Fraser & Nabeel explore what it means to build great products in this new world of AI.
Two former founders, now VCs, have an off-the-cuff conversation with friends about the new AI products that are worth trying, emerging patterns, and how founders are navigating a world that’s changing every week.
Fraser is the former Head of Product at OpenAI, where he managed the teams that shipped ChatGPT and DALL-E, and is now an investor at Spark Capital. Nabeel is a former founder and CEO, now an investor at Spark, and has served on the boards of Discord, Postmates, Cruise, Descript, and Adept.
It's like your weekly dinner party on what's happening in artificial intelligence.
Do we need a leaderboard for AI products
===
[00:00:00] Nabeel Hyatt: it's like watching the sitcom get built as you're watching the sitcom.
[00:00:03] it's going to feel like goofy toys and entertainment and community probably will be but it will at least be a good test of where the systems are today
[00:00:11] Fraser: I vibe coded this piece.
[00:00:13] It's so silly, but look at this other thing.
[00:00:15] Nabeel Hyatt: I want it sorted by interestingness, Fraser.
[00:00:19]
[00:00:19] Nabeel Hyatt: Hey Fraser. So, uh, a conversation we're gonna have today. It's, it's almost like a, like an open call, right? I, I do think there's somebody out there that has an interesting idea that will riff off of the conversations we were having at dinner last night and this week, and help us.
[00:00:34] Just increase the amount of interestingness in the world of AI apps with a product that probably we should be able to vibe code our way into the future.
[00:00:41] Fraser: Let's do it.
[00:00:42] Speaker 2: Yeah. Let's start from the beginning. Fraser, you said you've been checking out some stuff this week, and then we'll, we'll dive right into it.
[00:00:48]
[00:00:48] Fraser: I went back and used all of the, like vibe coding apps using Claude 3 point 7 to check, , but I don't know, like just works better,
[00:00:56] Nabeel Hyatt: the reason that would matter is because it's like, Oh, this cloud 3. 7 it's so much better at coding. Sure. Everything gets 10 percent better, fine. Or a hundred percent better, but in your read, when you went and tried out all these apps, it didn't move any company that you looked at from non violatable to viable.
[00:01:12] Because that would be the crazy thing. If there was stuff that was like somewhat janky product that couldn't quite make that parlor game or whatever it is, app platform or, some derivative of cursor going after some brother vertical or whatever. And then like, Suddenly 3. 7 is like, Oh, it can do that now. At least you didn't find that in the first cursory pass this week.
[00:01:31] Fraser: No, I didn't.
[00:01:32] Exploring Vibe Coding and Community Building
---
[00:01:32] Fraser: But have you seen , levels IO. Have you been following what he's done with cursor and sonnet?
[00:01:39] Nabeel Hyatt: no.
[00:01:39] Fraser: he basically vibe coded. A flight sim dog fight game.
[00:01:44] That's multiplayer. And he's gone up to having 26, 000 simultaneous people dog fighting in the air. he then had people saying like, Hey, I'd pay for this. So he integrated like lovably the concept of an F 16 that you can have as your premium dog fighting thing. And he has sold some number of those. And then a company reached out and said, can I advertise?
[00:02:06] So he, he, again, like vibe coded a blimp into the game that then has like their branding on it.
[00:02:12] Nabeel Hyatt: Did we just be quitting our jobs and building in public? Like what, what's the video game you would, you would vibe code yourself to do right now, Fraser?
[00:02:20] Fraser: I, I I'm too anchored on this. Cause it's so hilarious. What would I vibe code?
[00:02:25] My answers were too, mundane
[00:02:28] Well, so maybe this is an interesting conversation because before this, we were vibing, Asteroid or like brick breakers into existence or, you know, people are still excited about multiplayer, snake
[00:02:43] Nabeel Hyatt: Right, right.
[00:02:44] like the only thing that's not, great about this is the graphics. I mean, not the only thing there's, there's a lot, but, Well, there's also just the, it's vibe coding. Like obviously if this was a package, good, no one would buy it. Part of what you're doing is just following along with the levels IO guy. And you just, it's community building, right? You're like,
[00:03:02] Fraser: it's community building.
[00:03:03] Nabeel Hyatt: it's part of the entertainment.
[00:03:04] it's like watching the sitcom get built as you're watching the sitcom.
[00:03:07] Fraser: It is, what are the websim guys call it? self expression and entertainment through software development. it's not people who are into software development. It is like his, his fans.
[00:03:17] And then there's become a meme and like every great meme it's, it's taken on.
[00:03:21] Nabeel Hyatt: so maybe there are just ideas that we all are not thinking about right now that were too janky or unviable if you were playing in the play space two years ago, if you were just trying to build a company and prototyping that, that you should go back and, think through,I don't know what they would be, but
[00:03:36] It's gotta be true that some non viable things just became viable,
[00:03:39] Fraser: Yeah.
[00:03:40] Nabeel Hyatt: related to coding.
[00:03:41] yeah,
[00:03:41] AI Gaming and Future Prospects
---
[00:03:41] Nabeel Hyatt: a whole era of, you know, addicting games and all these other Flash website game generators. and there are a crop of AI gaming but none of them have quite gone yet, but maybe with 3. 7, we're going to see, we're going to be really surprised in 30 days. it might be the time.
[00:03:58] I think it's probably the time would be my guess
[00:04:00] could be, could be. My guess is that they, the AI built games are going to be in that first bucket that you mentioned a couple minutes ago. In that like it will be entertainment. And popcorn rather than games people who are building an AI games platform, I would be surprised if they're the ones who are the breakout,
[00:04:20] what do you think it's going to feel like or look like?
[00:04:22] Fraser: it's going to feel like goofy toys and entertainment and community more so than it's going to feel like a games platform is my bet.
[00:04:30] Nabeel Hyatt: Yeah.
[00:04:31] Fraser: Hey, you play with anything?
[00:04:32] Nabeel Hyatt: I've certainly been playing around with all the deep research stuff. the thing I want to try this weekend is, on the same topic as, and Conquer, the old game, the first kind of RTS game
[00:04:43] Electronic Arts was just released as an open source project.
[00:04:46] Fraser: yeah, amazing.
[00:04:47] Nabeel Hyatt: a very viable modern, know, and it's not modern in the sense of like Call of Duty from last year, but it's modern. that has an open source code and I cannot wait to get to tomorrow and pull that thing into Cursor. And just see what I can get done vibe coding inside of a really structured environment like that and see what happens probably will be but it will at least be a good test of where the systems are today.
[00:05:11] Fraser: I don't know if it'll be terrible. My guess is that you'll be able to do stuff. All
[00:05:14] Nabeel Hyatt: I had a fun conversation I didn't know last night. I almost turned into a. Almost like a, started prototyping a product live with like six designers.
[00:05:23] So, uh, I'm curious where it might go. If we talk about it a little bit here. the question is aren't there more weird AI experiments, right? if you look at the beginning of mobile or the beginning of the internet, or even the beginning of the Facebook app platform, there were many more weird experiments coming out. we had one just say like, maybe this, technology is, leans itself more to B2B SaaS like solutions. Maybe ChachiPT just leans itself more to B2B SaaS type solutions, which I don't buy at all.
[00:05:53] Fraser: not one bit.
[00:05:54] Nabeel Hyatt: Yeah, have you evolved your thinking on that at all? Like we're now a couple of years into this thing It is a very horizontal explosion that the opportunities are literally everywhere But we don't see as much weird stuff and it feels like the iteration cycles are actually kind of slow like it's, it's just a, look, it's another guy's got another model and maybe it's a little bit cheaper or whatever,
[00:06:16] Fraser: I don't know if I might think it's evolved into anywhere profound. Certainly the industrialization of startups and what we celebrate as a broader community has probably Change the way, uh, that people are, prioritizing what they do for at least some subset. I think we shouldn't underestimate like the, the cost associated with this as well.
[00:06:38] Like you could experiment in. The mid two thousands with web two, or you could experiment on, on the Facebook platform or even with iOS for, no marginal cost.
[00:06:49] Nabeel Hyatt: You're right. You're right. So, and maybe in that world, uh, a product like deep seek, just net beneficial to the amount of experimentation.
[00:06:56] And so it just increases the velocity, not just of models, which is a whole nother thing, but of just this whole ecosystem
[00:07:02] yeah. My read was that it's because there's not a leaderboard.
[00:07:06] Fraser: Hmm.
[00:07:06] Nabeel Hyatt: I, I was thinking about it more friends said he thought it was cause there was not distribution that you had this mobile distribution advantage with an app store where suddenly you have a new thing, new way to grow. And Facebook, certainly the Facebook app platform, part of its promise to founders was like, you could go viral and get in somebody's feed with your lonely cow and grow. ButI think it's the lack of a leaderboard, or another way of wording it is the lack of a common news outlet to talk about who's winning and who's not
[00:07:35] Fraser: Mm hmm. Mm hmm.
[00:07:36] separate cross talk
---
[00:07:36] Nabeel Hyatt: to our,
[00:07:37] Fraser: Mm hmm.
[00:07:38] Nabeel Hyatt: right now Is almost like trying to watch the stock market ticker in order to figure out the fortune 500 companies are going to be over time.
[00:07:46] Fraser: Yep. Mm
[00:07:50] Nabeel Hyatt: to ask you at the end of the day, like, Hey, which stock was up the most today?
[00:07:54] How'd it go? It's like, it's like, or this week or this month, like it's a terrible way to figure out things. And the thing that I remember as a founder in the Facebook app platform was like. All of the other founders, including me, you'd log in every day and you'd look at the top charts of the top apps and some new thing would spike up for the week and, you know, look, most of them were terrible, but it gave you a thing to kind of experiment with.
[00:08:18] and then you go see it and you see that, Oh, that guy experimented with there's log and flow a different way. And, Oh, that's kind of cool. He just drops you right into the game.
[00:08:24] I've never seen that before. You know, like, and then you're like, Oh, I could try that in mind. and it was same with mobile. you had the mobile app store. And so in those early days, you were looking at that every day and you were looking at new, types of games and new types of business apps and new types of travel apps that were trending. And it was just like inspirational juice for you to then, stand on the shoulders of those giants and then built your next thing. It's weird that in a world of everything being ranked, like there's no. You know, app store product hunting equivalent of AI products right now.
[00:08:52] Fraser: hmm.
[00:08:53] Nabeel Hyatt: Does that resonate at all? I saw a couple of head nods, but
[00:08:56] Fraser: it does. part of my response initially was going to be like, that sounds like you're making the argument that your friend made that you dismissed around distribution. But I realize that you're now talking about like, inspiration and discovery and, recognition more so than distribution.
[00:09:10] Nabeel Hyatt: I am talking about builders being able to see what other builders are working on
[00:09:15] like we're in a common salon
[00:09:16] Fraser: Yep.
[00:09:17] Nabeel Hyatt: we are critiquing each other's work, analyzing each other's work and iterating quickly. I understand that the output can also be that it gathers you more consumers if they see that, but that's not the outcome I'm talking about that's valuable.
[00:09:28] Fraser: Yep.
[00:09:28] Creating an AI Meta
---
[00:09:28] Nabeel Hyatt: Although having tens of millions of consumers come into your app is also valuable. Let's be clear.
[00:09:32] Fraser: that resonates mean, frankly, that's part of what so appealing about WebSim, right? the idea that there's this creative place where people are. Congregating and finding and forking and find inspiration and building. And then it's there and then you can contribute to it.
[00:09:49] Nabeel Hyatt: the best thing about Web sim is that it, it offers a version of that. it doesn't offer it for the whole ecosystem of ai, just for web sim things, but I'm really glad they, it.
[00:09:59] Like there's no version of that for cursor, right? There's no real leaderboard of, or explore page of AI apps, ,
[00:10:07] Fraser: right. Hey,
[00:10:13] Nabeel Hyatt: that you can find something that's close to what you're thinking about building as well. That's, I think, part of the, issue here.
[00:10:19] It can't be super constrained. Cause if you get that also like that, thing develops its own meta. I got into Flickr in the early web 2. 0 days. One of the very first web 2. 0 darling companies. And it's like a photo sharing website with an explore page. I can't tell you how many times with my first kid, I took photos specifically trying to think about the Flickr Explore page and the meta of that page and what would get me to trend mid journey did the same thing early on, like mid journey almost developed its own creative meta people were learning how to prompt by banging their head blindly against a
[00:10:55] and only looking at the slot machine poll to get what they want. You're exploring the discord and you're looking at the floor page, and then you're looking at how other people are using the product. And I think we just, grow a lot faster that way. we're not having that level of collective conversation at the level of playing with. Claude or playing with code gen or playing with AI generally. so instead we're left with like, well, what are the milestones that makes you look, it's like, I don't know, like they just raised 30 million series B like, okay, that doesn't happen often enough. And it's not
[00:11:27] of a milestone to be the creative nexus of what you should be looking at.
[00:11:31] Exploring AI Apps and Platforms
---
[00:11:31] Fraser: so, the two examples that you shared around the Facebook platform and, the iPhone app store, there was a solution where as the consumer, you got the discovery as well as you seamlessly used it on that platform. I think that's probably an important piece. what I mean by that is like, it can't be a product hunter, a Reddit type thing where it's like, here are the five cool apps of the day.
[00:11:53] And then you have to link off and use it. it is like a self contained platform and so, this is where that I go to my comment earlier that I think the next gen, uh, AI for games is going to feel something that's not like an AI for games platform, but it's going to feel exactly like what you just described and it will be some version, you know, cursor for like, where's this for cursor vibe created apps or Reddit created apps and like, it's, yeah.
[00:12:20] Nabeel Hyatt: this tomorrow.
[00:12:21] Fraser: Yeah, yeah, it's not going to be those, but you're going to be able to, like, go and find the playable or usable versions of the software that people have vibed into existence, and you're going to be able to, like, remix aspects of it, and you're going to be able to, like, go into the source code and take what you need and fork it into your own thing.
[00:12:38] Nabeel Hyatt: Yeah. I think you're onto something. you're right. part of the issue is this is all on the internet. And so your point is it might happen inside of a closed platform. Like the way it happens in Roblox, like people look at what's
[00:12:50] and it helps inform what you should build next on, Roblox. People look at the steam pages and steam spy to look at what's working and that influences what's going to happen next. I kind of think that's true, but then that leaves it up to each Canton village of vertical market to go do that. Their thing that they're going to
[00:13:08] their market. And I somehow don't think, you know, sure. So glyph is going to have an explore page and word, where it's going to have an explore page and Zapier is going to have an explore. Fine. I don't think that's the collective conversation we're looking for as a group of founders and investors and so on to move the whole ecosystem forward.
[00:13:25] So sure that can happen, but I don't want to give up on the larger goal. and so, yes, you're right. We don't have Apple. It's not a single platform that's pushing it like say Facebook or Apple, in a way, didn't this happen in the early internet as well? Like in the very early internet days Yahoo is a directory. And we also had a bunch of like cool hunting and products like this. And that was an earlier world. And we only had like 5 million people on the internet or what was it back then. But I think we had versions of it that were federated and, and maybe product hunt is an example, not a crazy successful example, but an example of a federated version of this. you just need signal, need real signal, not fake signal to make it work. Like, cause the thing that the app store has. that Roblox has is they can see the data, right? They can see number of views. They can see the things that are moving up in a way that it is not true when it's just product hunt.
[00:14:16] And it's a separate other rating system. That's just a dig or read it for the thing.
[00:14:20] Fraser: Right,
[00:14:21] a measurement of a movement, and more of a measurement of the community's endorsement of a thing, which is a different thing, right? it's the, uh, the. Fans of hunt as a community, like the object, which is a wildly different thing. so you just need a way to, to have some kind of objective measurement. Maybe we should build this Fraser. So I wonder if we could build it just by looking at traffic stats. Like if we paid for a feed of web traffic, and so you could see how many people were going to X domain versus a different domain. Would that be good data? I think I've lost the thread of what you're trying to do,
[00:14:58] Nabeel Hyatt: Just imagine the app store. I want to log onto a site and it's only AI products. And I want to see out of the last 30 days, spiking. And AI products. What should I be paying attention to? What should I go try now?
[00:15:10] Building a Curated AI Product Platform
---
[00:15:10] Fraser: the, the reason why I said, like, I'm, I'm losing the thread is like, I don't know if that's where you started. Like, you said that you wanted to see weird experimentation and, and what products are spiking that have AI within it might be like the, sales agent, the SDR sales agent.
[00:15:28] yeah, but you don't want to see that. That's not in this bucket, is it?
[00:15:30] Nabeel Hyatt: I'm not saying it's a deterministic list that is sorted by weirdness. Oh, that would be a different question. We should come back to, I'm just imagining what happened in the Facebook app platform days. I would log in on a Friday. morning, I'd look down the app store and the app store would be organized by the things that were new and trending. So not just, if I was trying to look at the top list, it's still going to be FarmVille at the top. It's not that helpful, but you look at new and trending. And what you will see there is, of course, if you just glance down it, you will see four different FarmVille and like, that's fine that you see that I don't need to click on those. But then you might, after that, see like. My pet turtle and you've never seen a pet game before. And you're like, Oh, that's weird. What's that? And like, why are people trying that thing out? and then you drop in and you take a look at it. so it's more like, you know, some version of, all of the things that are being released, which is fine.
[00:16:25] we can probably scrape Twitter And all the AI newsletters and all the rest of that stuff to get like launches that have happened over the last month or two months. But that is quite frankly too long of a list to ever try. And so it's just, it's getting a heat check against that group to get down to say 10 or 15.
[00:16:42] And then yes, I assume that maybe only half of them I want to check out, but I'm now looking at a list of 10 instead of a list of 400. Does that make sense?
[00:16:51] Fraser: Yeah, I mean, sure, but I think,
[00:16:54] Nabeel Hyatt: me in? You don't think it's going to get me in a list I want?
[00:16:56] Fraser: no, I think that the, field of you was narrowed on the Facebook platform. Dramatically because of the, the, the constraints of the platform and the nature of the audience and the people who are building on it. And like, if instead of that time, you had like a list of here's the trending web apps or people who are using a technology like, you know, HTML, then you'd be disinterested.
[00:17:20] if you do it just based on do they use AI? That's not what you want. you want that first layer of curation is what I'm saying that the Facebook platform instilled.
[00:17:30] Nabeel Hyatt: Um, maybe that's as simple as everything that I've been talking about, but then there's a little bit of automatic categorization or curation. So you can look at the, customer service oriented ones and I can look at the games oriented ones and another person could, you know, you know, something along those
[00:17:48] can get something that feels a little bit more constrained to the space.
[00:17:51] Fraser: I am surprised that there's not, a modern take on TechCrunch for curious AI experiments where like Michael's celebration of, entrepreneurship and These like crazy web app experiences was so pure
[00:18:07] Nabeel Hyatt: So your answer is it's editorial.
[00:18:09] well, that's, that's the easy first solution, right? Yeah, that's right.
[00:18:13] Fraser: this brings us back to our, our, your PRD that we need to write. Cause this is, what you're talking about. This is what you want to build. The curated place for this. here's a weird little app that I tried. It's cool. I vibe coded this piece.
[00:18:28] It's so silly, but look at this other thing.
[00:18:30] Nabeel Hyatt: So let me give you an example of something. And then you tell me why this doesn't work. And I agree. I don't think it's quite it, but, there's a product called taft, there's an AI for that, taft. com. You could drop into there.
[00:18:45] Fraser: Okay. Okay.
[00:18:46] Nabeel Hyatt: trending. So this is supposed to be a directory, almost Yahoo, like of all of the AI things being built right now.
[00:18:54] Fraser: Okay.
[00:18:55] Nabeel Hyatt: is, is supposed to be, in their own heuristic, the thing I'm asking for, which is hot stuff right now.
[00:19:01] Yep. But look at that list and tell me why, why did this go wrong?
[00:19:06] Fraser: cause they're, they're uninteresting.
[00:19:07] Nabeel Hyatt: Yeah,
[00:19:08] Fraser: they're uninteresting. They're shallow. They're shallow in an uncreative way. Like create AI course and create coloring pages. Those are like GPTs.
[00:19:18] but you're not looking for full things. You, you just want like, different, creative, new, novel, it could be as shallow as you, as, as it can be, but you just want, you want,
[00:19:28] Nabeel Hyatt: at, at Flickr when, when that Flickr was trending. they didn't sort it by views. or popularity. It was sorted by a word called interestingness.
[00:19:38] Fraser: yeah,
[00:19:39] Nabeel Hyatt: I want it sorted by interestingness, Fraser. yeah, yeah, well, that was a good exploration of the idea and I'm sure folks online will have some thoughts as well. And we'll loop back to it later. Thanks everybody. And as always, if you see an eye product worth chatting about or you have some comments on this, uh, let us know. See ya.
[00:19:59] Fraser: