Lenny's Podcast: Product | Career | Growth
01:08:45 · Marc Andreessen
[01:08:45] S0: My experience with really big technological transformations, and of course, I kind of lived this directly with the internet and I saw this happen, is the really big technological transformations, they take a long time to play out and there's all of these structural implications that just kind of cascade out over time. There's this rush to judgment upfront where people say, "Oh, it's therefore obvious that X, Y, Z. It's therefore obvious that this kind of company is going to be the company of the future, not that kind. It's obvious that this incumbent's going to be able to adapt and this other one isn't. It's obvious that there's economic opportunity in this kind of startup and not in these others. It's obvious that the moats are going to be in this area of the technology, but not in this other area." What everybody does is they kind of state those things with just an enormous amount of self-assurance where they really sound like they have all the answers. And then what happens is these ideas kind of saturate the media because the media naturally prizes definitive answers over open questions because it... it's like when CNBC is booking guests, they want a guest who's going to come on and say, "Yes, this is the way, it's going to be X." Not like, "You know, I think that's a really good question and let's debate it from eight different angles." What I found is if you look back on those predictions a few years later, and you can do this by the way, if you pull up coverage of the internet from 1993 through 1997, or for that matter even through 2005 or 2010, and you look at the kinds of confidence statements people were making in the first 10 or 15 years, I would say almost all of them were wrong, generally quite badly wrong. And so I think the process, I think there's going to be a massive amount of technological change. It's going to be like, I don't know, five or six layers of structural change that will play out over time. And again, we've talked about a lot of this, but the implications on what are the definition of products, what are the definitions of companies, what are the definitions of jobs, what are the definitions of industries. How does this play out at the national level? How does this play out at the global level? By the way, how does this intersect with politics? How does this intersect with unions?
01:10:52 · Marc Andreessen
How does this intersect with war? What's China going to do? And so there are just a tremendous number of unknowns, a very, very large number of unknowns, and I think it's just like really, really dangerous to prejudge these things. I'll just run this as a thought experiment and you can see what you think on this, but it's like, are AI models themselves defensible. Is there a moat on AI models? And on the one hand, you'd be like, "Wow, it certainly seems like there is or should be," because if something takes billions of dollars to build and you need this incredible critical mass of computing data and there's only a certain number of engineers in the world that know how to do this and they are getting paid like MBA stars. And then these companies have to deal with all these crazy political issues and press issues and reputational stuff and regulatory and legal. All of that translates to, okay, probably at the end of this, there's going to be two or three companies that are going to end up with like 100%, I don't know, whatever, 50/50 or 30/30/30 or 90/10 and one, or whatever it is, market share and then they're going to have whatever profitability they have and it's going to be kind of a classic oligopoly, or maybe one company's going to win definitively and it'll be a monopoly. And by the way, those outcomes have happened in software many times before. And so maybe that will be the outcome. The other side of it is if you had told me three years ago that in the Christmas of ChatGPT that within basically a year to year and a half there would be five other American companies that would have basically exactly capable products, and then there would be another five companies out of China that would have exactly capable products, and then there would additionally be open source that was basically the same, I would have been like, "Wow, the thing that seemed like it was black magic all of a sudden has become like commoditized really fast," which by the way, is exactly what happened. Within a year of GPT3 coming out, there were there open source GPT3s running on a fraction of the hardware that were available for free. And then there were five. Now you've got, fully in the game, you've got Google and you've got Anthropic and you've got xAI and you've got Meta and you've got all these other companies that are...
01:12:59 · Marc Andreessen
and then DeepSeek and Kimi and all these other Chinese companies. And so even at the level of LLMs or AI models, you can squint and make that argument either way. By the way, same thing at the level of apps. It's like one school of thought is apps are not a thing because the model's just going to do everything, but another way of looking at it is no, actually adapting the model is kind of the engine into a domain involving human beings where you need to actually have it fit for purpose to be able to function in the medical industry or the legal industry or whatever or coding. No, you actually need the application level's actually going to matter enormously, and maybe the LLMs commoditize and maybe the value goes to the apps. And again, you can kind of squint either way on that one, and I know very smart people who are on both sides of that argument. And so my honest answer on this is I think we're in a process of discovery over time. The way I think about this kind of structurally is it's a complex adaptive system. The technology itself provides one of the inputs. The legal and regulatory process is another input. Actual individual choices made by entrepreneurs matter a lot. The economics matter a lot. Availability of investor capital varies over time, that matters a lot. This is a complex system, and so we actually don't know the outcomes on this yet. We need to be open to surprises at the structural level of what happens. And of course, as a VC, this is very exciting because it means we're doing this now. We should make bets along every one of these strategies and see how this plays out. I would just say, there may be, I don't know, there may be like one particularly brilliant, I don't know, hedge fund manager or something who has this all figured out, but I guess I would say if they exist I haven't met them yet.
01:14:40 · Lenny Rachitsky, Marc Andreessen
[01:14:40] S1: So what I'm hearing here is don't over-obsess with moats at this point because we have no idea what'll end up being, and as much as it may feel like, okay, there's no way OpenAI will lose this lead, clearly we're seeing a lot of competition. GPT wrapper point is really great. It was such a derogatory term, I don't know, a year ago, just like, "You're just a GPT wrapper." Now it's like the companies that are the biggest companies, the fastest growing companies in the world. [01:15:01] S0: Yeah. Well, it's like a little bit like, I don't know, I mean, even just like with... this has been the holiday, three years ago was the holiday of ChatGPT. This last month or whatever has been the holiday of Claude, particularly Claude Code for coding. But it's pretty amazing because it's like, okay, there was Claude which is obviously a great accomplishment, but then there's Claude Code which is an app. It's a Claude wrapper. It's agent harness. And then they did this amazing thing where they came out with, was it Cowork? [01:15:01] S1: Cowork. [01:15:29] S0: Cowork. And remember what they said of Cowork, which is Claude Code worked Cowork in a week. [01:15:36] S1: Yeah, a week and a half, yep, 100%.
01:15:39 · Marc Andreessen
[01:15:39] S0: Well, and there's two ways looking at that which is like, "Wow, that's really..." I mean, obviously that's really impressive that Claude Code was able to build Cowork in a week and a half. That's great. That's amazing. The other way to look at it is Cowork was developed in a week and a half. How much complexity could there be? How much of a barrier to entry can there be in something that was developed in a week and a half? And then again, it's this push and this pull thing where it's like, wow, it's incredibly functional, incredibly valuable, and people all over the world and every day now are like, "Wow, I can't believe what I can do with this. It's like the most magical product ever." But at the same time, it took a week and a half. And so every other model company, I'm sure, you'd have to expect, is sitting there being like, "Okay, obviously we need to build an Asian artist and then obviously we need to build a Cowork thing for regular people." I'm not even saying I know anything, but just obviously they're all going to do that. And so how defensible is that? In six months, and we've seen this happen before, is Claude Code going to get lapped the same way that GitHub Copilot got lapped? The history in the last three years has been everything that looks like it's like the fundamental breakthrough gets basically replicated and lapped very quickly. Many of the smartest people I know in the field, when I really talk to them, kind of get a couple drinks into them, they're like, "Yeah." One theory is there really aren't any secrets among the big labs. The big labs kind of all have the same information and they kind of have all the same knowledge and they lap each other on a regular basis, but there's not a lot of proprietary anything at this point. And then again, evidence of that is DeepSeek came out of left field and basically was like a re-implementation of a lot of the ideas under American big labs and had some original ideas of its own. But, wow, it wasn't that hard for some basically a hedge fund in China to do it, and so how much defensibility is there. But on the other side of it, you've got, wow, these big labs are now paying individual engineers like they're rock stars and they're incredibly bright and creative people.
01:17:33 · Marc Andreessen, Lenny Rachitsky
Maybe there's a dozen nascent ideas at any one of these labs that is actually going to be a huge breakthrough that's going to be hard to replicate. And so again, it's just like, I think we just need... I don't know, my view is I need to put a big discount on my forecasting ability on this one. For me, it's much less interesting to try to say, "Okay, as a consequence, industry structure in five years is going to be X, the big winner and the category is going to be company Y, the big product killer app is going to be Z." It's like, I don't think I can predict that. I think a much better use of my time is being very flexible and adaptable at a time like this. [01:18:07] S1: So with all this in mind, do you feel like there's something you're paying attention to more to help you decide, okay, this is where we want to place our bet, or is the answer essentially the strategy you guys have, which is place a lot of bets? You guys raised the largest fund in history. Is that the way you win in this world?