OpenAI’s Triple Threat: ChatGPT Apps, AgentKit, and Sora

Speaker A: Foreign. Look, if you've been living under a rock for the past two weeks, OpenAI just dropped not one, not two, but three major product launches that could completely reshape how we think about product strategy and distribution in the AI era. And honestly, I'm, um, feeling a little bit vindicated right now because a couple of months ago we talked about the next great distribution shift on this show. I wrote a detailed post about how these shifts all follow the same cycle. And I predicted that within the matter of months, OpenAI would launch an app platform. Well, it's playing out exactly as we expected. And OpenAI just opened up their 800 million active user base and platform to applications. They've launched Agent Kit with the Visual Agent builder. And the surprise, which I did not expect, is that they put out a new social product in Sora. Today, I'm joined by two good friends. Ravi Mehta, who is a product advisor and former Chief Product Officer at Tinder and product leader at Meta and TripAdvisor, and Adam Fishman, most recently interim VP of product at Mozilla Vanilla and previously a Product leader at, uh, Patreon and Lyft. We break down these launches and talk about what product leaders should be doing right now, what steps you can and should be taking, and what the next shoe will be to drop in this cycle. As we've said before, this is the strategically most intense environment we have ever seen. The dial just got turned up a couple more notches. So grab your coffee, keep listening, and let's turn these insights into your unfair advantage. But before we dive in, let me tell you about how we're helping other companies navigate this. At, uh, Reforge. In the past year, we built an entire AI product suite because frankly, the old tools just don't cut it anymore. We used to divide product development into a dozen plus steps. From insight to research to PRD to validation, to design to testing. The list goes on. Each with multiple steps, their own handoffs, specialists, and a lot more. We've consolidated all of this into one AI native suite. Reforge Insights aggregates your scattered customer feedback from all your different sources and helps you analyze it and tells you what exactly matters. Reforge Research runs AI interviews and surveys so you can capture net new insights at scale, not just read old feedback. And then Reforge Build takes all of those insights and lets you prototype AI features on top of your existing product in a matter of minutes instead of weeks. And our newest release upcoming is Reforge Launch gives you the feature management infrastructure that you need for AI products like Dynamic Configs. On top of that, we provide the training through Reforged Learning to help your team go through deep transformation rather than just surface level. Additional adoption of these tools, we want to help builders go from insight to launch in one place. Check them all out@reforged.com. until then, let's dive in. Well, let's. Let's kick off with, uh, ChatGPT agents and agent Kit. So I think the quick context, if you have been living under a rock, right, is that, uh, OpenAI has enabled you to integrate and embed apps within the ChatGPT experience. So as you're typing something, you could say, Spotify, what are the top 10 or lists? It'll prompt up a little, uh, use Spotify for this chat. It creates a, uh, UI directly within the chat screen. So obviously a big move. They've made a couple lightweight attempts at something similar before with GPTs. They've had multiple attempts at this. This one seems to be deeper and, you know, richer, but, uh, interested in your opinions. A potentially much larger impact in staying power. But let's just start there. Hot takes and first reactions to this announcement. Ravi, we'll start with you.

Speaker B: Yeah, I think what's the most interesting about Agent Kit is the actual Agent Builder looks a lot like N8N. So I don't think there's anything too revolutionary there. But what's really cool about it is they released a chat, uh, kit which allows you to create a front end which taps into the agents. So the agents get fired automatically by the AI when they need to be, and then they go through the workflow and then they can generate a widget as a response which can then get displayed within the chat. And so I think, interesting thing throughout both the app, um, app platform as well as Agent Kit is the idea of how do you integrate additional types of user interface like widgets into the experience so that the AI isn't just generating text or generating audio or anything like that, it's actually generating user interface. And then you can interface with whatever it is, um, and then it goes back to the AI. So I think this is like the cleanest handoff we've seen between traditional software and AI software. And so for me, that's the most exciting piece about it is I don't think I haven't seen any other platforms go as far in that direction. And one of the things they released with Agent Kit is a gallery of all the widgets that you have available. They have calendars and a widget to send Messages and status updates and dashboards and things of that sort. And so I think we're going to see very quickly people be able to build AI based experiences that feel a lot more like a fusion between traditional applications and the uh, text first or voice first experiences that we've seen so far.

Speaker A: Wait, did I miss this gallery? How do you access that gallery? I have not seen this in the ChatGPT experience. Or is this.

Speaker B: Yeah, if you go into Agent Kit, they have um, a widget section. I'll send you a link. Um, and then you could see all of the widgets that are available. And in the demo day announcement they actually showed you can customize these widgets and you can associate um, any node within your agent with a widget that gets. It actually displays information in a structured way. And so you can have these really interesting workflows that combine both the text that's being generated as well as these more structured interfaces.

Speaker C: Yeah, I think the demo had like the classic like travel agent sort of situation where you could have it fork and look up travel plans and like do this thing. And you can imagine that you could have like a travel agent widget essentially. So if you're building a complicated workflow, you could have it like research my travel, then do this other thing, then notify all my friends, then like book the tickets and like blah, blah, blah, blah. So it's, it's pretty cool.

Speaker A: Yeah, yeah. And I should clarify because I think I screwed up the intro, which is because they launched too many things, it's hard to keep them. There's Agent Builder.

Speaker C: Right.

Speaker A: The visual workflow. There's Agent Kit.

Speaker C: Right.

Speaker A: Which is part of what you're talking about, Robby. And then there's the apps. SDK apps. Yeah, related things, but not actually the same thing.

Speaker B: Agent Kit and the Agent Builder is if you want to create your own chat interface and you want to put more advanced UI into that. And then their app SDK is if you want an app that'll work within the ChatGPT interface.

Speaker A: Right, right, right.

Speaker B: But they share some interesting DNA around like the, you know, the widgets, the more structured outputs, the workflows, that sort of thing.

Speaker A: Wait, sorry, say more about that.

Speaker B: So one of the most interesting things I think from their announcement is they released a, uh, design guideline which is basically how to design user experience for AI based applications dissimilar than what Apple would do for a new feature on the iPhone. And they talk about all these different ways that you can have widgets show up in the ChatGPT experience. And so you can have like, you know, an output widget, you can have a carousel, you can have a picture in picture widget, uh, you can have a full screen type of widget. And so in both the agent side as well as in the ChatGPT app side, you could tell they've done a lot of thinking about how do we create a better, more structured user experience around AI. Ah, that combines things that we can do on screen and in like a traditional software interface with the things that gen AI is good at, which is generating language and code.

Speaker C: Yeah.

Speaker A: Um, you know, before I move to Adam, Ravi, I mean, I think something we talked about in previous episodes with you as well as in AI strategy course is just this kind of relates to the topic of we're in the strategically most intense environment ever. One of those reasons is not just because incumbents have been moving very fast on AI plus, there's new startups in every category, but you have these huge new horizontal players in ChatGPT T and Claude as a possible vector point of competition. So I'm interested in how you think about these announcements in relation to how do you compete in this world? How do you actually build differentiation? Uh, I think this may be the

Speaker B: oh moment for a lot of the big incumbents because if you think about it, a lot of strong, almost monopolistic positions are built on how do you launch into apps? Google search is ultimately an app launcher. I search for something and then I launch into a content site or I launch into an application. That might be how I get into notion or other things. The iPhone, um, the home screen is essentially an app launcher. I go to it, I figure out which app I want to use, they control the app store, I go into that app and then I'm all of a sudden into an experience. I think the really interesting thing about ChatGPT is it completely obsoletes both of those as app launchers, neither one of which was great to begin with. Right. Search is not a great way to find an app. The iPhone home screen is not a great way to find an app. What is a really good way to find an app is you just talk about what you need and then the AI figures out, well, this is the perfect tool for you. Let me launch for you. Let's start a workflow within that app. And so I could see OpenAI launching a phone in the future that is essentially just an app, uh, just like a chat based interface into all of your apps. And any app that works on that phone has to go through their app SDK and that completely changes this position that Apple has had as like the mediating force in terms of how you go from, you know, being an app that wants to be on the device to how you get launched on the device.

Speaker C: Yeah.

Speaker B: So I think this is potentially a really big moment in terms of reshuffling around incumbents and it goes to what you've written about, which is the distribution shift. I think we're now at the very beginning of a pretty significant shift in distribution.

Speaker A: Yeah.

Speaker C: And don't forget, ravi, that they OpenAI paid six and a half billion dollars for Joni I've like six months ago. Speaking of people who have built phones before and other hardware, nice, elegant hardware devices. So I don't think that's that much of a leap, honestly.

Speaker B: And they would be in such a good position.

Speaker C: Yeah, yeah, yeah, they would.

Speaker A: All right, Adam, what were your hot takes and first impressions?

Speaker C: Oh man, I have so many, uh, more hot takes on Sora than Agent Kit. But we'll get there eventually.

Speaker A: We'll get there.

Speaker C: Yeah, I don't. I didn't have a ton of hot takes. I mean I think it's. I was, I guess, first and foremost, I'm incredibly impressed by just the pace of stuff that's coming out of that group. Like that is not a large company in terms of number of employees and they are releasing just an insane amount of stuff. I. It's like exhausting to keep up. Right. Like even us just trying to figure out what to talk about for the show was a lot, you know, to sift through it all. I think the other thing is, you know, I haven't used N8 N a ton, but I've used Zapier a lot. And you know, I was kind of struck in the demos that I saw. I've yet to really like deeply play around with with uh, the Agent Builder, but I was kind of struck that it's still a fairly complex thing to use. Like it's not really for the non technical user yet. Um, and I think what's still missing and, and in reality is missing from all of these things is like voice to agent flow stuff. So that was an interesting. Like, I want to be able to just say, hey, I'd like to build a travel agent, uh, booking, uh, system for myself and I don't know how to do that. Can you get me started and then just have it be like bloop, bloop, bloop, bloop and build all the things and then edit from there. I think something like that opens it up to a lot more people. But the downside risk there is, I think then you don't necessarily. That's not as attractive to the enterprise customer, which is who's, you know, who pays all the bills usually.

Speaker A: Yeah.

Speaker C: Um, and, and people want security and they don't want weird black box things happening. So I, I think they're in this like really interesting in between stage now with some of those products that they just released, which is like how far into the vibe agent builder direction are we going here versus you still have to have like a good head on your shoulders and some guardrails and um, and, and I don't know which direction they, they head in but. And then also I am super excited about the um, the apps in ChatGPT. I saw somebody do a Spotify playlist in real time based on their music interests and I saw the booking.com thing and this morning I had in my inbox about how to build a carousel in canva through a ChatGPT prompt. That's in ChatGPT where you pull in the Canva app. That's interesting. Why would I ever really need to go to Canva as a first party place to build something anymore? I could just do my editing there after I've already built the initial thing. So I think it could open up a really interesting acquisition vector for ChatGPT too where people are like, oh, if I just go in here I can do all this stuff and I don't have to create accounts in all these other places and like get into all these other systems. So that is pretty interesting. Again, on their journey to crossing a billion users and making you, Brian, the greatest prophesizer of all time.

Speaker A: There he is now. Now the profits just rain.

Speaker C: Right?

Speaker A: Like that's what happens. It's just like the money starts falling out of the sky because uh, I, I had a correct prediction. I wish, I wish that was the case.

Speaker C: Brian Nostradamus Balfour has joined the chat folks. The capabilities and the potential here is incredible. There's so many different directions that this can go.

Speaker A: Yeah, I do think one of the things I'll talk more about apps but uh, just real quick on the agent side which I haven't spent as much time with, but to uh, your comments Adam, is that I think on some of the previous attempts of what they did with uh, GPTs and the other things is it was targeting this middle use case of these no Cody types of creators. And if you're trying to build a platform that isn't the most powerful position to Be in there. Yes, there is a market for that. But I think you either want to enable developers to create very powerful, deep, you know, rich experiences, or you're targeting incredibly mass market consumer experiences with the front end, the stuff that sits in the middle. You have these trade offs. You can't enable the true depth in power, so you're limiting the use cases of what it's for. And those things also tend to be too confusing and complicated to use for the mass market. So if you are trying to build the next major, uh, distribution platform, it makes sense that they made this, both the apps and the agent stuff, more developer focused and I think really starts to signal this shift into um, a true platform. I think on the apps part, which is besides feeling a little vindicated from all the people who commented on the naysaying, is, uh, um, and I was worried at points there were a couple points where it's like, oh, like, are they not going to do this? And then I'm going to have to really eat my words. But they did it. And if you have not listened to the previous episode where we talked about the next distribution chip, you should go listen to that. We have clearly moved from what I would call the inception stage to the open phase, meaning we are entering a world where the rules are not completely written around this stuff. They just opened up a new way to access their 800 million active user base that's growing at, uh, an incredible clip. And I would probably say this move into apps probably pulls even more users and more engagement into their ecosystem. And so then the question becomes, okay, what do you do? What's the next step as part of this phase? And one of the biggest things that I've been trying to look at is one is what are the discovery mechanisms? That's incredibly important, which are still honestly fairly limited, but I'll come back to that. Two is the monetization opportunities, which aren't really baked into the system yet, but you could see that on the e commerce side that they're taking steps there and that feels like another natural next move of things. And then three, what are the rules of the game? Meaning, uh, what do they actually make available in the SDK and what are they giving you access to and what are you giving in return? There's some value exchange there. And one of my questions and one of my hypotheses when we wrote that article was this stems back to trying to collect more of the moat. And you can see in their terms when you install this application, one of the things that they are giving you is memory. So you see in the actual terms they say that they will give you access to some of the chats and some of the memories. They're giving the app access to some of your chat and some of your memories. And that's actually incredibly because for anything AI related, what do you need in order to actually generate a valuable output? You, you need, you need context, you need really good context and memory is a form of that context. And so what they're really doing is when they're giving you access to that, they are giving you a much better starting point than you can start on your own as a cold new application. Because people have been using ChatGPT, there's stuff that's stored up there. So hypothetically, AI apps should be able to generate a much better product if a user auths them into ChatGPT and gives them access. And what that's going to do is that's going to create more interactions, more chats on ChatGPT, which lets them collect more memories and more context. And so here you go, here's, here's the flywheel that you start to see spinning. And so I think in terms of people, of what are you, what are you trying to, it's not just like kind of what do they allow in the interface, but they are giving you access to a context in the data source. And so the question is, is what can you do that's new, unique, better now that you have access to that information versus starting cold as uh, like as a result. And there's gonna, there's a bunch of these micro exchanges that I think like, honestly I've talked to a couple of the early partners and it's very clear they're flying and building the plane at the same time on this stuff. So, so I think all of this stuff is going to change incredibly quickly. And then last note on this, and I'll get your comments in reaction to that is um, the discovery mechanism. So that's going to be incredibly important. Right now it' bit crude, you have to type Spotify at the beginning of your chat and then it pops up this little thing that says you want to use Spotify for that that really favors uh, brands that are well established and have common places in people's heads and really disadvantages folks that are just starting new or just starting to establish brands. But I imagine they will to incentivize more and more developers onto the ecosystem. They will enhance those discovery mechanisms too. I think Ravi, you were talking about before, which is if you don't like state a specific app, maybe they suggest one of a few or something like that. But I would say in this first form it really is benefiting the incumbents again, not the startups.

Speaker C: What do you think happens, Brian, with like. So if, if you're doing this handshake on memory between these two entities now, what happens to like the MEM zeros, those sorts of companies where they were doing that as a memory, as a service? It stands to reason that companies might not care about that anymore if they can just get all of the stuff that you've got in ChatGPT already because you're already using it every day.

Speaker A: Well, I did find it interesting that I'm looking at the page right now, essentially all but one of their early launch partners are consumer companies and not B2B. Did you guys, did you guys notice that?

Speaker C: Well, it's probably related to data sharing. We, we already give every consumer company everything on the planet and we don't care as consumers. But also that's the most data and the most people. Right?

Speaker A: So yeah, the, the only two B2B ish. Right. Are Canva and Figma. But out of the B2B world they are heavily on the prosumer end of the spectrum. I like the individual end of the spectrum, but I did notice that, that it was, it was heavy. Heavy consumer bet.

Speaker C: Yeah. Yeah, fascinating.

Speaker A: So, sorry, what was it? The.

Speaker C: What was the question? Well, I'm just wondering like what happens. There's. There was this kind of, this set of companies that was designed to try to create like portable memory or op. Yeah, like MEM M0. Like this open memory that you could then move to application. Application. And like essentially ch. If you just have Chat GPT as your default, you know, platform, then they become the memory source for all of these other partners and stuff. Does portable memory doesn't matter as much for. Yeah, for folks. I want to worry a little bit about those types of companies getting the squeeze here.

Speaker A: Um, yeah, and small disclaimer. I am a small investor in one of those, those companies. But uh, I forgot it doesn't necessarily. In the same way that everybody's saying Zapier and Aden, you know, all those like folks are dead. And I'm like, no, not really. Like if you look at it, it's slightly different types of users and these markets are so big that there's probably pretty sizable use cases. It's the same thing in the AI coding market. Right. There are so many use cases that AI coding have enabled and we're still seeing them like segment and emerge. And so it's, I'm hesitant, I am personally hesitant to say like you know, X killed Y on any of, on any of these things and similar their, you know, I, I don't think there's going to be room for ah, multiple big consumer chat experiences because the consumer doesn't want to move. Um, it. And if there is a third party portable MEM thing is probably more on the B2B side of things and there's probably only one of those out of that category, uh, that wins.

Speaker B: Yeah, I was just looking too. They have announced 11 more partners, all of whom are consumers. So TripAdvisor, AllTrails, Peloton, DoorDash, Thumbtack, Uber. And so yeah, I think you're making a really good point. I hadn't noticed this about the announcement, but they did make it a point at the beginning to say that they've, they're reaching 800 million users. And so I think in terms of who's the company that's sort of most threatened by the app platform, I think it's first and foremost Google. And then secondly Apple. I say Google because also if you look at their launch partners, many of them are very large spenders on Google AdWords. These are companies that advertise across, yeah, booking, Expedia, uh, you know, across millions of different keywords because they have all of these different things available for people and so they want to be the end of a person's, person's search. If those searches are starting to Happen now in ChatGPT, you know, that's where they need to be. I've also been talking to a number of companies and most consumer companies that depend on SEO are seeing pretty significant declines in SEO traffic. And so already we're seeing the traffic shift away from traditional Google search over to Perplexity and chatgpt and Claude and others. And so this makes total sense as a way to double down on that. And I think the other interesting thing about this market has been there is some commoditization that's happened between Claude and chatgpt and Perplexity. They're all roughly the same. A little bit better at some things, a little bit not as good at other things. I think ChatGPT now making a really bold statement around apps and getting apps into the product is something that could accelerate them and then give them a moat. Because, uh, as soon as you get into, yeah, I know all my apps are here, I know I'm going to get the best answers. I know I'm going to get the most functionality here that will just create this, this habit Flywheel. That's going to be hard to break.

Speaker A: Yeah, I hadn't noticed. But that's the second observation. I think you're right. Which is if you look at a lot of these initial launch partners, their dominant acquisition channels, either SEO or SEM. Right.

Speaker B: Yeah.

Speaker C: I mean, TripAdvisor and booking come to mind right away.

Speaker A: Booking, Expedia. Right. Are like a lot of SEM, Zillow, Canva, TripAdvisor, Thumbtack, all. All trails, all huge SEO. Coursera, huge SEO. Right. Like is their dominant, dominant one. So, yeah, it's like shops probably just went to all the companies, you know, being like, this is. This is your olive branch.

Speaker B: Other thing I noticed is Airbnb is not here and I was surprised by that because travel feels like such a great use case for this. But they're not a huge spender or not as big a spender on SEO and SEM as some of these other ones to get a lot more.

Speaker A: So what do you do? Like, if you're in the product leader seat of a company, it could depend your startup versus incumbent. What would you be doing right now?

Speaker B: I don't know.

Speaker C: I feel like Ravi should take this, given his product leadership chops here as a starting point.

Speaker B: Yeah. I think the first thing is to get on the phone with someone at OpenAI in the partnerships team and get to the front of the line in terms of building out an app. Um, right now it's a closed system. They're going to open it up. I think they said they'd open it up at the beginning of next year. The companies that are getting in now are going to have a pretty durable position. Like, I think if I was Udemy, it's now an uphill battle on education queries versus Coursera. So I think. No, it sounds very tactical, but I think that's the first thing. You know, a lot of times with these distribution shifts, it's who is first there that gets a lion's share of the reward. So you have to move real quickly.

Speaker A: Reforge is the exception. Yeah, no, no, that is very good advice and it's on my to do list. So. Yes.

Speaker B: And it's probably worth not just talking to OpenAI, but also talk to Anthrop, Topic, talk to Perplexity, talk to Google. Um, you know, you may not be part of this curated set for OpenAI, but there's also a ton of people using these other platforms. And so as quickly as possible, get into ideally a non Exclusive inclusion into the initial launch of the app platforms that these companies have. I think the second thing is OpenAI did release these app design guidelines. Part of what they did was they, they put together a punch list of things that you should think through from a use case standpoint. So they want people to start thinking about their experience through the lens of how does it work really well with AI? What are the types of use cases that are likely to come up when someone's having a conversation, um, with ChatGPT. And you know, how do you create a really great experience for that? So that's the second thing is to go through the documentation, go through those use cases, go through the app design guidelines, start to understand what parts of your app can, can be pulled out of your product and into the ChatGPT experience in a way that's really works well for someone in ChatGPT. And then I think the second thing is to create on ramps into your product. It was pretty clear in the demos that they showed is you can go into full screen mode, you can use the product in a way that is pretty engaged, but at the same time, at some point you do want to get people off of that. Like if I'm searching for a home, I don't want to do the entire search, um, on ChatGPT. And so there's off ramps to get to Zillow or to TripAdvisor or to other places. So thinking about where does the use case start, why does that happen in, in AI? How do you solve for that really well. What are some of the additional things that you want to support in AI and what are the set of things where you want a person to come into your full app? I think are really important.

Speaker C: Yeah, I'll, I would, I'll just double

Speaker A: click on that real quick because we've done a little of this work at Reforge. Uh, to say it more specifically is which workflows around your product are people starting in ChatGPT or Claude around and then take that output and copy and paste it into another tool. In our case, the product use case. One of the very common starting points is people start writing their PRDs in chat. So if you know those starting points, right, those very specific starting points, I think that starts to spring a bunch of ideas around. Um, you know, where do you start to uh, like insert yourself and, and what value you can add. But anywhere where people are like copying and pasting something from ChatGPT and trying to put it into another is prime workflow of where you could potentially Insert yourself.

Speaker C: Cause we talked about how this is very targeted at consumer companies right now. Like, if I were a B2B company, I would be on the phone with Anthropic saying, why don't you do the same thing for B2B companies? I'll be your first partner. Because, like, that seems like how those two kind of companies are dividing themselves up right now. And then Ravi, what you said about like, they've given you the like playbook for how to restructure your app to work for ChatGPT apps. A decade plus ago, Google did the same thing with SEO. They were basically like, if you rebuild your site so that we can crawl it and users can navigate it better, you will move up the search rankings. And I imagine that ChatGPT will reward the same, same sort of behavior. Because what's going to matter, what's going to keep people using it as a consumer is if they can actually be successful within ChatGPT with the thing that they want to accomplish. And so as long as the site gets tuned to be successful for that, then they'll be fine. But if people stumble over the outcome that they're trying to drive, then they're just going to go back to that core site.

Speaker A: I'll leave us with this before we move on to Sora 1. Is Ravi to, to your point where they publish the guidelines? I would say in every shift like this previously, like when, when Apple first launched the App Store and they published their guidelines, I remember Facebook, they even published their guidelines is that alpha is always found in taking some of those guidelines and mixing it with breaking the rules.

Speaker B: Yeah, 100.

Speaker A: You have to find the arbitrage opportunities. Right. And, and that's where the most creative kind of breakthroughs, uh, come from. The second thing I wanted to say is kind of, as I mentioned in the previous conversation is there's a bunch of folks kind of looking at this being like, oh no, like what does this mean? I don't want to participate in this game. And I just double down and say there is no opting out of this game. You do not want competitors to, to be earlier to the platform. They will gain not just an incremental amount more of the benefit, but probably about a 10x more about around the benefit. And then, sorry, last piece on this is everybody's asking who did read the article, who did kind of listen to the previous podcast, is they're asking for my prediction on when the next phase comes.

Speaker C: When's the other phase? When's the other shoe gonna drop? Brian, that's what everyone asks you in your, in your comments. Comments.

Speaker A: Yeah. Right. And I think that's just the wrong question to be asking right now. And it's not a very useful one because like, like Nostradamus Belfor. I can't even predict. I'm not gonna. I can't predict that one. We can start. We'll. We'll probably start to see the signals when they start to monetize it. Right. But they're not in that phase. But your first job is just to catch the wave. That's it. And that kind of goes back to Ravi's point, which is pull the strings. Like what? I like whatever strings you can to, to try to get in early and catch the wave. Once you catch the wave, then you can worry about the second problem, which is how do you come to shore safely and make sure that you don't completely get disintermediated. But job number one is just catch. Catch the wave.

Speaker B: So I think another important thing is the winners are made before phase two starts. Right. Like, if you were going to be a major OTA, you had to do that 15 years ago, like booking and Expedia did. You can't do that today. It's impossible to do it. So if you are just sort of clenching for phase two and worried about it, you're going to miss the opportunity to participate in the wave and to have a strong enough position so that you can actually withstand phase two.

Speaker A: Yeah.

Speaker B: You know, I think that we're, we're going to be in phase one for a while and now's the time to really lean in and try to capture as much of the opportunity as possible.

Speaker A: Yeah. Okay, let's move to Sora, my latest obsession. So in just a couple days before or the week before, I don't, I don't even know. It's things they're moving.

Speaker B: They.

Speaker A: OpenAI also launched and launched Sora, which was. And Sora too, which is so both, uh, a new version of their video gen model, but a, uh, dedicated standalone app that has kind of the look and the feel of TikTok, where rather than just like uploading your own videos, you can generate a video and there's some cool dynamics where they record a little bit of a snippet of you. You can insert yourself in the videos and you, uh, can also give access to everybody if you choose, or specific people, friends to use your cameo and generate videos with you. So I have access to Adam's cameo and so I could say, you know, I was doing things like generate A, uh, crossover podcast between startup dads and unsolicited feedback called the Growth Daddies. It dressed us up and gave us like baby clothes, and we were talking about all sorts of things.

Speaker C: You're drawing growth loops on a diaper. Like, it was like a whole thing.

Speaker A: Thing.

Speaker C: It was a whole thing.

Speaker A: So, anyways, I think I wanted to dive into this, uh, a little bit because I honestly, I personally haven't had as much fun with a new application and product as I have with this one, though. I know everybody doesn't feel the same way as me. And a lot of the reactions were kind of similar to Meta's launch of, what is it? Vibes or something that, like, they were like, oh, this is just more AI slop. I have a slightly different take, but, um, we'll start there of like, you're your hot takes and your first reactions. Adam, we'll start with you this time and then.

Speaker C: Yeah. Um, so I am also enjoying this quite a bit. It's very fun. I, uh, still think it's kind of a novelty. I haven't really seen a real interesting value, and I think the entertainment factor will fizzle out for me eventually. I already know, know. And Ravi, you talked about this in an article you wrote a long time ago. Like, if I don't know somebody in the video, I'm really not very interested in that. I don't find it funny. Like, of course, when it's you and me, Brian, like, doing a thing that's really funny because it's like, oh, they wouldn't be doing that. Look at them. And I know you and, like, it's. And it's fun to see my friends, but, like, it's not really that funny or interesting or whatever if it's a random that I'm not connected to. So as far as, like, feed dynamics go, I'm not sure yet, like, what's going to happen there. And I think this is definitely the problem with Meta's current offering, by the way. They're just nicely designed videos, but, like, there's no human connection to them at all. So that's. So that's kind of my. My hottest take. But I would say the other really interesting thing is, like, I shared this with both of you guys yesterday. ChatGPT has, or OpenAI has the number one and number two app positions in the free App store right now. Like, when was the last time a single company maintained, like, top two position for two products that they own? Like, that is like, insane dominance. If they haven't hit their billion DAO yet, like they probably did in the last week and a half with these two launches because people are just using this like crazy right now. And it's an invite only product.

Speaker A: Two, every new product starts by siphoning off users from another channel. That's right. So there you go.

Speaker C: That's right. That's.

Speaker A: I hadn't thought about that. Plus there's. It's invite only still, so that, that is kind of impressive.

Speaker C: And they were like number, maybe number one for a short period of time. But the fact that you only get five invites or four invites and it's still that high in the free app ecosystem is in like this day and age is incredible to me. So they just must be like vacuuming up users right now or at least potential energy for future users. So yeah. Ravi, what's your, what's your thought on Sora?

Speaker B: Yeah, I think even um, like another level to add to the fact that they're the number one and two right now is if you think about it, there's really only three ways to navigate the Internet. You can search, you can discover content in social media or you can install apps. App alone apps. Apple and Google own apps. Google owns search. Meta owns social media. Meta has had multiple apps in the top 10 with uh, Facebook and Instagram. But they've never been able to break into search, they've never been able to break into installing apps. They're trying to do that with all the Metaverse stuff. They want to own the app platform, not just the content platform. And so the fact that OpenAI has all three of those things actually covered and is getting traction on all of those I think is a real warning sign too. The incumbents that the um, the gatekeepers are really shifting right now. And you know, maybe Sora will work, maybe it won't work, but it's a great bet for them to take and it's probably one of the best ways. Like the app is the UX is clean, it looks a lot like TikTok, but that makes a ton of sense for what they're doing. The AI is really powerful. It's fun. I think they've really nailed the social, social element of it, which is where a lot of people go wrong. The article that I wrote talks about like the fact that entertainment value is a product, both, uh, of the social value of the content, how meaningful it is to you socially, and then the production value of the content, how good it is just in general, um, and so high production value content like you know, shows on Netflix. I don't have a personal connection with, but ultimately, um, they're interesting because they're just high production value. And so the entertainment comes from that. A lot of the stuff on Snap and TikTok, you know, is not that interesting or entertaining outside of its social context. But because it's from someone I know, it's, it's more interesting and more entertaining. I was originally thinking, and I think Brian would love to hear your thoughts on this, that um, this would actually flatten the curve because it's now possible to create really high production value content that has a lot of social value too, because individuals are creating it. I think what's actually going to happen is it just shifts the whole curve over. It just means that the production, the floor for production value has gotten a lot higher. And there will still be certain content that and certain platforms that are really focused on the social side of entertainment. And then there'll be other platforms that are really focused on uh, the production and the broadcast side of entertainment that will change. And that might lead to some reshuffling here in terms of which players have a space on this curve. But I think the curve, uh, still remains because ultimately it's a really human thing. We find things entertaining both because they're good and because they're really relevant to us.

Speaker A: Yeah, the shifting the curve is interesting. I haven't thought about that. My take from using it for a week is that they designed the product around this TikTok part of the curve and what is actually working and where the design should be is actually far closer to Snapchat. And the reason for this is, is all of the videos that we're generating, the ones that I find funny are with folks that I know and they're typically around inside jokes, things that are hilarious to. But to a mass market audience, people will just look at it and be like, what the wtf? I mean, because there's videos that I was laughing to tears on that you two would be like, uh, this is the most boring thing in the world. I don't actually mean this as a criticism. I think they kind of put this out there as an experiment. And you can even see in Sam's first update, he's kind of acknowledged this too, which is he said that there has been a much higher volume of long tail videos generated with much shorter amount of views. And that signals being more, more towards a Snapchat like experience where these are things that are acting more like DMs or group text messaging than it is a TikTok or even YouTube shorts where you have a very small percentage of creators creating mass market views and videos. So I think they have a mismatch between the actual uh, design of the product because sharing with friends and sharing with groups is in the application but it's buried and I don't even know if I get notified. So you see all these hacks, everybody downloading these videos and sending it in imessage or sending it in Slack DMs. And I think there's a flip here that a different design around these magic moments that probably unlocks it for a lot more. The, the other thing that I noticed is that usually in most social applications, uh, you get dopamine hits from the scrolling, right? And you get because, because of the variability and, and the dopamine hits when you create tend to come from feedback mechanisms, getting likes and getting content comments. Uh, the interesting thing here is that there is a dopamine hit further up the creation funnel which is just the generation. I give it guidance but I don't know exactly what I'm going to get. And sometimes it comes up with hilarious iterations or scripting, uh, that I had said never, like I did not, I, I would not have expected. And that dopamine hit on, on that variability of the creation is encouraging more creation. And you've seen them have to take down their limits pretty dramatically. They're now capping it at uh, 30 generations, um, per day. But so there's that product strategy question which is how do they change the design to potentially feed this specific use case? And then I think the second thing which Sam also alluded into is around monetization which is okay, well, well if people treat these generations more like DMs, I'm going to have a huge volume of these things with very low views. The monetization models of TikTok and YouTube shorts don't work. Something else has to take place. My other question is does this just get Snapchat storyfied? And Zuckerberg comes in and says thank you very much for being my R and D department.

Speaker C: I mean I think it's interesting. So I have a bunch of reactions to that. One is I think it's really interesting that they landed more, more out the door on the TikTok, uh, ask nature of this. If you think about like who is actually inside the company at uh, OpenAI building these products, it's a lot of people who probably felt very slighted by the fact that TikTok came around and just stole Facebook's lunch, right? And Instagram's Lunch. And so like a hundred percent, they basically built that. They're like, we're going to do the AI version of this first. Um, so setting that aside, the monetization stuff, Brian's super interesting because the way that all the bigger, bigger, uh, bigger reach platforms that aren't like one to one monetizes, they court creators to the platform. They get creators to make really big like videos that tons of people like and share around and stuff like that. And then they monetize via ads or you know, direct fan sponsorships or something like that and then take a huge chunk of it. But um, it's not clear what they would do now to move in that direction. The only thing I can think of is if you allow um, cameos of very recognizable people or you like other people's likenesses or recognizable objects or places or something. Which could be interesting if you think about the fact that immediately all of the like movie studios and everything were

Speaker A: like, yeah, I was just saying that

Speaker C: lawsuits, um, but if you can find

Speaker A: a way to work, the smart ones will find a way to embrace it.

Speaker C: Exactly.

Speaker A: Yeah.

Speaker C: Same with, with you know, sports celebrities, like everybody who is big on some of these other channels, if they found a way to entice them and say, well, if you allow pe, maybe you pay a dollar to be able to use that person's cameo or I don't know what it is, but they're going to have to sort that out if they want to move in that direction. And I think they're going to have to move in that direction, Brian, because the cost is just unsustainable as like a one, one to one application for, for me like DMing my friends, that's not a sustainable cost at all.

Speaker A: I know, but it's where the meta. I was just trying to do the math on this. So the, their APIs, the sort of standard is 10, 10 cents per second and so these are nine, so 90. That's a retail price. So what margin do you think they're making on that? Like what do you think their actual cost is per video?

Speaker C: Half of that maybe? Yeah.

Speaker A: Less than that. Yeah. Like so, yeah, you're talking 45 cents. Just call it 50 cents per per video, you know, so I'm costing them,

Speaker B: uh, what is it right now we may not want.

Speaker A: I'm costing them. Yeah. I'm costing them 15 bucks a day.

Speaker B: Yeah.

Speaker A: If it's actually like 50 Cent, you know, like, like that's, that is not sustainable. Their margin might be more, more on that from their, from their API. But I think the point is, is I'm probably costing them dollars per day.

Speaker C: Yeah.

Speaker A: You know.

Speaker C: Yeah. And then my other, you know, you asked the question, Brian, like what happens? Does this just get snappified or stories ified or whatever? And like, like, you know, I don't know. I mean, Meta already has the surface to kind of do this. They already have the Vibes product they could pull in. They're, they're, they already have the relationships with the creators and things like that. So they may have the head start, whereas OpenAI does not have those relationships. Um, but we'll see. I mean, could be, I don't know how challenging it is or, or whatever for them to do this. I don't think it's that much of a leap, but we'll see. See.

Speaker A: Yeah, yeah, I, I think Tick Tock is not well positioned for this, uh, for the points I'm talking about, which is the Tick Tock design does not actually wrap the magic moments correctly. And do they specifically design that experience to lean away from the social graph?

Speaker C: De.

Speaker A: Emphasize it right. Where actually we might be circling back to where the social graph is actually really important in, in this type of experience or a new type of the cameo graph. I don't know what you want to call it. Ravi, what do you think?

Speaker B: Yeah, I think another thing that's going to happen is they haven't quite figured out creator product fit yet. So some of them are good, some of them are more, you know, kind of personal. Um, and so they've got a lot of social value. But I think what's happened with each one of the platforms, TikTok, Instagram, vine before it is there's a set of creators that rise up, this is true with YouTube as well, who just really understand the form format better than everyone else and create can create things that large scale people find, um, entertaining. And so I think what they've done right now is they've cloned TikTok and ultimately if I'm going to scroll through 10 videos on TikTok or Sora, I'm probably going to be more entertained on TikTok right now than I am on Sora because they just have a bigger inventory and the people on TikTok have really mastered that as a creative medium. So they need to figure out like, what is the recipe on Sora. One of the things I think is really interesting here is because they're giving such powerful production tools to a much, much wider set of users, I think there's going to be sort of an Opportunity for a new era of creators to um, find the ability to create things that are really unique. An example here that I think makes a lot of sense is Roblox. Like if you look at the top games on Roblox, they look nothing like the top games for AAA game developers, right? It's not Grand Theft Auto, it's not Call of Duty, it's things like, like Adopt Me and Brookhaven, um, fashion shows and Life Sims. Because the people who are creating are in much closer proximity to the people who are consuming. A lot of times it's the older brother of the person who's creating for their younger sister or their younger brother. And here I think what we're going to see is people are much closer in proximity to the age and the psychographics of the people will be consuming are going to be able to create cinema quality content that is just purpose built for that particular demographic. And once that creator product fit starts to happen, whoever gets that really right and nurtures those creators, gives them the ability to monetize, gives them the ability to create an audience, is going to grow very quickly in this space. Um, so I think Sora has the right mechanisms there. But you know, so far the inventory is pretty light. Of course they just launched, um, and the quality is not quite there, but I think that can change pretty quickly.

Speaker A: All right, I think we got a wrap there. Thank you two for joining. I will see you on Sora, buddy. Brian, uh, you might regret that, but uh, I will see you there. Thanks guys.

Speaker C: Oh, uh, bye Brian.

Speaker A: All right everybody, thanks for joining for the episode. I'm Brian Balfour, founder and CEO of Reforge and today I was joined by Ravi Mehta and Adam Fishman. We hope you enjoyed all of those insights and, and when you're not spending all of your time on Sora generating hilarious videos, I do recommend Please check out reforge.com in the past year we built an entire AI product suite for AI native product teams. We've consolidated much of the product development process into one suite with Reforged Insights that helps you aggregate, analyze and act on all of your scattered customer feedback. Reforged Research, which helps you run AI interviews and surveys so you can capture net new insights at scale scale Reforge Build, which lets you prototype AI features on top of your existing product and design styles and system. And our newest release, Reforged Launch, which gives you feature management infrastructure you need for AI products. On top of that, we provide the training through Reforged Learning to help your team go through deep AI transformation rather than just surface level adoption. We want to help builders go M from Insight to launch in one set of tools as fast as possible. Check them all out@reforged.com until next time. Time. We hope you enjoyed the episode.