🎧🍌 The Past, Present, and Future of AI, Robotics, Venture Capital, and Crypto | Michael Dempsey, Managing Partner, Compound
Why deep tech startups mess up value capture, unprecedented state of hype in AI, venture slop, and why VC firms decay over time
Michael Dempsey is the Managing Partner of Compound, where he was the first investor in multiple AI unicorns. Neither of which were obvious when he led their Seed rounds back in 2017 (Wayve) and 2018 (Runway).
Our two hour conversation is loaded with well-reasoned takes on:
AI hype
venture slop
launch videos
humanoid robots
actual crypto use cases
why VC firms decay over time
short-term ARR growth tradeoffs
He hasn’t done much AI investing over the past few years, and we unpack why AI is max consensus obvious right now, the intersection of AI and the real world, the state of hyped startup launch videos, why so many deep tech companies mess up economic value capture, and his crypto thesis and use cases (which I found easy to follow along with as someone who’s generally skeptical of the space).
Compound is a thesis-driven firm, and we talk about their research process and how to replicate it, what private and public market investors can learn from each other, advice for anyone starting in venture today (be known for something), how to build a brand in VC, and why venture firms don’t compound and actually decay over time.
Big thanks to Kevin Kwok, Andy Weissman, Cristóbal Valenzuela, Blake Robbins, and smac at Compound for their help brainstorming topics for the convo
Presented by Ramp
If you're running a finance team, you know how much time gets wasted on expense management. Chasing receipts, categorizing transactions, waiting for expense reports. It adds up quickly.
Ramp handles all of this automatically. Its a corporate card and expense management platform that over 40,000 companies, like Shopify, CBRE and Stripe use to streamline their financial operations.
But here's what makes their corporate card different: every transaction gets automatically categorized and matched to receipts. No more wondering what that $47 charge was for three weeks later. You can set spending controls, get real-time alerts, and even block certain merchant categories. It's like having a finance team member embedded in every purchase.
The platform integrates with your accounting system and ERP, so everything flows through without manual data entry. Whether you're issuing cards to a few employees or managing spend across departments, Ramp gives you visibility and control without the paperwork.
Stop chasing receipts: check out ramp.com/ThePeel to get $250 and see what a corporate card can actually do for you.
Time is money. Save both with Ramp.
To inquire about sponsoring future episodes, click here.
Timestamps to jump in:
4:09 Leading Runway’s Seed in 2018
10:15 Short-term ARR vs long-term sustainability
16:20 Compound, a research-centric investment firm
18:41 Investing in bio, crypto, real-world AI, and healthcare
23:58 VC firms do not compound, they decay over time
29:27 How to build a research-focused investment firm
41:30 Current state of venture slop
45:43 Building a brand as a VC firm
52:31 Investing in Wayve in 2016
58:53 Why deep tech companies screw up economic value capture
1:04:57 How to approach massive funding rounds
1:08:36 Should VCs “play the game on the field”?
1:15:33 Compound is a forecasting firm
1:21:37 Advice for young people getting into VC
1:26:48 Public market investors underappreciate narratives
1:31:20 Michael’s crypto thesis + real use cases
1:40:07 Why crypto hasn’t seen mass adoption yet
1:49:00 Humanoid robots won’t work
1:54:42 Should you make a hyped launch video?
Referenced:
Find Michael on X / Twitter and LinkedIn.
👉 Stream on YouTube, Spotify, and Apple
Transcript
Find transcripts of all prior episodes here.
Turner Novak:
Michael, welcome to the show.
Michael Dempsey:
Thanks for having me.
Turner Novak:
I think this will be fun. It's been like a couple of years since we actually hung out and talked, so this will be we're just going in. We have a good list of stuff to talk about.
Michael Dempsey:
We're grown up. We're old.
Turner Novak:
Yeah, that's true. So the main question when I told people we're talking, I think three or four people said, "You got to ask him about this." You were the first investor in a company called Runway. I'm just looking up on Crunchbase there was a $2 million round, so we're going to assume 10 million posts or less valuation. They just recently, a couple of months ago, raised I think $300 million. A 3 billion post-money, something like that valuation. That's pretty impressive. So a lot of people were like, "Okay, Runway was not a super hot... AI wasn't invented yet back when you invested."
Michael Dempsey:
To VCs, it wasn't.
Turner Novak:
So I really just want to hear what's the story there?
Michael Dempsey:
So we had been doing a bunch of research in, at the time what was generative models, and that was these things called generative adversarial networks out of University of Tokyo. And originally started by this guy Ian Goodfellow who's a prominent AI researcher. And I think our view was like, "Okay, something's going to happen in creativity." Everyone thought AI was going to tackle very narrow image recognition use cases at that time, this was like 2016.
Turner Novak:
This is self-driving probably.
Michael Dempsey:
And then the other was like, "Oh, these enterprise use cases," which we see now in legal and stuff, but this was '16, '17. We had incubated a company that I was a co-founder of called Shadows, which was built around this premise, but it was a fully vertically integrated AI native animation studio. So imagine Pixar, but using AI tools. Turns out we were like eight years too early.
Turner Novak:
Which that's bad. You don't want to be eight years too early.
Michael Dempsey:
No. Well, we can talk about why I do, but yeah. And in that time we learned a lot. We ended up actually selling the company, but one of the things that we saw was there was a ton of creatives who liked using these newer AI tools, but they didn't really know how. And there was these three guys at NYU where I went to school, who were working on this tool and I just saw them posting videos on Twitter and it was basically like how do you get open source AI models into a nice UI such that you can use them as a non-technical person?
Turner Novak:
Which that seems like a pretty big deal back in 2018.
Michael Dempsey:
This is 2017.
Turner Novak:
2017.
Michael Dempsey:
Yeah. So it definitely was a big deal, but everyone was like, "This is a narrow use case and these models look like shit and this doesn't really look like a human that's generating."
Turner Novak:
And I remember when you'd post, you'd always post these research recaps of, "I looks like they're getting better." And that still doesn't look like a person.
Michael Dempsey:
So anyways, so I DMd them. I DMd Krista Ball who's the CEO of Runway today. And it was very clear, "Okay, this is going to happen. Someone's going to figure this out." I think the thing that we believed is that the tailwinds in AI were big enough that if you had a great product team, you could build products that would work around the quality of the models to enable creatives to do more and more things over time. So kind of betting on trend of model with team that can build around that trend.
We loved their seed round. We called some friends and the main learning was like, "Cool, there's a small group of people who wanted to use these tools." Companies started to scale and then basically looked up into the founder's credit. They had round a million of ARR and they said, "This is just not big enough and this isn't going to be a thing that everyone wants, like this tool to help non-technical people use this is not a venture business."
And they narrowed in on this video editing workflow with a few very specific AI tools that they had built. Green screen being one of them, how do you quickly green screen and in-paint and basically tossed out the million in ARR started over. And we raised an All Insider Series A and at that time I think it was $6 million or seven. And what they quickly then learned was as they were starting to build more of their own models, they could start to more tightly couple product with the model development to get these superhuman creative experiences.
And to fast-forward from 2019 till now, now they have the premier video generation model in the world. They do a bunch of image generation models and a bunch of other things. And the idea is how do you enhance creativity and allow more people to tell stories in a variety of ways. And Runway is an example of just being obsessed with a problem and waiting for a group of people that very much understand the problem natively and are very close to the customer, which is creatives, not removed as AI people and they're one of the fastest growing businesses I've ever personally seen.
Turner Novak:
Amazing. And so what is the current product today?
Michael Dempsey:
Current product today is effectively, there's a few different, there's a video generation model that allows you to go either image or text to video and-
Turner Novak:
Just type something in and it creates it.
Michael Dempsey:
Type something in and creates it. There's an image model called References which allows you to both create images like other people use Midjourney as an example, but also do things like using natural language to edit images quite well. And so they've basically built these world models that very deeply understand how worlds work and can be prompted with many different inputs and outputs and they're going to push a bunch of things on that in the future.
And most recently they started to preview something called, I think it's called Game Maker, which is the ability to use a similar creative oriented model to create your own games. And I think the main thing is Runway wants to own creativity in the future, and so if you were to think about all the ways in which people interact with the Adobe Creative Suite say, our view is that the primitives for the next generation of Adobe will look quite different. It's not going to be this tab on the side where you have 50 different tools, it will be something else. And in my view, which is the most biased of maybe anyone in the world, Runway is the best team in the world to build product around that.
Turner Novak:
And isn't there...? There's been an explosion of some of these AI creative tools over the past couple of years.
Michael Dempsey:
Yeah.
Turner Novak:
I know you have a pretty strong opinion on building products for today and what's going to get you really quick explosive ARR growth versus building a foundation over the long period of time. Is Runway maybe an example of that?
Michael Dempsey:
Yeah, I think Runway is a perfect example of understanding what is the long game you're playing. So something that Chris talks about a lot is, and he's said this to me before and it's stuck in my head and it really influenced a lot of how I think is, to have a short-term oriented view of what your product should be only limits the scope of ambition of that product. And so he's much more willing to know the North Star is enable creators and the in-between steps maybe are less prescriptive and more of letting emergence happen.
And I think that has helped the company a lot because model performance has had these step functions in which they're like, "Oh, wow. It's kind of crazy we can do that today." And then also has a linear, or maybe, I don't know, exponential, but smoother trend line up. And so it allows them to think about product as this malleable thing where about two and a half or three years ago, one of the things that they started to press on was there's a time, there's a very tight coupling between shipping product and shipping models often and-
Turner Novak:
Can you explain that? What is that?
Michael Dempsey:
Yeah, so to ship a product, you have a model that powers it and then you have a bunch of other UI UX interactions you have to go figure out. And training a model is something that is alchemy, you can do it. Figuring out all those UI UX and connecting them all together in the back end actually takes up a lot of time. And so if your goal is actually to run the maximum number of experiments possible to figure out the right answer, which is how Runway operates, you actually shouldn't worry about trying to build in that short term, this hyper connected product.
And so what they did is they shipped 30 features in 30 days that were all actually quite fragmented. And so at that moment in time they were having such performance on the model side that they just had a page and it was like, "Here's 30 things you want to do. You actually aren't going to be able to connect them all and it's not going to be this beautiful product where you can flow through them," or in Photoshop where you have all those tools, it's like these very purpose-built things. And it allowed them to start to see what are all the use cases that were working with different users, what was being heavily used versus not heavily used? What could they get better training data on to then train the model?
And so it was this idea that if you have an organization that's really good at being long-term oriented, you can ship a bunch of things in the short term that you think will work in the long term, but you're actually not super stressed about creating the most easy, beautiful product. I think in AI today, a lot of people sometimes try to... they get caught on both ends where they try to build both the terminal, beautiful, maximum viable product and on the other end they build these hyper narrow features but they don't ship enough features fast enough that they can out compete the noise of the space or the 30,000 other AI people shipping adjacent features to their core user base or core set of users.
And Runway was fortunate enough to be able to ship very quickly but also have this North Star that was driving them at all times. So I don't know, those are some of the learnings.
Turner Novak:
Yeah. What do you think you have to do in order to do that successfully? If I'm hearing this for the first time and I may be doing it the opposite way where I'm like one, super narrow, super scoped out. What would you recommend if I want to step back and try 30 features in 30 days? Is there a right or wrong way to do it?
Michael Dempsey:
I am just a person who sits and thinks and so I don't know if I have a great answer. I think the probable right way is understand that... I think, again, I'm a big believer of limits. Reason often creates limits in companies. And so a lot of people would say at the time Runway was I think 35 people when they did this, maybe 40, they'd be like "We're a product org that has..." I don't know, say at that time it was eight researchers and on the ENG team maybe 15 people. And so they'd say, "Okay, across these 23 people and then these 17 other non-ENG science side, how are we possibly going to move this quickly?"
And I don't know. Chris and Anastasius and Alejandro are just kind of don't... they're not baked in that Silicon Valley YCism culture, they just don't think about it that way. And so I guess my main view is often the lessons of the prior 10 years of startups, 2008 to 2016 was incredibly valuable for startup founders because a bunch of information became super legible to them through blogs and writing and YC and all this stuff.
Turner Novak:
I learned a ton from all those resources.
Michael Dempsey:
Sure, and it was great. It was like this mystical world is becoming demystified and then those things started to become ingrained in culture and that's how people thought about backing the right founders, how founders will behave. And we have new versions of that today, and I would argue a lot of those lessons are probably more useless than not now on a go-forward basis.
Both because of the amount of capital, the speed at which teams move, the tools available because of AI, all those things. And so I don't know, increasingly we say this thing a lot, which is creativity is the main constraint in startups now on a go-forward basis. And our entire view is the most creative founders will be the ones that win. And there's a bunch of stuff around what being a good creative founder is versus an artist that doesn't understand and build a company. But I think having that mindset probably is something that we orient more towards when we're talking to founders today.
Turner Novak:
So I want to ask you more about that, but maybe it'll be a good place to say, so what exactly is Compound the firm that you work at?
Michael Dempsey:
Yeah.
Turner Novak:
Did you join after it was started? Am I trying to remember this right?
Michael Dempsey:
I joined as we raised our first fund. So Compound was originally started by my partner David. I joined as we were closing our first fund and I took over running the firm in about 2019. We call ourselves a thesis-driven, research-centric investment firm. And what that means for us is we do a bunch of primary research in academia, in R&D groups, in... I don't know, government research. And we try to understand from a primary material sense, where's the world going, what is happening?
And then we build bottoms up theses across a subset of categories that are very prescriptive. And so we do the thing that most theses tell you not to do, which is like we very discreetly predict the future of what we think matters, where we think value accrues, what are the areas that make sense? And we typically do it across areas that carry large amounts of science or engineering risk or that carry, I don't know, we'll just call it societal shifts that are more long tail distributed than people or less long tail distributed than people appreciate.
Turner Novak:
So you mean fatter tails?
Michael Dempsey:
Yeah. So people would think a lot of the stuff that's happened in crypto over the past six years was really unlikely. And our view is actually the societal shifts of people losing trust in institutions at the rate that they did, people wanting to control their own capital data, whatever, all these types of things are actually far more likely to happen than the average person models.
And so our orientation is around that and the view is we do that at the very early stage at seed, pre-seed, and we do that at the liquid asset side as well, on crypto tokens in public markets. And everything in the middle we partner with founders through the Series B usually and then come off boards shortly after that.
Turner Novak:
But you're saying you invest inception through Series B?
Michael Dempsey:
Yeah.
Turner Novak:
Okay.
Michael Dempsey:
But we enter only at seed.
Turner Novak:
Enter only at seed.
Michael Dempsey:
Yeah.
Turner Novak:
Okay.
Michael Dempsey:
And then the public stuff.
Turner Novak:
Okay, makes sense. Do you do public market investing from the fund or this is just a personal interest?
Michael Dempsey:
Within the Compound management company but not in the... the funds are separate.
Turner Novak:
Got it. Okay. That makes sense. I feel like everything you just said, we're going to spend the next hour going a little bit deeper on each specific thing that you said. So I think one that's interesting, you talked about you take a bunch of research, you synthesize, you do a lot of your own, you start to think about what are interesting areas. So when you think about the world right now, what's interesting based on everything that you're seeing?
Michael Dempsey:
I think the vast majority of where we're spending time is bio, crypto and then how AI impacts physical areas. And so some of that is bio, some of its synthetic biology though and some if it's material science. And there's longer tail stuff that we know we've built a bunch of views on, but we actually haven't deployed money against as much like energies, probably another example of that. And then I have a nouveau obsession with I think healthcare over the next decade is probably investible again, where I'd say for the past four years that it's been borderline non-investible.
Turner Novak:
So when a category is investible or not investible, are there certain things you're looking for like when you say...? Like AI, right now, today, what's your feelings about it?
Michael Dempsey:
AI from 2022 to 2025 where we sit today sucked all of the air out of the room of tech. All the talent went and worked at AI companies and the main movement in that time was look at a bunch of areas and figure out what is the low hanging fruit to build new types of applications that are 50 to 500% more effective than the prior primitive of vertical software say. That's not what we do. We think that those ideas are very obvious and we are probably not going to be able to fund those because what Compound is great at is backing N-of-1 style companies.
We are not built to pick the best company across five that all look the same at the seed stage. And I think now people are starting to, one, have some sort of narrowing of the field in all these areas such that it doesn't make sense to go start the 10th legal AI business and it doesn't make sense to go work at the 10th legal AI business if you're talent and as an operator.
So that pushes people to move into what we call this next phase of technology development, which is like non-skeuomorphic/native businesses. And our bet is that over the next, call it three to five years, where is most interesting for us to be investing is in those. Things that look different in either how they approach go-to-market, how they approach product, or even how they approach UI than traditional software has and then the fast followers of this first wave of AI has.
And I think there's examples of those businesses. So you could make an argument that all of these AI software generation businesses are kind of that, like the Lovables and Cursor is kind of a version of that, it's like a little light version. There are probably next order versions that are... we've looked at stuff in AI therapy, we think that's really interesting.
Turner Novak:
But not generative AI chatbot therapy, right?
Michael Dempsey:
No.
Turner Novak:
So what's the thesis with therapy?
Michael Dempsey:
I think in general the main view is there is a clear need and a clear desire for people to interact asynchronously and 24 hours a day, seven days a week with something that has context on them as a human being and something that can help them understand themselves over long periods of time. And today people go to therapy and they pay a lot of money and they do it, let's call it once a week or every two weeks or once a month, whatever, and they do it a random time that they happen to be able to do it.
Turner Novak:
They might forget things, they might be feeling more emotionally or less emotionally charged when the conversation's happening?
Michael Dempsey:
All sorts of things. And even... So we were one of the first investors in Talkspace a long time ago, and Talkspace is a digital therapy company. It's a public company today. The main insight from Talkspace, and this is like 2014, was the founders were married and they were in couples therapy. And what the therapist realized is that after they would leave, they would both text the therapist a bunch of stuff because they didn't want to say it in front of each other.
Turner Novak:
That's the point of couples there.
Michael Dempsey:
Sure. But many things you realize it's sometimes easier to text someone because you don't have to see their face or you're not next to them.
Turner Novak:
And you can specifically say exactly what you want versus what I'm doing right now, I'm making things up versus specifically scripted out.
Michael Dempsey:
Exactly. And that is 11 years ago. A very similar primitive, which is sometimes these guardrails we put up are actually quite freeing for us. And I think that is also probably true with talking to a human versus not. And then the question is, "Okay, what's the product experience? What's the mechanism to make it not be talking to an AI chatbot as if it's a therapist?" Because that's a very skeuomorphic approach to this and that might work for very specific things.
There's a company in Europe that's doing this across... I believe male ED is a problem and they're scaling quite quickly around that. People have talked about doing this for very specific indications. Eating disorders for women was another one that people are talking about. "Why would an AI therapist be better?" But I think in general the product experience probably looks a little different. And so we are just in this phase of trying talk to as many people who have interesting views. We're pretty creative, but that's a relative part to VCs is not to actually people who build things. And so we're trying to talk to the builders on that side.
So maybe to get back to your initial question. I think AI is at its maximally obvious point today. There's massive consensus across the entire industry. And anytime I see that, what that typically means is that it's probably time for Compound to observe and not deploy too much money. And so you can make an argument, Compound has built a firm in some ways being early to AI in 2016, '17, '18. And from 2020 onward we've made three AI investments.
Turner Novak:
But that goes pretty counter to what it seems like the consensus strategy is you should be investing in AI because the companies are growing and you're getting markups, which always looks good. Why do you choose not to do that because it's so different from what most people are doing?
Michael Dempsey:
I don't think markups matter, rather they matter for sure because they help you raise funds, blah, blah, blah. I think we are fortunate enough that markups are not going to be what dictates success or failure for Compound at this point.
Turner Novak:
Do you have some...? You've returned capital, you've made money for investors. They're not like, "Oh, cool. We got a five X markup on this AIC, we don't care at this point."
Michael Dempsey:
Yeah. And I think that's a privileged position to be in as a newer/ish fund I guess. And then I also just think again, it's just not what we do. We just don't... I think so much of investing is about figuring out what is it that you're really great at and pushing that advantage over and over again because something I deeply believe is all venture firms decay over time. They don't compound over time as much as people think they do.
Turner Novak:
Interesting. That that's what most LPs would believe.
Michael Dempsey:
Most LPs would believe that because they look at persistence of returns, they look at the fact that Sequoia and Greylock and Index have added great returns for 20 years. And that's great. And also if you were to look at the universe of that versus the universe of investors from 2020 onward, it's just a totally different game.
And I also think we're going through the first generational turn ever in venture history, which is happening now. And that also means that these firms are probably decaying even faster than people appreciate. And so anyways, I just think we need to stay focused on what is the things that we're great at and the things that we believe we're great at is being a very generative firm that focuses on where non-obvious things could work. And we lose money 40% of the time is our model and if we're right on two out of the 25 companies that we invest in, we'll have very good returns. If we're right on one of them, we'll have pretty good returns. And if we're right on none, then we probably don't deserve to exist.
Turner Novak:
So you said something interesting there and made me think about maybe there's survivorship bias. So if you go back to the late '90s, early '90s, the '80s, I don't actually know when you go back to, but we think about Kleiner, Sequoia, these blue chip funds that they survived, they're around, they had good returns, but there was a lot of other funds at the time, maybe it's like a magnitude fewer than there were today, but there were still a lot. It's just like those are the ones that survived.
And then there's probably some that none of us know of that also survived that first .com bubble that survived through the Otts, the zeros and then died in 2010 and they've been forgotten again. And you think of maybe a blue chip fund that's persisted through a time, but it's really just a survivorship thing that they, not saying any specific funds, but I feel like that could actually be a piece of it, is that there is survivorship bias and thinking about this persistence of returns.
Michael Dempsey:
Yeah, I think if you were to think of it as percentage of ecosystem that dies maybe would be the way to look at it, which is like-
Turner Novak:
I could see that.
Michael Dempsey:
... if there's... 10 survive over 100 firms back then, 10% of firms survive. And if you do that today, it's like, yeah, there's 100s of firms that survive and there's 1000s of new firms that would suggest the same thing. But I think the difference is probably just the variance of returns is still quite high, even fund to fund against some of these legendary firms that is not the case.
But I just think it's meaningfully more competitive and the structure of the industry is so much more well understood such that you just have to be consistently trying to re-figure out what is your advantage? How has the world changed? I think for us, one of the things that we talked about at our annual meeting this past year was Compound was predicated on this idea, which is like we understand very complex things natively and that's a great edge, but the truth to that is that now anyone can put an academic paper into ChatGPT and be like, "Explain this like I'm 16." And so actually the edge of understanding complex things natively is no longer really that defensible for us.
It's not an edge, it's just another tool in our toolkit of maybe we can understand when something's a little novel. But actually what that means is you have to be very good at understanding second and third order effects of these technologies and how they cascade over time into value accrual into startups. And so I think there's a world in which I could easily just be like, "The thing we do at Compound is we understand hard things." And that's just totally bull shit today in a world in which anyone can put anything to ChatGPT and have it explained relatively easily.
Turner Novak:
That sounds like a great marketing tagline though. We understand hard things. So then how do you think about building a research driven institution? Because you guys, you do a lot of research. If you look at the team composition at Compound, there's no Stanford MBAs, it's not like an uber PM that's coming in. So how do you set this up and get this going? And then to your point about the ChatGPT stealing your secret sauce, why are you still doing it if anyone can do it?
Michael Dempsey:
So the argument, we're not still doing it, we're trying to figure out the next order of thing. So I think now you can look at the team, which is we have people who are obsessed with categories, have deep understanding of technical areas, and are incredibly generative and creative. And so what that means is we have Shelby who focuses largely on bio. She's a PhD from the University of Cambridge in bio, understands it very natively. And she can understand some of the dynamics of being a scientist, some of the complexities of building companies there. We have Smack who's a pseudonymous investor.
Turner Novak:
I love that it's just smack. I was trying to figure out what Smack's name was, just Smack.
Michael Dempsey:
And we met on Twitter, and he has prior background in traditional finance and structured finance, and now has been investing in crypto since 2014. And so is very native to the space. And then we have Mackenzie, who's a researcher on the team. And Mackenzie is just an obsessive person who sits at the intersection of understanding companies and understanding technologies, and loves trying to figure out how the entire world works. And I think all of us have this obsessiveness around where's the world going? How does science and technology impact that? And most of what we do is try and push on that idea of being generative, such that when we talk to founders, we can have constructive novel feeling conversations, we're not just asking them the same 10 questions everyone asked them, and two, such that when there are times in which we are maybe feeling like the error has been sucked out of the room for 24 months, we can try to figure out, but what would be the things that would get us excited, and why, and actually be more direct about those.
We have a thesis database that does that. We do thesis development sessions once a month where we sit for three hours. Everyone brings together two to three ideas based off of a bunch of research and says, "Here are types of companies that should exist, and how should we think about going deeper with those?" We host research days in very specific areas like biohacking and cryptography.
Turner Novak:
This is public or private internally?
Michael Dempsey:
The research days are open events of 50 people, top researchers in a space. The others-
Turner Novak:
Do you do open invites, or do you just close doors, organize these things?
Michael Dempsey:
Yeah, we put it on Twitter, and then we just pick and choose. And then there's usually six to eight presentations, and then we'll present something at the beginning and connect everyone after. And I think our whole thing is there's a bunch of weird ideas in the world that we think will matter long term, and we should have a grasp on all of the possible ideas that could matter. And I think that's how we orient our research process, which is reading a lot, wasting a lot of time reading things that probably will never be investable, and also meeting a lot of people who will never start a company. And the need to believe on compound, I always say is we allocate our time meaningfully differently relative to other firms. And the hope is that that time allocation plus how we think about the world and invest actually creates some form of alpha. There's other people who's like their time allocation creates alpha because they're incredibly good at networking, that scale the network, whatever. That's just like not what we do.
Turner Novak:
Yeah, Chris Runway, he said you have dozens of random esoteric substacks and blogs that you're subscribed to.
Michael Dempsey:
Hundreds, yeah.
Turner Novak:
Hundreds? Okay.
Michael Dempsey:
Not thousands though. Probably across the team, thousands, yeah. We have a pretty big apparatus to track research and to track people doing research and stuff like that. That's the main thing. And it's funny, now that we have all these AI tools, we've tried to basically bring in more signal versus noise. And I think the main learning thus far, and maybe we'll get better, but the main learning is you still need to just sit at the river and watch all the things go by. You can't try and expect someone to show you the signal, because I think maybe the special sauce is figuring out what is signal versus noise, and you have to be able to ingest the whole fire hose to figure that out. And so as a firm, it's very much like we just read all of the things, and hopefully we figure it out.
Turner Novak:
Yeah, I think Andy at USV said you seem to be good at looking at the stream and figuring out what is signal and what is noise and deciding those things, specifically some of the companies you invest in, really, really like. Runway, you could be like, "AI, whatever, AI is not going to be a thing. It's not worth building a product and a company around this." What do you think is the secret to actually figuring out what is worth even paying attention to in the first place?
Michael Dempsey:
I think it's spending a lot of time seeing all the things. It's like when people tell you, who's the VC? Trey Stevens, I think is always like, "The first six months you got to take 600 pitch meetings or whatever."
Turner Novak:
Yep, I've heard that.
Michael Dempsey:
"And you get the flavor of what's good and bad a little bit." I think reading lots of research and then talking to researchers about it and your founders and other people helps. I think the other thing too is people grossly underestimate how long these things sit out in the world. There's this woman who wrote a piece, I think it's called Discovering the Sleeping Beauties of Science, and it's very similar to something that we talk about, which is major scientific breakthroughs typically sit out in the world for a long time if you can parse it.
Turner Novak:
Like Ozempic, right?
Michael Dempsey:
Ozempic was the 1989 Stanford guy. Gwerne wrote about scaling laws in 2016, I think. And it was this random blog post and he was like, "The thing that people don't notice about GPT-1 into GPT-2 is actually, if you put more data and compute in, the performance meaningfully increases. And if you read these papers, it doesn't really look like this is going to stop anytime." And that was before anyone had the bitter lesson in scaling or whatever.
Turner Novak:
Open AI was doing ICOs around that time.
Michael Dempsey:
Yeah.
Turner Novak:
Oh, maybe was 18 that they did that.
Michael Dempsey:
Yeah. There's all this stuff. GPT-3 sat there for a long time, and the instruction tuned in and you got ChatGPT. And I think you could see that with AlexNet in 2014. You could kind of see that with some of the stuff on how Solana was starting to scale in 2023 from a transaction. There's data that sits, and if you're paying attention, the core thing that you start to notice is this is interesting and durable enough, and there's enough people talking about it. And academia does a really good job of taking an idea and then building on it. And so a simple thing I tell people who get interested in trying to do more thesis orientated research is, "If you find a paper that you think is really interesting, what you should just do is set citation alerts for that paper. And what you will do is you will get an email anytime anyone sites that paper. And you'll get to see in some amount of time in your inbox every day, or every eight days or seven, more and more people start citing it.
That's a pretty interesting signal where you're like, 'Wow, I used to get an email from Google Scholar every three weeks, and now I get one every four days. I wonder why.' And what that will tell you is a bunch of people who are ostensibly smarter than you are starting to care about a thing more, and are spending material amounts of time writing and talking about that thing." And so I think we have a bunch of signals like that in the same way that a lot of great VCs believe they have signals on how to tell if a founders amazing.
Turner Novak:
Yeah, interesting. Okay. Is there anything that you've built internally? There's a lot of things you can do with AI to build tools internally. Most VCs, when they're talking about how great they are, they've got all these data sources or they've got all these tools they built internally. Have you done anything like that on the research side to organize or surface it better?
Michael Dempsey:
We've done for searching across topics for how do we run topic modeling on academic papers. We've done some stuff for monitoring like esoteric writing platforms for mentions of companies more on the public side. But honestly, it's all incremental. None of it's like a silver bullet. You still have to spend a disgusting amount of time reading a lot of words every week.
And I just don't think any of the models are good enough to be like, "Hey, I read a thousand posts this week on this thing, and here's the one thing you need to know," because you're going to miss the nuance, and you're going to miss the, "Oh, that's an interesting idea." I don't know, there's this concept of model merging that emerged in AI maybe, this point, two and a half, three years ago. And the idea is you can effectively slice and dice models, put them together, and you can start to see some emergent interesting behaviors.
And if you were funneling a bunch of AI papers, you probably wouldn't have gotten the insight in 2023, 4 when that happened of, hey, more and more people are open sourcing these Franken merges of models that they're talking about on hugging face. You have to just be watching everything and watching what people are talking about on Twitter. And it's maybe the 10th loudest or the 20th loudest thing, but it might spike on novelty. And so you're like, "Oh, this is weird and experimental, and 10 high signal people are talking about it," where the other side would've been like, I don't know, test time compute is the most interesting thing now, which is running compute on inference instead of pre-training it for models. And it's like everyone was talking about that. A bunch of very smart people were talking about it. So if you're an LLM, you're probably like, "Hey, this is the most important thing to pay attention to."
And it's important if you care about what's happening in AI today. But if you're trying to invest in where does the world go in the future, the maximally understood important thing is actually probably by the time it's there, you're going to not make money as a seed investor.
Turner Novak:
Yeah. And it's interesting too, the way that a lot of the way that we consume things now is algorithmic powered and driven feeds. And most of those are weighted and sourced by time spent ultimately, because the platforms want to keep you on. So if there's topics that people are spending a lot of time on, that's literally what it is. It's like do you click on a tweet and do you spend time reading it? And then if you do that, other people will see it. And it's like a self-reinforcing thing, where if I'm a content creator or a founder or an investor and a specific topic is getting more airtime and I need to be surfaced, people think I'm relevant, I need to start talking about those things also, and I need people to spend time on my stuff.
So you get this weird dilemma where there's more content to fill that void. And then the algorithms think, oh, more people are spending time on this topic, so they'll push them even more. And then again, so it's this weird structure of how the internet works now where I feel like these algorithmic feeds have almost amplified some of these hype cycles in a way.
Michael Dempsey:
Sure, yeah. This is also important why everyone should have alts that are based off of different mindsets. I have different alts for the type of stuff that I'm trying to go into.
Turner Novak:
Do you have a bio-alt or something like that?
Michael Dempsey:
Yeah, crypto-alt, bio-alt, maybe a more speculative future alt, and then my feed is half venture garbage, some public stuff, and some, I don't know, memes.
Turner Novak:
Hopefully I'm in the memes at the venture garbage.
Michael Dempsey:
It's the Venn diagram.
Turner Novak:
Yeah. And you actually tweeted something, I think it was either today or yesterday...
Michael Dempsey:
It was a couple of hours ago, yeah.
Turner Novak:
... Yeah, a couple of hours ago, you said 70% of venture content is probably created by LLMs.
Michael Dempsey:
Or ghost writers.
Turner Novak:
Or ghost writers, yeah. Why did you tweet that? What's the rationale? Why is that a big deal for anyone listening who doesn't even know what that means?
Michael Dempsey:
Yeah. I don't know if it's a big deal. My Twitter is relatively unfiltered, and so usually tweets that are when I just get annoyed. I opened up an article or an essay and I thought it was going to be good, and then I see fucking M-, and it's not X, but Y framing and all these things, and that's annoying. And so I tweet it out. I also just think it's so cringe at this point. We all know what those writing styles are. And the fact that these people are not even trying to edit it a little bit to not have it be like that shows just a fundamental misunderstanding of how people think about content. And I think we do a lot of writing, but our view has always been like, you need to treat content like a first-class citizen in the organization if you believe you're a research oriented firm.
And obviously, people do it for content marketing. But I just think it's another thing about venture that's annoying, which is just a lack of thoughtfulness. Venture slop is a thing. And that can be investing, it can be how people are in board meetings, it can be how they are in pitch meetings, and it can be how they write. And again, I think if people were writing about novel ideas but using an LLM to help them figure it out, that's fine, but it's not that. And the ghost-writing stuff is there are who have Twitter threads about totally random stuff, and it's long and five times a day and all this stuff.
And I'm just like, either you have the time to do that, in which case you shouldn't do venture, or you're just paying other people and I don't get why. I don't know. I was on a pitch call with a fairly young founder a couple of weeks ago, maybe a couple of months ago at this point, and he was like, "I was reading your stuff and you're like a boomer VC." And I was like, "Cool. Yeah, I guess so." I think I am.
Turner Novak:
So what does that mean? Why did he say you're a boomer VC?
Michael Dempsey:
I think because I was writing about how companies need to think about value capture properly and competitive dynamics, and it's not just about doing fast ARR and expecting it to stick around. I don't know. And so I think
Turner Novak:
It's kind of annoying to be that.
Michael Dempsey:
And so I think I wasn't vibes based.
Turner Novak:
Yeah, you weren't vibes based. Yeah. I feel like part of it is because become a check the box exercise. So you just hire a marketing person, and they do everyone's stuff, or it's somebody really just likes working with founders, and this marketing thing, they have to make a LinkedIn post once a week or once a day or whatever, and they're like, "ChatGPT, cool, copy, paste, let's send it." So I feel like that's maybe part of it is because people don't treat content like a first-class citizen, like you said.
Michael Dempsey:
Yeah. I think I'm just a bit of a snob and also have too much of an ego around it to do it. I don't know. That's the reality.
Turner Novak:
I'm still so bad at using ChatGPT to write. I literally do not use ChatGPT to write at all.
Michael Dempsey:
Stay away for as long as you can.
Turner Novak:
Yeah, I have started using it for ideas...
Michael Dempsey:
Yeah, that's great.
Turner Novak:
... To get ideas on things. And I usually think about it as just it's a volume thing. A friend, he was coming up with new names for a product. And I just was banging ChatGPT for 10 minutes, just keep giving me more ideas. And he started to like all the ones I was coming up with because I was literally just ripping ChatGPT, like, "Make it more futuristic, make it shorter, make it rhyme or hit the silver." And again, it gave me hundreds of ideas, and some of them ended up being good. And I would not have come up with that.
Michael Dempsey:
I'm here for that.
Turner Novak:
Yeah.
Michael Dempsey:
Just don't put it into a 900 word blog post.
Turner Novak:
Exactly. And so then how do you think then about building a brand as a venture fund? How have you thought about it as Compound? And then if I was coming to you today, it's like, "Hey, I'm starting a fund, how should I do my brand marketing, if I'm doing content," what advice would you give me?
Michael Dempsey:
All people give advice for in the way of talking about the things that work for them. So I'm very aware of that. And so I'm always hesitant to give advice because I don't know. I think for us, the key thing that we're very precious about at Compound is is there a defined aesthetic across the entire firm, in the sense of you can understand what is a compound person, what is a compound company to a lesser extent, what is Compound, what is the style of it? And I think over the past few years, we've started to have that develop more. And I think some of that is being very intentional, we have the types of things we want to put into the world, the style of people we hire. And I also think there is some chaotic authenticity that people appreciate that we represent. So I think for us, we just continue to orient around, it's very hard to build anything in venture today and in investing today that stands out.
Every person has a venture fund, every person is an investor. And so we at least want to be known for something, otherwise we think you're known for nothing, and that's the worst place to be. And so we orient towards a lot of this more scientific, more thoughtful and prescriptive thinking in the world, hoping people will come and want to tell us we're wrong. But at least then the conversation we're having with founders is different. It's not the same 10 questions. It's like you might think I'm an idiot because I wrote some posts that directly contradicts how you think about building a company. And that's actually a little bit how Alex at Wave and I started talking early on. And I think at a minimum, you can continually hit on that.
I think the thing that we're maybe not as good about is trying to figure out how to continue to expand the universe of people that know about us. I think as a new person, I think there's twofold. One is, what is the credible reason why someone actually wants you on their cap table, is the question that we ask ourselves all the time. Our view is VCs don't impact the upside of companies. We think that they impact the downside. And so I think a lot of the lessons we've learned, which again, I know you can make all these expected value arguments, "That means the upside, blah, blah, blah."
But the thing that we believe is there's a lot of lessons to be learned around building N of one highly technical scientific businesses in areas that most people don't believe exist, and the way in which you can run commercialization long-term early on in some of these companies. And I think that's part of our reason. The other is, we're very high context. And so early on, people don't have to explain everything happening in the space with us, we can be thought partners a little more.
Turner Novak:
Like, "Oh, we also read that paper."
Michael Dempsey:
Yeah, we've read this, we understand all these companies. We get where technology is going. And thus, if you're building an AI product today built on models, we have a view of what capabilities will come out of these, whatever, a bunch of stuff.
Turner Novak:
Other funds do that, right?
Michael Dempsey:
Sure.
Turner Novak:
They have an opinion on things?
Michael Dempsey:
Yeah. I think our argument would be, our opinion is oriented very much around early stage, and probably just being net more thoughtful.
Turner Novak:
So it could be with AI, if you wanted to actually invest in AI today, inception stage investing versus should you just invest in open AI, or should you just invest in NVIDIA?
Michael Dempsey:
I was talking to a group of investors who are debating raising a fund of funds to do AI investing, and they want to do half fund of funds, half direct. And there's three people, and they're split on which to do. And one of them is like, "I want to do this passively. I feel like I can just play beta on this space and make money." And the other is like, "I want be more in the weeds." And my view to them was like, "Listen, it's going to take a long time to build up the alpha generating side of this. Go take out some open AI and Anthropic secondary, buy some meta and Google. And I don't know, if you want buy some NVIDIA and you'll probably be okay. And you'll perform on some way. And here's the waiting I would think about, just sit."
And we had a slide on our AGM this past year, which is we had all these ideas, and then the last slide was like, "But even with all these ideas, you should have just longed NVIDIA in 2020. You would've outperformed our fund, every fund 23x," or whatever it was.
So I don't know. Yeah, I think there are some moments in time in which you recognize beta is the play. And typically, I think when tech beta is the thing, that's when compound retreats. And when it's not, that's when we try to push our advantages. And I think any new fund has to be oriented around that idea of...
Turner Novak:
Alpha generation?
Michael Dempsey:
This is risk capital, yeah. I think if you want to go and try to compete against Founders Fund and Thrive and do growth stage, call it levered, beta alpha view investing, good luck, they're incredible at what they do. And also, as we both know, raising that sum of capital is non-trivial. But if you're a new person, you have to have an orientation of, "This is something I believe in, and I'm willing to lose all the money doing it, and it's not safe at all." And what we tell our investors is we should be the riskiest dollars in your allocation to investing, because, I think especially as so much capital is flooded in, in order to generate meaningful outperformance, you actually have to do riskier and riskier things over time.
Turner Novak:
Yeah, that's fair. I feel like a lot of these solutions to that has been, "Oh, we'll just invest earlier. We'll just do inception stage now instead of –“
Michael Dempsey:
We're past that. There's nowhere earlier to go.
Turner Novak:
Give college students high school dropouts.
Michael Dempsey:
Yeah. It's great. ProLot is great and ZFellows is good. They're very good at these things. But if you want to build an investment firm that can be durable. It's why I like the Gen Z firm orientation thing, is it's like a short-term advantage, but it's not a long-term advantage because these things ebb and flow. And to the point, one day you show up into a zoom and someone calls you a boomer, and you're no longer the young person in the room.
Turner Novak:
Yeah. As millennials, we used to be what Gen Z are. And now, we're like old. The kids on TikTok are wearing wide-legged pants. That's the thing, they come out though. I'm like, "That looks ugly objectively. Give me my just skinny or maybe straight jeans."
Michael Dempsey:
Yeah, you're a boomer. I don't know what tell you man.
Turner Novak:
I can't do this wide-legged thing. But yeah, we're just old. And you mentioned a thing I wanted to ask you about, you said Wave, that's name of it. You said you actually disagreed about something.
Michael Dempsey:
Yeah.
Turner Novak:
So what happened there, because you invested in their first round?
Michael Dempsey:
Yeah.
Turner Novak:
I think I saw they just raised, it was 800 million euros or something, was what I saw in Crunchbase. So I hope that that's been a good investment for you.
Michael Dempsey:
Yeah.
Turner Novak:
Yeah, so what's the story there?
Michael Dempsey:
Yeah, so I was fortunate enough to spend a bunch of time when I was living in SF for a brief amount of time with some of the early Waymo team, those five, 10 systems became Waymo. And I think as I spent time with them, the thing that I kept hearing was there was a very famous demo in which they drove over the Golden Gate Bridge. And that was mind-blowing at the time. And they were like, "Oh, if we can do this, we got this, it's game over." And I think they did that in 2014, 15. Then what they realized, actually they've done it earlier, but what they realized is they kept hitting new ceilings of like, "Oh, we actually can't do this."
Turner Novak:
Because there was really random edge cases, they just had to keep solving edge cases.
Michael Dempsey:
Yeah. They were just like, "We just have to keep hand engineering these rules and solving edge cases and computer visions getting better, whatever." And they would just throw out the code base every few year, start over, go. And I think the main takeaway that we had was self-driving was obviously a very valuable problem, it's very interesting, but everyone's building it in a non-human way, which is like you get on the road today, you go anywhere in the world, you basically can drive. You have two cameras in your eyes.
Turner Novak:
In your eyes, yeah.
Michael Dempsey:
And you have a lot of compute in your brain, and you figure it out. And you know the rules of the road, whether you're in China to New York, to London, wherever, you know, you know the rules.
Turner Novak:
And they can go, "Oh, by the way, in London you drive on the left side side instead of the right." Cool, I know that now.
Michael Dempsey:
And you don't even have to be told that, you can just look at the road and be like, "Oh, people are driving on the left side of the road. This is the way the world works." And so we had this framing of, okay, if someone is going to solve self-driving as a new company, they're probably going to do it in a much more AI native way. This is 2016. And at that time, crews had been acquired, Waymo was starting to scale. Tesla was still early, but they had autopilot. And we wrote a bunch of stuff around this, which is a lot of founders are going to spin out of these companies and raise short-term rounds with similar technical approaches, and then get acquired. It was very similar to AI today. These companies were getting acquired for 10 million in engineer back then, which was crazy and hilarious now. And I think we also said, okay, there also might be companies that sit as middleware or as different data layers on top of autonomy. So pedestrian understanding was one.
Turner Novak:
A single company focus on pedestrian?
Michael Dempsey:
Yeah.
Turner Novak:
Okay.
Michael Dempsey:
And I'd read this paper from this guy, Alex Kendall, when he was at the University of Cambridge. And it was his PhD thesis, it was called SegNet. And basically, it was a new type of computer vision model. And I emailed him and was like, "Hey, this is pretty cool. If you ever think about starting a company, let me know," because I was 24 at the time, that's how you thought about sourcing stuff is you read something and you email someone. That's how you still do it. And he responded. He's like, "I'm getting my PhD. Thanks, but I'm good."
Two years later, 2016, we hear from one of our venture partners who was at Uber on the self-driving side, like, "Hey, I met this guy, Alex Kendall, he's got this really crazy approach for self-driving. You should talk to him." And I was like, "The name sounds super familiar." Bumped the thread, and we ended up leading a seed round. And our thinking at the time was the wave approach was a single model, end to end, and you can teach a car to drive itself because you don't need a bunch of handwritten rules because humans figure it out themselves. Lots of data, you have emergent capabilities. And in retrospect, we all understand that now with LLMs, you've just toss a bunch of data and they actually start to reason quite well. But in 2016, that wasn't really the case.
And our view was like, look, maybe they're going to be wrong. Maybe this is a total miss. But if they're right, they are one of the only companies attempting to do this, and this makes way more sense as to why this could win as a final mover relative to Cruz, Waymo, Tesla. And Alex's main argument to me was, "All this stuff that you're talking about of pedestrian understanding and these little features, that's never going to actually be a thing. It all is going to live in the model. It's a single model. These things will be emergent." And so the first thing he said on the phone was like, "I read all your stuff, and I don't really agree with it, but we should talk." And I was like, "Cool, great. That's exactly the point."
Anyway, so we led their seed round, been on the board of the company since then. We've gone on, raised a bunch of money. The company's doing well. I took a ride in a car in New York, which was the first New York ride we've had a couple of weeks ago. And I don't know, I don't have kids, but it was one of the first times I've ever been so proud about an investment because I was like, "I'm crossing the Brooklyn Bridge in a self-driving car that for seven years, I've been wanting to do this, or nearly eight years."
Turner Novak:
Yeah, that's cool.
Michael Dempsey:
So I don't know. I think again, there's a world in which the company just doesn't work, but our framing is you have to take big risks on possibly great things. And if you do enough of those with really good reason as to why they can disrupt incumbents and capture all the value, which we think Wave has and will, I don't know, you got to be right one out of every 25 times.
Turner Novak:
I have a question on that, but one thing I want to say, in terms of having kids, the first time my daughter texted me back was a crazy experience, just as a millennial growing up, like, "Oh my god, she's on her iPad. She texted me."
Michael Dempsey:
How old was she?
Turner Novak:
I think she was seven at the time.
Michael Dempsey:
Wow.
Turner Novak:
She's eight now. But I was like, "Wow." It just blew my mind.
Michael Dempsey:
It's like a real person that can text.
Turner Novak:
Honestly, it was crazier than when she walked and when she could talk, but it was like texting me back. I was like, "Holy shit, this is nuts." So it sounds like when you're thinking about approach to starting a company and the opportunity in a space, I know you've said before that not enough people think about value capture, how do you actually build cash flow or enterprise value. What do you think people get wrong when it comes to thinking about how do you actually capture value for the business that you're creating, or what do they get right if that's an easier way?
Michael Dempsey:
Yeah. Well, I think the main thing people get wrong in deep tech areas is the commoditization curve. They think that the edge will lie in technology. And it almost never does, all technologies generally commoditize on some time horizon
Turner Novak:
So it's distribution or product uniqueness?
Michael Dempsey:
Yeah. Right now, it's like velocity. Right now, it's like everyone is sprinting. And if you're not consistently sprinting for the past three and a half to four years, you probably got disrupted by someone who started at a different point but could leapfrog you in some sort of product vector. Again, the way we frame it often is there's short-term, mid-term, and long-term moats. And I don't think it all correlates to revenue necessarily, or in the short-term. Long-term, these things accumulate over time. But I think short-term, your moats are ability to raise capital, it might be technology, you had a specific breakthrough. And you have to understand what's the half-life of that technological edge that technological edge that you've built? And then it's like, okay, talent, right? How do you compound just talent over time such that you have a moat on talent in your space? Especially in deep tech areas, typically there's 50 to 250 people in the world that are elite to the thing, and then there's going to be a tear down of great operators that are, call it, 200 to a thousand people. And that thus means there's ... Call it, 1,500 people in the world that could work at your company when you start it. And so there's a bunch of stuff you have to think about there. And I think, lastly, is probably just like how much of the given stack do you need to own such that you don't suffer death by a thousand cuts?
And this is actually more of a newer problem. So there's the Thiel-ism, which was competition is for losers. And that was a very prescient, albeit, obvious take. I think now it's actually you have to figure out what vector are you competing on that's different from customers or from competitors such that you can accrue value in your idea space. And the reality is is that probably in the first 24 months of your company, almost no matter what today, because the amount of people, the amount of founders, the amount of dollars, you will have competitors then. You have to very much think about why your value accrual is more durable than theirs, such that you can outlast them on a marathon. Especially because now ... Six or eight years ago, by the time a company got to series B or C, a lot of competitors would die out. Now it's like there's four to six companies that have raised a hundred million dollars going after the same thing. And you actually have to just continue to think about why you're part of the stack or why your product is positioned in a way such that value flows majority to you relative to competitors or to partners.
So the famous example of this is in synthetic biology and maybe bio broadly, the first generation of these next-gen bio businesses like Xymogen and Ginkgo, to a lesser extent, really messed up value accrual. They thought about product really poorly. They manufactured products that nobody wanted and they worked in these JV models that really didn't actually accrue much value. They didn't capture much value from. And I think what the next generation are now realizing is like, okay, you have to go after super high value products, that you have a lot of margin to be able to leak some value if you need to. Or you have to be able to get much further data in a data pipeline on a ... Call it a bio company. Such that when you go and partner with big pharma to do a scale up or to do a JV or something, you actually have a much deeper data package and you have more leverage in a given conversation.
You don't leak 95% of the value after spending, call it your seed to series B trying to get something to market. So it's very different space by space. And then cryptos's a whole nother world that we can talk about if you want.
Turner Novak:
So to the point of capital no longer being a moat, is it because there's just so much capital? If something works, there's a fund that's big enough to say a $50 million bet on something that doesn't exist yet, it's .5% of their fund and they can do it like an angel check. Like a-
Michael Dempsey:
Yeah. And there's enough capital around endowments, pension funds, Middle East money and sovereign wealth fund, whatever that are just looking to deploy such that the universe of investors that can write those checks is pretty large. And I think what's happening today in AI is pretty interesting because people are trying to king make companies. They're trying to foie gras these startups into just blowing as much on inference as possible to nuke a market and it just doesn't work.
Turner Novak:
Is it because there's so many other people doing the same thing?
Michael Dempsey:
There's so many other people and also it just isn't ... Again, the vector of differentiation is just not wide enough. The funny part is we saw this lesson. You remember this was 2017, we had Brandless, Wag. We had a bunch of these businesses that the original SoftBank vision fund blew money into and it was 200 million in this company with 10 million in revenue and at the time it was crazy.
Turner Novak:
That's like a pretty standard series A now.
Michael Dempsey:
And they all failed. They all failed because you can try to brute force distribution, but it hasn't worked and I don't think it's going to keep working. I do think that the thing that has happened in AI though is you have these late movers of really pedigreed people that then go and raise money and they're not in it to actually compete. They're in it to feel like they're not leaving a bunch of money on the table. And I think that also is these talented researchers get king made as these next generation labs that are going to steamroll the randos who could build it for them and it just has not been born out at all.
Turner Novak:
So what would you do if you're a founder right now? Should I go out and ... I have a couple million in revenue, I'm able to raise 200 million. What would you recommend? If you're on my board and I'm like, "Dude, what do I do here? What strategies or how should I think about it?"
Michael Dempsey:
I think there's a few different ways to think about it. One would be the obvious one. What do you actually need money for? Why do you do it? All that stuff. We'll call that this simple down the middle of the fairway. I don't think we need to talk about that too in depth, but do the reasonable thing.
Turner Novak:
Make a budget. How much money did you need to accomplish your plan based on your goals?
Michael Dempsey:
And figure out what are the experiments you would run if you had extra capital such that they could be asymmetrically skewed bets that you can take as a company.
Turner Novak:
Yeah. Accelerate getting to that. We built the competitive advantage and we're a public company and we have all the products and distribution et cetera.
Michael Dempsey:
Yeah. Or there's like we need to corner the market such that other people can't make offers to our talent because we can pay them two to three X top, which in today's market, unless you are literally Mark Zuckerberg, you can't out compete on comp so good luck. I think the other one would be think about where you are in the cycle and think about the lesson that a lot of people maybe are starting to forget and also learned in 2021, which is when the circus is in town, sell peanuts. And so if you believe we are approaching some sort of local top of the market or frothiness and you want to insulate yourself from dealing with that and you understand that your research trajectory or your spend trajectory is long go and raise a bunch and have clear alignment with the board, which is like this is what our two-year budget is and we are going to raise four X that because we believe either in two years we will have a clear inflection where we know how to deploy that money very aggressively. And we think being able to go heads down and do that is important. Or we think that there's a world in which the financing market falls off and the companies that survive will be those that took advantage of capital markets when they were ready.
But I think a lot of people use capital as a barometer for success. The only thing I'll say is as you start to build a scaling business and AI you do now today more than ever have to start to understand at least some of the dynamics around how do you get liquidity for employees as you scale and stuff like that because that is becoming more table stakes at growth stage companies. And so you might need money for that.
Turner Novak:
Interesting. Just in terms of a hundred million bucks to just top people off or buy their shares.
Michael Dempsey:
Like hey, we're going to raise an extra a hundred because or 50 because we're going to run secondary sales that we can guarantee to founders 10 million a year for the next five years. And that is our retention mechanism against Meta, Google insert other company that has effectively liquid stock. I think that's the thing that OpenAI has done really well. They figure out how to create consistent secondary. Obviously SpaceX has done that really well. And so even though they're private organizations, if you're an employee and you're doing the math trade off versus the public companies, you actually are like, my stock's like 85% as liquid as theirs and it probably has more ability to jump up in big multiples each time as we've seen with SpaceX and OpenAI such that the math is a little easier to stomach.
Turner Novak:
And as an investor, when you hear this phrase playing the game on the field ... I'm sure you've heard that before. Should you play the game on the field? To what extent should you play that game?
Michael Dempsey:
As an investor, no. As a founder, I think you have to.
Turner Novak:
So as a founder, you take the game, you play to the best as you can to whatever your strengths are. As an investor, you just don't play. You just sit and watch.
Michael Dempsey:
We raised our most recent fund in ... We closed it in April of 2021. We called Capital in September of 2021 and we effectively didn't invest until February of '23. And it was because we looked at the market and we said, "There's not stuff that we're super interested in and everything is so expensive and raising such crazy rounds that we don't think we can make money doing this." I think the hard part in what we've seen across different peers who maybe also had some feeling about the market is that doing that makes it harder to step back in and you typically need to at least have the ability to get back into the market in the right time and start to play a game.
Turner Novak:
Because people know you're not active and you lose deal flow.
Michael Dempsey:
No. Because you as an investor get validation for so long that you're right about the market structure that you're investing in and it usually takes ... There's more burden of proof to reverse your mindset. And so you are like, "Yeah, maybe things are coming back. I don't know." I would say the mistake we made was probably we could have started investing more consistently in that November '22 to March '23 period. And it took us a little longer to be like, "Okay. We think that the follow on market will come back." Because a lot of our thinking was in the stuff that we do, no one's going to do a series A in this thing in 24 months. And so if you're looking at April 2022, you're like, unless this is ... At the time it wasn't even clear. But unless it's an AI company that hits it, there's no series A capital still. And we spent all this time talking to follow on investors and you ask them like, "Hey, what do you want to see to get a deal done?" And they're just like, "I just deployed $600 million in 14 months. I have no fucking idea what I'm going to do." They're just totally lost. And so you're like, "Okay. Well, if that person who is ostensibly going to keep my company alive is trying to keep a bunch of other companies alive, they're probably not going to step up."
Turner Novak:
You have to be thinking about the game that's being played on the field even if you are not actually playing. So you have to observe still. You got to have the streaming subscription to still watch the games, but you just don't play.
Michael Dempsey:
Yeah.
Turner Novak:
So what is the importance? When you're making an investment, do you think a lot about follow on capital? What percentage? Is it like 50, 25% of your-
Michael Dempsey:
For our fund or what do you mean? Oh, what do you think about?
Turner Novak:
Yeah. So if you're like, "Okay. I'm going to invest in Runway. I'm maybe going to do the next round, but probably not the C and the D.", what percentage are you like, "Okay. I need to make sure that we can put 300 million into this company between now and the IPO?" Is that a big piece of it? Is it a small piece? Is it like Jesus, take the wheel? I hope that these guys crush it.
Michael Dempsey:
Yeah. I think we do a lot of understanding of what is the trailing six to nine month view on what milestones are needed to raise from a position of strength in every area we invest in. And so in bio that's very different. In robotics, it's different. AI, different. Crypto different. And so we have that and that's a big framing that we always tell founders, we want to align with you on the 18 months of your company and the seven to 10 years and everything in the middle, we don't know. But I want to know going in, hey, month 14 when we decided to go and raise capital, we're in agreeance on what we need to have done. And that is pretty important. There are some areas that this actually trickles even further. It's like bio right now I would argue there's a good amount of seed capital, there's a good amount of A capital and there's almost no series B capital.
Turner Novak:
Is that scary?
Michael Dempsey:
Yeah. For sure. But our view is like bio is so skewed to the upside right now. It's so hated at the growth stages. Again, unless you're building an AI foundation model for bio, which cool.
Turner Novak:
What does that even mean?
Michael Dempsey:
Yeah. It's so many of these. That we are actually just betting that the market structure will change four years from now, which is when a seed company that we invest in today might go and raise a series B. And there's a world in which we're wrong and we will start to see, hey ... We might recognize 20 months from now, hey, we're probably going to be wrong. Might recognize 36 months from now. But our job is to make sure that founders can be a little bit more aware of what's happening for that next round further out. But yeah. Sometimes it is Jesus take the wheel. And the interesting part about right now is everyone wants to do AI and most people don't want to do most other things. It's like AI, American dynamism, everything else. Some tech bio stuff. And I think as a seed investor, what that means is you just have to have confidence that markets will mature and or depth will move into breadth, which is funny enough a trend that we see in public markets too.
And so I think that unlike runway at the A back then nobody thought creative AI was interesting at a seed and we were like, who knows if anyone's going to do at the A. But back then venture was built around the idea of everyone invests in everything kind of. You do FinTech. Software is software is software. And I just want to understand who the buyer is, what the traction is, what the user looks like, show me some cohort data, whatever. It wasn't like I only invest in AI enabled software that does this, this and this. It was less prescriptive.
Turner Novak:
Yeah. So I think another thing you actually did really interesting ... Blake Robbins told me to ask about this. With the longer deployment periods that you have in the fund, did you do that on purpose or was it like because it was '21 and '22 you did a little bit of a different move there?
Michael Dempsey:
We always aim to do four years. Our first fund was 2016 to 2021 and it had a pandemic in the middle of it, and our second fund was '21 to ... We'll raise a new fund next year and that the dumbest bull market in tech's history in it. And so each will be a year longer. But our main view is we just don't get high conviction enough on a subset of ideas on a given year. And we're pretty confident that you need a certain end number of companies and we like the idea of being able to be very methodical and slow and not trying to raise capital and or deploy capital such that we can raise more.
Turner Novak:
Makes sense. You had a pretty interesting comment. You called Compound a forecasting firm.
Michael Dempsey:
Yeah.
Turner Novak:
Maybe self-explanatory this far into the conversation, but maybe expand on that a little bit. What does that mean exactly?
Michael Dempsey:
Yeah. I think one of the things that we do a lot of internally and then we've published two externally is we try to take year-by-year views of what will happen in a given area. So we've published two of them. One of them was in 2022, we published something called the Crypto Future, which was a I think 35, 40 page book that was looking at the next five years in quite great detail and saying here's what will happen across companies, regulatory, whatever. And we did that in bio hacking last year called The Biohacker Future. And we have this for a few different areas as well. And I think why we do that is it forces a level of rigor around time horizon and around confidence intervals on your beliefs across a bunch of different things. And to the initial point of Compound was originally built around we understand complex things to now is we understand how to model the second, third-plus order effects of how technologies move through the world. To do that, you actually do need to then draw direct lines over multiple years.
And so I think for us we view that as again, a partially academic exercise, but actually one that very much informs sometimes it results in actually this idea that we have that we think is a little crazy for 2025 isn't. We think actually the time to adoption or the time to this idea of proliferating the world is going to happen much faster to the other side of this cool idea that we think should exist. We actually can't imagine a world in which a venture-scale business is built over the next seven to 10 years in this space because we've reasoned through how we think the world is going to change. And I think it also allows us to have founders hopefully have some view of where they fit within these futures that we model and then either fit within them and say, "Hey, I'm building something that you talk about happening in 2028 and I think it can happen now." Or they can come in and just totally shatter and say, "This is never going to happen, but what is going to happen is what my company's doing." And that's equally strong signal to both.
We invested in a company called Tia many years ago which healthcare for millennial women. And our initial thinking was actually the right wedge was fertility because it was super high pay, very fragmented, bad experience. And those founders read a bunch of our reading on why we think the fertility market was quite interesting for a clinical healthcare model in a digital healthcare model. And Carolyn and Felicity came to us and said, "You guys are wrong. Actually, the way you should think about this is women don't build a relationship with their doctors until they have kids." Most of the time the average woman from the age of 20 to 28 in the US cycles through their primary care physician three, four or five times. And that's because you don't see them as often. You don't have the same type of connection with them. But once you have kids, you actually change a lot less. And so if you actually catch women upstream of that and you build an experience in which they are in your healthcare system from 20, 21, 22, you actually can then own them for the entire life cycle. And women's health is a very different type of healthcare than men's health. They actually go to the doctor more often, they spend more, all these things. And so that was their pitch and we were like, "Yeah, makes sense. Probably right."
Turner Novak:
We thought we had this great view. But you probably got exactly ... Like, oh yeah. But it's a perfect setup though because you knew why you were wrong then because you got it.
Michael Dempsey:
Yeah. You're like, "Oh, I thought this thing and here are all my priors." And I go, "I know exactly what you have voided and I know exactly what you've confirmed and I now have confidence that this was right, this is wrong." And I think again that's just ... I actually don't know how people do reactive venture investing because any company sounds fucking awesome when you sit down with them and they tell you about it, that I would just say yes to all of them. It's just like, of course this makes sense. Unless you don't believe in the founder. But I think outside of that, it's just so hard to have high conviction. I don't know. Yeah. That's how we do it, but who knows?
Turner Novak:
Do you think you can have success being more of a generalist investor not having a thesis?
Michael Dempsey:
I couldn't, but other people can, for sure. Yeah.
Turner Novak:
And it's just because you have these other advantages of you have the unique developed network or you have this great unfair advantage in understanding some technical things, so founders come to you and you get a fairly priced equity because of your expertise that you have or the value you provide.
Michael Dempsey:
Yeah. That's a great sounding pitch. I don't know. I think in general it's just like this is the playbook that we know how to run with relative success. And we think that a subset of founders really want to work with someone who has opinions on their space and know the space, whatever. And a subset or like I would never want Mike Dempsey on my board. It's annoying to have some person who reads a bunch of shit and wants to about it.
Turner Novak:
He wants to talk about everything.
Michael Dempsey:
Yeah. And I totally get that. I think that ebbs and flows, and I think so much about being a board member, so much about being a lead investor, which we normally are at the seed stage, is believing that you have a singular role to play in the company, not a totalitarian one. I think where a lot of VCs go wrong is they step into a board meeting, they're like, "I have answers for all of your questions." And I'm like, "I don't know. There's probably 50% of the stuff that you're going to ask and I'm going to have no idea how to answer. But I have a bunch of companies that in our portfolio we see struggle with weird complex things that often overlap with ambitious technical N of one businesses and maybe you want to work with us for that." And that enough times allows us to win. And then all the prior work allows us to identify and theoretically pick.
Turner Novak:
Okay. So I want to talk to you about crypto. I want to ask one thing first though. So let's say I'm 25, I just got a job at a venture fund. I'm an associate. I'm really excited about it.
Michael Dempsey:
Yeah.
Turner Novak:
What advice would you give me to help me understand what I'm getting into and do a good job?
Michael Dempsey:
You're getting into just a massive pile of egos. I don't know. My advice has become known for something so that you're irreplaceable. I think that most firms hire junior people to expand the bandwidth of the partners. And what they should be doing is expanding the bandwidth of the firms. And so when we've hired people, the thing that I've had to impress upon them is ... They in the interviews will be like, "I really want to understand what's hard in your world? What's hard in your life? What are things you think I could take off of your plate?" And I'm like, "I don't need you to take anything off my plate. I need you to go add a bunch of stuff to Compound's plate. That's what you should be doing."
I think good junior people figure out what is the thing that I will become indispensable for? And then what are the things that I am above replacement on for the partner I work with or for the team I'm on or whatever? Because you do need both where people view you as short-term valuable, which is like they do stuff I don't want to do. Which again, I think is the wrong way to think about it, but a lot of firms do. And then long-term valuable, which is like, "Oh my god, if they left would we wouldn't be able to do this thing." And I think building that framework across all types of founders is really helpful.
The other thing I would say is do not just try to anchor on talking to young people and all this stuff. Venture and investing generally is not graded on a curve. I do not care if you're young. Your returns have to be just as good as Pat at Sequoia's or Napoleon at Founders Fund returns to be a good investor. I do not give a shit if you're 25. And so that idea is something that people need to really internalize. It's why I hate when people in their Twitter bio at their age. It's like, I don't care about your age. No customer is going to be like, "Oh, this product sucks, but it's built by a 19-year-old, so I'm going to buy it." Versus this one by a 30-year-old is a little better, but they're 30. The world's not graded on a curve. So understand that very quickly and try to figure out why you can become above replacement to the GP at your firm in a certain area.
Turner Novak:
Do you think you should think about it as a long-term thing? If I'm getting my first job, should I be entering this thinking and I'm going to do it forever? Or should I be thinking about it as it's basically a sales job and a chief of staff type job? Should I just approach it from that perspective and embrace those skills?
Michael Dempsey:
It depends where you work. It depends on what you want to do. I don't know. We only hire people at Compound that hopefully want to be investors forever. And my view is everyone will just continue to own more of the firm. And that's great. I think some firms are like, "You're going to be here for two years. You're going to get experience talking to some of the smartest people in the world, and you'll figure it out." Venture is an incredible job. It creates massive amounts of optionality. It's a terrible job because you learn almost zero transferable skill sets. And so you have to figure out what your orientation is in it. I do think in general, all people, millennials and Gen Z in particular have this weird orientation now where everyone wants the next job to be the final one. Everyone always says that. You talk to them. I want the next job to be one where I can be at a decade. It's not realistic to try and make the next decision be the one that lasts forever. It's like how do I check in every six months, make sure I'm still on some rate of learning, and do I feel like I'm building a defensible long-term skill set?
I don't know. With the AI stuff, everyone's having existential crises. Every week I talk to a 24 to 32-year-old who has no idea what to do with their lives and think AI is going to steamroll their career.
Turner Novak:
Do you think that's true? I'm 34, maybe I slightly fit in that bucket. Is AI going to steamroll my career? Should I be worried about AGI or is it just a tool that I learn to use to take advantage of?
Michael Dempsey:
It'll probably steamroll parts of it. I don't know. To the point that I've said, it's steamrolled our edge in understanding papers. I don't know. And so what do you? You evolve to use the tool to do next order things better. I think the meme of the world will be split into people who can't use models, people who can is quite true. But I don't think I have a perfect answer. I will say I'm not ... I'm more in the Tyler Cowan bucket of diffusion of technology actually takes longer than people's tend to appreciate than some of the anthropic people who are like AGI next year we're all fucked.
Turner Novak:
Yeah. Well it's like with ChatGPT, I'm still meeting people. I just learned about ChatGPT last week. I'm like, "Oh cool. What do you think of it? That's fun. This is really fun, isn't it?"
Michael Dempsey:
Yeah.
Turner Novak:
So actually I lied. So I have two more things. So one of them, so you do the early stage in the public market and it's complete opposite really at the end of the day. But maybe there's some parallels. What do you think that the two could have learned from each other? If I'm a public market investor, what could I learn from learning to think more like an early stage seed stage investor and then early stage seed stage investor, what could I learn from somebody who does public market investing and has that skillset in that appreciation of the world?
Michael Dempsey:
Public market investors are starting to learn that idea motes and the framing of narrative capture is actually way more valuable than they ever appreciated. And I think private market investors have understood that quite well. Being the Uber for X is not nearly as valuable as being Uber. And understanding these companies that create categories, dominate the narrative for their space and act as what we've called them shelling point companies, we think that is materially valuable. That creates a multiple on your business. And I think that's something that public market investors have started to get. The tiger Cubs started to get that earlier. They were all about compounders and this idea that these things will just compound over long periods of time. And if you look at what they held, it was mostly tech stuff and Mag seven related stuff.
I think early stage investors or venture investors, the flip side of that is there are sometimes shooting stars and also there's gravity in markets. And what that means is that you have a maximally optimistic vision of a company. And I think we saw this in 2021, which is everyone took the maximally optimistic view of We Work isn't an office company, it's a new type of-
Turner Novak:
Data. Real world data.
Michael Dempsey:
Yeah. And it's like, well actually, if you understand how these companies generate revenue, what margin profiles look like, they typically will only ever maximally be worth X. I think that is actually quite important to understand. And the tension is that Mark Andreessen saying software is eating the world was incredibly prescient and actually defied a lot of the gravity of what people thought technology companies could be worth over long periods of time. But that is now a consensus view. Which is the consensus view is things will be bigger than anyone can appreciate. And I actually think it is more non-consensus to say that certain things have gravity and there's some stuff that will defy gravity. OpenAI, Anthropic, or businesses that defy gravity. Cursor is a business that defies gravity in the scale of which it's growing. I think that there are the vast majority of others that do not defy gravity, and they will reach some sort of terminal value. And thus private market investors should understand how to finance those businesses, how to exit those businesses and help founders understand how to navigate properly landing the plane on these companies. As public companies that continue to operate and generate cashflow or whatever, as acquisitions as not over funding them, whatever it is.
Turner Novak:
Yeah. Sounds like public markets, it's realizing that the upside is so much higher than you think, and in the private markets it's realizing that the public markets need to value these things at some point. So just remember how that works
Michael Dempsey:
Sometimes. And I'd say on the public market side, it's also that there are things that are worth more, not because of quantitatively legible things.
Turner Novak:
It's not showing up in the financial statements specifically. But this thing is trading at 500 times free cash flow, but it's because it literally has 90% incremental operating margins because of specific market structures, or scale economies, or whatever the thing is.
Michael Dempsey:
It can be that. It can be fundamentally driven, but also might be it's the only way to express this view in public markets and thus the flows are inherently dictated to go there.
Turner Novak:
So the multiple will just expand from 20 to 40 just because-
Michael Dempsey:
How else can you get pure play exposure to AI and government, and you long Palantir? How do you get pure play exposure to rockets when you see Elon at SpaceX? Well actually I can buy Rocket Lab. How do I get pure play exposure into robotics? Well, there's this random ass company called Serve Robotics. It was publicly traded via SPAC and on 4 million in revenue, it was worth 1.4 billion in public markets for a while.
There's a bunch of these. And I think, yeah, sometimes flows just disrupt any sort of fundamentals. And I think if you were to, one of our core beliefs is crypto today is 80% narrative, 20% fundamentals and how these things are valued. And public equities you could say are let's say 80% fundamental, 20% narrative. It's probably not that, but let's say. And those things are going to converge in the opposite directions.
Turner Novak:
Interesting. Well, it's an interesting transition into crypto. So just generally, if I've never heard of crypto before. Say I'm a smart person, I have the ability to understand it, just why does it matter? Why is it a thing?
Michael Dempsey:
I hate defending crypto at this point. So I will do it very lightly. I think there's a variety of things that have happened over with each year that goes by, I believe more and more in the core principles that underpin crypto. And those core principles are first order self-sovereignty, and being able to own and move and have transparency to the things that you own and move. And related to that is a distrust of institutions, and distrust in any sort of large actor in the world. And then I think third is just massive inefficiencies that compound over time in structured systems, and smart contracts inherently are good at that. Code is good at that. And immutable code is very good at that because then they can't be changed, and that works its way back into the distrust.
And so on the first order side, Bitcoin is like the greatest example in the world on this. It's you can move Bitcoin across borders, you can move billions of dollars and no one would ever know, except for theoretically on the ledger if you sent it. But you could just keep your hardware wallet with you and move. You can't do that with gold, you can't do that with cash. You also can't have assets that can be seized easily because again, there's no treaty that can happen between the US and insert other company, and they can take hold of your Bitcoin.
Turner Novak:
Your bank account or whatever.
Michael Dempsey:
Yeah.
Turner Novak:
You have a hard drive that has the hashed code that says that you have X amount of dollars, or X amount of value in Bitcoin, that you can plug in anywhere. You can go into, go to Japan, you can go to Brazil, plug it into a computer and you have access to it.
Michael Dempsey:
And this is the argument of people who are like, "Well, stablecoins don't matter because I can Venmo someone and it works tomorrow." And it's like you can, and also the US government can tomorrow be like, "Hey Venmo, you no longer can serve any person who lives in any country," insert country name or, "You can't serve anyone who has tweeted against the president or whatever." And being able to move a stable asset is also very valuable. And so decentralized stablecoins are very interesting. Tether is a good example of that in some ways. Dai is another example.
And then lastly, these smart contract components, which is if I today want to get a mortgage under collateralized, it's still quite hard. That is something that is built on that hundreds of years of trust. We actually haven't figured that out on chain yet. But if I want to do an over collateralized loan of any kind and I know it's not as efficient, whatever, I can go on and do that by pledging an asset into code, pulling out dollars against that asset in two transactions, and about call it six seconds for $3 of gas fees if I do it on Ethereum, maybe a little more. That is pretty elegant as a solution. And there's a lot of building blocks that can start to be built on top of that.
My favorite example, which is actually a largely failed protocol, but there's a protocol called Alchemics. And basically what it is, is let's say you have $100,000 of ETH, you can put your ETH in and you can take out $50,000 of US DC that you can go buy a car with. And what the protocol will do, it'll take your ETH and start to lend it, and earn yield and auto-pay back your loan. You don't have to do anything. So you now have pledged an asset, you've gotten money out, your loan is being auto-paid by generating yield elsewhere. And at some point you can come back and you can be like, "Okay, I'm just going to take my ETH out now. It's been all paid back." And it's fully locked in smart contracts in code.
And that is a mechanism that does not exist in traditional finance. There's no take a loan, put money market, have it route this way to pay back my loan. There's many layers in between, and there's people clipping 25, 50 basis points of fees all along the way that make that process even more expensive.
Turner Novak:
You're saying different asset management firms or different-
Michael Dempsey:
Verify-
Turner Novak:
... collateralized?
Michael Dempsey:
All sorts of stuff. And so I think on the defy part that's quite interesting. And then there's a bunch of other stuff too, which is capital formation matters actually across a bunch of areas. We think decentralized science is really interesting. And today if you wanted to do research on a rare disease indication, the pharmaceutical companies have largely punted on that, it makes no sense. And as a VC, it actually also probably doesn't make sense to back some of these rare diseases. But there's a cohort of 5,000 people that it really matters, and they're going to happily put money in to own a piece of the IP and watch this drug come to the real world. And doing that on chain is incredibly efficient and also allows people to coordinate capital in a trustless way, to then go and do trust someone will do something which is the science, but you can observe that science being done.
And again, we think that's very interesting. I think the problem with crypto is that it has largely been held back because I would argue the best builders in the world on average have chosen not to build in crypto because up until this year it was uncertain whether at some point in your life you would be dragged in front of insert legal body, because someone one day decided everyone who ever worked in crypto should go to prison. And I think now that in my opinion is largely off the table, and I think will be more clearly off the table in the next few years, you'll see a lot of more talented, less grifty people enter the space.
Turner Novak:
Yeah, because when I think of crypto, I just think, my general mental model is 99% of it's just frauds and scams. And there's maybe some interesting stuff, but too hard because just generally the incentives around it just lead to a lot of fraud. So it's just a void.
Michael Dempsey:
Yeah.
Turner Novak:
So an example with stablecoins, it's like the government says you can't send money to this place, but you can with stablecoins. There's a reason that the government is not allowing you to send money. Are you only using it to commit fraud or is there real reasons?
Michael Dempsey:
What if the reason is just we are having a bank run, and as a government we don't want our citizens to be able to move money out of the country?
Turner Novak:
So it's a way to make sure that you're not trapped in that-
Michael Dempsey:
Yeah, we assume that all governments operate in the interest of their citizens. Which is not true.
Turner Novak:
Yeah.
Michael Dempsey:
And in America it's not even clear if it's true.
Turner Novak:
Yeah, exactly.
Michael Dempsey:
In other countries it's definitely not true.
Turner Novak:
Yeah.
Michael Dempsey:
And I think that that component is quite important to internalize, which is this is the most ridiculous framing I've ever had. In the same way that you don't trust the random startup to off into your email and read all your email, there's a bunch of people who are the real world problem, not our first world problem, is like, "I don't trust the government to off into my entire life earnings and be able to see it and take advantage of it and seize it at any moment."
Turner Novak:
Yeah.
Michael Dempsey:
And why don't I do that? Well, because that startup has been hacked 10 times and their emails have been leaked. And they're like, "Well, because this government has printed money five different times and inflation has been 80% for the past seven years."
Turner Novak:
Yeah.
Michael Dempsey:
And so why do I want to hold it in this native currency? It's like, "Well, I know why they want me to hold it in the native currency."
Turner Novak:
Exactly. They benefit from it. Yeah.
Michael Dempsey:
But I don't. And so I think there's a lot of things like that, and I think you can go, this is a little bit crazier, but one of our beliefs is I think you can go further and look forward in the future and say what we have seen over the past few years is that monetary policy is actually fairly pointless unless you are one of the most powerful countries in the world. So if you control the yuan, if you control the US dollar, and then the Euro and the pound and some of these other more felt, but controlling monetary policy is incredibly complex and difficult. And a lot of countries are starting to just be effectively soft pegs to the dollar because it all that matters, or to the yuan. Some have totally ceded and they're just like, "We're actually just going to trade dollars now."
And I think more and more countries are going to start to be like, "Listen, the only thing that really matters is can you send a bunch of military into my country or not to control our citizens? And can our military control our citizens? Not printing dollars." And if that's the case, why don't I just instead use Bitcoin or use another asset or just use a US dollar or something else and I don't have to worry about this anymore?
Turner Novak:
Worry about trying to figure out my own-
Michael Dempsey:
Monetary policies. Yeah.
Turner Novak:
... rates and policies? Yeah. It makes sense.
Michael Dempsey:
And you saw a version of that one with the Euro. European Union happened. It's the Euro actually was great for, well, it was good for Germany in some ways, but it was great for other countries as well.
Turner Novak:
Great for Germany.And so why hasn't crypto just broadly taken over everything? Because the things you just described, a lot of them is you talk through them. "Oh yeah, it does make sense. Yeah, it does make sense." I get that. Why isn't every single financial service currently offered on chain in some capacity?
Michael Dempsey:
I think up until three years ago, it was pretty hard. The infrastructure is pretty hard. I think a large number of people don't want the accountability to be fair, but they don't want to have to have their codes and their keywords and their hardware wallets, whatever. We've started to solve some of that with privy and stuff like that. I think the regulatory environment up until 12 months ago was terrible. You could legitimately go to prison if you did something in crypto. And you also couldn't attribute value to tokens. It was illegal to be like, "This token is actually worth this $400 million that's sitting in this treasury. It's a claim on those assets."
And that makes it really hard to make these things feel totally valuables. Some people believed it, some people didn't. And so I think now that a lot of those things are starting to ease, and a lot of the infrastructure is starting to change, it will, but it's been definitely slower takeoff than we would want. But again, I don't know, effectively we're 13 years in, 14 years in. I know Bitcoin was earlier than that, but I don't know, 2008, 2012, I don't really take into account as much growth in crypto.
Turner Novak:
So you said something about how up until 12 months ago you could just randomly go to prison if you didn't know what the context. What exactly happened? I feel I've seen a little bit about, I don't know the exact nuance. It sounds like you're pretty-
Michael Dempsey:
There's been more, there's a bunch of stuff that we believe will happen over the next 12 months, but there has been more clarity around what is a security and what isn't? What's a currency versus not? Who controls the legislative body that would theoretically regulate crypto? There has been more of a signal from the government of we want to embrace this and we are working towards this. And that's what Trump and Sacks and a bunch of these people are theoretically working on. There's also been a disgusting amount of grift along the way with meme coins and other stuff. And now there are starting to be stablecoin bills that enable banking institutions to bank people with crypto to traffic and stablecoins, things like that.
And so I think there's a lot of, I would say there's not a bulletproof, and there's also now new legal frameworks as well that are enabling people to create entities that can be governed by decentralized groups effectively. And that took some time to be properly recognized at a state level. And so I would say it is still unclear whether if you want to direct cash flows to a token, which would make it pretty security-like, whether that would be totally embraced today legally. But the signals are, we are working towards embracing a version of that and giving clarity on what are the reporting mechanisms and the transparency needed in the same way that public tech companies have to, or public companies broadly, have to abide by a certain amount of rules in the SEC from a filing perspective? And up until then there was zero guidance and there was zero willingness to talk about what the guidance would be.
Turner Novak:
So you didn't even know what you were supposed to do in order to do things correctly?
Michael Dempsey:
No. Yeah.
Turner Novak:
So it was just you wouldn't do them because there's a chance that you'd be in prison for life?
Michael Dempsey:
Yeah.
Turner Novak:
You just-
Michael Dempsey:
There's very famous of multiple people at Coinbase Paradigm on the legal side, a bunch of Dows, Uniswap Team, whatever, trying to talk to the SEC. And they basically just refuse to tell them anything. And they're just like, "Sorry," we're just like, "We're not going to talk."
Turner Novak:
So we don't have an opinion or we're specifically not sharing things with you?
Michael Dempsey:
Just, "You want answers to figure out how can you legally do this properly, and we are not going to give them to you because we hate crypto."
Turner Novak:
So it was just maybe it was an irrational or is a rational from them for certain reasons to be against any sort of crypto option?
Michael Dempsey:
I thought it was, I think it was operating in bad faith is what I would argue. I think that there is a, if you have people who've been operating businesses to the best of their ability legally for long periods of time, and are actively documenting and trying to figure out what do we need to do for you to make you give us some guidance on if what we're doing is proper legal or not? And you're saying, "Not going to tell you," and at the same time you're throwing lawsuits at a bunch of these companies, it becomes like you're operating in bad faith.
And Hester Pierce, who was one of the SEC commissioners said that about Gensler last year. And there was a very clear dissent and disagreement amongst the SEC of this is clearly not operating in good faith. I think now that has changed and it's actually more of a bipartisan issue. And so we're quite bullish on how that can progress over the next few years.
Turner Novak:
So it's basically, "Tell us what we need to do in order to abide by the law and we'll do that."
Michael Dempsey:
Yeah.
Turner Novak:
And you think that will actually cause probably fewer, do you think there'll be less grift in crypto going forward?
Michael Dempsey:
Yes. There'll be less grift and also there'll be less people, victims of grift because it will be very clear like, "Hey, we are at least making a good faith effort to try to do things that are trackable, auditable, fundamentally legal, whatever, sit within some legal purview." And people who are like, "I'm not going to do that," are going to inherently probably suffer from a lesser amount of capital flows and interest.
Turner Novak:
It could be objectively clear, these guys are doing something illegal?
Michael Dempsey:
Yeah. Illegal or just there's some people in crypto who just refuse to believe that they should be under any law. And that's cool, but that's just not the way the world works. I'm sorry. It's you have to be a part of something. And so I think that's we'll start to see more of a separation of those things.
Turner Novak:
And it's interesting because the way I've always thought about crypto is my framing is just, it is its own separate economy, in its own world. Let's say you have $50 million of Bitcoin, you bought it a long time ago, and it's you now are a part of this whole entire economy. This whole entire society or world where you almost don't care what's happening in the US or Azerbaijan or Kenya, whatever, wherever you live, it's just I'm worth $50 million in crypto. That is my reality now.
So it's almost become its own separate economy, society, country, the whole nation. What's it called? Nation state?
Michael Dempsey:
Yeah.
Turner Novak:
The Balaji thing. How does he describe it? It's become its own sovereign state outside of geographic boundaries. It's an internet economy almost. Is Ethereum its own internet economy. Is Sol its own thing. I don't know if that's the right frame. That's in how my non-expert view on it.
Michael Dempsey:
I think of it as there is a view that I do believe is in the future there could be three factions. And what it's looking like today is there's the US faction and our allies. There's a China faction in China's allies, and I think there could be a third that is a more decentralized faction. And we have to order stuff from China and we export to them, and people from the crypto side transact and the real, all these things work together. But I do think there could be more split between those. I will say though, the interesting thing is because of the massive wealth creation, effectively on belief in crypto early on, this is a crazy thing to say, but there wasn't a lot of hard work to be done when you, the hard work was not selling your Bitcoin effectively. It wasn't like you had to build a company and do all these hard things.
There is a massive amount of philanthropy, and also more and more each day I talk to crypto people who want to do things that have impact on the real world. And so there is a desire to figure out how to translate or cross over in these worlds more and more, because I do think once you get past the wealth game and the I need to make a living game, people, they want the worlds to co-exist. And so we talk to a lot of people who want to do longevity, bio-investing, stuff that feels super, super tangible and important, and they're using crypto wealth to finance it. Which is an interesting signal to me.
Turner Novak:
Yeah. And I actually have one more completely unrelated question to any of this. So as Mack told me to ask you about, how are you thinking about AI and robotics?
Michael Dempsey:
I think robotics started as, the history of robotics for us, which is 2016 onward was that moment in time a lot of people thought that AI was going to make its way into robotics. And we backed some companies under that premise of using reinforcement learning and a bunch of stuff to have really intelligent robots. It was a little early. And what people then moved to was, okay, we have mechanical engineering, and that's you build machines effectively. And they're not super intelligent, but they're very repeatable, they're robust, they can do certain tasks.
Now I think everyone looks at the past few years of ChatGPT and says, "Okay, the main lesson we've learned is the bitter lesson, scale data, scale compute. You get emergent intelligence. Apply that to physical objects." And so everyone just is drawing the line and putting a piece of paper and tracing it with robotics. And they're like, "All right, so it's 2020 in AI time, but it's 2025 in robotics. And so next five years we're going to see an explosion of intelligence." And then they do the thing that is equally as not creative, which is they're like, "And what kind of robot should it be?" And it's like, "Oh, it should be a human." And it's like, "Okay, well why should it be a human?" It should be a human because that's maximally generalist. It's maximally easy to understand.
Turner Novak:
Think of the tam on that thing when you're raising money.
Michael Dempsey:
This is the point.
Turner Novak:
It's the economy.
Michael Dempsey:
It's a great narrative arc. I think the thing that we believe a little differently is this point on creativity is there's a bunch of different mechanisms in which you can build robots that can do things superhuman. We shouldn't be aiming for replacing humans. We should be aiming for building superhuman things. And so a laundry machine is a superhuman clothes washer. It's incredible at doing that. Same with the dishwasher, all these things. And so a lot of where we think about investing is intelligence will scale, things will become more multifaceted in their ability to move throughout the world, understand novel environments, but actually more purpose-built robots will be able to win, both because they're easier to manufacture. You don't need nearly the amount of actuators and motors and things like that to do this with your hands. And also they'll be able to get distribution ahead of a bunch of these humanoids. Humanoids just won't be able to scale at the rate that other types of robots will, again, because of the technical complexity in building them.
And so like all deep technologies, a lot of people build hammers and then look for nails. And I think in robotics, a lot of people have the hammer of AI scaling laws and humanoid robot, and they're looking for a bunch of nails. And my view is it's valuable to replace a human, don't get me wrong. Some human labor, physical labor can be anywhere from on a contracted basis, 30K to 200K a year. And the very valuable ones like Intuitive Surgical built a surgical robot, which is not autonomous, but for the sake of this argument, does a very high value task of surgery.
But the more valuable thing is actually not the human life. And that thus leads me to believe that if you were to build a humanoid, you care a little bit less about what is the dollar task of the employee you're replacing, and more about what are the situations in which it's really bad to have humans in that scenario because their life is at risk. And so that leads me to be like, okay, if you assume there's an amount of manufacturing of these humanoid robots that reaches some ceiling, which auto OEMs going manufacture a certain number of cars each year because the manufacturing apparatus is limited, you would then say, "Humanoid should be in military and industrial settings only."
Turner Novak:
It's like firemen or something.
Michael Dempsey:
Yeah.
Turner Novak:
Go into a home, carry people out.
Michael Dempsey:
Yeah.
Turner Novak:
Something like that.
Michael Dempsey:
As a city, as a taxpayer, I'm not upset about how much firemen get paid. I'm upset about if a fireman dies. And so the amount of value I'm willing to spend is materially more than the wage.
Turner Novak:
Yeah.
Michael Dempsey:
And so this idea of we're going to have humanoid robots in our home that do what mid-wage labor does is a little ridiculous to me if you try and reason through it in the near term. In the long term, sure, cost curves will come down, manufacturing scales, but I think by that time you will just have purpose-built robots that already have distribution today, such that I won't need in the same way that I don't have a server farm in my house because I actually, we figured out how to get PCs into the house before we put server farms into the house. There will be other products that emerge that are adjacent for doing the same thing, which is computation, that will get distribution and win out.
Turner Novak:
So is it just the narrative of the humanoid robots? It just takes over?
Michael Dempsey:
Yeah.
Turner Novak:
And it's-
Michael Dempsey:
It's great. Unlimited tam. You don't have to pick the utility today. You can say what's going to be general purpose, and so we'll be super malleable. And actually if you look at AI, the best models are general purpose right now. And so I don't know, would you rather be an investor in OpenAI or would you rather be an investor in 11Labs? They're both awesome companies, but I'd rather be in OpenAI. OpenAI is the humanoid robot and 11Labs is the defined robot.
And so I think that is a ridiculous argument when you have to bring things into the physical world. Again, markets have gravity in some ways, companies have gravity. But VCs and other investors get very lost in that. And some of the humanoid companies are incredibly good at marketing. And they make these sizzle hype videos, and a lot of investors love science fiction. They get nerd sniped into being like, "I know exactly how that world works because I saw a Terminator and I saw-"
Turner Novak:
I read Anders Gate-
Michael Dempsey:
... "Robot and Frank." And it's yeah.
Turner Novak:
Yeah. The foundation.
Michael Dempsey:
Yeah. And it's like, okay, cool.
Turner Novak:
Yeah.
Michael Dempsey:
I can envision that versus it's hard for me to imagine what a purpose-built retail robot is.
Turner Novak:
Yeah. Actually Max told me to ask you about this too. When it comes to the hyped launch, what's your opinion on, we're definitely in a phase where that is the strategy is get $100,000 launch video, or maybe it's less than that with AI now, but perfectly polished launch video. You put out this narrative and then you raise money and try to build a company.
Michael Dempsey:
Yeah.
Turner Novak:
What's your opinion on is that a good strategy? Is it?
Michael Dempsey:
It's probably a, from an expected value basis, a good strategy. Yeah. People will see the video, they'll be able to understand it, they'll share it, and they'll drive inbound. And if the first fundraise is all about maximizing funnel size, it does that well. I think it's at this point, so unbelievably not creative. And often the videos themselves aren't even that creative. It's-
Turner Novak:
There's a playbook.
Michael Dempsey:
Yeah. It's a playbook.
Turner Novak:
It's the same thing.
Michael Dempsey:
For sure. It's like the wood panels. It's the same thing.
Turner Novak:
Yeah, every podcast has one of these panels. Yeah.
Michael Dempsey:
Every podcast has the wood pans. It's the same. It's like stand at desk, talk about product, zoom in to screen to show product, whatever.
Turner Novak:
Probably an American flag somewhere in the background depending on who you're fostering to.
Michael Dempsey:
Yeah. And so I just think, I don't know, it's weird that we're the best founders are spiky, and they're a contrarian, and they are out of distribution feeling whatever. And then it's the people you give the most money to, if you're an investor, people I give the most money to are the ones who do the same exact thing, look exactly like all the others, act exactly like all the others. But again, I think it's reasonable. And I think that there's a lot of investors who make their way into these firms where the job is do the reasonable things, and if you do the reasonable things, you'll get promoted. And if you don't do the reasonable things, and if you do the reasonable things, you lose money, you won't get fired. If you do the non-reasonable things and you lose money, you're going to get fired.
And there's some next door to middle ground, which is if you do the unreasonable things and make money, you'll get promoted. But if you do the reasonable things and make money, you actually might not get promoted. And so there's a lot of very weird multifaceted thinking. And so usually people are like, "Do a bunch of reasonable things and then jump firms such that it doesn't matter how they turn out." And that's on the other side of the table, that's the expected value dominant approach to venture.
Turner Novak:
Yeah. That's fair. Because this performance based investing and AUM based investing. There's these two different, I feel like they're different strategies at the end of the day. How are you? Are you compensated from making money for investors? Or are you compensated from raising money from your investors?
Michael Dempsey:
Right. Yeah. Being Kathy Wood is the best fundraiser of all time and also has lost more money than anyone in human history.
Turner Novak:
Yeah. Which is maybe a story for the next podcast conversation. Well, this is a lot of fun. Thanks for doing it.
Michael Dempsey:
Yeah. Of course. Thank you.
Stream the full episode on YouTube, Spotify, or Apple.
Find transcripts of all other episodes here.