You were just at OpenAI Dev Day, so you can give us the scoop on that. Did you get to watch the survey results? Did you have any take? Anyone you really disagreed with the crowd?
Yeah, the restaurant thing, man. It's going to be OpenAI, Anthropic, and Resy. Those are our future juggernauts.
Yeah, 100%. That's big. That's Mag Seven, Mag Three. Do you think there is — I mean, just to take that one seriously — like, where, how do marketplaces shape out in a world of agents? Like, is it good to be a marketplace business where agents want to interact, or have you developed a thesis there?
I think the supply aggregators, it's not amazing for them. I think it's really good for restaurants, especially long-tail restaurateurs. Because, you know, being a restaurant, obviously you have to have like great food and a great experience, but you also have to be great at marketing, and then you have to do deals with these supply aggregators. So that's like one thing. So if you're trying to get discovery, I think hopefully these personal assistants will drive more discovery for the tail. I think the other thing is that, you know, if you're Carbone, you're kind of doing this linear optimization function where you want the highest AOV, the highest LTV, you want every table filled. And I think the kind of Resy thing is a poor proxy for that. So I actually think it's good for restaurants, but bad for the intermediaries.
Because they can just sort of send their business wherever they want.
Go direct. Personal agents can go direct, you know.
And they could, I guess, vibe code their own, you know, delivery. Is that possible? Do you think there was a meme for a while, like vibe code your own DoorDash type idea, or like...
I mean, that's insane. A network effect business with a physical aspect to it.
It's more just they can easily move between them, and the marketplaces have no leverage.
I think the big idea is zero marginal cost digital labor. So the personal agent can go do all — like, the value of the aggregator is it reduces the kind of cognitive overhead and the labor for the human, but the agent doesn't have that constraint. So it can kind of go direct to every restaurant and learn about it, and, you know, sort of do its own research, and then come back with a couple of recommendations. The other actually interesting implication is we've always had supply aggregators, but not really demand aggregators. So you can imagine like there's supply-constrained things like tables at Carbone, then there's demand-constrained things like people that want to fly to Cabo on Friday. I don't know, right? But you could imagine a world in which you get a bunch of people together and you say, 'Look, these people might fly to Cabo if the prices were $300 a ticket.' And then you go do this group buy with the airline. So it sort of reverses the polarity, if that makes sense, and it lets the demand be aggregated and exert leverage over the supply that's never been possible before.
What were your observations from OpenAI's Dev Day today?
I mean, I think the two words I'd say — one, I thought it was very coherent. Actually, it was like very much focused on enterprise and developers, and all the products fit together in a cool way. And the other is just the vibes were great, you know, it felt very high energy. Everybody from OpenAI that I talked to felt like a lot of momentum. I think two years ago there was just, you know, there were so many bets spread across different modalities of models and sort of expressions of those models. And this year it felt like a lot of that stuff actually came together. So it was really, really good.
Anything stand out to you? I mean, we've been talking about dots.
Okay, so I think there's like the futuristic and the mundane. I think the mundane is actually most startup-relevant. So there's like three, you know, what I'll call mundane launches. So one is sign in with your ChatGPT. So this is your user as a developer, your user signs in with their ChatGPT and uses their own token budget. Which I think is a big deal, right? Because now as a developer, you don't have to worry about the kind of economics of subsidizing tokens. So that was one. One is marketplace, where if you're a SaaS company and GCP and AWS do this, you can go and sell using OpenAI's budget at the enterprise without going through procurement. So anybody who's actually, let's say the enterprise has bought OpenAI, they can use a subset of that budget to buy some SaaS tool that builds on OpenAI. That was a second. And then they're going this thing called plugins, which essentially allows you to have your app live within the ChatGPT app, which could be the new — let's see, we've been saying this like App Store for AI, but like maybe this will be it.
You know, you're obviously an investor at Andreessen Horowitz, sort of an enormous firm. You guys are on a good run. We just had the World Labs exit. And obviously Cursor is with the largest M&A sale of all time. What I mean, for founders just like looking to navigate Andreessen Horowitz, can you talk about sort of the companies you're focused on and just how you would guide them to think about the firm? It seems like what don't you want to invest in, yet you obviously don't invest in every startup.
I mean, I think there's kind of two core theses of the firm from 15 years ago. One is that, you know, founders make the best long-term leaders of their companies, which now seems consensus, but was non-consensus 15 years ago. And the other is just that these markets are going to be way, way bigger. So I think as the firm has extended, it's taken like the original Fund One construct and said, 'Hey, you know, four to six GPs with all the supporting networks, let's replicate that for all these areas.' And let's think of the founder as a customer. Let's build a product for them. You know, like if you're working with our infra team, they've got access to GPUs and, you know, networks around that, and very different for applications. I focus on applications. I sort of think of apps as like, you know, intelligence like cloud...
He's most bullish on applications.
There you go. There you go. I'm your man. So like, you know, intelligence is a primitive and it needs to get productized, packaged, priced, and sold into industry, and that's what applications does, consumer and enterprise. It's been all B2B for two years, but post sort of Instinct and some of the things that have been happening in the last month, I'm excited, tentatively excited for that.
The story type in Newcomer is consumer's back. You know, it's like always a random reason. It's like literally sometimes we're like investors have just decided it's been too long for making a push. Like, I don't know. So I get fatigued on the consumer's back, but what we've seen with Instinct, what are the other data points that you think there are? There's consumer opportunity.
Well, I think that sign-in with ChatGPT will be a big deal because I think one of the challenges for consumer is that people don't want to pay for things, or at least not to try things. And you know, you can go bankrupt very quickly with a free-to-try AI product. So if we have a way to solve that problem, then you just have a traditional consumer kind of product design distribution challenge. So I think that's one reason to be bullish. Also, I think just culturally, when things like Instinct happen, hopefully founders in this room are like, 'Whoa, I could do something like that too.' The Instinct and Muse companies are bringing an ecosystem with them as well. I don't know if people saw Muse connectors last week. Like, I think that's going to be a big deal, especially for, you know, vertical agents, fintech, financial services. There's like a lot of startup opportunities that that enables. So...
There obviously, there's a graveyard of startups that depended on Meta for distribution. You know, it's...
It's hard to — I mean, that is a challenge of like the AI world that we have. We have this obviously, everyone agrees, sort of like an amazing new technology...
...but what's like the distribution mechanism to go with it? Or what do you see a playbook emerging for how — you know, there's so many different types of companies — but how startups are getting distribution?
So there's sort of been two different models for the eras, right? Web 2.0 was you had to create distribution alongside product. There was no channel for it really. Maybe Google, but not really. iOS spoiled us all because you had the App Store. So you kind of had this channel. We haven't had a channel yet. Maybe it'll be Muse connectors. Maybe it'll be this like plug-in thing in ChatGPT. But no, I think we all have to build distribution alongside product. Like, with that said, the products we can make now are so crazy that you should be able to grow through word of mouth. Like, we tell this to, you know, our teams that we work with a lot, which is: you don't have a distribution problem. There are no distribution problems. There are only product problems. And if you're not seeing growth, then maybe the product's not ambitious enough.
What, you know, you spent three years, I think, at Credit Karma before Andreessen Horowitz. You know, this event, you know, it's like I think it was the Newcomer Banking Summit, Breaking the Bank. Like it sort of reflects some of the existential crises of like the fintech sector. It's like what's hot? Like, you know, crypto. I don't know. What advice would you give to sort of fintech founders in this moment where it feels like you guys, like the investors in some ways, have cooled on it and it's still the energy?
I mean, we have a lot of energy for it, but a lot of people are not building in it. So, I mean, I'm sure you've all read these like mind-numbing blog posts, many of which I've written, which are self-driving money. It's going to come. It's going to change the American consumer. Like, and none of that stuff has happened. So I think the mission was right, but the work is highly incomplete. And now, ironically, we have the technology to fulfill the mission, but nobody's working on it. So you're saying like Credit Karma as an agent is a way bigger business than Credit Karma. So please work on it and then come talk to me.
I mean, you know, I've already talked about on stage. It feels like some of the fintech opportunity is helping consumers spend less. Like, are there good businesses to be built there where it's about sort of auditing your spending and sort of making sense of...
I just don't think that works, dude. I think the American consumer does not want to spend less, right?
They love spending, dude.
100%. And it's very paternalistic. Like, my big lesson from consumer is that paternalism never works. And instead, we should be like, 'Look, let's enable the consumer to do the things they're already doing, but more effectively.' Sheel, you know this. We've solved some of these problems together. Like, 'Look, there's free money, which is, hey, what's all the money that you're actually spending on your loan and money that the government owes you that you haven't come and collected? Like, let's go get you that money.' Then let's maybe help you do more things to like generate more money. But I think saving money is the wrong framing.
Are you bullish on sort of like AI as your friend, and are you making investments in — this isn't sort of a pure fintech question, but I don't know. How much do you see the world in 10 years being one where, you know, I can be like, 'Oh, one of my best friends is an AI'? Or like, what is the consumer on...
