I talk to people and record it! Some call this sort of thing a podcast. I call mine Moth Minds and the premise is “interviews with high-agency humans."
Subscribe to hear the latest on a bi-weekly(ish) basis: Spotify, Apple, Google, Amazon, and Pocket Casts.Highlights:
As a thesis-driven, research-centric investment firm, Compound spends most of their time discussing research and developing theses internally. They also use their collective online presences as Rorschach tests for finding people who align with their approach and care about topics they’re discussing at a similar depth.
Michael defines deep/frontier tech as areas that start as vertical technologies and then gradually become more horizontal — indicating they are now in the next phase of commercialization (e.g. mobile, AI, likely in the future: VR, AR, crypto). These spaces are verticals today but will become platforms tomorrow.
Three areas in which Compound has been spending significant time lately are robotics, decentralized energy grids, local software, and AI. The firm has been investing in AI for many years and believes that the key question to ask to discern what will win shifted recently from ‘What do you believe that no one else does?’ to ‘What can you believe, consistently ship, and run this marathon as if it’s a sprint for the next few years as capital flushes its way through the system?’”
MM: Hi everybody. Today I’m joined by Michael Dempsey, the managing partner of Compound, a thesis-driven, research-centric investment firm. Prior to founding Compound, Michael worked at the predictive analytics startup CB Insights and at Crane Partners, a multi-strategy hedge fund.
Our conversation covered how Michael views venture capital and frontier technologies, how Compound develops theses, and how he thinks about several of those theses in the areas of robotics, decentralized energy grids, local software, and AI. I hope you enjoy.
MM: Hi, Michael. Welcome to the podcast.
MD: Hey. Thanks for having me.
MM: Of course. By way of introduction, my first question for you is what do you think has informed your outlook on investing and what drew you to venture specifically after spending time researching private markets?
MD: Yeah. I originally got my start at a hedge fund and I was doing a bunch of different types of investing. I was doing long-short and public market investing and then I was doing private equity from Asia to the US, and a lot of stuff we were focusing on around the private equity was these tech companies that were possibly disrupting companies we were buying and helping. I think a lot of that kind of lesson and learning of working with companies that were being very majority owned was really compelling to me relative to public markets where it felt very zero-sum, but I wanted to learn and understand a bit more about how the tech ecosystem worked. As I joined CB Insights pretty early on, got to see just a total funnel of data every day and try and make sense of it all related to private markets and also got to talk to a lot of really interesting people who wanted to understand data on the venture investor side, whatever, and then on the startup side, as we were writing more research-centric reporting, wanted to help educate us on these various spaces.
The thing that I always say was people didn't... It never was clear to me why venture was so kind of consensus-driven. Hedge funds is all about information asymmetry. You think about your best ideas, you work on them, once you have a position on you really kind of then talk about it or venture felt a little bit the opposite. That I think kind of informed a lot of how I've thought about venture and just the right to win and where there were open spaces to develop I guess what we would previously call long-term alpha or long-term edge in investing. I think investing in general is really, really hard, and it should be really hard. I think we lost that the past few years, candidly, but I think that a lot of seeing all of these things across both kind of the public market, private market side, and then on the startup side led me to just think about a very research-driven and kind of thesis-driven way. We call it thesis-driven research-centric investing, but yeah. I think that's probably the main way in which it informed how I wanted to approach investing, and then, I don't know, the past decade has changed a lot of that.
“It never was clear to me why venture was so kind of consensus-driven. Hedge funds are all about information asymmetry. You think about your best ideas, you work on them, once you have a position on you really kind of then talk about it or venture felt a little bit the opposite.”
MM: Yeah. I'm curious to hear more about that. How have you evolved your thinking on both markets and then venture and investment more broadly?
MD: I think probably have come to appreciate market participants more. Mostly the simplistic way of framing that would be at a certain point of time, you realize that you're essentially participating in a market with a bunch of people and you need to understand how those people behave, how they act, and how that impacts a given market. In public markets you have a bunch of different styles of investors who do all sorts of different types of things, fundamentalist, momentum, activism, et cetera. In venture, you have just a bunch of different personalities and these personalities all have different heuristics for what they think make a good company, for what they think make a good founder, and they all have different incentives.
I think the incentives are wildly different across venture than it is in other asset classes where you have big funds that are sitting on large fee structures and the incentive is to index a space, you have small funds that the people managing them are making very little money and their incentive is to try and grow and build a firm and generate outsized returns, and there's a bunch of things in the middle of that. But I think understanding the game you're playing and who you're playing that game with is really important and has changed over time where you have to be realistic and comfortable with the idea that what you view as the best way to do this investing thing is not necessarily the way everyone else will behave, and that matters. I think that's changed a lot for me, and it's also been a point of maturation, of realizing there's a bunch of different ways people invest in venture and all of them are correct and incorrect at the same time.
I think that as that's changed, you also have to be realistic about where people are in their kind of life cycle and how that influences how they behave and what they're willing to do. We talk a lot about firm building versus fund building, as an example. I know we've talked about this one-on-one before as well. Again, just these things change. I think one of the things I believe is over the past few years as markets have corrected, people believe that everything's going to go back to a more responsible area. I think if you just look at who the market participants are and how they've learned how to invest, I'm not entirely sure that responsibility is ingrained in the vast majority of venture investors today. That will mean that the game being played could look very different than what a prior generation might think it should look like. I would say I'm more aligned with the prior generation often than the current.
“One of the things I believe is over the past few years as markets have corrected, people think that everything's going to go back to a more responsible area. I think if you look at who the market participants are and how they've learned how to invest, I'm not sure that responsibility is ingrained in the vast majority of venture investors today.”
MM: Where do you think it's headed? Do you think it's headed more towards where it was a decade ago?
MD: Probably not. I think you'll see consolidation, churn, all the stuff that everyone's talked about. I think that you have compression of returns, which is what happens when any asset class institutionalizes, and I think that likely you'll have a lot of people that maybe have very different frameworks. I've written about a post on this called Get Rich or Die Tryin’ and this idea that people are just continuing to go as far out on the risk curve as possible and doing it in a very kind of chaotic way. I don't think that behavior is going to change.
I think that there's also... There might be talent flows in and out of venture, for example, but that talent might flow from someone being a GP to becoming an LP and how they think about the world will still be the same, and so they'll just allocate capital to people that have a similar viewpoint. So I don't think we have some... I think we have a big reckoning in terms of people's performance is going to get destroyed and that will create more churn. I don't think we have meaningful shifts in behaviors outside of maybe sector-specific or maybe some apathy-specific things where people are like, "Software is late cycle, so I want to do crazy things and go the American dynamism route and things like that." But I don't think it's as Back to the Future-type thing as people are talking about.
MM: Yeah. All these different theses on where the alpha is. What is an opinion of yours about investing that you think the average investor would not agree with?
MD: Yeah. There's a lot. I would say the two that come to mind are we take a very prescriptive approach to investing at Compound. What we do is we look at a bunch of categories, we build collections of futures we believe in, and we have founders fit within those or shatter them. I think there's strong signal to both. We don't think we're geniuses in these areas, we just think we're pretty smart. Most VCs in my entire career are always like, "You are not smarter, better. You need to let founders show you the way of the future," and are very anti this kind of more prescriptive approach. We deeply believe that this is the best way in which we can partner with founders and the best way we can build a view of the world that we can deploy capital against and help our founders by having high context.
“We deeply believe that this is the best way in which we can partner with founders, build a view of the world that we can deploy capital against, and help our founders by having high context.”
The second would probably be I fundamentally don't believe that all founders are good founders, meaning this idea that this person is a great founder because of who they are, no matter what they do, I think is insane. I think it's insane because founder market fit or founder product fit more specifically is very real. I think that especially in this world in which we used to have much less capital to do bridge rounds and the bar for series A was incredibly high, we used to see a much higher velocity of pivots, so this idea of back a founder, let them pivot around, I actually think made more sense. There was also just less capital and way fewer founders. I think now there's just a glut of all of those things, and so what that means is that founders pivot less and also pivot less radically for a given entity. They might shut down their company and start a new one and go raise but cap table might be different, et cetera. So this idea of, "I'm going to back a founder because they're incredible and let them figure out what kind of startup they're going to build," I think is largely insane sans a few people.
MM: It's a very good one. How do you define what you invest in more broadly and what do you think that people often misunderstand about the category that you invest in?
MD: What we invest in at some point was called frontier tech, and that was something we made up at CB Insights. There's now called deep tech. There's a bunch of different definitions of this kind of feeling of company, and I think that that can be defined in two ways and they're kind of mutually exclusive in some ways. One is the framing of where in the commercialization phase is a given technology, and so that's where I think I would say is emerging technologies that have yet to have material commercialization or clear go-to-market paths. Sometimes they're past some sort of singular moment like Oculus and VR, Cruise and AVs or things like that. Typically the way you can kind of think about a lot of these areas is that they start as verticals, and when something is a vertical, it is... A vertical technology, rather. It is early, and when it becomes a more penetrated or more kind of horizontal technology, it is now in the next phase of commercialization. That's mobile, that's AI, I think that will be VR, AR, I think that will be crypto. Those are things that are verticals today, but they become platforms tomorrow.
“Typically the way you can kind of think about a lot of these areas is that they start as vertical technologies and when it becomes a more penetrated or horizontal technology, it is now in the next phase of commercialization. That's mobile, that's AI, I think that will be VR, AR, crypto. Those are verticals today, but they become platforms tomorrow.”
That's one, which is we get to things really early that we find weird and interesting and then we try and learn about them. We have a really strong view that... There's a famous saying like, "The way to learn about things is study them when they're coming apart." We think by studying things as they break down, and we've seen that in AI, we've seen that in crypto twice now, we've seen that in VR, we think that's where you really learn a lot to have kind of a long-term understanding of where value can be captured and accrued over various times in these phases of these technologies. Eventually they reach late cycle and hopefully you've built a defensible brand network effect in that area over decades, but I think for us it's about getting that early and exploiting it over many fund cycles.
The second framing is the type of risk in which you are underwriting at the earliest stage. We're first money in usually to founders, and that means you're doing things that have never been done before is one way to frame it, and that never been done before can be on a science risk side or an engineering risk side. Science risk would be things in bio or healthcare, engineering risk could be both bio, healthcare, AI/ML, robotics, a bunch of other areas as well. That's kind of our definition of the categories and the things that we like.
Your second question was around founders and founder types. I think, again, each of these types of businesses actually necessitate different types of founders where if you are not going to be going to market with customers for a long period of time, like bio companies for example, you maybe have a different profile, maybe you're super high attention to detail, very high-risk tolerance, very high velocity of learning, but not great at outwardly selling or narrative or building your storytelling in the same way. I think that could be one version of a trait. I'd say overall the version is super ambitious, very attention to detail oriented, fairly technical and always an understanding, not necessarily an ability, but always an understanding of what is the bleeding edge of their category, and the ability to as the time between academia and kind of operationalizing technologies compresses more and more, which is something we believe in deeply, that ability to know what's happening and bring it into a company framing I think is really, really important for almost every category we invest in.
MM: Mm-hmm. Yeah. I feel like that's the rarest skill, knowing how to turn it into a company. Do you think that's true?
MD: Yeah. I think it is, but I think my framing and my thinking on this sometimes has evolved based on some categories and founders I've worked with where I generally overbias towards, yes, the founders that can navigate the idea maze in an incredibly thoughtful way and have a very well-trodden path of having a hypothesis, trying to prove that hypothesis correct or not, and then moving on and working through that very quickly.
There's another framing that Cristóbal at Runway talks a lot about, which is... I'm blanking on the name of the actual framework, but the idea is that the way to get the right answer is to have the maximum amount of ideas possible and just put them into the world as quickly as possible and that allows you to actually then find the right answer, not being incredibly slow-moving and thoughtful. I think to his credit and to their team's credit, in AI they've recognized that for a period of time and especially in this moment of time they have. That's a framework that is a little novel to me and a little terrifying to me, but I think that those are very different skillsets of founders of velocity of ideas and being willing to be wrong quickly versus being super methodical and taking tons of data in and forming a view, which is much more where I bias towards, but it's not clear that's actually always the dominant approach.
MM: Yeah. Comes back to founder product market fit. Yeah. Interesting.
MD: Yeah.
MM: Okay. My next question for you, I'm curious what is your process for developing theses for Compound? Maybe mention a few of them and then we can dig deep.
MD: Yeah. We spend way more time reading and thinking and talking about these things internally than anyone else I know. That also means we spend way less time in meetings and attending kind of networking events than a lot of people we know. I don't say that as a like, "We're cool and other people aren't." It's literally like that's just how we prioritize our time. A lot of people do ask, they're like, "Well, how do you guys go about this? How do you read all these things, blah, blah, blah?" It's literally just hard work and time. A lot of that is because we always are reading research. We always are reading weird kind of visions of the future from engineers, researchers, operators, whatever in these R&D groups.
“We spend way more time reading and thinking and talking about these things internally than anyone else I know. That also means we spend way less time in meetings and attending networking events than a lot of people we know.”
We spend... Every few weeks, we have a session called Compound Learning Together where we sit as a team, we bring two to three what I would call 50 to 75% baked ideas with a bunch of research. We send this the night before. We talk through these ideas, and then that kind of creates a flywheel of us figuring out what we need to learn more and to dive deeper on. We do that pretty consistently. Sometimes we invite people from the outside as well. Then I just have an insane apparatus around reading where I have a bunch of things that I'm topic modeling and then having flow into an RSS feed. Really, it's just this perpetual cycle of read, talk, listen, write, learn, and that just kind of continuing to go. That's the process, but a lot of it is like, "How do we get to a point where we have a formed view that allows us to move quickly and talk at a level with a founder that maybe your average investors, a more generalist, kind of wouldn't be able to? How do we make sure we set up our schedules and our time so that we can do that and not go entirely crazy?"
“Every few weeks, we have a session called Compound Learning Together where we bring two to three 50 to 75% baked ideas with a bunch of research. We send this the night before. We talk through these ideas, and then that creates a flywheel of us figuring out what we need to dive deeper on. “
MM: Yeah.
MD: So there's that, and then that kind of feeds forward into like, "Okay, what are versions of the future we believe in, which is what we view our portfolio as a collection of?" That's sometimes also driven by external worldviews. Right now we're spending a bunch of time on energy, as a lot of people are, but more on self-sovereignty and decentralized energy grids as an area. That also is kind of related to another area of local compute and personal cloud computing. I think both of these are formed under a premise of a lack of trust at scale perhaps and more of a desire to have always-on computational systems.
In the energy side, we've done a lot of investing in crypto over the years and generally crypto is at odds with centralized entities. Energy grids is maybe the only one in history I can think of thus far that is net beneficial for the centralized players in the sense that we have very clear understanding that energy infrastructure is not going to keep up with demand in the near term and there has to be something that cracks within that. We've looked at a bunch of ways in which one could kind of invest against this, have value accrue within different types of companies. We also think technological shifts across a bunch of these areas are happening. Those are a few of the areas, and there's a bunch of others like robotics and kind of AI as well that we can talk through.
MM: Neat. Cool. Okay. Starting with decentralized energy grids, I'm curious to hear where you look for opportunities that would make sense for an investment from Compound, specifically venture-backable. I'm curious how you think about that.
MD: Yeah. I think there's steps. The first order step is something like Tesla Powerwall or some of these companies doing other types of engineering through heat from the Earth's core. I'm blanking on some of the names of those companies, but how do you just get sovereign energy broadly, which is fully independent energy? That's a first order thing. It's pretty centralized still, but I think over time what you're going to start to see is, okay, there's financing issues. We backed a protocol on the crypto side that is still stealth that's focusing on bringing renewable energy financing at a certain scale of 5 to 20 kilowatt and sometimes bigger and smaller on chain to enable better liquidity for these assets. That's one where there's a bunch of people who would love to build more renewable energy infrastructure but just don't have the capital and there isn't formed institutionalization of that. If you can bring capital, then that starts some flywheel spinning.
Then I think you have secondarily you'll then have a bunch of technologies that can sit on top of ways in which you securitize, which is what this company is also doing, tokenize the cash flows and build these things into a true alternative scaled asset. That creates a new incentive. Then you'll have broader understanding of what we think is groups of people that don't want to be tied to basically under the thumb of a monopolistic energy and/or utility provider across a bunch of different countries. Theoretically, which this is a very non-consensus, but you could have theoretically those entities themselves, as we talked about, also not wanting to deal with the infrastructure investment and the supporting of a large number of groups of people. You'll have different pockets of ways in which that will manifest, but we think that that generally will point into a broader view of people will want to have the ability to be somewhat self-sufficient across a variety of areas in their homes and/or in pockets of homes in a similar way that communes have operated in certain years. Obviously different but similar.
MM: Yeah. Yeah, that makes a lot of sense. It also leads to the second point you mentioned about local compute and personal cloud computing. I'm curious in that case, kind of a similar theme, what do you think it would actually take for something like that to go mainstream and be desired by the average person?
MD: Yeah. The realistic answer is some really terrible event, and that's the first-order thing that creates lack of trust at scale, and lack of trust is what will lead to more people wanting to have local compute. The second maybe more optimistic version would be a breakthrough in performance driven by whether it's constant retraining of models or something that is economically not possible. One of the things we believe that's really compelling is if you look at housing, there's a bunch of costs built into a home. Today if you wanted to run a fairly large language model or multimodal model generally, you would probably need about $24,000 of compute about, and that's going down pretty materially. But if you bake that into a home purchase and you run a cost curve, it's actually not that crazy to have a compute stack in your home in a similar way that you have a bunch of other core utilities that make your home run, whether that's heating with a boiler room and things like that.
That then opens up a bunch of ideas of like, okay, you can have personal models that consistently retrain, you can have privacy that sits in data that sits in your home versus not, you can have a bunch of systems that need to be managed but without the ability to be shut off. Again, if you have power that can't be shut off maybe then you can have compute that can't be shut off. Then you'll have probably some continual window that shifts across a bunch of different countries on how they view the rights of their citizens. An example in the US would be we have these insane states that believe women shouldn't be able to get abortions and/or they want to impact how people make that decision. You can imagine in a world in which you have a constant on biology monitoring system in your home that is running insights on a bunch of things about you, you might not want that beamed up to the cloud and run on compute somewhere else that could be potentially subpoenaed or taken. You might want that data run locally that you can control who go in and out of it, especially in a world in which possibly there's two to three major AI providers that do that.
We think that there will probably be more and more things that people will want to, either from a lack of trust side or from a performance side, will want to kind of have done locally. Then the last is there could also be some short-term pressure from an ESG perspective on some of the large compute centers and cloud compute centers, but that's something that's more of a short-term issue.
MM: Mm-hmm. How do you go about trying to find founders that are working on this problem?
MD: I just chaotically tweet a lot.
MM: That's good.
MD: Writing, tweeting, asking people like you.
MM: Bat signals.
MD: Yeah. When I go places, this is stuff I'm thinking about. I think in general, all of our online presences are like Rorschach tests for a bunch of people to figure out if they align with you or not and think it's interesting, and we just try and talk to as many people as possible. I've yet to find a way that is more scalable and higher ROI than writing, and so that's the main approach but also talking to the people doing some of the academic research. The thing you realize is not a lot of people care about a lot of the fringes of academia, if you might even want to call it that. These people are usually very willing to talk to you about and love the fact that you're interested in something they've spent years thinking about, but it's totally unscalable. There's no good system for it and we're just banging our heads against the wall until we can find people.
“In general, all of our online presences are Rorschach tests for a bunch of people to figure out if they align with us or not, and we just try to talk to as many people as possible.”
MM: Luckily, that's kind of the name of the game. Yeah. Moving on, I'm curious to hear more about how you think about robotics over the years and what you think is next in that category.
MD: Yeah. We think... Robotics I would say is one of those things that everyone has been promising for years and it's never really worked, and it's also been really misunderstood by a bunch of different types of investors. If you were to look at robotics in the past 10 years, I would say it's largely been a mechanical engineering problem and there's been a bunch of people who wanted it to be an AI problem. What that means is you have to set up systems that are not super intelligent and dynamic but can do things very precisely over and over again, and maybe they can have some simplistic things like computer vision but there isn't material intelligence. There was a wave of companies that tried to do this and largely haven't worked out.
I think what we've seen now is we're at a clear turning point. Our view is we are in... If 2016 was the early kind of transformer days, I think we are there in robotics with a lot of the stuff we're seeing from large language models, a lot of the datasets we've had in a similar way that image recognition had ImageNet. There now are these big datasets being opened for more complex robotic understanding of the world. I think what we're learning is these large multimodal language models do have some form of world model, and we've seen this through some of the work where our portfolio company Wayve has been publishing on being able to say like, "Hey, why did you stop at this light?" It's not just like, "Because I've seen that data a million times." They can reason and understand a bunch of things about the world. I think we are in the phase of it is clear robots are going to be very intelligent and work very well over the next decade plus.
I think what that then means, which is a really hard part to be at, is you have to then think about really a unique and beautiful products. For the first time ever, robotics is as much about technology as it is about product, which is really exciting but, as we do more and more work talking to these people, really terrifying because those people do not talk to each other at all, no one. There is such little overlap. I think a lot of it comes down to how do you ideate around where these things should be, what are the use cases that make sense, and in a world in which it's not all just humanoid robots, how do you build really unique experiences that create tons of value and also maybe are just first of a kind? That's really hard for us to do because we're creative, but we're not that creative.
MM: Yeah. Absolutely. It seems like it'll probably start with niche industries, very specific use cases that you probably would never have even heard of.
MD: Yeah. I think... There's a company called Matician, which we're not investors in but I'm fascinated by, and they're a great example of... Their first product is a autonomous vacuum robot, and it's the obvious thing. Everyone's seen that. That's the only robot that people actively buy today. Their whole thing was like if you apply a new layer of intelligence and understanding, you can do things like point at something on the ground and be like, "Hey, clean up that trash over there." That's a beautiful UI/UX. That's something that is so nuanced and so creative from a product perspective that I think that we just need more and more of those types of conversations happening. It's kind of like the Disney Imagineering vibes of these people that are starting first from a delightful experience lens and then working backwards into what that means and how it looks and how technology plays a role in that. I guess there's the famous Steve Jobs thing also of when someone interrupts him when he's talking about the Mac of saying, "How do the technologies work?" And he basically says he doesn't think about technology that way. I think we need more of that in robotics. Yeah, it's inevitable in my framing and it's just is it three years, five years or seven years? We'll see.
MM: Yeah. It feels like a lot of the paradigms of how we interact with robots, so it's the part that needs to be innovated on. We just don't really have them established yet.
MD: Yeah. I mean, it's all driven by... All of our robotic understanding is driven by science fiction and media.
MM: I was just going to say, yeah, it's like all movies, which is not usually the best.
MD: Yeah.
MM: Some are really cool, but-
MD: It's not.
MM: Yeah.
MD: But it does... I do think what we're going to start to see is more of the... As more paradigms from software-driven AI allow us to be a little bit more creative and think about new types of products that should exist that haven't before, that probably will feed forward a little bit into robotics. I think the difficulty is when you're constrained by the real world and physics, it becomes really hard and your iteration cycles are much slower. But I do think there's going to be a ton of people that have built a bunch of fun digital products in AI and kind of hit a wall of saying, "I've built a bunch of features, but I haven't been able to build a company." The way you can build a really interesting company is you have a physical moat by being great industrial designers as well and all these other things. That's really-
MM: Yeah. Owning your hardware. Yeah.
MD: Yeah. That's a really interesting world to live in. I do think right now it's like the hardware narrative is a little too artistic and not actually venture scale, but we'll get there.
MM: Yeah. Yeah. What ideas do you think that we're still not ready for in robotics or culture-wise?
MD: Ooh. Yeah. I mean, the culture one is always the sex robots and things like that. People love talking about that. I'm like, "Whatever. Sure." Honestly, I think it's just going to end up being... I think there's... The way I would answer that, which isn't a good answer, is it's more about the brain damage that exists within the robotics community of never do in-home robots, agricultural robots only lose money, all these types of things-
MM: Mm-hmm. Interesting.
MD: ... that I think those are the limiting factors more of how do people break out of the modalities either being a six degree of freedom robotic arm or a humanoid and nothing in the middle. I think like that, and then the others are very sensitive but fully automated things. There's a few people working on surgical robots that are not remote controlling the robot, it's entirely autonomous, or childcare robots is a pretty interesting one where it's like of all the things you don't want to be the first user of, putting your child in front of a robot is probably number one. But if the main limiting factor on people over the next decade or two is labor and a need to make more money and likely spend less time with their families for better or for worse, likely for worse, you can imagine a world in which these robotic systems are going to need to push the limits on a bunch of stuff that humans do from an emotional standpoint as well because our time is just going to be more constrained. But again, I'm mostly just trying to spend time with really interesting creative people because this is one of those areas where I definitely don't feel like I have deep conviction outside of simple enterprise use cases where robotic companies could exist.
MM: We just need all the kids to get really into robotics-
MD: It's true.
MM: ... and then come up a new wave of thinking it's the coolest thing to work on.
MD: Honestly, a realtime robotics simulator in Roblox, put that in and-
MM: Yeah. That would do the trick, honestly.
MD: ... how to be physics dependent and you'll be there. Yeah. I think someone will figure something like that out. Yeah.
MM: Good idea. Yeah, that's so cool. You mentioned AI. I'm curious to hear briefly how your and Compound's thinking has evolved on the subject and sector over the years because I know you've been focused on this for a while.
MD: Yeah. I think... There's a framing of all emerging technologies that we have which is that emerging technologies eventually reach a point where there's a net human and/or talent retention and it goes above 100%, right?
MM: Mm-hmm.
MD: So as people leave an industry, more and more people and waves come in to replace them. I think once you see that, that's a clear sign actually ahead of technological breakthroughs of, okay, there's something here. It's kind of like the Chris Dixon framework of studying what smart people are doing on the weekends, but I think it's a little bit deeper of an understanding of these ecosystems. I think if you look at AI, basically once we got the GPU and AlexNet in 2014 and once we got ImageNet being dominated and solved and object recognition being solved, we kind of reached that point of there's a consensus view that AI in some form or machine learning then would be effective and narrowly better than humans.
That created a floodgate of capital and talent that was going to persist, and the talent moved across certain areas. A lot of people in autonomous vehicles churned out because they were like, "I've been working on this for a decade and nothing is changing and I can't do this anymore." But it all stayed around that kind of AI vertical and computer vision and deep learning. I think for us, a lot of it... The evolution kind of moved from understanding that new AI-first products would be created that think about the paradigm of the UI and UX and how that changes with AI. I think Runway is an early example of that where they kind of looked at the bar of everything in Adobe on the side and they were like, "This is insane. There's all these little tabs and tools and that's not how the world should work. You put a text box and things change quite a bit." All the way to a bunch of other companies in the spaces that basically were built around that primitive, and it's a very simple... It's mind and framing.
I think the other thing that we realized early on was the best algorithm is not going to be the thing that wins. I think we've forgotten that. Not we, but collectively the industry has kind of forgotten that over the past few years as the companies with incredible algorithms have risen, namely like Anthropic and OpenAI, but I think that now it's about, okay, how do you think about perpetual value accrual and really this framing that I think a lot about in crypto, weirdly. Crypto teaching AI things is not where I thought I would end up, which is you... Because you're building open source software in crypto and the commoditization curve is effectively instant in a lot of these areas, you have to have an execution ability that is persistent for many years as you're building some form of moats and as the other people who are trying to compete against you in these death by a thousand cut ways die, basically.
I think what that means today in the AI world is it went from "what do you believe that no one else does" to "what can you believe and consistently ship and run this marathon as if it is a sprint for the next few years as capital just flushes its way through the system". I think our framing has kind of shifted in how we think about teams even from before it was like Alex Kendall at Wayve, this wunderkind PhD from the University of Cambridge who's one of the most brilliant people in the world and solved a bunch of these problems around segmentation to today can be really talented product people that understand the bleeding edge of AI but are never going to be the researchers to implement it. They can build a really interesting product that just continues to sprint across as other people do a lot of hard work for them.
“What that means today in the AI world is it went from ‘what do you believe that no one else does’ to ‘what can you believe and consistently ship and run this marathon as if it is a sprint for the next few years as capital flushes its way through the system’”
That's a very simplistic view, but I think more and more teams are like that. Runway was kind of weird because it was an incredible product team but also really talented on the AI side, but I think we are overestimating how legible credentials are in AI and what it actually means to be a talented AI builder. I think the fact that people lean on pedigree so much in this space is probably very wrong and is very much something that should be early commercialization cycle behavior, and if we are in inning three instead of inning one, it's likely a false signal.
MM: Mm-hmm. That's super fascinating. So you look more for really strong product people, basically.
MD: I think it's both, right?
MM: Okay.
MD: You can have a really nuanced view of... We have a company in our portfolio called Orbital Materials and they're a spin-out from DeepMind, and in their case they're the world leaders in foundation models for chemistry. They have a lab and they have a bunch of these really hard go-to-market moats and technology and that's a unique company, but on the software side I do think a lot of it is like how do you have a unique view of what AI can create from a UI/UX and product perspective and how do you know how you can sprint against probably some form of commoditization curve and death by a thousand cuts for the next three years as you build distribution moats, talent flywheels, brand moats, and a bunch of other things?
I think we're getting to the point of brand moats mattering more and more in AI because the consumer and user, whether that's enterprise or consumer, is getting a little tired of downloading the 50th new AI tool and switching and trying it out where nine months ago we were all like, "This is awesome. I'm doing this. I'll download every AI thing. I'm ready. Any image generation, any tech, whatever." Now I think people are kind of like, "I have my things. I like them." Midjourney is incredible, and you're going to have to do some really amazing stuff to get me to ever get off of Midjourney image generation. DALL-E 3 might be that, I don't know, Imagine might be that, but those brand moats are starting to manifest and that's pretty interesting when you think about what it then means to be a world-class AI company.
MM: Right. It reminds me of what happened to note-taking tools.
MD: Yeah. Yeah. Eventually... Note-taking tools, yeah, eventually you just become this standard and you then have people who try and do niche adjacencies.
MM: There's something for that, but it's just not going to be huge. Yeah.
MD: Yeah.
MM: Interesting. I'm super curious, what do you want Compound to become? What is your long-term vision?
MD: I think the thing that destroys all venture funds and maybe people in venture is scope creep. I think that that is the... You can see that for, I don't know, the past decade plus. So honestly, my answer is what we do at Compound is research-centric, thesis-driven investing. That's our tagline, that's how we talk about it. I think we want to be world-class at that. We want to be world-class at partnering with founders as early as possible and we want to always understand where... And have very defined viewpoints on where we think the world is going and deploy capital against those views with incredible people.
I have very little desire to meaningfully scale the fund size or try and be this multi-stage institution. I want to prove that we're really great at what we do and in a specific type of company, a specific type of founder, with a specific approach. I think to prove you're really great... I think we're doing okay right now, and I think you have to be pretty good for multiple funds. We're on that journey now. It's a pretty simple answer, but I don't want to scope creep into doing other things despite at times in the past few years, you look up and it's like 2021 and you wonder if you're the idiot and everyone else that you thought was dumb is smart. Then you look up in 2023 and you're like, "Maybe I am a silly idiot, but they were idiots too." It becomes like... You just have to be... Venture, there's no feedback loops. You have to very much trust the approach and the inputs at all times. I think we feel really good about our inputs, and we'll see if those outputs pay off for our founders and our investors.
MM: That's a great answer. It's just such a long-term game. Curious to hear more about your philosophy on writing. When do you know it's ready to share? In what form? How have you thought about it?
MD: Yeah. There's a principle called Cunningham's Law, which is the best way to get the right answer on the internet is not to ask a question, it's to post the wrong answer. That is-
MM: Extremely true. Yeah.
MD:... extremely true. Some of it is that, which is like, "I have done all I can and I want to now put this in the world and have people tell me I'm an idiot, and that will make the thought process even better." There is one which is, "I have a view on how something could change or something that's interesting, and I'm starting to hear more people talk about it. I really want to make sure that my view is early and timestamped and there and can be scrutinized." I think some of it is, "I don't see anyone else talking about this thing and I just think we need to." That's another version. Sometimes it's just like, "I am tired of researching something. I don't feel that strongly that my view's great, but it's good enough that from a VC lens, the bar is so unbelievably low I can just fire it out there and hope someone thinks it's interesting."
"‘I have done all I can and I want to now put this in the world and have people tell me I'm an idiot, and that will make the thought process even better.’”
But I think a lot of it is... Time and time again, I just come to realize, especially in venture, there are so many styles and the style of investing we do is both incredibly replicable in the sense of there isn't any... Our moat is how we approach the job. It's not some big tech stack, it's not some algorithm, it's whatever. Also, it's incredibly defensible because people in venture are just miserable at putting the right inputs into long-term oriented things and doing certain types of work and time management and these firms decay over time. Not by them doing anything wrong, it just happens. Our view is just the more we talk about things, the more we dive deep, and the more we can connect with founders, researchers, interesting people, we should try and do that.
We're trying to experiment now with more private centric sharing as well on top of our public stuff. We launched this thesis database, which was more public a year ago, where we are very prescriptive on company types, et cetera, but it's always evolving. But the main thing is we just want people to feel like we have done a good amount of work and thinking on our own, and that I think goes a really long way when you then want to talk to people because they can feel the genuine passion about and the work you've put into something versus pinging some person being like, "Hey, I'm digging into this AI. I'd love to steal 45 minutes of your time to hear how you think about it because you work at OpenAI." Those people are going to be like, "Great. You're the 10th person. Awesome."
MM: Yeah, 100%. I remember a framing that you mentioned, if you don't agree with someone's worldview or something expressing it that way like, "This is how you see the world. This is how I see the world. There's a discrepancy, and that's the reason that I'm not interested in pursuing it further." I think that's a really good way of putting it which you can only do if you've put in the work beforehand.
MD: Yeah. Yeah. I think also just this job is so... So many egomaniacs in venture and just approaching it with this idea that we easily could just be totally wrong. The reality is is the beauty of our structure as a firm is we need to be right to two times every four to five years, which is how long we invest each of our funds, and then we're going to be wrong 30 other times. It's pretty incredible. The thought process really should be just how do you maximize being deep and generative and low ego and high EQ? You might totally go down in flames, but maybe it'll work out. Who knows?
“The thought process should really be just; how do you maximize being deep and generative and low ego and high EQ? You might totally go down in flames, but maybe it'll work out. Who knows?”
MM: Totally. Yeah. Enough shots on goal. My last question, I'm curious to hear more about the inputs that you mentioned you feel good about and what are your favorite ones that you return to and challenge your thinking most or theses?
MD: I'd say the challenge thinking is the team, spending time with Shelby, David, Smac, a bunch of our venture partners, that... Being able to have people that are obsessive about the craft of investing in deep tech and talk about things. I can come with an idea on some deep plant biology thing and Shelby, who's a PhD in plant biology, can be like, "You're an idiot, but also this is an interesting idea." The culture of being able to do that is awesome. I think... So that's one. I'd say we... Again, I kind of mentioned I have a very tuned RSS feed. I'm obsessed with RSS as a concept. It totally died. I still use it all the time where I have a bunch of topics that get automatically brought in from a topic modeling funnel across a bunch of different sources, some of which I don't even know.
“I have a very tuned RSS feed… I still use it all the time where I have a bunch of topics that get automatically brought in from a topic modeling funnel across a bunch of different sources, some of which I don't even know.”
I spend a bunch of time on archive and social sciences research just to see how people are trying to understand the world and from a research and scientific view. Often that's technology, but sometimes it's not. Then there are other areas where people are just forever... Maybe they're a little crazy, maybe they're really deeply thoughtful, but stuff like the Effective Altruism Forum and LessWrong and these random subreddits of people who are just obsessed with thinking is really fun. It's a totally different part of your brain, but it's a level of rigor that we don't get much on tech Twitter, which I also love but it's totally different. Then weirdly I'm a daily active user of Tumblr. I have a very tuned Tumblr feed.
MM: It's great.
MD: I love that for all sorts of random stuff, but Feedly is probably the main thing, that's my RSS reader and everything funnels through that, and then just an insane amount of Twitter. But all those things then funnel into me being able to... I try and send, I don't know, 10 to 15 probably realistically cold emails a week to random people that I think have written something or researched something. Andy Weissman at USV is an incredible investor, very close friend, one of my favorite people in the world, and he, as many people know about him, he just cold emails everyone. He'll read a book and email the author and they'll respond sometimes or watch a movie and find his way to the person who made the movie and just send them a note. He sends super thoughtful, deep emails.
I'm trying to incorporate that more into my learning process because I do think that serendipity, especially when you're a little bit unknown and moderately early in your career, is... You have a right to be able to talk to people or rather people are willing to talk to you a little bit more than once you're more senior or more well-known or viewed as more successful. I think that decays very quickly, and that's what I tell all people early in their careers is people will talk to you just because you're young and early and it's a beautiful thing, and just push on that until they're tired of talking to you.
MM: Yeah. Such good advice. People always think that they lose the student card when they leave school, but they do not.
MD: Yeah. Yeah, exactly. It's insane.
MM: It sounds like you've built a serendipity machine. I really respect that. I think that's great.
MD: An introverted trying to... I don't know if... Yeah. Trying to engineer serendipity, which is not a real thing, but I think it's just trying to find people that you can connect with them on some feeling of how they think about the world and things you care about or whatever. Yeah, I think that's how we met. We just definitely have a view on the depth at which we care about things, which is an insane statement because that's so general, but you feel it so quickly and at times that's why I think places like SF and New York are so special is because you just have really high alignment on a depth of caring about things, and then you can go a layer deeper on what are the things you care about. But just that first order handshake we all make is pretty incredible because it just helps make any topic of conversation more well respected and digested, I guess.
“I think it's just trying to find people that you can connect with them on some feeling of how they think about the world and things you care about… We definitely have a view on the depth at which we care about things”
MM: Yeah. Is it sincerity, earnestness? I don't know what it is.
MD: Yeah. It's like-
MM: Something like that, yeah.
MD: It's like some form of insane optimism, some form of sincerity, some form of we all spend too much time on the internet.
MM: Yeah. Curiosity.
MD: I haven't quite gotten what it is yet, but it's something.
MM: Lovely. Well, I think that's a great place to wrap up. This has been so lovely, Michael. Really enjoyed it. Thank you so much for coming on the podcast.
MD: Me too. Yeah, thanks for having me. This was great.
Subscribe for more episodes to come: Spotify, Apple, Google, Amazon, and Overcast.
Share this post