You have a situation right now where the consumer, the two of us, we want to own our memory in context. We don't want it to only live in OpenAI or Anthropic or anywhere else.
This is the core thing that USV is obsessed with.
The idea that we want to own our own data. It's actually the core root of the thesis on blockchain. I would argue that at least up to now that has been a philosophically interesting idea, but the utility of it has been limited.
By the way, I 100% agree with you because the utility has not been accessible. I mean, that's why I think there is a bit of a browser war coming because it is the ultimate. It's the only other operating system. We're either in like Windows or Android or iOS or a browser. Yep.
And a browser can have its own rules and profile and memory and everything else about us.
Well, thanks for doing this.
So fun. I love coming to USV. We are a repeat asker of you to come to USV. So we appreciate it every time. So here's what I wanted to start with. Where are we in the consumer AI cycle? Is it we're waiting for it to hit? Is it that the chatbot is the atomic unit and we're there? Is it that consumer has gotten absorbed into the model and this is what it's going to be? Let's start with what's your view on where we are in this?
Well, I definitely think we're early. I'm asking the same question you are and I'm sure many others. I do happen to believe that whenever there's a platform shift, there is this sort of skeuomorphic era where the new technology tries to emulate the old ways to help us bridge the gap or cross that chasm, if you will. Back in the early days of iPhone, we all remember the stitching in the address book and the things that were in some ways patronizing, but were helpful in making us understand that this is almost like that analog thing I once did. I think the analogy is the chat. I happen to believe that the chat is a very reminiscent natural language way of helping us identify with this new technology. But when you start to see agents interact with each other and you realize that they don't actually talk to each other with long formal pauses, they actually just go straight to it. I think we get a little bit of a glimpse of what our relationship with AI may be in the future, which is that it has all the context that we have at the moment, where we are, who we're with, what we're looking at. It has all the history and the memory of everything we've ever done. And it's in some ways beating us to our own thoughts as opposed to us being like, 'What do you think of this?' So, I think we're in this skeuomorphic very early stage era.
Yeah, I like that because your thesis there is we're familiar with chat. We know how to chat. And so, in some ways, it's an entry drug because it gets us familiar with something new, but it will lead to something else. I read Sarah Tavel, our mutual good friend, posted something recently which I thought was interesting, which is her thesis is you're not going to see the jump until you have a new interface, right? Like either a neural implant or a new device. And she went as far to say like a full new device, right? Not like a new atomic unit, but a new physical thing or maybe it's in the brain. Like do you think we need to go that far or it's more gradual?
Well, I think to quote another mutual friend Dave Morin with the devils in the defaults, you know, I think is another way of backing into the same conclusion which is that the devices that we have today, they can't really change the default structure. I mean, Android, Apple, it's all apps, right? And there's just a way we think about apps where it just can't be transcended without an entirely new operating system. And so I do wonder, imagine this magical device that knows all the history you and I have, everything we've ever talked about together, where we're sitting right now, who we're with, what time it is, what else is going on in the world, and then with all that intelligence or super intelligence starts to suggest the things we want to do or want to know now. That is like a totally different experience than what we have today. And so, I think that a new device is a forcing function for us to have a new operating system and a new set of defaults that we're accustomed to. So, maybe that's what it takes. Now, of course, if I were Apple, I would be thinking about not even a new variant of iOS, but like an entirely new operation that goes back to 1.0. Totally.
But it's hard to do that when you're big. It's hard to do it when you're a big... I mean that brings up there's so many threads I want to pull but one of those is I think we're in the kind of craziest moment of incumbent versus new entrant, right? Where in some ways incumbents are more advantaged than they've literally ever been. You look at Google, they could not have more data to do what we are asking them to do. They have all of it, and yet a lot of what we're asking for requires the biggest players to burn the boats. And that's the hardest thing to ask a big player to do, especially the business model boats.
Totally. It's the business model boats. Our partner Brad, he has always said something that is ingrained in my brain whenever we meet a company, which is we can wonder about competition, but the hardest thing to do is burn the boat on the way you make money.
Yeah. And that's happening, right? So, how... what do you think like is this Apple and Google's game or is this the best time for new entry? Not only Apple and Google but Adobe and Figma, right? This is true across so many different categories.
Well, I think there are advantages and disadvantages on both sides as we both know. And the question is like how do they stack up? I mean, one argument that Google might make is that they're going to be the only and first full stack AI company, right? Where they own the app layer down to the chips, down to the cloud, and they can unlock magic both at a product capability level, but also a cost level, right? As opposed to like all these third parties in the mix that need to get paid somehow. At the same time, to your point, when 90 something percent of your money comes from the thing that you're trying to disrupt, it is super damn hard, right? And I imagine them trying to trickle it as opposed to their competitors who are full on like we're just going to do it the new way. And I think that's going to be... though they have... I don't know if I'm Google, I think Google in some ways is the most interesting use case of this right now. There are a bunch of others. I'd be curious of your thoughts on Adobe and Figma too. But Google, the fact that they have Waymo brewing, the fact that they have AWS like all of these things. Shouldn't that make it easier? Like we all think about ads, but actually it's like they got a lot of other bets in place, right? Like shouldn't that make it easier to make the big jump because they have a lot of things to hedge them and actually more than you would think, right? And I do believe that having that portfolio where some things counter other things. I remember some investment banks they thrive in investment banking when equities are doing poorly and vice versa. And similarly, you look at a company like Adobe has a digital marketing business, a creative business, you have enterprise business, a consumer business. And I think that these sorts of things can cancel each other out. I do sometimes think about consumer as having these major operating systems that we're somewhat prisoners of, to some extent happy prisoners in some cases of the OSes that we use on our devices. On the enterprise side though, what are the OSes is an interesting question. Adobe has an operating system of creativity that is quite embedded. You know, I'm on the board of Atlassian. Atlassian has a system of work. And it's funny, people saw the recent news about the browser company and Jira.
To me, the disruptor of Jira is a new interface where you don't even have to necessarily interact with Jira, but Jira may actually still be the ultimate source of data. And the permissioning graph and the graph of who works with whom on what projects that may be very essential. So I think that the incumbents have to not be a prisoner to the interfaces they've shipped in the past. They have to now go down a level to the data and say how should customers interact with this data.
Totally. If you know what I love about what you're saying is it feels like there's two pieces that are true and it actually kind of all ties back to context, right? Which is you have to own the data and you have to own the interface and the question is how married are those two? Like one question is if you're OpenAI, are you just dying to build a browser, right? Because you're hitting a limitation on interface and you just have to build the browser or do you just go deeper into the data stack which is kind of what Anthropic is doing, right? And say we're going to own that. How quickly do you need to do it all?
I think it's a great question. And one of the things that I've been trying to figure out and I'm sure you're thinking about this a lot as well, and I know everything kind of goes to context and memory, but so pardon me jumping around here, but you have a situation right now where the consumer, the two of us, we want to own our memory in context. We don't want it to only live in OpenAI or Anthropic or anywhere else. We want to have it be portable. I mean, just sort of like so far to this day, typically our status in places is actually not portable. If you're a Delta Medallion, diamond medallion, or whatever it is, like unfortunately that status only works at Delta and JetBlue is like, get in the back, right? So, I think that it's an interesting world where consumers are going to want to own their memory and the status that they have with brands and stuff and not have that live within the LLM. They're going to want to have their own.
Can we talk about that? Because in some ways, this is the core thing that USV is obsessed with. The idea that we want to own our own data or at least the idea that we should want to own our own data and we have this long history of being obsessed with this idea. It's actually the core root of the thesis on blockchain, right? Like this idea of data gets beholdened by the big platforms and how do you unlock it and unlocking it provides power to the individual but also massively unleashes innovation and prevents monopolies and all these kind of things. So we really love this idea. I would argue that at least up to now that has been a philosophically interesting idea but the utility of it has been limited.
By the way, I 100% agree with you because the utility has not been accessible.
So here now we're in a world though where conceivably you could be known by every digital brand in the world. I'm a vegetarian. When I walk into a restaurant, I don't want to hear the specials on beef. Like everyone has their longtail preferences. And by the way, every brand wants to know you and you want brands to know you, but at least in your control and for the right reasons and with privacy settings, etc. So, I just think it's interesting when we talk about memory. Consumers want to be known by everyone. And therefore, that's another way of saying, by the way, with utility that we want our data to be portable, aka belong to us.
And then the brands also don't want to be limited to one LLM to know their customer obviously. Like they want to own the memory with you as well.
Yep. And then you have the LLM in the middle that's going to want to own this for both.
I would argue they actually want to own that more than they want to own anything else.
The issue with context is it's the most important thing and it's also desired by everyone, right? So how do you create a world where the individual owns their data and it is portable when the platforms where it finds its utility want it so badly?
Yeah. And also the end provider of something also wants it. So there's solutions like memory as a service. Is there some layer and do you think that works?
We'll see. I mean I think that it's in the interest of the consumer but the question is how do we get the utility to invest in doing it in the first place? It has to be seamless and I don't think anyone's really figured that out yet. There's some companies, Mem, a few others that are tinkering in this space. On the company side, everyone's building their own agents and their own customer data profile systems. But those treat us mostly as strangers and their dream is to be able to kind of get quickly populated with who you are and give you this amazing experience. And I do believe that we're going to look back at today years from now and say every website treated us as a random person.
Like Nike asked me what my gender was. I had to sign in to every individual website with individual credentials. It's bizarre. The web should be every place you go should be like, 'Welcome Rebecca.' Like, 'This is what we have for you.' And you should know how they know that. And you shouldn't have to only be using one LLM underneath to have that.
But that's... I always say that the biggest reason we know that the internet was built bottom up and not top down is because we don't have a universal sign on and knowledge, right? Like if you architected the web, you would never architect it without it. But because it was built in this kind of beautiful but terrible bottom up way, you wind up with this hodge podge architecture. So, here's one idea, right, which is that we have an agent that works on our behalf that is almost like this conduit to all the other agents and AI enabled capabilities that we encounter every day. And that agent that works for us negotiates with all these others. So this agent knows and is the sort of source of truth of who I am and what my preferences are. And whenever it comes in touch with an agent that's powered by Anthropic or OpenAI or anyone else for that matter, it has this like sort of back and forth. And maybe that's why I think there is a bit of a browser war coming because it is the ultimate. It's the only other operating system. We're either in like Windows or Android or iOS or a browser. Yep.
And a browser can have its own rules and profile and memory and everything else about us. Who has advantage to build that agent for me?
Yeah. Well, I think that it's a different question about our consumer life and our work life, right?
You think those are separate?
I do think they're separate. And the reason they have to be separate is because the work agent will have certain security and other things that your enterprise will request, right? Whereas the personal agent is yours. On the personal side, if these big OSes got their act together, we'd have it now and we don't.
Yep. I, you know, that's more to your department than I like. Which startup is going to solve? Please someone solve this problem. On the enterprise side, I actually do believe that whether you look at the companies in these verticals, they have the most permission, they have the data, and also I don't... one thing that people don't... when we say data, I also mean the work graph of who works with whom, like all that context. When you search, people connections I think is a huge signal and huge limitation into what currently agents have good access to.
That's right. And companies like Glean, Rovo from Glass, and others are trying to figure out how to give you results when you search for some generic file. It should figure out you and I work together often. You've sent it to me. I sent it to you. It should surface that first over other ones that have the same name. At scale, that is an incredibly important thing to have in an enterprise and you only have that with all the data around who's worked with whom and similarly privacy settings. I think the nightmare a lot of companies are having right now is they're turning on these new AI search things. Oh yeah, every public file can be searchable and then you realize that two years ago you made this random Google link and made it public to your organization but you only shared it with me. And then suddenly it's like accessible by everyone and it's like whoa, that's wrong. So that stuff at scale is going to be a huge thing to figure out and it's actually a real incredible advantage that incumbents have.
