AI Data Wars & Claude's Growth Loops: Who Owns the Fuel & Wins the Race?

Speaker A: Hey everyone, it's Brian Belfort. And welcome back to another episode of Unsolicited Feedback. I'm joined today by my co host, Fareed Mossavat. And in this episode, we take two big swings at bubbling topics. The first is the data wars. We tear into the escalating fight over who really owns AI's crown jewels, the data. From publisher lawsuits to API chokeholds from Slack and Atlassian, every publisher and platform seems to be slamming the gate shut while AI giants keep racing, ramming it open. We'll unpack the legal skirmishes, the product strategy, the API throttles and Cloudflare's bold default off move. Plus why all of this could shove even more power toward Google and OpenAI's of the world. The second big topic we dive into is Claude's artifact growth engine. We flip to the Anthropic's latest launch, AI powered mini apps called Claude Artifacts. Think mini apps that run on Claude, use Claude's authentication and tap the user's own API quota. But they set up a unique growth engine of self perpetuating growth loops. We break down the growth model step by step. The unique growth loops that were enabled, what constraints they will have to work around and the possible results if it all works. Grab your coffee, hit play, and let's turn these insights into your unfair advantage. But before we dive in, while the data wars rage and the platforms wrestle over access, your own customer data is just sitting there begging to be turned into an advantage. And that is exactly why we created Reforge Insights. You can pipe in all of your support, tickets, reviews, calls, everything, and let the AI surface patterns faster than you can say robots. Txt. No more manual tagging marathons or staring at dashboards that pretend to be answers. Insights flags the churn bombs, pinpoints the upsell gaps, and helps you decide which feature to ship next fast. Join the teams turning chaotic feedback into revenue fuel@reforged.com insights. I hope you enjoy the episode. One of the things that you've been paying attention to is it seems like every major player has realized the value of their data and they're going into defense mode. Can you just give a quick summary of the players that have made defensive moves around data and, uh, what some of the differences are?

Speaker B: Yeah, so I think there's sort of three categories of ways in which companies are trying to defend their data of various kinds. The first is legal. So this tends to be content and media companies using legal methods to sue or otherwise file injunctions, et cetera, against the use of their data by major LLM providers. So we have book publishers suing anthropic, uh, they lost that case. We have the New York Times, we have the BBC and perplexity. There are a number of these cases but they're all roughly the same which is they claim that these LLM providers are using their data without permission to train and that that violates some set of statutes. I'm not a lawyer but probably mostly around copyright, fair use and those kinds of things that's been going on for a while. The second is regulatory so trying to convince regulators to do something about this. This tends to be around things like monopoly, antitrust. Google tends to be the main that's being fought against on this. It's just an extension of the long time search wars with Google around their monopoly power, their use of things like AI answers, AI mode, traditional you know, search interventions, things that prevent traffic from hitting uh, you know, and websites and keep people on Google either with paid uh, opportunities or things they get paid for or in their own like organic results, things like you know, shopping or you know, location, AI answers, et cetera. And then the third that I think is the most interesting right now is what I would call enterprise terms of service or API levers to prevent not just the major LLM providers but primarily enterprise search. Glean seems to be the one that a lot of these companies are fighting and defending against right now. So this is examples like Slack, Atlassian and now I just read that Figma have all either throttled, pulled back or changed their terms of service or throttled their APIs in ways that make it hard for these centralized search providers primarily Glean, but also many other players and also Google's products from accessing and being able to index their data in meaningful ways. I think there's one last one which is sort of a combination of those things m that we should get into which I think is is really connected to the first around copyright and media companies which is Cloudflare just announced that they are default off for all their customers on their setting robots txt and other crawling permissions to prevent LLM um companies and AI bots from being able to crawl their customers data by default. And what they're really trying to do is put a layer in between all of that data and the LLM scrapers to put the companies that create that content in a position to be able to charge or allow access if they believe it's valuable for them without it being the default which is open by default which has been sort of the way LLMs worked, you know, really because no one knew they were doing it at first, but also because they didn't understand the trade offs. And the primary value proposition of the Internet and search for the past number of decades has been I put interesting data out there, Google or other search engines, crawl, index and rank that and I get traffic in return which I can monetize on my own property in some way. And LLMs have put that on its head and now we have a bunch of different things that those content providers are doing and data providers but in order to try and change the equation or disrupt uh, things a little bit in this intermediary period. So right now I've just. This all kind of sums itself up. This isn't a unique framing but there's a data war. There's a data war between people who have interesting data, either systems of record on the enterprise software side or content and media on the more general knowledge side. And the companies that index, scrape, modelize and create end user experiences which I would call a system of action layer the places where people are actually doing and taking action on things. And so there's a lot going on right now. And the question I wanted to pose to you Brian, let's first talk about enterprise data because I think it's the more emerging one is do you think it's a good idea for these companies to block things like Glean or maybe Gemini or Microsoft products copilot from being able to index, access and store their data to provide these great end user experiences? Supposedly I haven't used Glean, so I don't know if it's great, but let's assume it's great.

Speaker A: Yeah. Well, just a first couple of quick clarifications. I remember on the first kind of tier that you talked about around the lawsuits you said they lost. But to clarify who actually lost in specific of the book publishers versus Anthropic, because my understanding was that was a little bit of a mixed. It's amazing. So like my understanding and correct me if I'm wrong is that um, basically the ruling found that the use of those copyrighted works as inputs to training was okay, but the problem was is that Anthropic used a library of pirated content versus buying the actual books. But had they just bought the books it would have been fine. And so it would have been fine. Yeah. And so this kind of really second

Speaker B: they, I uh, think they were disallowed from keeping copies of those books around. So my understanding, and again I'm not a lawyer, is that this stuff falls under fair use. You buy a book, you're allowed to use it for things. And that it's transformative, that you're not just taking the book in order to resell it or give it away for free, but rather that the model, the process of training a model is transformative in some way and creates unique outputs, unique things. And I think that's an interesting take. I don't know if it's the right one, but certainly is, I think will be set some precedent in terms of other forms of media as well. Images, video, et cetera.

Speaker A: Uh, yeah, it certainly starts to establish the precedence and starts to clarify like the inputs versus outputs, right, that it's pretty clear they're paving the way for the inputs to be used for training. But a lot of the uh, clarification will come around the use of the outputs, which is if it starts to output Mickey Mouse and Disney characters, I think that's one level. But the second level is then are people using that to monetize in some way, shape or form. That's where the real damages come in. And uh, it probably starts to separate those things. In the enterprise case, oof. Like I read a lot of these moves as defensive moves in sense of, hey, we know this, this data is valuable. We don't know exactly what to do with it. Therefore we're locking this down temporarily until we really figure out the rules. I imagine what's going to emerge is some form of taxation around getting this data out of the system. And I already think you see that in other places, right? Like amplitude as an example, as a product analytics tool has this thing, I believe they call it cohort sync, where you can kind of sync your data with other platforms, but the users have to, the customers have to pay for it, they have to pay on the cohort sync. So once you get your data in there, they're essentially taxing you. They're monetizing to get it out. And that disincentivizes massively broad use. The customers actually have to think about where they start to uh, export that data and it puts up controls in the system. I think some of these players have looked at folks like Glean as an example and have started to realize like, wait a second, I'm in position to capture that uh, type of business and therefore I'm gonna, I'm gonna pull the levers in order to do that. Certainly Slack is in position to enable a very similar type of use case so is it last year, the vision

Speaker B: for Slack for the longest time was Always all your stuff will be here and you'll be able to search across it. And now that's not 100% how it's been used, but it certainly is. Glean is a clear competitor to that long term use case.

Speaker A: A hundred percent? Yeah, 100%. And Atlassian's been making moves around this with their, uh, Rovo chatbot and you can like integrate a lot of the same sources and all those pieces. And then also you see, you know, OpenAI and anthropic doing these preferred partnerships with uh, some of these systems of record. Uh, they launched deep research with HubSpot and a few others. And so I imagine there's negotiations going on in the background. So Atlassian is probably also sitting there thinking, hey, like this, I could monetize this as well. And so I need to put some, some safeguards and protections in there. It's going to be interesting to see like how customer forces play into this. There's kind of two opposing forces that I see here. One is, which is customers saying, hey, you know, wtf? This is actually my data. Yes. Why are you leveraging this control? And there's backlash around that. But on the opposite side of the equation, enterprises also don't want to like make their data shareable with a ton of different platforms either. Right. So there's, right, there's both of those dynamics going on which it's kind of talking both sides of the equation.

Speaker B: And there's real to steel, man, let's just say the Slack argument because I know that company better than.

Speaker A: Well, I was going to say like you spent years inside that company. So were you. If you were still in a product leadership position today, what do you think you would be arguing, uh, in the conference room in these discussions?

Speaker B: Great question. So for one, if I were to steal my unlike, why be defensive here? Why is this good for customers? I think is the question that you have to ask first because clearly it's good for the company to be defensive here and maintain your ability to build that kind of experience yourself. Right. Everybody wants to be the central hub. This is the tricky thing. If every company wants to be the central hub, then no one ends up being the central hub because nobody gives their data at anybody else. Everybody wants to be in it in jester, nobody wants to be an exporter. But I think there's a real customer reason for these things. Now whether this is just a justification or a real thing internally, I don't know. But I think why this is good for customers is allowing data from your platform to Be synced, copied and left in some other platform indefinitely is dangerous. Like you as an enterprise customer of Slack. Pay Slack because you trust Slack to maintain the security integrity and privacy of your message. Of your message data. For instance, there are lots of permissions and complications about who's allowed to see what. When you type into a search bar in Slack, it doesn't just go look across all the messages. It knows what channels you're in, what channels you're not in, what channels you're a member of, what ones are private, what ones are public and has like meaningful control. Now look, the logic for that is not extremely complicated, but it ends up being more complicated than you think. And if you just let someone copy all of the data over, how can you ensure that Glean is going to follow those privacy things the same way you would. For instance, if you have message retention set, meaning all my messages get deleted after 30 days in a particular channel. How do I trust if some other company has just copied all of those messages into their own index for, uh, search that they are going to properly delete those messages as well? Now there's some court case two years later and you're like, all my messages are deleted. And it's like, nope, just kidding. Random Startup X with a full copy of all my message data, happens to have it somewhere else when I didn't expect or want that to be the case. Right. So I think there are real security concerns around this. I think the downside is, is that as the company is, it does, it does hurt your customers. You're telling them that you're not allowed to use this other tool effectively and so they have to create some other pathway. Slack's approach right now has been to build, I don't know a ton of details about this, but some kind of enterprise search API that allows companies like Glean to be able to run, um, searches in some unique and interesting way in order to be able to provide the same end user experience without copying all that data. They'll probably charge substantively for that to

Speaker A: both the third party. Yeah, they're going to monetize this to the customer.

Speaker B: They're going to monetize it in some way. I think that's the right play.

Speaker A: It's like it's the enterprise version of what Reddit is doing. Right. With their data. So. Yeah. And they're trying to monetize the hell out of that.

Speaker B: Yeah. I think the thing that stinks is if you play that out and every company does that, they're all going to want you to use their search box and imagine a world where the way you use the Internet is you went to New York Times and search, then you went to BBC.com and you searched, then you went to, I don't know. Yeah, uh, the information and search. And if you were looking for something, you had to like type the same stuff into a bunch of different places and see where you get the right answer. And what customers really want is a single pane of glass experience. They want a single place that can integrate and not just answer questions, but actually take action across all of these places. And um, that's where the customer experience is headed. But if everybody tries to do it, no one's. There won't be an actual single place. Like they will all have some subset. And I'm very concerned about this as a net result.

Speaker A: Well, I think this is just like another reason why I think somebody like OpenAI wins it. They've, they've got the consumer distribution, the pull, the habit and also the funds to basically probably do deals with all the ones that are important. And at some point they have so much of the gravitational pull that the mid and the long tail of people who own this data will just integrate for free and not monetize it.

Speaker B: So uh, the winner ends up being the person with the most market strength in these situations, uh, which is probably OpenAI or uh, Anthropic or Google or you know, and not one of the smaller players, which is bad from m an innovation standpoint.

Speaker A: Yeah. Well, I think we should talk about the Cloudflare thing for a second because this is like an interesting counter move. And so just to re explain it, Cloudflare powers about. I was just looking this up, uh, approximately 20% of websites globally in about a third of the top 1 million sites. And so they sit in an interesting position in the stack. And what they turned on was by default companies that use Cloudflare will not allow LLMs and automatically crawling and scraping for the training. So they're putting in a defensive move for this. So, uh, customers have to go in and turn this off to allow LLMs to start indexing and or training on their model. And they've clearly stated that they want to enable some type of ecosystem or some type of economy to allow some type of monetary exchange to allow for this quickly.

Speaker B: Apparently this is step two of a strategy that started with allowing their customers to click a button. Okay, turn it off.

Speaker A: Yeah.

Speaker B: And they got a lot of positive adoption of that. They got a lot of excited especially, I'm sure media companies and so they decided to flip it to default off, which I think is an interesting move.

Speaker A: Yeah, there's a bunch of interesting stuff about this. I think the big question on my mind, does this work? Like, does this actually have a substantial impact on slowing things down? To be clear, before I explain, my reasoning and positioning for this is that I have been a content creator my whole career, including reforges with a content business. So to see, you know, to have seen this and play, like, just, just to be clear is like, I don't like it like that. I don't like it. And it has 100% negatively impacted my ability to, quote, unquote, monetize or gain any sort of benefit from all of the content that I've personally put out there over, over my career. So that, that's my personal take. But do I think this works? I'm pretty pessimistic about it for a couple of reasons. I think, once again, in isolation, like, looking at this, if I'm an individual player looking to turn this off or to disallow ll to crawl independently, it makes sense. It's like, oh, why would I let them do this without some sort of monetary exchange? But we don't live in that world. We live in this very competitive, uh, environment where people are making choices based on the competitive. And I think it ends up as another prisoner's dilemma. And I think the reasons for that is, I don't know if even though Cloudflare is like 33% of the top million sites, which is some substantial market share, I don't think it's enough. Because at the end of the day, those that are not using Cloudflare, where unless a bunch of other similar players like Cloudflare follow suit, right? And all of a sudden you've got like 80% or something turned.

Speaker B: Every CDN does something like that, right?

Speaker A: Like, but without that, then all of a sudden all these folks that are on Cloudflare are going to start watching their traffic, uh, decline either from Google or these AI search experiences. And then they're going to ask why? And then they're going to end up in this competitive situation and they're like. And they're essentially going to reason them ways to, well, okay, we should probably turn, we should probably turn this on. And, and that's probably where this, all that, that's probably where this all ends up. And so I actually think, interesting, in the short term, there's rather than this hurting the open AIs and the anthropics, the big model, uh, I actually think this Hurts more of the smaller tools using AI. Like I'm thinking about things like Clay or the agent builder Nan, because there's a lot of people are like using Clay for example to set up all these workflows to Enric and content. A lot of that is based off of using AI to crawl websites and other things in real time to enrich this data. But all of a sudden if a third of that enrichment, uh, starts to go blank or is bad, that has a much more significant effect on Clay's value prop than it probably does on OpenAI's or anthropics, which has already scraped and indexed a massive amount of content and is already using synthetic data. I actually think this probably impacts other people way more than does the big LLM players.

Speaker B: I think the other really unique situation here I saw this first talked about, I think in the information is that um, Google is a complicated one because you can't turn off Google. You need them for your normal everyday search traffic. Do you have the fine grain control to say I want to be in AI answers but not in Gemini more broadly or I want to be in this thing but not that thing. And how much trust do you have that this isn't going to impact your search rankings in some meaningful way? Look, I don't think Google's being malicious, but these things are complicated black boxes with a lot of little details and publishers tend to be very, very careful about what they are and aren't willing to do, especially when it comes to their existing search traffic which is already probably under fire in some way. You maybe can block OpenAI because you don't have much traffic from OpenAI today and hope that it gives you some leverage to work with a bunch of your peers to, you know, fight for a good content deal. Or maybe cloudflare will create some sort of pay per crawl, uh, system that's automated. But Google's always going to be a problem. So does this just maintain or enhance Google's existing monopoly on search? I think is another thing that I would be concerned about. It hurts the small players but also maybe helps Google because they're the one you're going to be the most wary of shutting off or controlling. And then last robots Txt is just a handshake agreement. It's not legally binding. I'm sure there are ways in which there will be regulatory moves around this, but ultimately I don't think the LLM providers are going to pay per crawl. Like I just don't think they that the incremental value of another page is all that high on their training and model. So I think it's going to have to be more complicated than that. Uh, and I think ultimately needs to be that we need to build the incentive. It can't be. The content itself does not have much value. You have to find some other way to build a cohesive ecosystem of incentives that make it make sense for you at Reforge, for instance, to put interesting stuff on the Internet.

Speaker A: Yeah, look, I think at the end of the day, like I said, uh, unless and I don't know what the rest of the CDN market looks like and how fragmented it is, so I don't know how hard this happens, but unless everybody else follows suit and we all of a sudden get to a super majority of uh, all the LLMs being blocked, it just doesn't feel like it works as a broad based mechanism.

Speaker B: And I think going to default, default off, whatever you want to call it, default block for. This was a really interesting move by Cloudflare to have it be the default is. We're all in this together which I think solves some of the prisoner's dilemma problem. But yeah, they don't have a monopoly on CDN and a lot of people don't even use CDNs. It's maybe better than nothing for some level of these players who are concerned. But I think long run there have to be organic incentives.

Speaker A: Yeah, fair enough. Well, should we move on to Claude?

Speaker B: Yeah, let's move on to Claude. Brian, why don't you tee this up because this is something you've been thinking about a lot over the last week is Claude's new, what are they calling it? Apps? Is it that simple? Cloud apps?

Speaker A: Yeah, like, like artifacts or um, uh, like AI powered apps as part of artifacts. This, this announcement came right after we started talking about, you know, new platforms. Emerging from their announcement was um, Claude can now create artifacts that interact with Claude through an API, turning these artifacts into AI powered apps where the economics actually work for sharing. So a couple quick things here before I go on is so they always had artifacts in the sense of uh, like document creation but or like co generation but now what they're saying is that uh, these artifacts can also create many AI powered apps. From my experimentation it's not as powerful as something that you might create with a uh, replit or a um, V0 or a lovable.

Speaker B: It's definitely an MVP. Like yeah, you storage can't make external API calls. It's basically like little cloud apps.

Speaker A: Yeah, yeah. But the second thing that I found, found Way more interesting was this next part which was they were essentially detailing out what the value exchange was for building these on Claude. And so they said when someone uses your Claude powered app, they will authenticate with their existing Claude account. Their API usage counts against the user subscription, not the creator's subscription. So you pay nothing for their usage and no one needs to manage API keys. So what they're essentially saying is, uh, create this app on Claude authentication will be taken care of for you. We will charge the end user against their usage so that you don't incur those charges in exchange. Now what I think is unsaid here, and part of the reason I think this is interesting, goes to our previous conversation, which is now there isn't this incentive mechanism for people to drive users towards authenticating through Claude and using Claude and through that usage. That gives the advantage of Claude to capture more of that usage, more of that memory. Which goes to all of the things we've talked about on previous episodes which might help establish their moat. But the thing, the next thing that I thought about was a couple things which was this starts to set up a pretty interesting and unique growth model and types of growth loops. So growth loops is something that we've gone into depth in on at Reforge. They are compounding systems uh, that create growth and there was a lot of work done by Casey Winters and Kevin Kwok to categorize and label and go really deep on all these different types of growth loops. Now the thing that this immediately sets up is uh, a new type of growth loop. It's almost a content type of loop that were used by tons of tons of different players. Where a user comes in, they create some content, that content is distributed in some mechanism which drives more new users and engaged users in the cycle repeats. And this is a type of that, uh, where a new user comes in, they create one of these Claude powered applications, they are then probably distributing that application either internally or externally which is driving more new or more engaged users to Claude who then some percentage repeat the step and create their own Claude powered apps. Pretty interesting. But I think the second part of this growth model is even more interesting, which I'm not sure I've ever really seen before. I guess you could say in, in some other platform cases, maybe like a Shopify or something else, which because of this Claude authentication, each of these apps that are created on Claude have their own mini growth models built into it that also has their own ecosystem of loops that drive usage towards Claude. So if I go in there and I create some type of mini collaboration app, like a personalized version of Asana or something we specifically want to use for some workflow inside of Reforge. Then what that has is all sorts of its own other growth loops where I'm inviting and collaborating right. With other people in my company, which is driving people to my Claude powered app. But they're authenticating through Claude and so they're either signing up and engaging in Claude indirectly. And so you take this to the extreme and you could say, okay, well, if millions of these Claude powered applications existed on Claude with Claude authentication, each with their own set of unique set of growth loops, that potentially creates an incredible compounding machine.

Speaker B: Right.

Speaker A: That I think is a little bit unique in this AI world.

Speaker B: So it's very different because they're apps. Right. And they're dynamic and they have their own mechanisms to drive user growth and bring people in. Like, I could be really good at invites and I'm helping Claude at the same time, which is, I think, the uniqueness of that second growth loop that you're talking about. But in a lot of ways it looks a lot like Eventbrite, which is basically I'm a event, uh, creator, I'm running a party, I'm going to go bring all these people because it's good for me to bring all these people to my awesome meetup event. And every single one of those to buy a ticket needs to sign up for an Eventbrite account. So it's like they're not coming for Claude, they're coming for this app and as a result become a Claude user. And now if you do it right, you can sequence this into other kinds of network effects where you actually have cross distribution, where, because I use Brian's cool app, I went to Brian's meetup event. I also see there's a stand up comedy event or whatever on the Eventbrite example or some other app that I find valuable. And now all of a sudden each app is growing more quickly because there's an ecosystem of users around it.

Speaker A: Yep. And you can already see that like very V1 versions of this, they have a page now called Artifacts. It's almost like a mini app store. It's really templates to get people started on these applications. But I imagine once they have enough of this usage, they could very easily turn this real estate into almost its own mini application store and start to enable some of the dynamics that you're talking about.

Speaker B: But I think it's smart not to start there because then the promise isn't hey, we're going to drive distribution for your app. Their initial value promise right. To a, uh, creator is we're going to make it really easy for you to host, manage and build some tool that's really valuable for you and it make it easy for you to bring users onto it. It Right. Without having to think about authentication API keys cost to you. It's basically free. And then you can sequence that into the App Store concept later. I think the question mark that I have, and I'm sure they will solve this, is who is an artifact a fit for like, who's going to build artifacts?

Speaker A: Yes.

Speaker B: Can't really charge money on it. So it's got to be like, cool, free, interesting stuff. Yet there's no persistent stuff. It mostly it's for apps built that use Claude, uh, to do interesting things. You know, it's, it's relatively limited. I think that's okay. But then also it's sort of like who's going to want to give away all their authentication to enough to Claude. So there's some subset of people for whom I think it's utilities. Like, cool, interesting utilities. For instance, I built an app for one of my kids to do SAT studying. Kind of a pain in the butt in something like lovable. Uh, because I got to get an Open API key, OpenAI or Claude API key. I got to write all the code to call that API and do the prompting and do the prompt engineering and go back and forth. I need to create accounts and all this sort of stuff. I'm not planning on charging for it. I could see that as being a really interesting fit for a Claude app because it's like, hey, I want to build a SAT tutor. Here are the criteria, here's who it's for, here's the tone of it. Make it really simple. Boom. Sign in with Claude, just go use it it. And now it's like, look, I paid 20 bucks for your cloud account. Go ahead. Hey, you can share it with your friends. And it doesn't cost me anything because right now that app, if she were to share it with all her friends, would cost me money. So I think there's an interesting. It's like starting as a toy and I think that's a great way to get going on and bootstrap network effects. The question is, is there enough there? And that we don't know yet. So I am excited to see what artifacts get built. I'm excited to play around with it and try to build one. I am curious to see where it goes.

Speaker A: Yeah, I Just want to highlight the point that you made, which was the initial value prop they're starting off with, which is, isn't about discovery and distribution. And that's interesting because they can't play that card assuming that OpenAI will be coming out with something similar because ChatGPT has I can't remember how many more millions but uh, like 10x plus more millions of users, they had to go to market with a very different value prop to incentivize that creation. It's an interesting comparison and pretty smart creative entry point in value exchange that, that I wouldn't have necessarily thought of on the surface. I think what you're talking about is that every type of growth loop has a uh, constraint. And depending what type of growth loop you're using, the constraints tend to concentrate and follow patterns in different parts of the loop. And certainly in a content loop the constraint is almost always on the creation step, uh, and how much you convert those new and engaged users to that creation part. That step can have very high conversion but low like distribution per piece of content or it has very low conversion but very high distribution per piece of content created. But, but those numbers have to net out to make the loop a compounding system. You can't end up with low conversion and low distribution per um, content. And so I think you're right. The question is, is once you whittle down the, the Personas and use cases that this can and should be used for to generate an AI powered application versus using different more powerful independent platform like a V0 or all the way to like something like Cursor, the question is do those numbers work out? Do the percentage of enough users actually create these applications? Um, and are those applications of the type that get enough of that, that next step, that distribution to make all of these economics work? Because certainly even if you to your example, if you're creating mini M education app for personalized for your daughter or your son that doesn't have high distribution effect, you're creating that for one or

Speaker B: two people, hopefully 100% conversion on two, you know, people. Right. Is that going to be the kind of app that's being built here?

Speaker A: Yeah, that, that's right. So certainly they can build things to decrease friction for that creation step or enable a broader set of use cases to start to make those numbers work. But I agree like that is going to be the biggest question of whether this net's out for them. And they're certainly also in a position of which isn't in this initial release, but similar to what we've Talked about with ChatGPT and OpenAI is that they have a ton of integrations as well. So to app creators they could start to enable, hey you can use your end users integrations and memory context now

Speaker B: I don't have to manage your Google not exactly like oh, you can connect your Google Drive and now we can do things with it. I think automation and actions will make these apps very, very powerful and the fact that they're built on top of Claude could start to be really, really valuable. I mean my understanding is this and. Is that how you say it? I have no idea how they.

Speaker A: I believe so. Yeah, yeah, yeah.

Speaker B: Nathan and Aiden uh, is kind of blown up like a lot of people are using it. I've been tinkering with it a little bit here and there but honestly I'd much rather do that inside of the place that I already do all the other things I do. So the, and, and be able to customize it. So Claude's code generation is really, really good. So if I could do a little bit of code and a little bit of you know, uh, drag and drop automation take actions. I think it could be really interesting for micro apps. Like I don't think there's going to be million user Claude artifacts, at least not the way they've constructed it today. But like you described, you don't need that, that for this to be really valuable from a growth perspective for anthropic for Claude.

Speaker A: And I think one of the other dynamics here is these applications. It's not like you can access and manage all of the users and export all those users from my understanding yet. And so there's this almost like does it continue to be a closed platform or more of like a substack platform? And I imagine we'll see some differences of these platforms but I think what we're starting to see is this AI powered application marketing it the spectrum of things. You kind of have these mini AI Claude powered apps which feels like the longest of the long tail. It's the lightest weight but the easiest, most personalized use cases. You've got the middle of this tail now which is the lovables and stuff of the world. And then you've got the heaviest, most powered like applications probably being built with the cursors and the wind surfs of the world. But this AI powered application creation market, it feels massive and you can slice it that way. Plus on a category basis it's you

Speaker B: uh, start internal tools, external tools.

Speaker A: Oh gosh yeah.

Speaker B: Taking action. You've Got size of product, how standalone is it as a business, how integrated does it have to be with your existing code versus being net new? We're seeing this market chopped up into a million different chunks and pieces,

Speaker A: and

Speaker B: it feels it all comes out of the fact that these things are really good to write in code and the market for code is massive.

Speaker A: Yeah, yeah. Each one of these segments feel like they can be $100 million revenue businesses. Right. And so, uh, it'll be interesting to see whether this actually lends itself to more of a monopoly duopoly, or if it feels more like SAS, where there's a higher quantity of billion to $3 billion businesses or very few hundred billion doll.

Speaker B: I like what they've done with making this very, very constrained. Now, the question mark is, is it over constrained to be able to build anything meaningful? But I like where they've started. I like that it's pretty small and narrow. I like that it really encourages tinkering and ease of use. And they're basically turning code into content that can be distributed, shared, and drive new users, which I think is unique and new and very cool.

Speaker A: All right, everybody, that's a wrap. We hope you enjoyed the episode in today's deep dive on the data wars and Claude's artifact engine sparked a few aha. Uh-huh moments. Please do me a solid. One, share this episode or mention us on LinkedIn, uh, with a teammate or anybody battling the AI chaos or trying to plan their AI growth engine. And two, give Reforge insights a spin. If you're ready to turn all of that raw customer noise into roadmap ready intel, it's the fastest shortcut from what's going on to here's what we need to ship next. I'm Brian Belfort. That was Fareed Masavat, and you've been listening to unsolicited feedback. Until next time, keep shipping, keep learning, and stay one pivot ahead. We'll see you next time.