The Exchanges

Q Lenses works well for me, but what is the three categories of AI startups and how should we think about that?

A Uh, this actually is a framework that came in the mobile revolution for us at Spark and then reapplied. So this isn't, this is an old lens reapplied, uh, which is adaptation, evolution, and revolution. And, and there are versions of this that have existed as people have talked about AI generally, but for us, Can I use the mobile analogy to kind of get you there? So adaptation is the obvious, like, I'm going to take the thing, and I'm going to make a copy of the thing for AI. And so in the mobile revolution, this was the New York Times makes NewYorkTimes.com on mobile, and that's the product, right? And that's obviously a world where, 2023 was kind of the big adaptation push, and that was when Adobe Firefly launches, it's when Spotify DJ launches, it's when Canva create, I think it's called. It's when, like, you know, the big boys came to town with their AI products, and everybody had, like, a year to go think about what they were going to do after the GPT era, and, like, ship their incumbent advantage stuff. Like, that's all adaptation. Evolution, I think the easiest way to kind of separate it is it's when there's a new workflow. It's when the behavior has changed slightly, and so the good example of this is in the mobile era is Instagram, Where you're suddenly doing a different behavior than you used to when you think about Flickr or prior photo websites. It's a new behavior t…

AI assessment note: “this is an old lens reapplied, uh, which is adaptation, evolution, and revolution.”

Q Lenses works well for me, but what is the three categories of AI startups and how should we think about that?

A Uh, this actually is a framework that came in the mobile revolution for us at Spark and then reapplied. So this isn't, this is an old lens reapplied, uh, which is adaptation, evolution, and revolution. And, and there are versions of this that have existed as people have talked about AI generally, but for us, Can I use the mobile analogy to kind of get you there? So adaptation is the obvious, like, I'm going to take the thing, and I'm going to make a copy of the thing for AI. And so in the mobile revolution, this was the New York Times makes NewYorkTimes.com on mobile, and that's the product, right? And that's obviously a world where, 2023 was kind of the big adaptation push, and that was when Adobe Firefly launches, it's when Spotify DJ launches, it's when Canva create, I think it's called. It's when, like, you know, the big boys came to town with their AI products, and everybody had, like, a year to go think about what they were going to do after the GPT era, and, like, ship their incumbent advantage stuff. Like, that's all adaptation. Evolution, I think the easiest way to kind of separate it is it's when there's a new workflow. It's when the behavior has changed slightly, and so the good example of this is in the mobile era is Instagram, Where you're suddenly doing a different behavior than you used to when you think about Flickr or prior photo websites. It's a new behavior t…

AI assessment note: “this is an old lens reapplied, uh, which is adaptation, evolution, and revolution.”

Q And there's, there's one part of Oculus, which always fascinates me. And that's, Uh, the ability to hire a CEO in Brandon so early. So how do you think that affected the process? And what do you suggest to early stage startups about this with founders who maybe aren't right?

A Yeah, that is just an amazing piece of self-awareness by Palmer Luckey as a founder to say he didn't want that job, especially in the kind of Hollywoodization of startup land. Uh, we can, we can often think about the CEO job as the, the, the top of the ladder and the glamour job. But I think everyone just needs to understand it's a job like any other. It's very much akin to, you know, the engineer who's the best engineer on the team, but might not make the best VP of engineering because he'd much rather, frankly, be coding than managing a bunch of people. And it's just a job with a certain set of disciplines. Most people don't view it that way. I personally do. I've held the, you know, role of CEO in a startup, and I've held the role of COO and head of product and others, and all of them work. We're, you know, kind of equally difficult, quite frankly. Um, in Palmer's case, he had made that decision actually before ever taking any outside investment. When he raised the A round, uh, Brendan was already the CEO. Uh, you know, he had already decided to bring on Brendan and put him in that role. And, uh, that was, it was just amazing piece of foresight for him to have no ego attached to that role and simply say, what is it that I can do best to make this company successful? And what was it will make me happy to do? And then hiring for the other roles around him.

AI assessment note: “amazing piece of foresight for him to have no ego attached to that role”

Q It's gonna make us dinosaurs if we don't move with the times. How do you think about this and the mindset shift that needs to happen in investing?

A Well, I think it's happening already, whether we like it or not, right? I, I think we can't really preach that a founder is supposed to adapt to a market and understand that the market is there. There's a thing called founder market fit. And there's also, frankly, a thing called VC market fit. And this market for AI is Is wildly different. I don't think anyone would argue it's not wildly different. And then the question is, in what way is it different? And, uh, we had a B to B SaaS, you know, amazing, wonderful bull run in 2000 and, uh, 21 and, and a little bit afterwards. And I think we got really good at like, um, uh, uh, Greg Treverton uses this phrase of puzzles versus mysteries, which is just like puzzles are this thing that you can like, you know, use that raw horsepower to solve. And mysteries are, you know, there you have to go on the journey. There's like fog of war and you cannot work it out ahead of time. And many, in many ways, like the BDB SaaS blow up of that era was all about like the industrialization of venture capital. It was all about figuring out all of the puzzles needed to hire a hundred associates to do all of the work, to figure out exactly the right SaaS metrics, and then grind it all out. And no one has any idea what a model is even going to do in a week. So I don't know how that isn't a mystery. And so I think you have to build a firm with that set of…

AI assessment note: “I think it's happening already, whether we like it or not, right?”

Q Who tells me the opposite. Why do you think that they are very different, and actually it requires different teams?

A I just watch what our growth team does, and they do their job incredibly well. We talk about deals all the time. We talk, and again, there are seven people. They're also a small team, and, but they can be a little bit more hierarchical. They can have principals. They can have associates. They can do more diligence. You can look at the numbers. You can call 25 customers. Like, I'm not calling 25 customers. Most of the time I'm investing, there aren't 25 customers, and so it's a different process. It's a different muscle. I'm not saying I couldn't do it, but in a world where, as you said earlier, Just doing the job simply at the highest possible level is incredibly hard. Why would I try and play two sports? Like you can, you can try and be Jordan and play basketball and baseball.

AI assessment note: “it's a different process. It's a different muscle. Why would I try and play two sports?”

Q Okay, one to two a day. How long do you have?

A I think the one hour call is the worst call anywhere, because it's like too long to get a real read, and yet, you know, it's like, sorry, too short to get a real read, and, and, and too long to get the kind of speed dating version of the world. So a half an hour, I'm just trying to figure out Whether I like the person at all, and I want to have a second call, right? And then I'd rather go from a half an hour to two hours. So you do the first one, and you're just trying to get a read. Is there a kismet? Is there a connection? Is there any chemistry here? Do you feel it? And then after that, it's like, yeah, we should just, let's just go for a walk, man. Let's go have a conversation. Let's go really talk about everything.

AI assessment note: “I'd rather go from a half an hour to two hours.”

Q And then on the other end... Can you paint the bull case for me with Anthropic?

A Sure. The thing to understand about Anthropic and OpenAI, ChapGP, they're direct competitors, obviously, is to think about it in the terms of this adaptation, evolution, revolution framework. I'll just use this thing we just talked about a minute ago as a way to make this point. If you are trying to make the next best model and you're running out of data, then what do you need? You need to understand how people want to use your model. If you want to understand how people want to use your model, then you need a lot of people using your model. And so the fact that those two companies have a user interface, which has both insights to how a user would use it, but also frankly data exhaust, On how people are trying to navigate a model and get through it gives you insight that no one in some academic lab somewhere just popping up a model is going to be able to advance as fast as you. You can make a faster algorithm, but you can't get new data and new data exhaust from consumers.

AI assessment note: “gives you insight that no one in some academic lab somewhere”

Q Gosh. Of a business school have just gone out of business. Um, what about, I'm thrilled. I can't work in Excel spreadsheets. Like, you know, teams like, hi, if you do equal some, I'm like, ooh, came up with a bracket. Um, what about the way that your parents brought you up? Did you do differently with your kids deliberately?

A I don't know that I did very much differently because I think my parents did not understand me at all, and yet were incredibly open to me walking my path. You know, Harry? Like, I, I, my mother was a first generation immigrant. Like, she just wanted me to be a doctor or a lawyer or whatever, except unlike a lot of first generation immigrant families, she never told me that, and I never even felt it. She could tell that I was going to walk a weird Path. She didn't even know what entrepreneurship was. She didn't know any of that stuff. And she just like, just wanted me to find my place. So no, I, I'm more trying to mimic my parents than I am the opposite.

AI assessment note: “I'm more trying to mimic my parents than I am the opposite.”

Q Did you make much money from selling your company? What I'm going at here is how did making money change your mindset?

A I don't think that making money changed my mindset very much. I don't actually think I execute that differently than when I was, when I had a term sheet pulled on me and I had to sell my car in order to Make payroll for my team and those stages of early entrepreneurship. I don't know that I process the world that differently, mostly because I like playing the game for the joy of playing the game, whatever the game is, and the score takes care of itself, and I get that we all get measured on a leaderboard. Like, we don't get to get, I don't get to keep doing this if we don't make a lot of money for our LPs and our founders aren't happy and all the rest of it, but I just enjoy the work, and I still enjoy the work, and I can go do other things if this doesn't work out.

AI assessment note: “I don't think that making money changed my mindset very much.”

Q Did you make much money from selling your company? What I'm going at here is how did making money change your mindset?

A I don't think that making money changed my mindset very much. I don't actually think I execute that differently than when I was, when I had a term sheet pulled on me and I had to sell my car in order to Make payroll for my team and those stages of early entrepreneurship. I don't know that I process the world that differently, mostly because I like playing the game for the joy of playing the game, whatever the game is, and the score takes care of itself, and I get that we all get measured on a leaderboard. Like, we don't get to get, I don't get to keep doing this if we don't make a lot of money for our LPs and our founders aren't happy and all the rest of it, but I just enjoy the work, and I still enjoy the work, and I can go do other things if this doesn't work out.

AI assessment note: “I don't think that making money changed my mindset very much.”

Q When you reflect back on your prior portfolio in the last decade, was that a pattern matchy? Approach that was successful. Have you had to change?

A No. I also think I was really badly shaped to be an investor in twenty-twenty-one. I mean, I think we got lucky. Spark was started in the early web two point O era, like right at that age, uh, in the same cohort as USV and benchmark two point O that the beginning of the girly era and a handful of other firms that I think all treated mobile really well and did mobile really well, which felt similar. You didn't know what the metrics were supposed to be. You had, it was a wide open, crazy world. And you're like looking at Something that was maybe a fart app in the morning, and then, and then Uber in the afternoon. Like, it was an insane situation, and I think our DNA was very fixed by that. Our values were set by that navigation, and I'll be the first to say that, like, I don't know that we navigated the B to B SaaS era four years ago, this kind of industrialization. We didn't do the things that a lot of our peer firms did. Like, we had been very successful. We could have very easily raised five billion dollars. We could have very easily You know, tripled or quadrupled the size of the team. We didn't do that. We stayed seven people, six people partnership. We all write checks. We all do work with our founders. We like the service work. And so I, I would argue that made our job a lot harder four or five years ago, to be honest. And it makes it a lot easier now because we feel very …

AI assessment note: “No. I also think I was really badly shaped to be an investor”

Q Who tells me the opposite. Why do you think that they are very different, and actually it requires different teams?

A I just watch what our growth team does, and they do their job incredibly well. We talk about deals all the time. We talk, and again, there are seven people. They're also a small team, and, but they can be a little bit more hierarchical. They can have principals. They can have associates. They can do more diligence. You can look at the numbers. You can call 25 customers. Like, I'm not calling 25 customers. Most of the time I'm investing, there aren't 25 customers, and so it's a different process. It's a different muscle. I'm not saying I couldn't do it, but in a world where, as you said earlier, Just doing the job simply at the highest possible level is incredibly hard. Why would I try and play two sports? Like you can, you can try and be Jordan and play basketball and baseball.

AI assessment note: “and so it's a different process. It's a different muscle.”

Q Okay, one to two a day. How long do you have?

A I think the one hour call is the worst call anywhere, because it's like too long to get a real read, and yet, you know, it's like, sorry, too short to get a real read, and, and, and too long to get the kind of speed dating version of the world. So a half an hour, I'm just trying to figure out Whether I like the person at all, and I want to have a second call, right? And then I'd rather go from a half an hour to two hours. So you do the first one, and you're just trying to get a read. Is there a kismet? Is there a connection? Is there any chemistry here? Do you feel it? And then after that, it's like, yeah, we should just, let's just go for a walk, man. Let's go have a conversation. Let's go really talk about everything.

AI assessment note: “So a half an hour... And then I'd rather go from a half an hour to two hours.”

Q Do you engage in secondary markets actively? When we look, no, why not? With the huge influx of like private late stage capital and the continuing delays of Public markets. We need liquidity. At some point you have to deliver cash to your investors. I do too. Do you not think that becomes an ever more important part of our role?

A I, I'm not saying you never sell secondary. I'm not saying it never happens. It's just that I think the, again, the primary job of trying to figure out a little bit about the future, listen to those founders who have that little glimmer of the future, have a beginner's mind enough to be open to it, so that when somebody comes into you with some cockamamie idea that was way off piece from how you thought the world was gonna work, that you're open to shifting to it. Like, that takes time, energy, research, It takes trying every product. It takes curiosity. Like, the question is, like, where are your hours coming in the day? And so, sure, there's secondary that happens. Sure, there's later stage valuations that happen. Sure, you can decide to figure out you want to do growth. Sure, you could run, you know, a conference every month. Sure, there's a thousand things you can do. But doing the simple thing at the highest level takes time and energy, and I don't even think I'm good at it yet. So I'm still just trying to get good at my first job before I do the second, third, and fifth job.

AI assessment note: “I'm still just trying to get good at my first job before I do the second”

Q And then on the other end... Can you paint the bull case for me with Anthropic?

A Sure. The thing to understand about Anthropic and OpenAI, ChapGP, they're direct competitors, obviously, is to think about it in the terms of this adaptation, evolution, revolution framework. I'll just use this thing we just talked about a minute ago as a way to make this point. If you are trying to make the next best model and you're running out of data, then what do you need? You need to understand how people want to use your model. If you want to understand how people want to use your model, then you need a lot of people using your model. And so the fact that those two companies have a user interface, which has both insights to how a user would use it, but also frankly data exhaust, On how people are trying to navigate a model and get through it gives you insight that no one in some academic lab somewhere just popping up a model is going to be able to advance as fast as you. You can make a faster algorithm, but you can't get new data and new data exhaust from consumers.

AI assessment note: “the fact that those two companies have a user interface, which has both insights”

Q have ever thought, and so an example is that Happy Robot, this Andreessen company, which is brokers calling truckers To organize loads and transport. Before you're like, that's not very interesting. But actually, if you replace the broker, that's a multi multi-billion dollar market. How do you think about the future of an agent economy in that respect? Are you excited by that? Do you spend time thinking about it?

A I spend a lot of time thinking about it. I think most of them How do I put it? I think most of them fall into, if you just really think through the second order effects of it, fall into kind of like near-term arbitrage, which might just take that whole market to zero, especially if you are meeting the market where it is today with the models of today. Like, you, you, you need to assume that you still have a second, third, and fourth act in your business, because you're going to need to keep innovating, or else how are you not going to get lapped Marked by 25 other competitors who are also gonna build call agents into your vertical market tomorrow. And so I think you do have to ask some of these questions. Sensibly, how hard is the job to be done? This is a direct contrast. If I'm, if I just took my 200 K check, and I just joined an incubator, and I'm trying to show 10 or 10% week-over-week growth so I can raise my seed round in three months, then I want something that the models can solve for tomorrow, right? And so I might go into a market that I don't know, a little bit of transcription from AI, uh, it, like, solves it and we're kind of done. If that's really the extent of your innovation, you're probably gonna be awash with 50 other people who also joined all the other incubators and are also doing the exact same thing. And so your marginal benefit to the world is zero. If y…

AI assessment note: “I think most of them... fall into kind of like near-term arbitrage”

Q Gosh. Of a business school have just gone out of business. Um, what about, I'm thrilled. I can't work in Excel spreadsheets. Like, you know, teams like, hi, if you do equal some, I'm like, ooh, came up with a bracket. Um, what about the way that your parents brought you up? Did you do differently with your kids deliberately?

A I don't know that I did very much differently because I think my parents did not understand me at all, and yet were incredibly open to me walking my path. You know, Harry? Like, I, I, my mother was a first generation immigrant. Like, she just wanted me to be a doctor or a lawyer or whatever, except unlike a lot of first generation immigrant families, she never told me that, and I never even felt it. She could tell that I was going to walk a weird Path. She didn't even know what entrepreneurship was. She didn't know any of that stuff. And she just like, just wanted me to find my place. So no, I, I'm more trying to mimic my parents than I am the opposite.

AI assessment note: “I don't know that I did very much differently”

Q And I'm also intrigued to dig in on the team aspect there of Oculus and ask how important do you think it was for them being so small in terms of kind of agility and nimbleness? Was that a key factor to their success?

A I think team size was a key factor to their success. I, I think it I think the thing that Oculus had, you know, the right mix of being nimble because they were capital constrained, but enough money to try and execute on their vision. It is something that sometimes in these, you know, as these bubbles, uh, are forming and, and lots of capital is flowing in. I do worry sometimes about a company raising too much capital and the team getting way too big and ostensibly getting so large so early and That they don't have the nimbleness of a startup, even though they're, you know, two years old, and they still are figuring their life out. You know, Oculus, I think, had the benefit of being, I would call it like right-sized, which is they obviously felt a tremendous amount of constraint. They had so much to do on the software and hardware side in order to bring virtual reality to the masses, but weren't so big that they couldn't kind of change direction and, and adjust with the times as, as new technology was coming out along the way. That makes sense?

AI assessment note: “I think team size was a key factor to their success.”

Q And then final question, your most recent investment and why did you say yes?

A Uh, the last announced investment is a company called Fig. It is, I said yes because of the founder first and foremost, but the product itself is transformative. Um, it is, in short, um, reinventing the way games get made, and I, uh, long have hated the way that creative content comes to market, which is generally some VP of marketing or some mid-level biz dev person at some large game publisher or movie studio says, I think we need more dinosaurs, and so somebody makes a dinosaur movie or a dinosaur game. Fig is trying to reinvent, kind of take the best of Uh, crowdfunding, equity funding, um, kind of the whole thing, uh, data-driven approach to creative content and kind of rebuilding the way games get built. And, uh, and so I was really supportive, not just for thinking it's a good business, but quite frankly, hopefully, uh, changing an industry for the better.

AI assessment note: “the last announced investment is a company called Fig. It is, I said yes because of the founder”

Q And so, what do you think Cruise's acquisition then means for the autonomous car industry going forward?

A Well, it's hard to talk about the material stuff because Cruise hasn't announced exactly what they're doing, uh, and the public will find out in a, in, you know, relatively short term, uh, it will be able to talk about it, um, but not just yet. I, I will say it probably represents A shift from phase one of self-driving into phase two of this ecosystem. And what I mean by that is, look, when we invested in Cruise, there were very few people investing in self-driving. Everyone believed that Uber and Google, probably, you know, the big companies were going to run away with this, and it wasn't really a startup scheme. A billion-dollar exit means that all the VCs are now scrambling to try and put some money to work, and a whole bunch of entrepreneurs are Um, are now thinking that they can start a self-driving startup. And so in one sense, I think that's wonderful. There's still a whole bunch of technology problems that need to get solved in self-driving. There are other market segments other than the market segment that Cruz is going after that also can bring this technology to bear. On the downside, it also means that it's a pretty treacherous time to start a self-driving startup because there's going to be a lot of funding in the space, a lot of attention in the space. Talent is going to be very hard to come by. And so, you know, if you're starting a I think you have to think back…

AI assessment note: “represents A shift from phase one of self-driving into phase two”

Q And I'm really intrigued then to merge your operations in your VC life and to ask what were the biggest lessons that you can apply from your time as an extremely successful founder and CEO that you can apply to being a VC at Spark? What are those kind of transferable tables?

A Oh, it's, yeah, it's really easy. Um, it's not transferable. That's actually, like, I actually, the conversation I have a lot with, with, with CEOs that I don't think, if you have not been an operator, you know, and you go into the VC life, I think the thing that you don't understand or can't really internalize is, as a CEO, how many people give you advice? You know, it's, it's, you know, I, I think most VCs think they've had some blinding insight to Um, and they're going to impart that advice onto a CEO without really internalizing the fact that most advice that is given is horrible because it is basically, Hey, I have this other company in my portfolio, or I started one company in my life and it did a good thing. And so I'm going to give you the advice to do the same thing as that company. And of course, like every individual startup and individual founder is unique. And so any kind of patent advice I would give right now, Is, uh, kind of like by its very definition wrong. All I really try and do is read every situation, take the time and energy to read every situation for what it is. And I know that's kind of, uh, Abstract, uh, and high level, but it's also the advice I give CEOs. Whenever you get advice, try and think about where that advice is coming from and why as much as you are, whether it is seems like off the cuff valuable or not.

AI assessment note: “Oh, it's, yeah, it's really easy. Um, it's not transferable.”

Q And was the deal itself very competitive? Because now we look back and think, oh my God, everyone must have been fighting to be getting in there. How was the deal itself in terms of competition with other VCs?

A Oh, it was definitely competitive. Um, but there were also a lot of tier one, you know, Silicon Valley types that passed. And, you know, when you talk to them later about why some of it was the same, you know, reaction that I had, that first reaction I had before I'd actually gotten the demo, um, Which is this feeling that they, oh, that's not even a category that is hot or not, and so I should pay attention to it, which is, you know, the complete wrong reaction. And the second one is, I think VCs generally, despite whatever they say on their websites, are market-first investors. So they want a market that they can model, you know, that they can give to an associate, and an associate can say, this is a one billion dollar market or a one trillion dollar market. And, and I think most VC firms, that is actually, for them, more important than They get to team and product and business model after that, and we've just never been that kind of firm. We've always first been a product-driven company, uh, and we kind of lead in with that. If you look at all of our big wins from Oculus to Twitter to Tumblr to Cruise, like all of these companies were, at the day that we invested, were very hard to actually model the exact market size of that company, And so there are a lot of firms that kind of had a harder time competing for them. It doesn't mean they weren't all competitive deals. Of cour…

AI assessment note: “Oh, it was definitely competitive. Um, but there were also a lot of tier one”

Q Will you ever invest if you haven't met them in person?

A If I've never met them in person? Probably not. There's some world where You met them five years ago. You knew them quite well. You know, Chris from Granola is a prior Spark founder, and so he was one of the first people I met after I joined Spark, um, because I was supposed to go sprinkle growth fairy dust on him in New York and talk about growth marketing, growth hacking, and all the rest of those metrics things that I don't really aspire to now. Um, and, and I, I said, you know, Discord, Jason, I knew for seven years. We were founders together before, before investing in Discord, and also smartly passed I was twice on Discord before investing. So like, you know, like a lot of long-term relationships and a lot of short relationships where you passed and then spent time. Cruz, back in the day with Kyle, I passed. We did a huge deep dive on why I thought his business wasn't going to work. He disappeared for nine months and wouldn't return my emails, and then he comes back nine months later, and he's like, yeah, this is like, we've re-pivoted. We've gone from trying to put, you know, aftermarket things on top of Audis, and we're now going to build a full self-driving stack, I'm going to show you a demo. You and like five other investors a demo because I really liked our last conversation, and so you knew somebody for eight, nine months. You've thought of how they internalize inf…

AI assessment note: “If I've never met them in person? Probably not.”

Q Do you engage in secondary markets actively? When we look, no, why not? With the huge influx of like private late stage capital and the continuing delays of Public markets. We need liquidity. At some point you have to deliver cash to your investors. I do too. Do you not think that becomes an ever more important part of our role?

A I, I'm not saying you never sell secondary. I'm not saying it never happens. It's just that I think the, again, the primary job of trying to figure out a little bit about the future, listen to those founders who have that little glimmer of the future, have a beginner's mind enough to be open to it, so that when somebody comes into you with some cockamamie idea that was way off piece from how you thought the world was gonna work, that you're open to shifting to it. Like, that takes time, energy, research, It takes trying every product. It takes curiosity. Like, the question is, like, where are your hours coming in the day? And so, sure, there's secondary that happens. Sure, there's later stage valuations that happen. Sure, you can decide to figure out you want to do growth. Sure, you could run, you know, a conference every month. Sure, there's a thousand things you can do. But doing the simple thing at the highest level takes time and energy, and I don't even think I'm good at it yet. So I'm still just trying to get good at my first job before I do the second, third, and fifth job.

AI assessment note: “I'm still just trying to get good at my first job before I do the second”

Q I am building a firm, When we think about reshaping our firms to a new world of venture and startups, what needs to be done?

A I think you need to think like a founder. I think you need to be okay with a small team that makes subjective bets and go think about the craft of what a founder needs today, which is that, honestly, what a founder needed four years ago was a lot of playbooks. You needed every single arbitrage-y, simple way to make everything move a little bit faster. The conversations I have founders calling me about now, um, after a board meeting, before a board meeting, they are all unknowables. They are all trying to, like, figure out, um, what is the new user experience that I should use when models go multimodal? And, like, there are explorations. Not that, not necessarily I have the answer for. They're not coming to me for answers. It's a sounding board, right? And they need a sounding board of board members that are actually, actually using their products and actually have a curiosity to use the rest of AI products and get more native. Instead of just, you know, watching Twitter and looking, look at the markups.

AI assessment note: “I think you need to think like a founder.”

Q When they've lost their way, what did you not see that you should have seen?

A It hasn't gone well. It hasn't gone well for mostly Two reasons. The first is that they're conflict avoidant, and I didn't pick up on it early enough. And, and it's very hard to pick up on early because you're going through a period where ideally you love them, they love you. There's not that much conflict, maybe until late term sheet or something like that, but like there's not that much conflict. And then you go through 25 conflicts in the first month of the company or three months of the company, and you realize that they're conflict avoidant. They're not facing the problems of that company because every company has a thousand problems, obviously. That's the first one. You try to read for it, but you can get it wrong. The second one is that every founder is like, um, this is one of those things that I did not have as a framework 10 years ago, but after you make like a bunch of mistakes and you look back on things, like they can become more clear. Um, every founder you want to move really, really fast. We all talk about execution speed. You want to say, you can imagine somebody on the very, very far end of execution speed. We also want them to have taste and judgment, right? We also want them to be, Especially in the world of AI, only taste is going to matter in the future because execution is just going to happen, right? So these two things are directly in conflict. If you a…

AI assessment note: “The first is that they're conflict avoidant, and I didn't pick up on it”

Q Does DeepSeat change how OpenAI and Anthropics should operate? I just had Jonathan from Grok on the show and he said, if I was Sam Altman, I would open source today. You will die if you don't open source.

A I don't know that DeepSeek changes that much for the way that I think about the future, strangely enough, and maybe that's an odd thing to say when everybody in the world is freaking out about DeepSeek this week. Um, I, I don't know that I ever really believed, personally, that the hundred billion dollar, or five hundred billion dollar, or one trillion dollar training was the only barrier to entry for making these models. And so, like, it was, in fact, if we had believed that only capital was going to win, Then we would not have invested in Anthropic, because you, you know, surely Sam was telling us and everybody else, like, we're going to win the capital game. The capital game is the only way to win, and so there's no reason to build a competitor. So we didn't believe that back then, or else we wouldn't have invested in Anthropic, and so it's still true today.

AI assessment note: “I don't know that DeepSeek changes that much for the way that I think”

Q have ever thought, and so an example is that Happy Robot, this Andreessen company, which is brokers calling truckers To organize loads and transport. Before you're like, that's not very interesting. But actually, if you replace the broker, that's a multi multi-billion dollar market. How do you think about the future of an agent economy in that respect? Are you excited by that? Do you spend time thinking about it?

A I spend a lot of time thinking about it. I think most of them How do I put it? I think most of them fall into, if you just really think through the second order effects of it, fall into kind of like near-term arbitrage, which might just take that whole market to zero, especially if you are meeting the market where it is today with the models of today. Like, you, you, you need to assume that you still have a second, third, and fourth act in your business, because you're going to need to keep innovating, or else how are you not going to get lapped Marked by 25 other competitors who are also gonna build call agents into your vertical market tomorrow. And so I think you do have to ask some of these questions. Sensibly, how hard is the job to be done? This is a direct contrast. If I'm, if I just took my 200 K check, and I just joined an incubator, and I'm trying to show 10 or 10% week-over-week growth so I can raise my seed round in three months, then I want something that the models can solve for tomorrow, right? And so I might go into a market that I don't know, a little bit of transcription from AI, uh, it, like, solves it and we're kind of done. If that's really the extent of your innovation, you're probably gonna be awash with 50 other people who also joined all the other incubators and are also doing the exact same thing. And so your marginal benefit to the world is zero. If y…

AI assessment note: “I spend a lot of time thinking about it. I think most of them”

Q I am building a firm, When we think about reshaping our firms to a new world of venture and startups, what needs to be done?

A I think you need to think like a founder. I think you need to be okay with a small team that makes subjective bets and go think about the craft of what a founder needs today, which is that, honestly, what a founder needed four years ago was a lot of playbooks. You needed every single arbitrage-y, simple way to make everything move a little bit faster. The conversations I have founders calling me about now, um, after a board meeting, before a board meeting, they are all unknowables. They are all trying to, like, figure out, um, what is the new user experience that I should use when models go multimodal? And, like, there are explorations. Not that, not necessarily I have the answer for. They're not coming to me for answers. It's a sounding board, right? And they need a sounding board of board members that are actually, actually using their products and actually have a curiosity to use the rest of AI products and get more native. Instead of just, you know, watching Twitter and looking, look at the markups.

AI assessment note: “I think you need to be okay with a small team that makes subjective bets”

Q When they've lost their way, what did you not see that you should have seen?

A It hasn't gone well. It hasn't gone well for mostly Two reasons. The first is that they're conflict avoidant, and I didn't pick up on it early enough. And, and it's very hard to pick up on early because you're going through a period where ideally you love them, they love you. There's not that much conflict, maybe until late term sheet or something like that, but like there's not that much conflict. And then you go through 25 conflicts in the first month of the company or three months of the company, and you realize that they're conflict avoidant. They're not facing the problems of that company because every company has a thousand problems, obviously. That's the first one. You try to read for it, but you can get it wrong. The second one is that every founder is like, um, this is one of those things that I did not have as a framework 10 years ago, but after you make like a bunch of mistakes and you look back on things, like they can become more clear. Um, every founder you want to move really, really fast. We all talk about execution speed. You want to say, you can imagine somebody on the very, very far end of execution speed. We also want them to have taste and judgment, right? We also want them to be, Especially in the world of AI, only taste is going to matter in the future because execution is just going to happen, right? So these two things are directly in conflict. If you a…

AI assessment note: “The first is that they're conflict avoidant, and I didn't pick up on it”