm not missing”?

Yeah, Sunday with my dad and my mom. That's a big part of the reason that we're still in Pittsburgh, honestly. Although I tried to convince my 15-year-old daughter to move to San Francisco recently. It didn't work, but I think we're on the fence.

A big part of the reason we're anchored there is aging parents. My dad has heart failure. I get to see them, and I get to bring the kids to them. It's the highlight of their week, and I get to do that every week. I'm happy to live on airplanes to make that happen. That's the thing I'm not willing to give up, and it's an incredible opportunity and privilege that I get to do it.

Final one before we do a quick fire. I think San Francisco is the worst place to start a company because it's so impossible to acquire and retain talent as a startup. Think $5 million in seed funding going up against Abridge's $300 million fund and OpenAI and Anthropic. Do you agree, or am I short-sighted and wrong? There are so many cracked early-stage teams over there.

The reason why I don't agree is that this moment is so insane. The amount of people, the amount of ideas, the amount of creativity, and the amount of energy right now that's concentrated in that area is just wild.

We went against the grain when we started. We started with a lot of Carnegie Mellon DNA. We started in Pittsburgh, and we built in Pittsburgh for years. Then, with this AI moment, it just became so clear that we had to move in 2022, 2023-ish.

Should you have moved earlier?

14. Quick-Fire Round

No. We did it exactly at the right time. As we were inflecting, we just needed to be in that moment. The amount of ideas that you osmotically absorb by being there, the kind of talent that you have access to, and the kind of mentorship that you have access to—I think it's super important.

I'd love to do a quick fire because I could talk to you all day. What have you changed your mind on in the last 12 months?

I've changed my mind on how a company can operate—how it should be built and how flat it can be. Now I'm in this very idealistic mode that we can be a very, very flat company in the near future.

What does that actually mean in reality?

Just fewer managers, a lot more ICs, and super ICs taking on that much more. Everybody, with these tools, is able to move in lockstep and be coordinated in a way that was impossible before.

I tweeted the other day that my single best advice to a graduate is to be full-stack. If you're a marketer, be the one who creates the copy, ads, and video.

Yeah. Super IC. Totally agree.

What's the kindest thing anyone's ever done for you?

I think, because we're talking about venture and raising, one of the most generous things—maybe not kindest, but one of the most generous things, in retrospect—was people betting on me when I had nothing really to show for it.

Is king-making real in terms of an investor brand denoting a winner in a space?

It's less the brand; it's more the placebo effect. It's more the psychology. When you feel inevitable, the more you can feel inevitable, the more you will be. When you're able to spend time with folks who you really admire, who've got incredible track records, and they've bet on you, that starts to seep into your psyche.

Do you feel inevitable now?

I feel our mission is absolutely inevitable. I think we're in a war to make sure that we're the ones to make it happen. On some time horizon, it's absolutely inevitable: making health care cheaper, better, faster. But we're at war to make sure that we're the ones to do it, and that we do it in the next 2 to 3 years.

You said, “I know Abridge is your mission for life.” You start a new company, and you can only take 1 investor with you. Who do you take with you?

If I start a new company right now, I would call Elad—Elad Gil. We'd go big: single GP. All of the wisdom in his head, all that I've learned from him, and his ability to go every stage and be incredibly valuable.

What's your biggest lesson from Elad?

There are so many lessons. I always feel somewhat like just a pupil with him. I just want to learn. Recently, we've spent time talking about corp dev and his lessons learned from his days at Twitter: how to do that effectively, how to think about corp dev, and how aggressive you should really be if you decide you're going to do it.

You need to have people fully focused; that needs to be their entire job. You can't split people's brains on it. If you're going to do it, you've got to be binary. I've learned a lot, but there are lessons like that that always come out of him.

Who do you think is the underappreciated CEO today?

That's a good question. Honestly, because we talked about that Costco CEO, I think it's that guy. What's his name? Ron Vachris. He's a guy who went from the forklift to the executive office—all the way up. He's a lifer over there. He carries the torch of culture.

He's managed to continue to grow a business that he inherited—or he was already awesome—but they've only had, what, 3 or 4 CEOs? That's one of those businesses that you don't read about enough. I think it's absolutely amazing.

Obviously, I have my idols. I have people like Ali Ghodsi, who I think is an absolute legend in the way he plays his chess and navigates his market. I just try to learn as much from the outside in, but that guy, I think, is pretty awesome too.

Final one for you. When you look at the next 10 years, what are you most excited for?

Where I'm focused right now is health care. That's all I live and breathe. That's all I think about. Health care is just an absolutely messed-up market.

You remember that XKCD from years ago—I think it was about Conway's Law: you build what you look like.

Yeah.

They had Apple there, and it was concentric circles, like their headquarters. There was Google, which looked like a neural network. There was Facebook, which looked like a graph, maybe. And there was Microsoft, and it looked like silos pointing guns at each other.

That's what health care is in the United States. It's providers and health systems pointing guns at and getting pointed at by insurance companies. Then there's also life sciences companies. Those 3 stakeholders, in some way, shape, or form, are misaligned. They're not only misaligned with each other; they're misaligned with the people who matter most: the patient, the person.

The opportunity that we've got right now with AI is not to deploy the latest model and make science fiction happen in a clinic. It's to change the business model. This technology is going to do it. It's doing it right now.

We're going to see that on both ends of the spectrum. We're going to see, to your point earlier, a lot of AI doctors and a lot of automation of high-frequency, low-stakes workflows. But then, on that enterprise, that health care system, when you get really sick, you're still going to want to see an expert.

We're going to see new models that actually align folks and incent prevention. Prevention meaning care, and not health care—not sick care. I think that's what I'm most obsessed with right now, and I don't think it's even a 10-year horizon. I think we'll start to see the beginnings of that in the next 3 to 5 years.

Dude, listen, this has been such a joy to do. Thank you so much for doing it in person. It makes it so much more special, and for putting up with my meandering around.

No, it was amazing. Thank you, Harry. It's a privilege.