EPISODE 1645

[EPISODE]

[0:00:00] ANNOUNCER: There is a revolution unfolding in biotech, the confluence of new biological methods like CRISPR, virtually unlimited computational capacity, and machine learning has fundamentally transformed our ability to engineer biology for wide-ranging applications. Andreessen Horowitz or a16z is a venture capital firm that was founded by Marc Andreessen and Ben Horowitz. 

Vijay Pande is a founding general partner at a16z, where he leads the firm's investments focused on the cross-section of biology and computer science, including areas such as digital therapeutics, cloud biology, and computational medicine. He joins the podcast to talk about innovation in biotech and healthcare, the biotech startup landscape, the impact of AI, and much more.

This episode of Software Engineering Daily is hosted by Sean Falconer. Check the show notes for more information on Sean’s work and where to find him.

[INTERVIEW]

[0:01:06] SF: Vijay, welcome to the show.

[0:01:09] VP: Thank you so much for having me.

[0:01:10] SF: Yes. I'm really excited to talk to you. You have a very interesting background. You studied physics. You moved into biophysics at Stanford for a long time and now you're a general partner at Andreessen Horowitz for the past eight or so years. I also know, you have kind of a long history of interest in startups and tech, going back to when you were a teenager. I guess, like what sort of has motivated the movement from academics to being an actual investor on that side of the tech space?

[0:01:39] VP: There are a couple things. As you mentioned, like I was involved in a startup when I was a teenager. This is Naughty Dog software, which is still going strong, with HBO shows, and all this stuff. The Last of Us on HBO, and so on. But it was very early and a very small one when I was there, just three of us. But I love startups even then. When I was thinking about faculty jobs, going to Stanford was high on my mind to be in the startup ecosystem. That was explicitly on my mind. Actually, that happens automatically. VCs on Sand Hill reached out to me within the first few months of me joining Stanford. 

So, that was always on my mind. The drift to venture was something I've been thinking about, even once I started working in startups, because you work with VCs and EIRs and so on. The reality, though, is like, the different venture firms are fairly different, and I didn't find one that was like a good match for you, even though I hung out at a few places. But like, I think a16z was really fundamentally different. And so, I knew, that once I started hanging out there, that things could be different. But the draw to venture is very natural for academia, for the type of academia that I did, and that with Folding@home, almost itself, kind of a startup, and I did deals with Google and Sony, and like, we shipped product and try to really build something significant. So, that was very familiar.

I also, like doing lots of things. As a faculty member, I had 25 people in my lab with a bunch of projects. We've got, like, I've got maybe 25, 30 investments that I'm shepherding a16z. That all feels very, very similar. Actually, in fact, I would say, there are more similarities between academia and venture that people might realize.

[0:03:17] SF: Yes. I mean, there's a lot of research involved to be a competent investor. And you also need to be pretty up-to-date on kind of where the world is going and thinking about the future, just like you would, is if you're doing academic research.

[0:03:27] VP: Yes. Also, I think there was a shift, though. I would say, around the time I started a16z, a shift from venture being maybe a generalist game to being a specialist game, and even maybe a sub-specialist. And part of that is just the startup ecosystem is so huge. Now, that as specialists have an advantage 20 years ago, there just weren't a number of investments to have specialists or subspecialists. When I got to a16z, I was a life science and healthcare specialist. Now, I run a team of 50 people, and I get to be a subspecialist amongst other strong investors.

[0:04:00] SF: It's just like a natural progression for anything that gets big enough. If you even look at, I know, the medical profession, and you went to a doctor 100 years ago. That person probably did everything from like surgery to birthing a baby and so forth. But now, you have like, highly specialized people who do just this like one thing on this one piece of organ, and you see the same thing in engineering.

Now, when it comes to like investing in startups, it would be very, very difficult, I think, especially when we're talking about the biotech and life sciences, as we'll get into today. If you didn't have any experience, could you gauge like does this make sense? Are these people credible? How is this different than something else? Is there a real problem here?

[0:04:39] VP: My area that I've double-clicked on is AI and life sciences and healthcare. So, I suspect pretty soon subspecialties within that or some subspecialties, and that's kind of fun.

[0:04:49] SF: So, was this something you were originally like Professor in residence at a16z? Then, you moved into being an actual partner leading this group that's making these investments in AI, in life sciences and biology. But was this something you had to actually like convince Marc and Ben about? Or was this something they were interested in doing? You were kind of the right person for them?

[0:05:10] VP: Yes and no. It's interesting. So, they were certainly interested when I was this professional resident, because I think they were seeing and I was seeing, basically the transition from sort of tech companies in computation and life sciences, like tech and healthcare going from like, “Oh, let's do research on it”, to like, “Let's build companies.” That's totally different. I think, a lot of the companies in life sciences were single-asset companies. If they were platform companies, they weren't sort of a tech-driven platform. Same with healthcare, analogous to healthcare, they were, let's say, very services heavy, not very tech-heavy.

Actually, Marc and Ben famously said, when they launched the firm, what they would do and what they wouldn't do, and it made it clear, they wouldn't do life sciences and healthcare. I think what they really meant was a single asset, style, life sciences, ultimate life sciences, and high services, health care. I think when they were seeing these companies, these companies were basically tech companies that are being built by tech companies. But I think Marc, and Ben would be the first to say that you needed a deep expertise. You needed to have a PhD to understand these life sciences. Companies needed experience with the healthcare world to understand these healthcare companies, and it's something that they had to bring on. I think they were excited because this is a huge space. It's like almost a quarter of GDP. And tech has not really had a huge impact in it to date, although, that's changed a lot in last few years. But that was the opportunity.

[0:06:27] SF: Yes. I think that there's this noticeable shift that's happening in terms of more companies in the health space, being kind of like technology companies, and also adoption of technologies is still maybe like, slower than ideal. But there's a lot of sort of momentum that's going on. I think in many ways, if you look back at the 20th century, that century was kind of the age of physics, or maybe you know, chemistry to some degree as well. There's work and impact that Einstein had, nuclear bombs, space travel, all these types of things. I'm not an expert, but it feels like we're on the cusp of the 21st century, sort of being like the age of biology. Part of that feels like it's being propelled forward by the innovations happening in the technology space.

I guess, what are your thoughts in terms of does that feel ring true to you, that we're kind of moving in this direction? And what are some of the impacts that we might see in the world of like healthcare, drug discovery, and so forth?

[0:07:21] VP: Yes, for sure. So, just a reminder, my degrees are all in physics in both undergrad and PhD. When I got to MIT for the PhD in 1992, I think there were roughly 100 physics undergrads and 10 biology undergrads. When I got my degree in ‘95, it actually flipped, which is kind of amazing. There are 100 biology undergrads and very few physics undergrads. I think that was due to genomics getting hot, but I think a lot of us could feel that the interesting areas in biology were super fascinating, and that physics was maybe hitting limits for what you could do experimentally.

Also, when you think about last century, a lot of the big solutions to problems were things like nuclear bombs, and so on that ended World War Two, and as maybe made sure we haven't had a world war since then, and so on. But the challenges that we have today in terms of life sciences, health care, feeding a billion, maybe soon to be 10 billion people, climate change, all these problems are really problems that are problems in biology. And that the solutions will probably come from biology.

[0:08:21] SF: Then, how do you think AI will play a role in this role? Clearly, there's a lot going on in the last year, with the hype. But AI, of course, existed before that. And I think that things like ChatGPT is really what has provided the UI to make AI feel accessible, and wow people to realize, “Oh, this is what we can actually do with this thing.” But in the context of like healthcare, or in life sciences, what role do you think AI is going to play in terms of bringing biology into the 21st century, and being able to really sort of innovate in the space beyond what we've been able to do in the past?

[0:08:56] VP: What's interesting about ChatGPT is that I think, for a lot of people that weren't studying AI, doing other things, that was their moment where they finally got connected to some AI something, and they can see like, “Oh, my God, there's something really amazing going on.” But that's almost like just seeing one glimpse of a much larger trajectory. As you know, since you've been involved in machine learning, and all this other stuff, that there has been a huge progression in evolution over the decades. If you think about AI can do a couple of things, one is that you can actually first do some sort of dimensionality reduction, which is take something really complicated, and sort of be able to understand it. And then you can do classification, and you can say, “This is cancer, this is not cancer”, or something like that with high accuracy.

Then, you could do generative, which is then we can go back up in the high dimensions and make things. People are seeing the general part because it looks kind of like a person although, there's a bit of magic trick to that too. But I think in terms of healthcare, the ability to just understand biology and do the dimensionality reduction and the classifier part, that alone has been in progress and heavily used even in startups for the last decade or so. That's the part that will be useful for making drugs, especially understanding human biology and human disease biology. I think what people tend to forget is that biology is complicated, but you can do experiments to understand things. But you can only do experiments on people in clinical trials. Even that sounds a little squishy to say it's an experiment on person. But in a sense, that's what it is, and it'd be unethical to do anything more.

Because of that, that's what stymies drug design. Because we don't know that much about people. AI allows us to take experiments on different things, whether it be organoids, or the human data and so on and build synthetic models of people that are way better than let's say, animal models would be for people. They're not perfect, but then, like any system, we can iterate and the key thing is we can iterate in an engineering fashion, and they'll get better and better and better. So, that ability to understand human disease biology will hopefully allow us to hit the right targets, to make sure the clinical trials are successful.

When clinical trials are more successful, the price of drugs will go down. The speed of things will go down. The efficacy will hopefully go up. All of those things are very natural once we understand it, and AI is probably the most natural way to understand it.

[0:11:14] SF: Yes, I mean, there's a huge cost involved with like running a clinical trial where you're having human test subjects. If the majority of those fail, then that's like going to drive essentially out the cost of designing and bringing a drug to market. But if you can use AI to have a better confidence, essentially, that the clinical trial is actually going to be successful, then you're significantly reducing that cost, and you could run those things much faster in that scale, than you can run multiple human-based clinical trials.

I think, it even goes beyond that. I think, at first, we want to build a model for a typical person. But in time, I'd love to have a model for me, because I don't want anyone to experiment on me. But I'd love to know what's best for me. A lot of the dreams of personalized medicine that we've been all talking about, haven't really come to fruition, in large part, because they're driven by genomics, not about my current phenotype, which may even not be completely described by my genomics. So, that gets really interesting. In some cases, especially with terminal diseases, so if you have cancer, making sure you get the right drug for you in this moment, could be the difference between life or death. That will be an obvious win. But also, I would also prefer just not to get sick. So, if I can understand myself and my biology better, we can do things to get out in front of the disease before it gets bad.

If you look at health care, especially in the United States, I think it's something that just like continually getting more expensive for people. Unfortunately, if you look essentially at the trend lines of the average lifespan, so there's the pandemic in the United States, like such a wealthy country going in the wrong direction. Do you think AI and tech is essentially the key to transforming healthcare and a meaningful way?

[0:12:57] VP: I think there's a couple of things we need to do. So certainly, there's a technological change, there’s a policy change, and then there's a go-to-market change. Part of what makes healthcare complicated is that you can't just fix one part and have the old system be the same, and it will just magically get better. You can't put like a new part into an old car and suddenly it’s a new car. So, in terms of technology, for sure, I think we're already seeing cases where AI could be co-pilots for doctors, maybe even replace some aspects of nursing, even today.

The technology there, actually, maybe not even the hardest part of the things I'm talking about. Then, the next part of it is actually the policy part. I would love to see much more acceptance of AI and understanding that we have a regulatory framework that can understand software as devices, and we need to embrace the technology, rather than be paranoid about it.

But then the final part, and this is the part that may be the hardest part, is to really change, at least in the United States, how we think about paying for health care, and being able to build value-based care rails will facilitate all this because then we have the alignment, that adding technology to keep you healthier will in a sense be worth something, rather than just treating you when you get sick. So, there's many different ways that companies are working on the value-based side. But I think when you can bring all these things together, then we have a fundamentally different healthcare system.

So, I think it would be over-simplistic to say any single thing is going to fix it. But we see people working in all these different aspects. Actually, I think there's a lot of reason to believe that in time, maybe not so far from now, even like five years from now. Maybe those pieces will be in place.

[0:14:36] SF: Then, outside of looking at the ways like technology, AI could innovate and improve health care, in terms of just like biological research. There's so much data that's involved in biology. Do you think that without essentially the aid or the tools available that we have through technology that we wouldn't even be able to reach the next limits to biology, because there's just too much information for like a human to possibly process and keep in their head. And that's like one of the limits essentially of like discovery and innovation in this space?

[0:15:09] VP: Yes, I think that's spot on. One way to think about is actually, we've been through this before. People like to say, “Oh, predicting the future is really hard.” But actually, we predict the future all the time. The sun rose at 7:30 this morning here. It's going to rise at 7:30 tomorrow.  I know when it's going to rise a year from now. I know what the weather's going to be like a year from now. There are cycles. So, we could look back at previous cycles and what that went like. I think a great recent example is computers in Wall Street.

So, 20 years ago you start talking about having a computer do what an expert trader could do. That sounds crazy, right? This person has been studying this for decades. They're like the best person on Wall Street. How's the computer going to beat that? But then, once you understand that this is really a data game, and you start thinking 20 years from now, how could a person do what a computer could do? That's kind of ridiculous. You see this falling all over the place. Chess is another good example. I went from, like, “Oh, you could never beat a grandmaster at chess.” Now, it's ridiculous to even try to beat a computer.

We're in the middle of that early, maybe first 5, 10 years of that maybe 2025-year arc, where I think people are starting to realize like they're shifting from, “Oh, a computer could never do that”, to like, “Oh, I'm starting to see how a computer could be really useful to that.” But there'll be the list of caveats. And those are things that have to be addressed. But like over this 20-ish arc, it will be addressed and we've just seen this over and over again. I think we will get to the point where 20 years from now, maybe 15, 10 years from now, it's hard to imagine like asking a human being to do what we asked doctors to do. We asked them to incorporate all this genomics, and all this blood tests, and all of this sense, and all this biology, which is constantly changing, expect them to be geniuses and all of this, understand me in like five minutes, they've read my chart, and then tell me what to do. That's crazy to expect a human being a tool to do that.

But I think in time, just as there's still people on Wall Street, I think there'll be plenty of people in the medical profession, but there'll be using computers to do what they can't do today.

[0:17:12] SF: Yes. I remember hearing related to the chess analogy is like, the best human chess player is going to get annihilated by AI. But if you actually combine AI with a human expert, then you'd like reach a new level, essentially. If we can do that with doctors, where they can essentially have the advantage, or value, or the power of AI paired with their own training, expertise, intuition, and knowledge, that suddenly you have like a super doctor that can do way more and do it way better than any single human could do.

[0:17:42] VP: For sure. I think we'll also need people, you could imagine how we talked about like doctor specialties as an analogy to investing earlier. You can imagine that actually, for the real doctors’ specialties, AI could be a new specialty. Because AI is going to be everywhere, but who's going to assess that AI? Who's going to make sure that it's working the way we want it to be working? Who's going to understand all the changes. There are other doctor professions that go with new technologies. There wouldn't be radiologists without imaging that didn't exist 100-plus years ago.

So, I think we'll need doctors who understand medicine and also understand AI, that will be a huge part of it, because I think it will be a collaboration for years to come.

[0:18:22] SF: Do you see in the research world, the biology, or even in the healthcare, is there like a resistance to the use of technology in some of these like newfangled approaches from like the old guard? If you look at even older generations like physics, like the older generation was resistant to quantum, when that was first introduced, or black holes. I remember even I had a CS professor that you their default was always like, I saw this research in 1976 by so and so, and that was always sort of like default reaction to everything. And you could dispute it, essentially, because I wasn't alive during that period. But just this dismissive behavior of whether this stuff could actually outperform a person, and what value it actually brings.

[0:19:06] VP: Yes. I mean, for any major change, it usually starts with some degree of pessimism, and even the pessimists are right, to some degree. Because this thing is not as great as the hype in the beginning, and there's all these flaws that we point out. I remember like in the late nineties with the Internet, all these people talked about how it's ridiculous that you'll ever buy clothes on the Internet, or shoes on the Internet, or dog food in the Internet. Malls will never go away and all this stuff. There's a famous Newsweek article on this. The pessimists are often right in the beginning. So, I think it's a natural point to take it because these things do have flaws. But one of the things that surprises me, especially, I would say this has changed just in the last few months, like the end of 2023, is that we are seeing people on the provider side and other sides of medicine, really excited about AI.

I think part of is that what we have given them in terms of software tools over the last decade has not really gotten them excited. You think about like electronic medical records that came from the HITECH Act, they're kind of billing systems. Does it really help them help the patients? Maybe a little bit, but it's also a nuisance for them. Software is very skeuomorphic. Instead of a whiteboard, you can have something on the computer that's a whiteboard or a physical notebook, you can have computer notebook. But that really like changing their world. It's probably not and maybe just adds burden sometimes.

But now, what we're talking about is that AI is co-pilot, or even AI as nurse, those are things that now it looks different. It feels like almost like a type of staffing. It's like a resident or a nurse. Actually, what's interesting, is we're seeing people think about it from a staffing perspective, not just from how do you implement it, because you could implement it from SMS, the way you’d implement people. You don't have to hook up your software to it or so on, or from telephone. Not just from an implementation point of view. But even from a budget point of view. This isn't coming out of the software budget, which is maybe 5% or 10%. This might be coming out of the personnel budget, which is maybe 50%. That's fundamentally different.

Now, what we're telling you, instead of giving you something skeuomorphic that you're going to have to mess with, we're going to give you help, at a time where there's huge nursing shortage, and general practitioner shortage. We're going to get you help. That's a very different proposition and I think something that is getting people's attention.

[0:21:27] SF: The user experience too, for like a co-pilot interaction is way more transformative too and natural, I think, than trying to learn some new complex piece of technology that you used to do on paper. You were doing that for 20 years, and suddenly someone comes and says, “You have to do it this way.” That feels like counterproductive than maybe the way that you did before. But if you could just ask something, the way that you would ask a colleague, and then you get a response that actually feels like the response a colleague could get, that's tremendously valuable, and overly low barrier to entry in terms of adoption.

[0:22:00] VP: Yes, I 100% agree with that. I think, while existing LLMs may not be the panacea for healthcare, because I think we may need much more specialized data and all this stuff, and there's work to do. Just even what we could do today with LLMs, as a user interface is very interesting. Because to your point, I mean, you don't have to learn how to use the thing. You just talk to it. As long as it has human-like latency, you may even like literally voice talk to it, because you're starting to see, you don't have to text it. You can just literally talk to it on the phone or someone. That's starting to be very, very different. Then, there's no learning curve. Actually, if you think about from that point of view, it’s very rare that software has had zero learning curve.

[0:22:41] SF: Yes, exactly. It's the idea of like, when they introduced the iPad, that you could give it to someone that had never seen technology before and they could actually use it. That's a rare once in a generation type of technology that comes along that, anybody can essentially use. It doesn't matter what their sort of text –

[0:22:59] VP: And as human beings, we spend our whole, like, maybe our whole lives learning how to deal with people. So, in a sense, maybe we've already learned how to deal with it.

[0:23:06] SF: In terms of building technology in the space, what are some of the challenges that exist for a biotech, like startup, growth company, versus like a conventional technology company?

[0:23:18] VP: Yes. A couple different things. One is that, I think, it depends on what type of tech company. There's some tech companies that are only in the world of atoms, and then there's some companies that are tech companies that sort of straddle tech in atoms and bits. Lyft or Uber has to get into the world of atoms. Instacart has to get into the world of atoms. There's plenty of stuff that plays between the two, whereas other companies maybe you can only – are just all data companies. So, they live in the world of bits.

I think first is for tech companies that deal with atoms, that finally requires probably more capex, more work much different things. That part is similar to what we see for tech companies in bio that, of course, have to do with atoms as well. That part's not radically different and be the capex isn't radically different. The other part, though, that sometimes is different is the regulatory side, that we obviously have the FDA and so on. The counter to that would be as there are plenty of companies that have to deal with regulatory aspects. So, like Lyft, and Uber, and Airbnb have regulatory aspects they have to be thoughtful about. But the counter to that is that, well, they don't have federal regulatory. They have just local. So, you could have regulatory arbitrage.

If one county or one area doesn't like Airbnb, but another one does, that's fine. Then, one county doesn't get Airbnb, and that's just the way it's going to be. So, I think the challenge is that in America, there's only one FDA. So, our FDA is often the one that other countries look to. Even if you said we can go to another country, though the markets are different, and we'll look to the US FDA. Actually, if that were something that were different if it were federated, or if they were other options, that could be interesting from the company point of view, because then you could have regulatory arbitrage. Who knows what's coming?

But I think we're starting to see sort of a different regulatory posture. Also, because of the IRA Act, a lot of former companies are looking to other countries that have other regulatory environments, so that may start to look more similar as well.

[0:25:19] SF: Then also, I would think that for at least not necessarily for every company that’s operating the biotech space. But for a lot of them, the learning cycles are going to be drastically different than a conventional technology company, in part because of the regulations. And essentially, it's just like slower to learn and test something. That might take multiple years to go to market versus being if you're in like the e-commerce, consumer space, maybe you hack something together, release it, put it out on product hunt, and get feedback immediately.

[0:25:47] VP: Well, this is something that I've been challenging founders on, which is that, can you think of new go-to markets that the rapidity of your, let's say your AI's ability to come up with a therapeutic relatively quickly, can be matched by the go-to markets rapidity, as well. There's a few areas where we've seen this. We've seen this with vaccines. Certainly, vaccines can go to market relatively quickly. I suspect there actually will be a renaissance of vaccines. People are talking about cancer vaccines, beyond just like infectious disease.

The funny thing is, like, there's been so much politicalization of vaccines as well. But if you go way back 50 years plus into thinking about vaccines, they're kind of amazing. We don't have therapeutics that we use for polio, because no one gets polio. I'd rather actually just never get polio, than to get it. But being an iron lung, and I take some drugs that may eventually gets rid of the polio. So, there's all these diseases that we just never get and that's the ideal scenario. It sounds magical for that to be things like cancer, in part, because we're just used to that being the situation. But from a biological point of view, there's many reasons to think that doesn't have to be the case.

In certain areas, I can imagine that we will see new go-to markets that will allow greater rapidity, and rethinking of actually even what sort of life sciences contribution to healthcare should be, that it's not just about treating disease, it's about avoiding disease, all of these different areas. So actually, one of the things I really encourage founders when I talk to them is that often the people I meet on the life science side are coming from great scientific backgrounds, or just genius, creative scientists. I try to really encourage them to take all that hard work and passion and creativity and genius, and apply it to go to market side. If you can be creative there with this novel technology, that's where the big wins are going to come.

[0:27:38] SF: If you're a founder in the space, and you're coming from sort of genius biology, life sciences side, so should you be thinking about pairing up with someone that has maybe more of the traditional like technology go-to-market lens and expertise?

[0:27:52] VP: Yes. So, there's a couple of ways to do it. In general, my favorite founders will have all that in one person. For instance, it used to be more rare to know biology in, let's say, machine learning. But that's becoming more common. Then, it used to be more rare to maybe have that and the go-to-market side. But I think some of the best founders have all of it, but maybe they're the rare unicorns in the group.

Yes, so in lieu of having all of it yourself, putting together a team is a very natural thing to do and that's a tricky thing, though, because that matchmaking is some of the hardest part. Especially, the go-to-market people we're talking about, or maybe people that really want to do innovations and go to market, not just do with whatever he has been doing over the last decade.

[0:28:32] SF: In terms of companies that you're involved with and what you're seeing in a space like, are people actually developing new approaches in terms of AI or other technologies specific to biology? Are they kind of adapting things that exists maybe in other areas and applying that to biology based on their domain expertise?

[0:28:53] VP: Yes, I think historically, a lot was adapting, for sure. But I think that's very much changing, and we've got people now who are AI icons coming into life sciences, and they could do anything. They're the ones that have developed transformer architecture. They're the ones who've sort of really pushed AI architectures forward. And now they're applying that understanding of the life sciences or healthcare area.

So, I think that's very much changing. I think that's where it gets really interesting. Because purely adapting is nice, but making the sort of architectural changes will allow you to make big leaps forward. Oh, with that said, like, even just the base architecture that people use today for other areas in general AI, like transformers, they're pretty versatile, and you could have vision transformers and use it for different things. You can use it for protein sequences and DNA sequences. I think sometimes we're talking about is now not a wholesale architecture change. But that if you look at the architecture of these things, you can see the little Lego blocks that are in it, and you just maybe change some of the Lego blocks rather than starting from a clean sheet of paper. And that evolution, that's the fun thing now, is we're seeing a lot of that and that evolution will have now very specific architectures, very specific different biological problems.

[0:30:04] SF: Then, what about in terms of the data? We're in an AI revolution, but we're also, you still have that essentially, without it being a data revolution at the same time, and we don't have foundation models without massive amounts of data, where is this data coming from in life sciences? Is that part of, essentially, building like a moat for businesses how good your data is?

[0:30:26] VP: Yes, for sure. So, I think one of the things about life sciences and healthcare is that most of the data of interest that exists is not on the web. Let's take life sciences first. It's a couple of form. It might be sitting in a biotech or Pharma. But a lot of that data wasn't generated with the intention for going into AI. It was intended to run some campaign to find a drug. Often, there's various issues that limit their applicability for AI. The second type is where actually you start from day one, where you're generating this to build the best model. I think, that is my strong preference and I think that's what we're seeing a lot of startups doing. Especially, they're building, like robotics architectures, and other automation architectures for the data side.

So, I think really, a lot of the best AI in life sciences companies will really be great at the data and biology side, as a means towards the end of building these models. Because if you use just what's out there, you're not going to have a huge differentiation. You can maybe have some differentiation algorithmically. But I think that's becoming less and less the case.

What's interesting on the healthcare side is that, it's very natural to train with like emergency medical records, and all that stuff. But obviously, that's not on the Internet, for a variety of reasons. That amount of [inaudible 0:31:41], it gets really interesting. I hear about all these interesting things people are doing, like you train models on medical records for, let's say, all but the last 10 years of the record and see if you can predict the last 10 years. If you think about, it's pretty dramatic, because we're able to sort of predict for it, and that's a very natural thing to do in any AI task. You need a lot of data for that to handle a long tail on all these challenges. So, I think this will become very much a data game, who can actually either generate the data, which would be preferable in life sciences, or get the data and healthcare would be ahead.

One final thing is that I think there are plenty of companies that generate their own healthcare data by providing healthcare. So, healthcare services companies, which may or may not be all that exciting, from a service point of view, if they have the strong data play, they actually may be in interesting positions to be able to use that and become AI companies because they're generating the data.

[0:32:34] SF: Yes. There's also you mentioned, like, a lot of this data is not available like digitally. There's also in the context of something like a wet lab. There's also a lot of noise involved with the data, because you have essentially, humans working in the wet lab, collecting data, and observations, and then giving that to essentially the data team. Then, there's like, a lot of things that can get missed along the way. Is that something that you've seen? And what are some of the tools or technologies or innovations that are happening? Because we’re addressing some of those problems.

[0:33:03] VP: Yes. So, some of the best companies, I think what they do is that they don't have siloed data teams and biology teams, and the two are very integrated. Part of that then, is that you have data teams, thinking about what needs to be done in the experiments. So, measuring temperature, very carefully, understanding just the room temperature, and the variation. Just really coming to the heart of why a lot of biology data is reproducible, because there's hidden variables that are involved. When pipetting by hand, there's going to be a ton of hidden variables that you just can't even measure. Robotics, at least standardize this and allow you to rerun the whole experiment and see. But being able to measure everything with the intent for extreme reproducibility, that is, I think, the foundation then for being able to do ML. Because obviously, if experiments, it reproducible, the ML is not going to be probably all that useful.

Finally, though, I think, let's say the experiment is reproducible within some precision, then that's also interesting from an ML perspective, and ml can incorporate that. So, understanding limits of your experiments or variation from experimental type to experimental type, and the impact on your models, that's important part for models to understand such that they're not taking this data sacrosanct and understanding its limits. Again, when you blend these two groups of people, where you want your AI people to be really deep into the biology, and ideally vice versa, I think that's where the real big wins are going to come.

[0:34:31] SF: Yes. I mean, you could probably apply that thinking to like any great company. If you have a company that every function is running independent, and in a silo, it's probably not a great recipe for success.

[0:34:40] VP: Well, and the challenge though, is that when you're small, if I were going to try to give the counter when you're small, it's easy to do. So, the question will be is, how are people going to do this at scale? But maybe they're doing at scale, and then instead of having data silo and CS silo and bio silo, you have it by therapeutic area or something like that, and then the teams are integrated. This is also not just a revolution in the technology, but it's a revolution on the people side. How do you get people to do things very differently? How do you get this space that was kind of artisanal to become engineering? Artisans don't become engineers overnight, or maybe never. So, if you're like an incumbent with the best artisans ever doing a bespoke process, can you one day, turn them into engineers? That might be hard. So, it could be that the people side of it may be where the startups have the advantage, because they're starting from a clean sheet of paper.

[0:35:36] SF: Yes. They don't have these necessarily preconceived notions about the way things need to be done. I think you raise a good point about this transformation to things like drug discovery, becoming more of an engineering discipline. If you look at history, there's all these like famous accidental discoveries of drugs and vaccines, smallpox vaccine, or penicillin. I guess, how is current-day drug design, like not discovery, but like the actual design of drugs like moving to being more of an engineering discipline than this kind of like artisan thing that you spoke about?

[0:36:09] VP: There's a couple of different ways. So, part of it is that through new modalities, so like a lot of the gene editing work, that's about sort of engineering and editor that has a strong efficiency. When it edits, it edits with high probability and also high accuracy for targeting. It doesn't edit the wrong thing. Those are things that you can very much engineer. Then once you've engineered that, then it's about just designing the guides to be able to see what you want to edit. So, that framework feels a lot more like engineering, both from the work you do to improve the editor, and then what you do with the editor, than like most of life sciences that's come before it. Obviously, it's still early days. We've just had now the first editor therapeutic approved.

But I think we'll see a big shift there. Especially, in cases where sometimes it's the biology that trips us up. But in cases where the genetic origins are causal, and the beautiful thing about genetics is that that's one of the few areas where the causalities can very clearly be sussed out. We've got the biology solved. We've got the engineering part for the editing solved. And then, what you edit usually is straightforward and is a data science problem as well. That's a very different paradigm for more conventional things like we can work backward in complexity like antibodies. Now, that's a very natural thing to be designed. And all the work in generative AI carries over into that and AI in general, since it's now a language, just DNA language, not English language.

But then even for small molecules, I think, we're seeing examples where you can actually drug really difficult things to drug, and that's a combination of simulations, and machine learning, and experimental data, and how do you fuse all that together? That's a very natural AI problem.

[0:37:52] SF: In terms of like building and scaling companies in space, we have a fair amount of experience building and scaling like conventional tech companies. But even then, it's easier said than done. A lot of companies fail at that. It’s a hard process. But what is different from your experience of like trying to build and scale a company in the biotech space?

[0:38:09] VP: Yes. I think one of the things is just the cultural differences between a lot of the people who might hire in the wet lab versus the CSI. So, that's definitely something. It even comes down to even pay scales and how they view equity and comp and all that stuff. So, that was, I think, more new territory 10 years ago, and now I think, almost a decade in, there's plenty of best practices for that. But it's still something that comes up.

Actually, I think now that the space of like machine learning and comp in life sciences has been kind of rolling, it feels for like, almost a decade. Out of those, I can point to examples for how to do things, and for companies that have been able to get there. I think, though, the interesting thing is thinking about what the next 10 years looks like. One of the areas that we're all looking to is how AI changes the company itself, and that, can you scale what a biologist could do? Could you integrate in helping a biologist with programming in ways that they couldn't do before?

I've seen multiple companies now working on that, and it's just the ability to scale the people you have, rather than just adding more people, it gets very, very interesting. So, we talked about how it's easier to do things when you're a small group. You might be able to stay a smaller group for a lot longer.

[0:39:28] SF: Yes. I mean, if you could make everybody 5x to 10x more efficient at their job, then in theory, you should need 5x to 10x less people to get to like similar places of business.

[0:39:38] VP: Then, I mean, the fantasy is that with the capital that's there, we do 5x to 10x more companies. We do 5x to 10x more drugs, more therapeutics. It's not like they're going to be all these life scientists without jobs. I hope it's going to be five to 10x better healthcare, essentially, in terms of the output that we can generate.

[0:39:59] SF: Yes. I mean, if you speed up essentially, like time to market and lower, essentially, the cost a company needs to operate at, in order to rescale and bring a product to market, then that just means that there'll be more companies and products. I mean, you see the same thing with all kinds of technology. I mean, the introduction of the cloud is a great example of that. Suddenly, you didn't need to set up your on-prem servers and spend the first two years of your life as a company, like buying a bunch of hardware and figuring out how to do this. You go and you click a couple buttons on AWS, and you're spinning up EC2 instances, on run in Kubernetes, or whatever it is, and you don't even have to think about it. It’s like you're up and running in no time. You can just start building.

[0:40:37] VP: Yes, or mobile. The shift to mobile, you could be making money very quickly with a very small group of people.

[0:40:44] SF: Yes. Absolutely. What about burnout? It's hard to be a founder in any company. But I know a number of really smart people that tried to do stuff in like healthcare, and they kind of just got burned out on it, because of some of the challenges from the history there, and barriers to adoption. from a technology standpoint. Is that something that you have seen?

[0:41:04] VP: I don't think I've seen more burnout in healthcare, life sciences, more than tech. But it is obviously a lot more complex for short, and much more complex go-to-market. And a lot of what we've tried to build is ways to help founders who come from one, but not the other, integrate that in. And if you look a lot of our writing and serve intended for either audience and pull them in. I think, a lot of times too, the burnout is particularly rough, if the team has to learn things the hard way. So, if you can build a team that has deep experience in both like – doing a startup is probably the hardest thing that any human being really ever does. So, it's never going to be easy, for sure. But I think we're seeing now these startups being built like tech startups similar in this way, too. So, it's not easy, but I wouldn't say there's more burnout on the healthcare side.

[0:41:56] SF: Okay. Well, that's good, because I feel like this could have some of the most like positive impact in the world. It was also like a very attractive reason, if you're coming from software engineering, or, like you mentioned, some of the best people in AI are moving in space, I would guess some of that has to do with wanting to have like a positive impact on the world.

[0:42:17] VP: I think there's a couple of different things. So, for sure, it's harder to find a more exciting and satisfying mission than like curing people of diseases, helping feed the world, dealing with climate issues, and all these things on the biological scale. So, it's hard to find a more rallying mission than that. It's something real. A lot of our companies, healthcare companies, life sciences companies, often the board meeting will start with like, even though they helped so many patients, and you can cite the numbers to have this specific story. That stuff is just so powerful to hear these stories of these people that wouldn't be alive without this technology that saved their life.

But then on top of that, and the ability, actually, that mission allows us to get really great people, it's an area that tech has not been able to uplift. If you think about major sectors of the economy, probably the last sector that tech has not been able to permeate in a significant way. I think AI is leading to that sort of, I think, permeation for reasons that we've talked about, like that it's not seen as much as software as seen as other things, almost like an alternative to staffing.

Because of that, I think a lot of the savvy tech founders are looking at what their next things are. Sometimes there's people who have built billion-dollar companies already, and they're looking for the next gig, and they're looking for where the big opportunities are, not necessarily the easy opportunities, but where can you build a $10 billion, $100 billion company. The market here is so huge, that it's like if you compare the healthcare market to the market that Google taps, let's say, for ads or something like that, it's like 10x bigger. It would be able to support 10 Google scale companies, or 10x the GAFA scale companies. So, that's something that it's just not familiar to us when we think about the space. But I think that's very much possible over the next decade-plus.

[0:44:09] SF: There's a lot of stuff in healthcare, that's still using a lot of legacy technology. Not that long ago, I got an MRI where they gave me a version of my MRI on a CD ROM where I'm like, “Okay, well, thank you. I could use this as a coaster. I don't have anywhere to put this, essentially.” And there's a lot of still-like print material. Every time you go and you see a new doctor, you have to fill out multiple pages on paper, physical paper, that they enter into a system and so forth. How do we overcome some of these like legacy systems? How do we sort of propel some of these systems forward? Because I feel like that adds significant cost to something that can like – there's better ways to solve these problems, essentially.

[0:44:49] VP: There's a couple of aspects of cost. So, one cost is the time, not just patient time but like provider time to get all that stuff done. Second bit of cost is like all the errors and the lack of data, or incorrect data, and the downstream cost of that. I think one of the big differences is when we can have a value-based system, then all of this stuff will take care of itself. Because in healthcare, especially as an investor, the first question you kind of have to ask is like, “Okay, this is great, but who's going to pay for it?” Let's say there's some system where it'll take care of everything you just said. Who's going to pay for that? In a value-based system, they'll pay for it or build it themselves, if it actually can lead to the value and let's say keep people healthier, or improve patient retention, or whatever.

I think that shift to where we're now consumers, instead of the product in the healthcare system. Right now, the payers are to some degree or your employer is the consumer, and keeping them happy is different than maybe what I want for my own personal care. Only when you're out of pocket right now, are you truly the consumer. I think that shift to value-based care really sort of puts the patient in a much more important place. Then we'll see all that stuff happen automatically. And for companies that have created value-based care rails, I'm already seeing examples of that, where they're buying or building tech to be able to do this.

[0:46:13] SF: How does investing in companies in the space look different than investing in conventional technology companies? How do you evaluate? a16z, I know they invest all the way from very early stage to pre-IPO and stuff. So, that's like a huge range to try to evaluate companies. But how do you evaluate companies in the biotech world?

[0:46:32] VP: Yes. So, it starts with having a team that has deep experience on the tech side, and the healthcare life sciences side. So, all the people that are brought on have that criteria on the investing side. Then also, in some degree, it's really just doing your homework, and a lot of our process firmwide, but certainly, on my team on the bio health side, is that we want to look at all the different spaces, look at all the companies in that space. We can't predict which spaces are the ones that are going to be the ones to hit. But we want to be able to really have done our homework for each one of them and try to find the one that we think is the best company in that space.

But it requires constantly being up to date on all of it, and so much is moving quickly. Just gene editing in AI as a combination, both of those areas are going very quickly. So, it helps that we have a team large enough that we can even go deep into it, and even through journal clubs and have advisors who are scientists or academics that can advise us. But it's the benefit, though. We talked earlier too, about being able to have specialists and subspecialists. We can have some of the people that are the smartest worlds in these specific areas to guide us.

But then finally, I think, the technology may be great. The product may be great, but if the go-to-market is flawed, that company may fail against the company that has subpar technology and subpar product. But the really crystal perfect go-to-market. So, we spent a lot of time debating different go-to markets. And actually, in principle, investments, different go-to markets could be different investments. And to really, in cases where it's not obvious which one's the winner to try to invest in all the different spaces. 

But because I think that's some of the hardest parts. That's in some ways, also some of the most frustrating part, because you can have these brilliant people that made brilliant technology and brilliant products. But if it just can't work within the healthcare system, that company aren't going to succeed. It's just painful to watch all around. So, hopefully, we can figure out from the beginning, what the right go-to markets are and help guide our companies along the way. If they can do that, then that's usually a key part to understanding whether the investment will be a success or not.

[0:48:37] SF: What's go-to-market typically look like? I mean, I'm sure it depends on who you're selling to. But it's not like you're just throwing up Google ads, and someone's like swiping a credit card and signing up for something when you're talking about like drug discovery or some sort of innovation in that space.

[0:48:50] VP: Yes. I mean, so some of it looks a lot like B2B, because you're selling payers or providers, but you have to understand how to sell them, and that's actually a really complex thing. So, it's helpful that I've people on my team that have done this on both the life sciences side and the healthcare side. Some of them actually used to be at the pharma, or used to be on the provider side so that they have that experience of being on the other side, and knowing the way they think.

But in an overly simplistic way, it's a sense, a lot of it is a variant of B2B sales, where also though, these are typically long sales cycles. Nine months is not unusual. Then, the integration can take some time. So, it's definitely, when you get that right, it works well. But also, when you get it wrong, you may not know what's wrong, until you have a couple of cycles in, which might be a couple of years in, and that's perhaps one of the most painful aspects.

[0:49:41] SF: Yes. I mean, it's like a true enterprise, like sales go to market motion. But the downside is a deal might take 18 months, and it's hard to make those adjustments essentially, because you're not able to just like learn. It's not like you're putting out a canary version of your product to 5% of users and millions of people are going to hit it, you're going to find out. Is that button effective or not?

[0:50:00] VP: Yes. Well, and overly simplified in a way, there's like four Ps. You can sell to the pharma, to the payer, to the provider, or to the patient. The patient looks like consumer, right? So, we have some companies that are direct to patients, direct to consumer, and that at least you can understand with all the consumer tools. But the rest of it, very much is enterprise. 

[0:50:23] SF: Then, as we start to kind of wrap things up, what are your predictions, I guess, for like, the next 10 years in the space? What's going to be the big innovation?

[0:50:33] VP: Yes. So, we touched on a lot of it, but I think – so where are we? I think we are very much in an industrial revolution for life sciences and healthcare. People toss that term around without thinking about what those words mean. So, the revolution part, that's the obvious part, but what do we mean by industrialization? What is the industrial part of revolution? Well, the industrial part is the bespoke to the automated part. Before we had industrial revolution for shoes, you had a cobbler that would make shoes by hand. Now, you have this whole process for making shoes. Shoes went from being this bespoke, expensive, rare thing that maybe mostly rich people had like a lot of shoes, but the rest of us, we have like one pair of shoes or something like that. So, like, what we have today.

So, what is this process of industrialization look like? Well, in life sciences, we’re actually industrializing discovery of new biology, and then industrializing understanding of human disease biology, and then industrializing, coming up with the therapeutics to attack that biology. Then, of course, even engineering and industrializing the clinical trials themselves. Similar transformations on the healthcare side, industrial revolutions don't happen overnight. This whole process requires a new technology, requires a new way of thinking, which is new changes in people, which is the hard part. We knew people would come in and handle these new ways of thinking about things probably take 20 to 25 years. You could argue depending when you want to say it started, we’re 10 years in, probably have at least another 15 years to see the arc of it.

But on the other side, we'll have healthcare where, like we talked before, it'll be insane to imagine not having a computer do this stuff. It's going to be a lot cheaper, a lot higher quality. And I think the part that people will get most excited about is that there's a huge disparity in healthcare, in access and quality. A computer is easy to copy. So, the best doctor, in principle could be available to everybody, much like billionaires, and everyday people have the same Spotify, or you can all get an iPhone. But right now, there's a huge disparity in healthcare. That democratization and access, that part is just going to be so different for that world we’re going to be in. As you could tell, I mean, I'm super excited about it. It feels like the last thing that has not been industrialized in our world.

[0:52:52] SF: Yes. I think, there's huge unlimited possibilities in this space. As someone like, who moved from Canada to the United States, the disparity in healthcare was like one of the most jarring experiences as kind of like an outsider. Canada's healthcare system is not perfect by any means. But everybody's at the same sort of B minus level of care. Whereas in the US, there is this huge disparity where if you have the means and the funds, you can have the best healthcare in the world. But if you don't have those things, then you're really, really struggling and unnecessarily for a country that has so much wealth and so many things going for it.

[0:53:28] VP: Yes. So, I think this is now gotten from a place of talking about science fiction to where people are building it right now, and I think the ChatGPT moment was a big eye-opener for everybody. I think people are now asking how, and I think the startups are starting to answer that. So, it will be a very, very fun decade-plus to come.

[0:53:46] SF: Yes. I'm excited for it. Well, Vijay, thanks so much for being here. I really enjoyed this, and I think everybody listening is going to enjoy this as well.

[0:53:53] VP: Thank you so much for having me.

[0:53:54] SF: Cheers.

[END]