Speculative Obsession

A conversation with Michael Dempsey of Compound.vc, recorded April 8th 2025 and published April 12th on Kyle’s channel. Roughly 53 minutes. Watch on YouTube.

Summary

Compound describes itself, in a deliberate word order, as a research-centric, thesis-driven investment firm — and explicitly not as venture capitalists. The whole conversation is an argument for why that ordering is the strategy rather than the branding.

The structural case comes first. Dempsey’s starting question is not “how do we win deals” but why should this firm exist at all — he does not think the world needs more venture capital. If the only comparison available is to other funds of the same vintage, “you’re just beta to the market,” and that is not what hefty fees are for. The answer he lands on is smallness used as leverage: $800K–$2.5M checks, leading or co-leading at pre-seed and seed, deployed over five-year investment periods (2016, 2021, 2026 vintages), in areas carrying real science and engineering risk. One of his LPs told him “we don’t want you to manage risk — we diversify our risk outside of you”, which he takes literally: Compound should be the riskiest dollars anyone deploys.

The research playbook is the substance of the episode. Read, listen, write, as a cycle. Private one-pagers shared with operators to find unknown people. Thesis development sessions every three to four weeks where everyone brings two 70%-baked ideas. Research days that mix founders with academics — one of which produced two researchers arguing in front of the room. And a public thesis database, which Dempsey frames not as giving away edge but as a targeting mechanism: most people treat writing as top-of-funnel, “it’s really more like middle to bottom… more like a sniper than some wide net-casting thing.”

His biosecurity example is the clearest illustration of how the method actually pays: no commercial companies were working on it, and the reason turned out to be a legible gap — easy to reason your way to $10–50M of revenue, very hard to reason your way to $150–300M. Naming that gap is what lets you have a useful conversation with a founder about years three through five.

On propaganda. Kyle quotes Dempsey’s own line back to him — that the job is to make the world believe what Compound believes, 18 to 24 months later, and that you could call the research organisation a propaganda shop. Dempsey’s answer separates the two muscles: the quality of the first-party story, and the credibility behind it. He describes a sequence — the firm believes, so talent believes, so later investors believe, so customers believe, and it is the customers parting with real money that feeds credibility back. Hence the instruction he gives portfolio companies: every update should state how you are positioning to hires, to investors, and to press, because “there’s nothing worse than someone misunderstanding your company and then communicating it.”

Where they disagree with the market. Dempsey’s read is that Compound’s original arbitrage — being the firm that tracked and understood research while the gap between research and commercialisation collapsed — has been competed away. “A firm that tracked research maybe exists; understood research no longer exists. You dump a paper into ChatGPT and ask it to explain it like you’re 16.” The edge has to move to second- and third-order reasoning. On moats, he is blunt that models and technology were never it: brand and pace are the moat, with Runway as the worked example.

The last third is the part that will age most interestingly. Kyle presses on capital destruction — his worry is not that a technology fails but that the sheer scale of money behind it, when it fails, kills public appetite the way it did for nuclear energy. Dempsey largely accepts the framing and declines to be upset about it: “if people want to blow themselves up chasing low-conviction, high-consensus areas, I am all for it.” His defence is discipline — stay small, stay wrong cheaply, and you can be wrong and still return more than people who were right at scale.

The title comes from Kyle’s own aside mid-conversation: people see the writing and conclude they should blog more, which misses it entirely. Obsession is not incidental. Obsession is long-term by definition.

Transcript

KH: All right, Michael, thank you so much for joining me to chat. I am very excited about this conversation.

MD: Thanks for having me.

KH: I wanted to start first by thanking you for exposing me to one of my more recently added sort of favorite quotes. I saw this quote on your website where it says, “Research is to see what everybody else has seen and think what nobody has thought.” And I was like, that’s — I love that. That’s perfect, because I feel like what most people in our industry are trying to do is figure out, how do I think exactly what everybody else is thinking as quickly as possible? Like, I don’t want to be thinking what they’re thinking a couple months from now, I’ve got to be thinking it as soon as possible. And that is the opposite of what we should be trying to do.

So I really appreciate that. And I also feel like it is very representative for Compound. I love how you guys frame the firm as a thesis-driven, research-centric investment firm. What I wanted to start with was a brief biographical sketch of Compound as a firm, because I honestly feel like you guys are an example of what I wish more people in venture were. So I want to get it directly from the source: what do you feel like it is about Compound that is unique and special — that’s worthy of my idolatry? How do you think about the firm?

MD: Unclear if it’s worthy or special, but I think in general we’re very intentional with our wording, and in the general aesthetic we put into the world. Research-centric, thesis-driven investment firm. We don’t call ourselves venture capitalists. We put those things in that order. We take a prescriptive view.

Compound generally is built around this idea that in order to be in an increasingly crowded venture capital world — and an increasingly hard area to be consensus and make money — we should be prescriptive. We should be deeper. We should be focused on our own process and be willing to go down with that ship.

It started in 2016. There was a previous fund that existed, and when I joined we kind of rebranded the firm. Compound One, our first fund, was raised then. Our approach was trying to build this more deeply research-oriented approach to investing, where we could say the thing that all investors tell you not to do — which is, this is what we think the future is going to look like, and these are the types of companies we think should exist — and doing it in areas that felt non-obvious or harder to understand and parse. Because that’s how we could make money relative to other really talented venture investors that were candidly just ahead of us, and bigger than us, and maybe better than us at the time.

So our approach really is: do a bunch of research, figure out where we think the world is going, figure out the ways in which we want to invest against that. Largely we do that at the seed stage by leading or co-leading rounds, and we do it slow, steady, and at small scale. I wish a lot of people in venture did that, and I’m also sometimes glad that they don’t.

The next order is just trying to figure out, as venture evolves and as some of our areas become hotter — we spent a bunch of time early in crypto in 2016 and AI in 2016, and now those have gone through various cycles — figuring out what is our role in those ecosystems as they mature, as they go through ups and downs, as they become more consensus versus less consensus. It’s a perpetual pursuit. I talk about this a lot, but venture is best when it’s a craft. Our view is all venture firms decay slowly over time, and we have to fight against that decay by continually improving at a rate that outpaces it. That’s what craftsmanship is.

KH: A couple of things I like about how you frame it. Number one, you use that phrase a lot — these are ideas that we’re willing to go down with the ship on. Literally the definition of conviction. I want to talk more about that, and the convictionless nature of a lot of investing today.

But I also feel like the way you structured Compound vibed very well with your background, because you spent a couple years at a hedge fund, you’d spent a couple years doing private markets research at CB Insights. So obviously you’d been very research-centric. It’s funny, I don’t know that I’ve thought about it this way, but I had worked at a hedge fund as well, at Coatue. One of the ways that we framed what we often did was like big idea arbitrage — if you believed in something that had a larger potential nominal value at the end, you can pay higher prices, you can be more aggressive, you can do all these things. That is a very hedge fund mentality, but you’re applying it to venture investing.

Most VCs try to avoid saying what they believe, because they say, oh, we just bow down to the founders and we let them define the future for us. But that often leaves them in the passenger seat of I’m just along for the ride — they said the future was going to be great, let’s see how it turns out. Were there analogues that you used to say, this is why we think we should do it this way? Or other firms or philosophies that informed how you guys thought about the world?

MD: I think honestly just my first job being at a hedge fund, trading long-short and derivatives. You are forced to take a point of view on the world and have high conviction, such that when things run against you, you either know you were wrong, or you still believe you were right and buy more.

I also think there’s a more structural question, which in 2016 was starting to become prevalent and in 2025 is incredibly prevalent, which is: why should this firm exist, or any venture fund exist? Unlike what some other people believe, I don’t think we need more venture capital. I don’t think that we don’t have enough risk dollars at play. So a lot of it is just, where’s the alpha to be generated as an investor, and where’s the alpha to be generated as a partner for founders? Why are we above replacement to another venture firm, to take a framing from sports?

A lot of that comes down to: if there is some way to have the possibility of generating alpha, to see things that other people haven’t, the way to do it is to be a better partner to founders with smaller scale and theoretically less people on the team, and thus less human capital. What is the way to do that at super high leverage, and why does that matter?

There’s a view in venture that you have to play the game on the field — we’re vintage-based investors, so the comparison theoretically is to other people in the vintage of the year your fund was raised. I think if you view that as your role, you’re just beta to the market. I don’t think that’s really why someone should pay very hefty fees, which is what they do to invest in venture funds. We take the idea of managing people’s money very seriously, and don’t think that — an LP told me once, we don’t want you to manage risk. We diversify our risk outside of you, so you do what you believe most. And that’s fine. So I kind of think we often should be the riskiest dollars that anyone deploys, period. And hopefully that does well.

KH: That’s a solid LP. That feels like somebody who understands more the systemic potential that the asset class is supposed to have, as opposed to the very idealistic view that most people take.

MD: They’re not in any of the big funds, so that’s why. That’s not surprising.

KH: Not surprising. So I feel like the firm is set up in a very deliberate way. I also feel like the team that you’ve built — speaking of how you think about really specific leverage — your team is very unique to any other firm. Everyone on your team has a very atypical background as an investor, but are usually the smartest people that I come across. It’s a pretty solid collection of people.

We talked about your background a little bit. There are folks on your team that have PhDs in biological sciences. There are folks that have gone super deep in crypto and cryptography. But you also have a deep bench of venture partners who have really deep experience in automation and robotics and AI. Walk me through the framework: what’s the special sauce that you want to see in someone to want to bring them onto the Compound team?

MD: For a long time, from 2016 to 2020, we didn’t hire anyone. Then in 2020 we made our first hire, a woman named Nicole, who is amazing. I found her on Twitter, basically just obsessively and chaotically tweeting about AI research papers and other things. She was debating whether or not she wanted to go work at a big tech company or do something else. I DM’d her and said, why don’t you not work at a big tech company, and figure out maybe you want to do venture — and here are some books. And she read them all. We hired her without having ever met her, because COVID happened.

From that year onward, we’ve hired one person per year, pretty deliberately. For us, we know that there are a few core things that matter. The first that we always say is obsession. We are an obsessive firm, and you can view that through the lens of our Twitters, which are highly chaotic but highly obsessed with the areas we spend time in. It’s also shown through — we’re always online, always talking about different ideas. I don’t think this is a firm where we view our job as not fun, and done at any point. So I filter for that.

Second, in some of these areas the dynamics are so nuanced and the rabbit holes are so deep, you want people who can specialize while still having a frame of reference. People frame deep tech as sometimes the adjacent possible, or for us, the ability to speculate on futures that they believe in.

Biology was a big one for us. We felt that biotech broadly was going to be one of the largest beneficiaries of automation, both on the software side and the hardware side, over the next decade, and we knew we wanted to continue to have expertise there. That’s why we brought in Shelby, who has a PhD in bio, and also has that lack of risk aversion that normal PhDs have, and a certain attitude toward the institutions. We think that’s important to be a good, successful biotech investor at the seed stage at a small fund.

Similarly, we had been tweeting and talking on Twitter for a long time. Smac [?], an anonymous investor, has been around crypto for a long time, and has a background in traditional finance as well. For him it was very easy to understand how he would bring a very different lens, both on the crypto side and on very prescriptive views of what he believes the future looks like in the world.

Both of those people embody the thing of Compound, which is that we are a very generative firm, and our view is: how do we continue to generate creative possibilities within technology and science? And McKenzie, as a researcher on our team, similarly untraditional background — has worked in traditional finance and other types of investing roles in the past, and really was just an obsessive researcher who blogged a lot on his personal blog before coming to Compound, and has the ability to traverse all sorts of areas. For his interview process, we asked him to do a research project on what comes after transformers. We gave him 72 hours, and what came back was 30 pages on the future of AI over the next decade from an architectural perspective.

So it’s very clear that those types of areas allow us to do what other firms are afraid to do, which is speculate on specific futures on a reasoned, sequential basis. And then the venture partners help with zooming back down into how do you operationalize these companies that are trying to do things that have never been done before.

A principle of Compound that we talk about privately a lot but don’t say as much publicly is: we think we’re pretty good at investing in N-of-1 businesses at the time — companies that don’t really look like other companies in the moment. That’s rare and getting rarer. But we’re not very good at picking the best of the five companies. We don’t have a legal AI company that we’ve invested in. We don’t think that we can make money that way — or rather, if we can, we might just be lucky versus having an actual edge. So our venture partners help us and help our founders reason through these N-of-1-style problems that a lot of these businesses have. Those are people who are highly specific, maybe not always the flashiest names, but are just great operators in areas like data science, machine learning, consumer, bio, stuff like that.

KH: I heard a founder once describe you guys as probably the only Twitter-native firm out there. You literally can see your thought processes bleeding out into different posts and threads, and if you map that with the research that you publish, you can see where the atoms form into a specific cohesive idea. Which is super rare. Even the interview question of go speculate on the future of this specific infrastructure — that’s so different from let me give you three companies and you tell me which one of them is good. It’s a totally different framework for how someone’s brain works.

Before I talk a ton about research — we’ve talked about the firm, we’ve talked about the team — I wanted to give you the chance to articulate the fund specifically. How do you think about fund size and check size? And maybe a teaser as well: how do you maintain your fund strategy in a world of increasingly louder and bigger funds? I think just recently it came out that a16z is raising a $20 billion fund. How do you exist in that world?

MD: Our core principles are: we deploy over really long periods of time. We love time diversity. Our first fund was a 2016 vintage, second 2021, third will be 2026. These are five-year-long investment periods to build portfolios. We’re highly concentrated. We write checks between $800K and $2.5 million to lead or co-lead rounds. We do it all at the pre-seed and seed stage, and we do it across these core areas that are taking large amounts of science and engineering risk, largely in North America and Europe.

The other thing we believe is that this approach is highly alpha-generating at the very early stage, where there are people who just don’t want to spend time doing non-consensus things, or things that look weird or have a lot of need-to-believes and second- and third-order reasoning. And then also on the later-stage side — we have some investing things that we do that are more on liquid assets as well. That in part is my background, and partially is just where we have started to notice increased dispersion and increased impact on larger, at-scale businesses in the areas that we spend time. We think those are only going to materially accelerate.

In terms of as venture grows, we continue to believe in just raising small funds and doing what we do. I don’t think we have the ambitions to go and try to compete at Series A like a lot of firms do. It’s less profitable on a fee basis, but hopefully the carry side makes up for it. For us, we’re pretty boring. That’s what we do — we focus on this craft. We don’t care about accumulating power. We’re not going to try to make our way into the White House at some point in the next decade. We just want to help really ambitious people, often who maybe have slightly off-center backgrounds, or are first-time founders doing very ambitious things.

Hopefully why founders want to work with us is because we are pretty high EQ and relatively high and specialized IQ in these areas, such that they don’t have to give a bunch of context ahead of time when they want to work closely with a partner. I think we are something for someone. We are not everything for everyone.

KH: I love that. You mentioned liquid assets. Is that primarily on the crypto side, or will you do stuff like take positions in public equities?

MD: Both. Long-short public, and liquid assets in crypto.

KH: And is that all out of the same fund, or different vehicles?

MD: Different vehicles, for sure.

KH: That’s good. I didn’t know the diversity in terms of the asset side. It’s always interesting to me, because classic venture can feel like really trying to nail a couple of different rings with one very small arrow at the same time. It’s not just about having the right idea — it’s having the right idea with the right person at the right time at the right round that fits your fund. You’re hitting everything exactly right. But if you really are, to your point — the wording is very deliberate — if you are thesis-driven, the thesis comes first. Here’s this idea. If you find an opportunity to put capital behind that idea, even if it’s not at the seed stage, that’s fine, because the thesis is in the driver’s seat.

MD: We are at a moment in time, in our opinion, where there’s really high dispersion — even before this past week. And there are a very high number of dark forests to navigate across economics and technology and science. Our job is to prescriptively navigate those dark forests, those idea mazes, in terms of how value will shift. Our entire process is oriented towards that, and working with founders to do that.

Sometimes underwriting on the founder side is about the people who can take the most shots on goal to figure those things out quickest. Other times it’s people that have incredibly precise views and can execute over long periods of time very precisely, with very high cadence and very high competency. We think about a lot of those things depending on the space we’re operating in.

KH: So we talked about the first part of being thesis-driven. Research-centric comes next, with research at the center of everything you do. People talk all the time about building in public, and I genuinely do believe — having now written my blog for, this is my fourth year, I haven’t missed a week. My only goal was, I just want to write something every week and put it out there. And it has 100% demonstrated the power of this idea of putting vibe out into the universe and seeing what responds to you, because you just don’t know who’s thinking the same things.

You guys have this really awesome thesis database where you’ll codify these ideas — here’s how we’re thinking about it, here’s some of the reading that’s informed us. It’s this big Notion database of a bunch of different things, which feels totally antithetical to what any VC would do, because they’d say this is our edge. But you’re putting it out into the universe. Walk me through how you think about the research playbook. How are you dictating your time? Where are you going deep? What are the best sources? Are you getting those from people, versus just reading a ton of research papers? What does that look like in practice?

MD: I guess I’ll scattershot it. The general process is read, listen, write — those are a cycle. Usually we try to build foundational views on our own, and that comes from — we have some apparatuses that we use to track research papers, that sometimes comes from looking at GitHub repos, spending time in Discords, whatever it is depending on the category. We then try to build one-pagers or little databases or mini-views on a private research repo, and sometimes we’ll share those with friends or other researchers or operators at companies, and that will help people point us in the direction of finding unknown people.

And then it’s having conversations and saying, hey, why you want to talk to Compound is not because we are going to “pick your brain” — instead we’re going to have a point of view, and you’re going to want to spar with us, or you’re going to be like, wow, this is great, I’ve never found someone who wants to talk about this thing.

A good example over the past year: for the past 18 months now we’ve been pretty interested in biosecurity as a concept. One of the most interesting things is there’s no commercial companies really focusing on this. Obviously Ginkgo has a small group that continues to operate in this, Palantir has a group within bio. But we built a doc that outlined a bunch of the areas that we thought were interesting, and the philanthropic community actually took hold of it and pointed us to a bunch of different people. That allowed us to further refine our thoughts, further figure out what the right customers are. And then there eventually does come the traditional — how do you understand customer development, how do you understand sales cycles, how do you think value accrues.

The biosecurity one is most interesting because you get to a point where — what are the need-to-believes? If there isn’t a venture company really working on this, usually the reason is because there’s a high level of uncertainty somewhere in the journey. In biosecurity the obvious answer is: it’s very easy to reason how you get to $10 to $50 million a year in revenue. It’s very hard to reason how you get to $150 to $300 million in revenue. That means you can talk to people about the logical gaps in company building — hey, if you want to build a company here, we’re going to figure out in year three, four or five how this actually becomes venture scale. And that probably means there’s a bunch of funds that won’t touch this area for some time.

A lot of it is just that process. And then we do these things internally where we have thesis development sessions. Once every three to four weeks we block off three and a half, four hours on our calendars, and everybody brings two 70%-baked ideas around where companies should exist, or here’s a specific thesis I have. We try to talk through them and reason through them as a team. We invite people from the outside to come in. Sometimes they bring ideas, which opens up our aperture more. Sometimes they push on our ideas because they’re an operator at a company that we’re talking about.

That helps us do the thing that I think is also really helpful, which is it allows everyone on the team at any given time to have context of what everyone else cares about. When you have a team of specialists, and a team that is trying to be highly generative, you have these end artifacts that are a little more polished, which is writing. You have these middle-funnel artifacts, which is the thesis database. But you don’t have a lot of shared understanding of the top-of-funnel stuff happening in all of our brains. This allows us to be sitting in a meeting and be like, oh yeah, Mike actually is interested in human challenge trials because he talked about this two weeks ago. I don’t have all the answers, but I know this, and this is why he’s interested in it — you should talk to him.

In a lot of firms there’s just a lack of shared context at that layer, the early ideation layer. USV is probably one of the firms that does this best, and we’ve actually done shared partner meetings with USV where we’ve talked through a bunch of theses independently. We have a close relationship with them because I think we operate in somewhat similar ways, and obviously sometimes very different areas.

The last thing we do is we host these research days, where we bring together purely — 70% people who are not building companies yet: researchers, operators, maybe people in academia — and say, here’s an area we’re thinking about. We want to bring some startups that are interesting and early, working on some ideas, but we also want to hear your ideas around where you think the industry is going, maybe abstracted away from your job. That creates a node where we get to see the weirdest people hanging out and talking about stuff, and also sometimes debating stuff. We held a biohacker research day last year with USV, and in the middle of one of the presentations two of the researchers got into an argument in front of everyone.

It’s really interesting to see, and I think the beauty of venture is we have the ability to partner with founders on both sides, or various parts of this idea maze or value creation path. Seeing those debates play out in real time allows us to draw clearer lines of where the different paths of the industry might lie. It’s a chaotic process, but we try to introduce consistent cadence such that there’s some organization.

Maybe the last thing, which I think is maybe most unique, is we try to do these speculative fiction writings that we put out. The reason why is that as industries mature they become very consensus at a high level and very unclear at a meta level. A version of that could be AI today. Everyone kind of knows AI makes sense and it will be big, but people have shifted their conviction on what part of the stack is going to accrue value, on time horizons, etc. Crypto has always been that — highly cyclical, highly low conviction, high consensus.

So we try instead to understand this idea of, on a year-by-year basis, how will these industries change? What will happen? What won’t happen? What will the big companies do? What will the small companies do? We put out speculative fiction that allows people to have some frame of reference for how we think about things on precise timelines over the next five to six years. We did it in crypto, we’ve done biohacking, we have some internal stuff on robotics and machine learning.

It also helps us sharpen our view, because of the canonical wisdom that being too early is the same thing as being wrong. So we try to take that prescriptive viewpoint to a local maximum of, hey, this is exactly what’s going to happen in 2026 versus 2029. And it allows founders and other people who are ambient to the spaces that we care about to come to us and say, this is ridiculous — or, wow, it’s so cool that you guys have this point of view, let’s talk about it.

KH: I’m salivating. There’s so many pieces. The first thing I want to talk about is the USV stuff, because that’s exactly where I was going to connect to next. And then I want to talk about the speculative fiction stuff too.

On the USV front, a couple weeks ago I shared a video that I thought was great. It was an interview with Fred and Brad, and they were talking about — they’ve been around for a while, and they’re still first in the office, excited about the job. The reason they get excited about it is because they are allowed to be. The way that I tweeted it maybe sounded meaner than I meant it to, but they like it because they get to be the dumbest person in the room. They get to come in and have really smart people come in and want to talk to them. And they’re not just doing it out of the goodness of their heart — they’re doing it because they can write a check.

But there is this really interesting dynamic that you want to try and build with these very intelligent people, where the currency that you’re bringing is not always hey, we want to invest in what you’re doing. The currency you’re trying to bring is we have gone deep on various aspects of this thing. We want to engage with you — not as “I don’t know anything, come teach me about it,” but as an equal. You guys have a very similar worldview, where you let the thesis drive, and as the thesis drives you into these different pockets you try and engage with folks.

But that can be the most difficult part. These people are obviously very sharp folks, but also very busy, highly sought after. A lot of people want to pick their brain, to your point. How would you describe the mechanism that is most successful for you guys when you’re trying to engage? Is it just, here’s the corpus of our work, let’s chat — and they almost always respond to that?

MD: Most people in venture think about writing as a top-of-funnel process, and it’s really more like middle to bottom. It’s more like a sniper than some wide net-casting thing.

Sometimes you do things like, you’ll tweet about this very strange idea that we’ve put in the thesis database, and people will come out of the woodwork and they want to talk. In those cases, those people are actually starved for connection, because nobody has cared about what they care about. To know that someone does is much easier. And to know that someone cares about it in a relatively translational way is also helpful, because we love science, whatever, but we are pure-play capitalists. We care about value capture at really large scale.

On the hotter areas that everyone’s trying to dig into — listen, you can go talk to the investor that claims that they’re super excited about creative AI, but here’s a 90-page book on animation from 2016 talking about GANs, and here’s these posts. All things being equal, who do you think you’re going to have a more pleasant and/or productive conversation with?

There are some founders that candidly don’t give a damn about what venture investors think, and they’re like, I just want someone who’s going to leave me alone and give me the highest price. Weirdly, I think those founders are more often first-time founders than not, which actually makes intuitive sense — to the point of Vinod and all these other people, venture investors are often net negative to companies. If you’ve gone through it a few times, you can understand what having someone who doesn’t give a damn can do on the downside, and what someone who gives a damn but doesn’t know what they’re doing can do on the really bad downside.

Our view is, hey, we think we know what we’re doing — you, who knows, we’ll see. But we care, and we are knowledgeable about the area you’re spending time in. At a minimum, you don’t have to explain the same thing you’ve maybe had to talk about with a bunch of other VCs, and you’re not going to get questions that are the same. So it’s more of a way and reason for people to talk to us. As our firm has grown in some sort of brand, it’s worked as top-of-funnel a little bit more. But there’s so much random writing on the internet now that our goal really is to be able to send links that show that at a moment in time we had prescriptive views. That’s also why we try not to publish things that are like “this space is interesting, talk to us to find out,” and more “this is why this is interesting, and these are the types of companies that should be built here — and if you think we’re morons, we’d love to hear it, because you probably are smarter than us.”

KH: We get a similar question. Contrary thinks about talent as a key piece of what we do, where we’re trying to identify folks as early as possible — years, in the majority of cases, before they start companies — and have a platform that is valuable to them throughout their career. We get questions all the time from other investors that say, I’d love to build a little scout network, what’s the easiest way to do that? And it’s like, I don’t know. The only way we know how to do it is to spend the better part of a decade building relationships and letting them compound over a serious amount of time.

It’s the same thing — people will be like, oh yeah, those Compound folks, they write a lot, I should blog more. And it’s like, you don’t understand. I keep thinking about what I want to title this video in my head, and I think the thing I’m landing on is this idea of speculative obsession. Obsession is not incidental. Obsession is long-term by definition. And so you guys have to have demonstrated that, which is huge.

The last bucket I want to click on before we move on: you mentioned speculative fiction. There are definitely some really interesting examples that I have seen play out. The first one I think about — we’re investors in Anduril, and there was a book written a couple years ago called 2034 that basically laid out a hypothetical future case scenario of how a large-powered global conflict could play out. A lot of it has eerily — the framing of how different powers have sided in certain ways, different moves that people have made, how technology has evolved — is eerily playing out in that way. And that’s not a function of “he guessed well.” The authors of that book had spent a lot of time thinking about those driving forces and what it would likely yield.

I’m curious if you have analogues for, oh, this is the type of fiction we’ve seen elsewhere — whether it’s popular fiction or more of the type that you’re doing, which is more tactical fiction — that you look at and you’re like, that’s a really good example of how we like to do things.

MD: We read a lot of science fiction stuff. I would say the closest example is probably LessWrong’s technology forecasting groups. There’s a bunch of people on there who are obsessed with this idea of forecasting generally, and obviously prediction markets play an adjacent role to that.

But I think it’s more of: how do you be precise in near- to mid-term futures, and/or bet on — I would argue that the first wave of Compound was a realization that the timeline between research and commercialization was collapsing. If you could see breadcrumbs in research, you should actually expect it to reach its way into production-level engineering capabilities, or company building, at a far faster time than historically people had thought about in venture. That meant you should be looking at research and betting on an arbitrage that other people didn’t see.

I think now that’s actually over-rotated. There are too many people funding research projects, and we’re privatizing — I wrote this thing called the venturification of research. I think that is now a different problem. Instead it’s: okay, you then need to have a better reasoning of second- and third-order effects. The edge or alpha that Compound had as a firm that tracked research maybe exists; understood research no longer exists. You dump a paper into ChatGPT and ask it to explain it like you’re 16. So you instead have to have much better reasoning, and that reasoning has to be more precise, with more degrees of complexity. And with more people investing in deep-tech-ish areas, it has to be, I think, probably more correct and more risky.

So it’s just trying to find people that are willing to, with precision, write about unpopular and/or strange things. The weird part about science fiction — and you actually can see this if you try and query some of the AI models to have it write fiction — is that it’s far more formulaic than many people appreciate. So you have to play a translational role between what people believe science fiction is often structured as, and what the strangeness of reality that we’re in today could create.

It’s not a very precise statement, but it gets at something, which is that it’s weirder on very different dimensions than your traditional science fiction, where it’s crazy technologies. The technology part — we have this saying, which is we’re only creativity-constrained on a go-forward basis. We actually don’t think technology is a limiting factor anymore. And science fiction has often been built around the idea that technology was the limiting factor to a lot of cascading changes. So that’s kind of how we look at the different areas of speculative fiction. But really for us it’s just higher precision, higher risk, and continually understanding why we believe something that other people don’t.

KH: There’s another element of your research function. In your conversation with Bryce at Indie, you had a great line that I wrote down and have thought about frequently, where you say: part of our job, and why we write so much and talk so much and do a lot of research, is we want the world to believe in what we believe — but 18 to 24 months from when we believe it. And you could frame our research organization as a propaganda shop.

I’ve been talking about this a lot with our editor at Contrary Research — this idea of memetics, and why certain ideas catch on in certain ways at certain times with certain audiences. It’s interesting because they are structurally different muscles. When you think about pure research versus the propagandizing of those ideas, they are definitely related, because the quality of the idea impacts how capable you are of making it viral. But they’re not the same muscle.

So when you think about a lot of what we’ve talked about — how do you do good research — how effective do you think you are as a propaganda shop at getting those ideas out there and convincing people to believe what you believe?

MD: I think there’s two parts of it. One is the quality of your actual propaganda from a first-party basis: how good are you at writing, communications, telling the story, helping people understand the sequencing to a larger vision? I’ve talked about this idea of pseudo-secrets back in the day, which is: how do you have an idea around a company that makes people feel smart about the company, like they have a secret that other people don’t get? Because I think that’s what drives a lot of buying in dark pools of capital, which is what venture capital is.

The second is, how do you have credibility behind those statements? The latter is just a function of — as we’ve had more time to operate, as our companies have done better, people think we’re smarter. Whether or not we’re smarter or lucky, who knows? But that helps. The first is about how you have precision around the initial belief, and then continued precision around why these beliefs have changed over time, and how you communicate those in ways that enable a small set of people to latch on to them.

For us, we look at it as: we want to believe something so that we can make an investment and partner with the founder, who obviously believed something on their own. We then want to make sure that we can push forward an idea such that other talented people believe in the founders and the company they’re building — they can acquire talent. We then want a very small number of people to believe, such that they can further finance a company to then run some sort of go-to-market motion. And if you can do that, you can have some sort of credibility around team, talent and capital — and often that is what gives customers some sort of confidence. Weirdly, you intuitively would believe that’s not the case, but strangely it is. Once you can get the customers bought in, that creates a different level of credibility that feeds back into the propaganda, such that people don’t need to believe just Compound, they don’t need to believe just the founders — they’ll believe the customers, who theoretically are willing to part with their hard-earned dollars.

So for us it’s about sequencing the propaganda and realizing the layers of abstraction. Whenever we invest in a founder, one of the things we often tell the teams is: in all of your updates you should have three things. How do you want to position the company to hires? How do you want to position the company to investors? How do you want to position the company to the press or media or the general public? You should continually, in every update, have that written out, because it will allow you to understand how your narrative changes, and it will allow other people to have a top-down view of the propaganda that they should be pushing forward. There’s nothing worse than someone misunderstanding your company and then communicating it — and if they have a bigger megaphone than you do, which in investors’ cases sometimes they do, that can be very bad for the founder.

So I think it’s those things. But the worst part of our job is we need people to finance our companies after us. That is the only reason I dream about having a bigger fund, and the only reason I could ever imagine doing it — for that reason of not having to deal with the other investors in venture.

KH: It reminds me of a line from Palmer Luckey — you don’t have to convince everyone, and in fact if you are convincing everyone you’re doing a terrible job. You have to convince a very small group of ride-or-die stakeholders. That’s true of employees, investors, customers. But outside of that you actually don’t care how you’re communicating it to everybody. So the more capital and influence you have, or the ability to influence that company’s success, you actually limit the aperture of people who you have to bring along for the journey. Which I think is super interesting.

MD: I also think the last thing on this is it just shows how unbelievably terrible most public companies are at investor relations — and then there are some that are too good, and those companies are usually the ones that you short. The ones that are maybe sometimes better longs are the ones that just need to figure out how to understand the market structure shift when you go from a private company to a public company.

KH: 100%. On this note of storytelling becoming an increasingly prevalent part of these deeply technical areas — I’ve heard you talk before about the barriers to entry in these areas. Historically you’d have maybe 100 people in the world who were really capable of building something in a particular space. And then suddenly — this is true with every hype cycle — while also enabling some of these very capable people, it also attracts a lot of the hustlers and opportunists. This happened in crypto in 2021 for sure. There’s a ton of people who had never been interested in the space and suddenly were able to raise hundreds of millions of dollars.

When you think about these categories you spend a lot of time in — it can feel like Twitter made all these people who are like, oh, I’m an expert on this. And now those people are not just tweeting about it, but they are in real life raising hundreds of millions of dollars, trying to put out products. How do you respond to that? And is that good or bad? On the one hand people will try and say, no attention is bad attention — if we’re bringing attention to a space, that is great, it raises the public view of this world. But definitely it feels like there are dangerous elements to that type of inflow.

MD: I think there are reasons why it’s bad. The first-order reason is that many really hard problems are not solved on the first go. When the companies that rise the fastest are often the ones that are best at accumulating capital and creating a lot of hype, they kind of destroy the financing landscape and the customer landscape for those that are trying to build in a more sequenced or levelled or measured way.

Second, the thing that I think is a reality even though it shouldn’t be: you do have market distortions that take place that hurt founders. You have founders that look around and say, hey, these other people have been able to raise a bunch of money and I can’t. What am I doing wrong? And I would argue nothing, sometimes. But again, markets will market, and those things happen. You have to figure out what is the game that you’re playing. If you’re going to play the game of perception, you need to be a master at it. And if you’re not, then you need to play the game of execution. If you’re anywhere in the middle of those things, you’re probably going to die.

So I think it’s hard. But I’m just past the point of complaining that things get too hot and it’s too hard and prices are too high. None of my investors and none of our founders give a damn if those things are true. What matters is if you can execute. So we have to continually figure out what are the things that we do — and likely that means we need to take more risk. We need to help our founders understand when they’re going to inflect on moving towards telling their story more publicly, perhaps earlier, such that they can box out other competitors that fast-follow them, such that they maybe will erode what people used to believe was IP early on.

A lot of this we’ve talked about in the framing of AI companies, which is: people used to believe models were the moats, technology was the moat. In our view, long term, brand and pace is the moat. One can look at Runway as the best example. They’re the best product team in AI, in our opinion, and they ship at a rate that is absurd for a company of their scale. That’s what you need to do in this new world in which more people are able to get capital for less than ever before, if they play a certain game.

KH: I tend to agree. The area where I struggle is the poisoning-the-well dynamic. One area you can think about this is the top of funnel of innovation as a catch-all term — where people are doing the thinking. This is a key insight that you guys had; you talked about this a little in the venturification of research — this belief that the time between commercialization and research was collapsing, so it’s being pulled forward.

It reminded me of a conversation I had with a Cambridge professor in 2014 who was really struggling with the incentives inherent in the academic institution. He said that the vast majority of research papers effectively get read by three people — the author, the editor and the publisher — and then they just kind of die. But that’s what you’re incentivized to do: create research that effectively dies, because all that matters is whether you’re getting published.

But now it feels like people have almost weaponized the commercialization engine — they’ve reached into the research apparatus to rip forward the work that’s being done into the commercial world. And that can change a lot of the incentives. On the one hand, you make this argument in your piece where it may in some cases just change — where it used to be that you would be funded by a larger company or research facility or nonprofit grants, now more of that is coming from venture funding. There’s an argument that even the technology readiness curve you write about may not change; it’s just different money coming from different pockets.

But it feels like there is a world where — I don’t know if there is a pure science that needs to happen in pushing some of this stuff forward, that now, because commercialization is so front and center so quickly, you may actually end up doing the work of innovating more poorly because you’re so focused on the dollar. Do you think that the scientific well is being poisoned to some extent by that focus on commercialization?

MD: Probably in some ways. At most points of a deep tech company you build something that is off, or not directly aligned with, your product roadmap in order to show some sort of distillable proof point to investors — which is inherently decelerating to research. You’re doing something such that other people can be like, okay, I get the idea that this company is circling around, because the core thing would actually take longer than the 24 months that venture capital allows for. And as an early investor, the other problem is these things get flagged [?] so much that you do take along the way.

Is it bad that there’s a framework now that people can accelerate what happens in academia by getting a consistent team working on a problem every day in a capitalistic way? No. Is it bad that those people for whatever reason get thrown really large amounts of money and then usually die because of it? Yes.

But again, you make the decisions that you need to make as an investor. We have passed on a bunch of companies that are really awesome because we thought that they were not going to really prove anything with tens of millions of dollars, and that’s just not our model. But if I had $20 billion or $3 billion, whatever, I need to figure out where to deploy those things. You could get those grants from the government, or you could get them from a big venture fund that gets to charge fees on that money. And who knows — maybe with a bunch of the government changes, there won’t be any funding for any of these things anymore, and then we do need it.

So I don’t know. I think it’s probably net negative for a lot of the founders, but I’m too biased to try and make the argument that everything should stay in research until it’s ready. Because maybe that then just looks like a bunch of Mira and Ilya-style companies fundraising, where those things were effectively kept in research for seven years by the funding of OpenAI, and they spun out and they raise at $500 million to $2 billion post.

KH: Do you believe that sentiment that the volume of capital does not materially drive forward the curve of technology readiness? If you really are trying to figure out what is the cutting edge, whether you have $10 million or a billion dollars, that’s not going to materially change the progress that you’re making.

MD: I don’t think it does in many cases. I think people have been kind of saved by the idea of compute being this pretty finite resource that dollars create outcomes in. The bitter lesson was actually the sweetest lesson for multi-stage funds. In a lot of areas of maybe biology and science, there’s obviously some minimum viable of — can you do more things in parallel with high resolution and high precision, such that you can figure out the premise that you’re trying to prove faster? Sure. Are there still limits on that? I think so. But again, everyone has a counterpoint to every point I can make on that, so I kind of don’t really care about debating it as much anymore.

KH: It’s a good segue into the last bucket of things I want to talk to you about. This is something you and I have vibed on before, thinking about the potential exacerbation of capital destruction. You touched on this earlier — the danger is that if you pour a ton of capital into the hucksters, the opportunists that are chasing a category, and they sully the good name of the category by not producing results, and they destroy capital and they don’t produce product, that makes it harder.

I’ve also heard you describe the time that we’re in in venture as the lowest-conviction time that you’ve seen in venture. This idea of these wandering, passionless souls — people are just like, I don’t know what I care about, and then there’s a loudness over here, a sound, and they’re like, I guess I could care about that. And everybody lurches towards that thing and looks for things that look and feel similar.

I always get pushback, because anytime I express skepticism — which is definitely dangerous for any investor to do — it’s usually not necessarily about a technology or specific progress. It is about a specific company, a specific fundraising round or dynamic, or what that sets up. And I get a lot of pushback where I’ll say, hey, I just don’t think that this approach makes sense. And they’ll say, oh, well, I’m just really excited about the technology. And I’m like, don’t get me wrong, I’m excited about the progress in robotics or AI or whatever. What I’m nervous about is that if this thing doesn’t work, the level at which we have heightened the risk profile because of the amount of capital — and it’s still very risky, there is a lot of risk inherent in this thing.

If that thing fails, the volume of capital destruction — we might think, hey, these big multi-billion dollar firms, if they write a $50 million check and that goes to zero, that’s not a big deal. But what if they write tens of billions of dollars and that goes to — eventually there is some scale that it does hurt. My anxiety is always that you look at nuclear energy 70, 80 years ago, and we had people envisioning a future of every car being nuclear-powered. That was the progress we thought we would see. And then because of certain things that happened, there was a loss of public appetite for it.

My theory is that the same thing could happen — we get really excited about this prospect of a hundred billion AI agents running, automating all these aspects of the world, but we don’t get there because billions and billions of dollars get destroyed. That capital destruction is a fear that I have, especially when you are at the cutting edge of things. What’s crazy is that you guys are early, but suddenly it goes from you were early to these things and you’re developing a real expertise, and then suddenly you have these lumbering giants tripping over themselves to get into it and throwing billions of dollars at it, and you’re like, well, crap. Does that describe the dynamic that you feel, or would you frame it a different way?

MD: I think it’s a very reasonable feeling. The premise for Compound was that we’d get to areas early, we would understand them deeply, we’d make non-obvious investments — and as those areas matured, we would have institutional knowledge to continue to own and exploit those categories in the same way that Bill Gurley did for marketplaces for 20 years.

The reality is that once those things become consensus, the dollars deployed and the scale at which people haphazardly throw money is much larger, such that we usually just don’t play those games. We have to think about instead what are the non-obvious approaches or next-order things that we think will be valuable that other people don’t. Which is disappointing, and is a lesson we’ve learned now a couple times.

But if people want to blow themselves up chasing low-conviction, high-consensus areas, I am all for it. Venture will continue to cycle and largely be an asset class where the returns are terrible. Eventually, if people continue to do that and get it wrong, you either hope that there’s no money after you go to the Middle East — eventually someone is going to have to stop funding these things if they don’t work. And if they do work, then great, we were wrong. And if we were wrong, as long as we maintain discipline and kept scale down, we can be wrong and still make significantly higher IRRs than other people, such that we can stay in business and help support our founders.

So it’s a disappointing lesson. I think it’s true, but I don’t think that an increased burden of proof for founders is a bad thing at this stage of capitalism. We are at a point where we are willing to take a single data point and extrapolate it so far that it wouldn’t be bad if we knocked it down a little bit. I think we started to see that in 2022 as multiples compressed, as markets started to readjust, and as people didn’t know what to feel or think about the next generation or wave of technology — and AI changed that dynamic and actually stopped a reset from really happening.

The last thing I’d say is people also just don’t appreciate the market structure shift that’s happened in all of investing, which is that people our age have largely operated in the greatest bull run in history, in which every single time anything goes wrong you just buy the dip of the thing, and you have been right. You do the most obvious consensus trade and it has paid off significantly more than anyone has ever thought — the compounder, Tiger cub type thing. That market structure shift probably means that eventually there will be a different one.

Our view is just: continue to pay attention to what are the dynamics that are creating all of these behaviors. Largely we are never going to be able to play the consensus dynamic, and that’s what people give us money for. That’s why we are small. That’s why we’re the riskiest dollars. So will there possibly be some companies that don’t get funded, or have a harder time raising, because of scar tissue from other investors? A thousand percent. Does it mean that they need to build real businesses faster, or they need to execute, or tell their story better? Yes. I guess I’m okay with that — but some founders are probably going to be annoyed that I said that.

KH: But it goes back to your point about you’re not trying to build everything for everyone. It is a very specific audience for what you’re doing.

I think that’s a great place to wrap it up. Normally I say, where’s the best place for people to go to learn more about you — but we’ve talked about it, that you guys are a Twitter-native firm, and so digging into your Twitter and your writing on your website is obviously the best place to get smart on what you guys are doing. So I appreciate you walking me through the playbook.

MD: I appreciate it. Thank you for asking questions.

Connections

• Michael Dempsey · Compound.vc — the guest and his firm; the episode is effectively a full account of how Compound is constructed.
• Conviction — the spine of the conversation. Dempsey’s phrase is ideas “we’re willing to go down with that ship” on, and his description of the present as the lowest-conviction moment he has seen in venture.
• Union Square Ventures — named as the firm that best shares context at the early-ideation layer; Compound has run joint partner meetings and a biohacker research day with them.
• Contrary Capital — Kyle’s own parallel: talent identified years before people start companies, and the same unanswerable question from other investors about how to shortcut a decade of relationship building.
• Anduril — Kyle’s example of speculative fiction predicting well, via the novel 2034.
• Runway — Dempsey’s example of brand and pace as the real moat.
• Bill Gurley — the model for owning a category on institutional knowledge for twenty years, and the thing Compound found harder than expected once their areas went consensus.
• Palmer Luckey — Kyle’s line that if you are convincing everyone, you are doing it wrong.
• The venturification of research — Dempsey’s own essay, referenced as the argument that too many people now fund research projects.