Uncapped #23 | Martin Casado from a16z

23:00And I think at least that gives them the power to decide who it is, but not kind of hamstring investment efforts. That makes sense. It's also funny because I think a lot of times, two companies that look the same but are serving very different segments of the market. Those are actually, they might sound like competitors, but they're actually never going to bump into each other versus two companies offering different products to the same customer and much more likely to bump into each other. Well, in AI, it's even crazier than that, which is the market is so big and it's growing so fast. Even companies that seem like they're competing end up in totally different places, just because so much white space is being created, but they're all competing, like totally different companies are competing for the same talent.

23:39So the first time I can remember where the actual talent competition is like way more fierce than the market cup. Well actually, it's funny about that as I've heard of people getting upset with their investors because, and they're like, I know this company has nothing to do with it, but we were interested in that candidate and told them to be a partner sold that candidate on one of their programs. It's a very real thing. What's interesting is like, you often don't even know. I'm like, you know, like they'll say, oh, I'm talking to a healthcare company, or whatever, and you're like, okay, well, what's tough, but I guess there's also like a blessing overall.

24:13I think there's a lot more good ideas than there are talented people to work on those ideas. And I think one of the hardest things right now is like clustering talent densely enough behind a good idea. Yeah, it's also, this happens when there's these large infrastructure buildouts. It's happened with the cloud too, which is there are these moments in time where to build the system, you have to have experience with a system at scale. This happened with the internet. This happened with the big cloud data centers. And this is the case with AI, which is like it's one thing to go to school and know AI and be a good researcher.

24:43It's nothing to have actually trained a very large model. You know, maybe what? There's 30 teams that have ever done it. And that's part of what you see is like mega, aquahire, opposition things. Like certain experiences are just worth a huge amount. Yeah, yeah, 100%. And like the market always normalizes these things. By the way, I mean, this is all ancient history now, but the exact same thing happened in the internet I remember once there was like basically one guy that ever wrote a BGP stack Which is like it's a way that kind of routers talk on the internet and he was like the one guy that can make it work And so he just basically got this crazy out at the time crazy offers and he'd jump between all the router companies and do that And there were a few teams that could do this and so we've always seen this episodically in the industry We're just kind of seeing the new version and the listen the businesses are working and they're doing great we're kind of seeing it on steroids.

25:28Yeah, I mean, if you're a fang -size company, what's it worth to have like the one or two or three people who really know how to do something huge before? Yeah, I also feel like tech always figures out a way around kind of regulations and markets, like the late 90s, it was like, you know, you could IPO a company for very little, with not a lot of market traction, remember like the whole spec craze? And then now we've got these weird aqua -herat things. I think the reality is is in hot markets, people know that there's a lot of value to be had. Nobody knows exactly where it is. And so there's all sorts of things.

26:14The markets try to do to get access to the talent or the companies or whatever it is. And we're kind of seeing our version of that now. Whether it's these like I will hire an individual one, and I'll do it. We're going to acquire. But again, I feel like this is all normal in this sense of, you know, we've seen it in different shades in the past. Yeah, I told it's like an evolutionary response. What are the markets right now in AI that you feel most confident are totally working? What are the ones where you feel like they're on the horizon and you know, should be working very soon? And then what are the ones that any that you maybe have low confidence will work period?

26:47Yeah. So the diffusion markets are all working. So any area where you bring the marginal cost of creating something a piece of content to zero is clearly working and creating an image. Creating music, you know, creating speech. And we don't think about these markets as much because we're also focused on like the frontier labs, but like it's cheaper to build these models because they're smaller. and then people need content. And actually, the economics are so simple. Like, I, you know, like whatever. Imagine you're an artist and you're like, okay, I'm gonna draw a picture of Martin, right? Like, how long will that take you?

27:25Oh, well. Whatever, three hours and it costs you 400 bucks, right? But if I have a model, do it as a hundred of a penny type thing, right? So you've got a four orders of manual to different economics. So that's why we've seen those types of companies, you know, I think like, you know, a lab in labs or whatever do very well. So like that's clearly working and this is content creation where the market cost of creation goes to zero. I actually think the whole kind of low -wingness companionship stuff is definitely working. It's just this very fragmented market. So I think the unit economics are fine.

27:57I'm not sure like from an investor standpoint how you think about it, but like it's a use case that will be solvent. Well, we'll do fine. Code seems to be working incredibly well. Yes. And you know, we, you know, you see this in, you know, in in in cursor and in the whole thing. The areas that I don't know, I mean, they're working, but I don't know how the economics actually pencil out are the enterprise use cases at this point that are kind of a bit more agenticky, automating. These are ones you're putting in the middle bucket. This is like what you're saying is like kind of working, but not 100 % sure yet.

28:32No, no, no, no, no, that's so the ones that are So the middle bucket was actually like the friend, the emotional, like the character that AI is, there's a long tail of companies that are basically emotional support at our friends and our entertainment. It's probably also a big component of the usage of the main models. Yeah, 100%. That's clearly working in the sense that people willing to pay for it, the engagements, grade, et cetera, from an investor standpoint, tends to be kind of long tailed and fragmented and kind of spread and stuff. And then that enterprise agentic work flow type stuff. That was the fourth one I mentioned.

29:07That's like chatbots. I mean, clearly it's working. And there's companies that are doing it, but it tends to be, you know, if you look at the companies like, you know, there's a lot of like bespoke work going on. Like it's just a different type of economic model than the contract creation one where it's like, totally. It's just a model. So like those ones were still trying to understand. Yeah, I mean, one way to think about that is like how confident than are you that highly skilled work will get replaced in, let's say, legal finance accounting tax, like those kinds of areas. So I think the way I do it is actually very simple.

29:42So if the use case is the model is creating content, and that content is whatever, it could be language, it could be image, that clearly works. So OK, if the model is automating something a human being would do. And we conflate these two things all the time. That's totally different. Right? So if I'm like, model, do this thing instead of me, that's not content creation. It's somehow has to like mimic exactly what I do. That area still needs a lot of work. It seems to me. And so they, they clearly can do some work of a human. But you know, you know, not as exact, and they need a lot of guidance.

30:22And I think that's an area where there's still holds a lot of promise. But like the economic case isn't as obvious, right? Like make me a picture versus like go browse the web for me. And so it's kind of that second one that automating what humans do, where we're still like, we've got lots of investments, we're very excited about it, we think there's a great future there, but like the economic cases that nearly is good is, make me a picture. Totally. Let's to double click on code for a second. Obviously you know a lot about it through cursor, you were technical CTO. Like, where do you think we are right now?

30:51You know, like I, I just posted one with Guillermo who, you know, who obviously knows a lot too. And it was great. It's like this is both the future and it's also at the moment, you know, it's not obvious that it is in today's You know incarnation. It's not necessarily producing quite as much value for engineers is even they themselves experience But clearly it's gonna get there. So yeah, I'm curious. So just you know AI in general has this problem Which is so dazzling people conflate. Oh, this is dazzling whether this is useful right? Yeah, that's for everything right? It's not just cold, right?

31:24It's like you're so impressive. These things were magic. And somehow that dopamine, you know, hit. Well, it's really funny. So I posted this on X and you know, like this study that was saying that people experienced their own programming as plus 20 % and like the the observed results are minus 20 or whatever. And what was funny was there were a bunch of reply threads where somebody was like, no, no, no, this is crazy. I've been using it and I'm so productive. And then the reply to that is like, that's what the study's saying. Yeah, yeah, yeah, yeah, yeah. Which I'm sure for some people there, but you know, there is this thing.

31:53There is literally, there is an endorphin head. These things are absolutely magic, but I don't think it makes it very hard to think clearly about the actual utility. Yeah. Now, so I think you can say a few things like, you know, like there's a lot of things that it does very well that programmers don't like to do that like is pretty routine, right? Documentation. It's great writing documentation, right? There's a lot of boilerplate stuff it knows. The thing that I use it for the most is, you know, writing code isn't just writing code. Writing code is like understanding the frameworks. It's knowing how to deploy it.

32:25It's knowing how to run the toolchain. And like, there's no first principle way of knowing that type of stuff. There's not like some like core computer science fundamentals on deploying to not let it find. Like, that doesn't exist. And so it has all of that knowledge, which is clearly very useful. The writing code itself, I think it's still clearly very early days. I mean, if you constrain how you use it, it can be very effective. And then if you don't, it may not be as effective. And so I think that like studies like this that are purely observational are hard to read into because, you know, it's just like this is like you can use it for anything.

33:05And because they're so magical, I do think people tend to use them for stuff that they may not be so good at. Yeah. Just because the experience feels good. And so my guess like any new technology will develop best practices. I feel very strongly we're going to get a 10 X in productivity. Yeah, but like it'll take us a while to get this. I remember an other epochs of development productivity. Like when the IDE came out, when it higher level link, when OOP came out, like we get so enamored with the toolset, like object -growing and programming. I'm going to do everything in it. And it turns out these were great advancements in computer science.

33:36And they helped the best practices and they helped architecture and engineering. but at the time they came out, they were just so cool. I just kind of took a lot of our focus. So I think we're seeing that. So it's like, this drop of productivity is not like, necessarily could requires you to drop productivity. I think it's just so cool. I'm gonna try it with everything. Yeah, it does seem like it's on the path to, like, you know, in the same way that, like people have been saying this about self -driving cars and it seems like it's gonna become true. It just seem like we'll get to a place where like, you can pretty clear.

34:01It's so clear. I mean, this is incredibly obvious. I mean, you can literally just go up into a subset of things that are obvious. I guess really get a writing test. It's really get a writing documentation. It's really good at like dealing with a bunch of like, you know, long tail framework stuff that you don't know. It's good at teaching you things. I mean, this clearly, these are obviously good at. I just think we get so enamored with it. Maybe we start using it for stuff that is not so good at or not so equipped to deal with it, etc. But yeah, this is very clearly a, it's going to change software.

34:29And, and, and sorry, I don't mean to ram along. I just have to say, I have been software for a long time since the late 90s. We disrupted everything, right? We disrupted the back office, we disrupted hotels, we disrupted everything. This is the first time I say that we're probably getting legitimately disrupted as a discipline. Like what it means to be a software engineer is changing pretty fundamentally. I think it's because of AI. So it's kind of fun to like actually be the disrupted for a change. That's awesome. Yeah. Another topic I wanted to get your thoughts on was like, Why is open source so important to you?

35:04And I saw you were really excited about opening as open source model. And I know this is like thematically and spiritually important. Why do you think that it is such a critical part of the way that this plays out? So I think open source is historically one of the best mechanisms that shows a healthy ecosystem. And what normally happens is somebody does something close source, it turns out to create a market and then somebody releases open source and it stops a monopoly from forming and enables everybody else. And then it kind of keeps the people that are close source to continue to be innovators and allows everybody else to come in.

35:46So it's just been very, very healthy. And the thing that really worried me last year, and the past, academia, VCs, startups were all very pro -opened source because they understood that it was a very important part of a healthy competitive ecosystem. And the thing that really, really worried me last year was the people that should be championing open source like VCs, like startup founders like academia, were decrying how dangerous it was in relationship to AI. The implications of this to me are huge, right? I mean, of course, the national security implications are pretty straightforward, which is like if somebody else does the open source of proliferates and that's not good for US interest, but for the industry it's terrible, right?

36:29I mean, this is kind of how you actually create monopolies if, you know, you're not allowed to create something that enables everybody else. And again, it was very dramatic when like Vinod Colossus like open source is bad and founders fund is like open source is bad and he had academic and open source is bad. So I was much more interested in like trying to reset the discourse than like any specific open source release. Yeah, and I guess a lot of that probably came down to like how how potentially dangerous was the technology Yeah, you know, and so like if you thought it was extremely dangerous for example Then there is a argument for like massive containment.

37:04I got or what was the seal man in your mind? You know at the moment when close source was like such a strongly argued like if you had to You know argue why using people were saying that like what would have been I really think it's the legacy of both strong Right like so, yeah those from both the book superintelligence in 2014. Terminators waking up, we got it. But this is before all of these things, right? That like, like, both terms book was a thought experiment. It was like this platonic ideal of AI, and then somehow that created this, you know, very interesting kind of intellectual journey on the perils of AI, but then, you know, like, GPT -2 lands, and these two things got totally conflated.

37:44You also got, there's a lot of incentive for people to be doomerious, I think. Like it gets a lot of clicks. It was like, yeah, but not a store. But it's the weird thing historically. Like let's take the internet as an example. So like I was there during the early days of the web. We had a lot of examples of like how it actually changed the dialogue when it came to risks. Right. We run a critical infrastructure on it. We had totally new types of attacks like the Morris worm, which had actually taken down computer systems. So we're at this space where like, okay, well, it makes you more vulnerable if you use it.

38:16And we have new attacks that you can attack with it, right? And yet, academics are like, this is great. The technologists for this is great. So you had this very even -handed debate. What was so weird about the AI was it wasn't even handed. Like I'm all four both sides of the debate. Like I'm not just, you know, I'm not just, you know, pro -innovation at all costs, but like that's not what was happening. And so, like, Maybe you're right, maybe it's like the doobers got more click, but I think it's something more than that. I just think that there was an existing intellectual legacy that came from Boastron that had some very influential people, right?

38:50Like Elon was piled by that. You know, Eric Schmidt, Moscow, it's like, and they had very legit concerns, but they were kind of already primed and like, you know, there was already kind of that So when it happened, I just think they, they were already ahead of the game. It took a while for the rest to catch up. And now I, now I listen, when I listen to the discussion, it just feels more even handed, which is that God, I don't have to like be on Twitter. Yeah. Talking shit about this. Like I did for so long. Yeah. I feel like the right voices are in the room for sure. And obviously this isn't like, you know, I don't think you have us saying that like there aren't real rest.

39:27No, totally, really, which is a totally lopsided debate. It's so crazy when VCs are talking against open source. I mean, to me, I mean, like no academics are talking in defense. But now we have a bunch of academics in defense. I mean, like I think the right people got mobilized, but it really took some rallying to get the right folks back into the conversation. Yeah, yeah, absolutely. I want to ask you just a couple more questions about sort of the structure of the firm and sort of your own work outside of the specifics of what you're investing in. One that I thought was a really interesting point from Mark, when I talked to him on the podcast, was basically around, you know, like, how do you sort of drive the right overall aggression of the firm?

40:08And he said something to the effect of like, when you're in a market moment like this, the right answer is to just encourage people to do more. And so many partnerships are trying to get other people to know, and like that is like the common function. And, you know, he was basically describing something where like, it was like, how do we get people to a yes? And like, how do we take advantage of this? I'm just curious how you feel and, you know, your experience, in the last couple of years in AI lab. So one of the reasons Mark is such a great leader is he has intuition on the temperament of people is almost perfect.

40:43And he will drive like the right behavior relative to that. And so like if he thinks people are being too conservative, of course he'll drive them to be more aggressive. There's definitely more aggressive. I mean, the reality about AI is there's been a lot money that's already been lost in absolutely record time. Right? Right? So just because like the upsides aren't great, doesn't mean like the down like there's been a lot of money. You know, and so, you know, as a firm, we tend to be fairly disciplined and do a lot of market analysis. And you think you're coming. And so like, you know, he's very good about like pushing the team.

41:19And I think he's absolutely right to do that because, you know, there is already a foundation of discipline that like is not going to be eroded or compromised by doing that. On the other hand, there are individuals who are like, don't shoot from the head like everything. And then he actually tempers his messaging a lot with those. And so I would say again, I mean, you know, I don't know what was in his head or do you saying that. But I just think if I was going to add a little bit of nuances, my observation of Mark is he's very good at pushing when people need to push, but he understands the situations when and that's probably not the most appropriate thing.

41:56And so I think this is a core issue of leadership, which is you need to provide kind of the right kind of macro,

42:05macro shift in order to get the right one without being too overbearing. Yeah, I think some of that conversation was in relation to like fun sizes and what's possible. And I think some of it was articulating just like, these companies could be enormous. This opportunity is like, you know, quite oversized relatively. Yeah, but I mean, but again, I just think it's important to know which is a perfect thing. If you take him at his word, it would be an infinite fun deployed infinitely. So like this is a, it's calibrated. This is an intellectual landmark which is set and is the 100 % the right one to do.

42:43But it's important. He's so good at it. Yeah, it is in relationship to the mindset of the people that he's talking to and he knows that and what flagposts you have to put out to get 100 % 100 % and then I have seen him very subtly depending on who is the audience. Yeah, yeah, yeah, you know, you know, you can like move that flag post to different places. Where do you normally experience yourself there? Are you do you feel you're often needing to get pulled into more aggression or into more conservatism? I think I'm a seven out of ten for aggressiveness. I would say my team is five or six out of 10.

43:18There's people on my team that are 10 out of 10. There's people on my team that are three out of 10. So I think like if you didn't know I'm gonna just reach that, say we're six and a half to seven. And so for us, you know, he, I mean, he, he pushes pretty hard, but I know he knows that he's getting like, you know, like a step four that's opposed to like. Yeah, and I think that's like one of that mean being a leader of people is always in relation to, you know, pulling, you know, minds and assertions. So that's always how it goes, which is, you know, it's a rare thing. It's rare to be able to do multiple versions of that at the same time with last time.

43:48Exactly. It's just, it's just, it's hard to appreciate from the outside because you have to actually see the conversations. Yeah. It's something that he's just phenomenal at. And I'm, and they're very kind of nuanced different takes where he kind of knows where people are, you know, kind of nudge him in the right direction. To the extent that we are in somewhat of like a gold rushy in the good sense moment, you know, overall. Yeah. Do you think, um, does it change your perspective? active at all about what you need to see to want to make an investment. So like maybe a specific version of this question is in a somewhat stabilized time, I think most, you know, most people would agree that you should only back extreme special founders.

44:27Yeah. Is there ever a version in these kind of moments when a good founder and, you know, a great market with like exceptional traction or some configuration like that where you say, actually, you know, what that works here and that can produce something really big. So, do you know how we think about investments? Because I think it answers this question. So, it's very simple. So, the way that we think about investments, and the reason is the only way you can scale because you can actually distribute this kind of algorithm to a team is the only sin is picking the raw company in a certain space because it's a conflict.

44:59Because that conflict, because you're conflicted out of the winner. Like, investing in a space that doesn't work is fine. I mean, like, there's no way you can actually predict whether a space is going to work or not. I mean, that's like weather prediction, right? But in a given space, if you know all of the companies you do the work, you can most likely, at least tilt it, like you can do the work to determine if you think one is better than the rest. Like something you can actually put. So the way that we view the world is, first you have to identify legit spaces. We think that founders are smarter than VCs.

45:32So I don't care VCs, they get interesting if there's five founders in a space. And they're good founders. and they're betting their families, their fortunes, you know, their time. So it's probably a real space. And then we do the work to understand the space and all the teams that we make a pick within that. I mean, that's really how we think about it. That's true within AI is anything else, right? The thing that's harder is, and I've evolved so much as an investor, I used to think like, oh, we're getting good deals, like price matters, outcome matters, and more and more especially with AI, that's what you have to throw away.

46:08The market is the market. And that's what matters the most, your saying? It doesn't matter at all. I'm saying, so, so, so his, so we don't know what the tam is because it's growing so fast. Like nobody knows valuations. So I think this is, it's the contrary to like common belief. I think in these times where you don't know the tam and things are moving quickly, you definitely want to pick the best team. You definitely want to pick the best team. I don't think you should overthink the space. But like asking questions about like, tam or valuation or value makes a lot less sense, because that's actually what's uncertain.

46:44You just need to be in the best one. You have to be in the best ones. And the market will just be in every corner. And every corner. The market is the market. And listen, either you believe that the stuff is very quickly in market. It's our efficient or not. And listen, having been through the dot com boom and bust, the reality is the market was actually pretty smart And if you put the bets in the right companies, they would have been generation one up the same thing with cloud But same thing with mobile and so the goal is finding the right companies and like it really there's one change I would say is you need to throw away too many thoughts about market sizes in town is the rational sort of behavior then to You know, obviously you want to invest as early as possible but you want to invest as early as possible when you know that you've got the right one.

47:27So does that ever push you to being like a rather weight -around? Oh yeah, all the time, right? Yeah, all the time. So I mean, this is all very rough heuristics, right? Like, you know, we get it wrong all of the time. You know, I've made so many mistakes. And so you're just trying to beat the market, right? You're just trying to have. But yeah, very often, like our discussions are like, do we actually know who the winner is? Yeah. like, you know, and often we wait for that reason. Basically the earliest that you feel confident you can pick it. Yeah, and so like listen, when we do see stormy, it's like the person that did the thing in the big company and now I was doing the thing and it's the world expert.

48:05And the thing is very tactical thing and like somebody else is just going to wake up and decide to do it and who's a good founder, right? Is it like the pool of people that do it are like five, like this one's the best of the five. Like that's kind of like for early investments the way that we think about it. And then most everything else is actually the result of a lot of market work. And then this is our best guest like this is kind of the best approach best team best market and then in these types of way And this is where market this is this comes from mark and mark is totally right like if you think you can ask about the market Like I think you know, it's very tough and so just being the best deals As a final topic.

48:38I want to just hear your perspective on the board relationship and board roles in general Yeah, and maybe it's like a prompt on this I feel like You've done something which I find very impressive, which is it seems like you're able to manage successfully many more board seats than a lot of people. And I often hear common wisdom that it could be 10 or 12 or 15, but it seems like you found a way to do more than that and be very effective with those founders. And so I just want to hear your perspective on what's that relationship, what do you think is the limiting factor here if any, how does it all play out?

49:15Yeah, I mean, I do think that like a lot of the common wisdom on boards came from like that earlier year in DC when you know, it's like people would literally choose VC for like a life choice or whatever. Right. And I think if you come from like pretty serious operating like you do. And like I do like just got a lot of hours of the day to throw at it. BC is also involved that we've got better platforms and I actually really help with these things. And so between like actually the hours in the day, I mean, how much does that take? Like board worth take. I mean, you're a board member. I actually find that like, can I just take a step back?

49:50Just because let's talk about what boards mean. Like I always ask when I invest in a founder, like what is a board for? What do you think they say normally? I'm actually curious what the most common answer is, but I couldn't ask them. What do you think it would be? I would think they would probably say it's something like governance and approval or something like that. No, that's what they should say. That's what you and I would say. They say to provide guidance, to help with hiring. I'm like, no, that's not a board, right? And so a lot of, there's a lot of this belief that a board member is somehow helping with company building.

50:25And almost like, it's actually hard for a board member to be like the best friend of a found her for those types of things because like, you know, you're, you're a fiduciary and you do governance. So like the actual boardwork itself is just not a lot from the fiduciary governance standpoint. So often implicit in this question is, is like non -board stuff. Like how can you be helpful to a company and everything else? Because that can take a lot of time. And I do think this is where like, you know, having a big platform to lean on, being totally available helps a lot. But like, it's not the boardwork.

50:56It is the other stuff. And so I would say to anybody listening who is a VC, you can take as many borders as you want. The actual boardwork is not. The question is, can you still be available to founders and add value, whether you're on the board or not? A lot of the companies I spend the most time with are not even on the board. Yeah, that's what's funny to me is, you know, people talk about this with me. There's some board seats I have that are, you know, the founders asking quite a lot less than, you know, some seats. No, there's no board seat. And there's sort of decoupled. Yeah, exactly. So I feel like we need to be very clear like a board is to keep everybody out of jail and to like do the right thing for the shareholders Yeah, and like like the actual work requires that a relatively little yeah That's actually not the hard thing But we use it as a proxy for the hard thing the hard thing is like how do you add a lot of value?

51:37And for that I do think like this and you have to like in my case Listen, I have the pleasure of working with the best team. I've ever worked with my life They're fucking amazing. We've got like a phenomenal platform Like it's fucking amazing and I get to leverage all of that and it's not a board thing is like the help the company type thing. And I just think this is like the new era of VC. Like you help these companies with more than just one person who shows up, you know, off the golf field. Like, I don't know, every Thursday. Yeah. Well, that's great place to leave it. Martin, thanks for making time for this.

52:05I really enjoyed it. It's a pleasure. That was great. Someulate when you set it up. Lets try thisfig.