Future of Aggregator Market

Hello and welcome to another episode of App Stories. Today's episode is brought to you by Mercury Weather and Claude. I'm John Voriz and with me is Federica Vitici. Hey, Federica. Hello, hello, John. How are you? I'm doing really well. We've got a fun topic today . We really do, I think , because you know we've talked in the past quite a bit about LOMs in general and kind of the big players, you know, Claude who happens, you know an,thropologic and claud, which happens to be a sponsor this week , and open AI's, chat, GPT, Codex, all that kind of thing. But today we wanted to talk a little bit more about apps that aggregate various LLMs, whether it's from those big players or some of the open source things that can be run locally , some of the other specialized models that are out there . And there are a lot of apps out there in this category. It's a very fast moving category, I think, and one that has already evolved quite a bit and attracted some bigger players. So we thought it would be interesting to talk a little bit about those. And so I guess maybe the first thing to do, Federica, before we actually get into some of the apps that bring together all these different models, we should talk a little bit about what these apps are and sort of the general pros and cons we found from using them ourselves . Yeah, I think it's an interesting market, right? These services that basically they let you they make you pay another subscription to have access to frontier models usually and maybe some of the open source ones as well . I think lately we've seen a lot of buzz around the new GLM five point two model from ZAI out of China. And obviously there's chemi chimi k two six and K two seven . All of these aggregators as we call them, I mean we're referring to the likes of perplex maybe and notion AI to an extent, even Raycast is going to be one of the apps that we talk about. The idea is that rather than using those like the official clients and the official chatbots and apps from those companies, you're using a different frontend, a different client that aggregates a bunch of these models and usually they let you pick between these models, right? And so right there, I think the first advantage maybe of this approach is that you usually get like a list of different models that you can try and they make you choose. They let you chsoose between different LLMs with different versions of those models, different reasoning levels of those models. And so you get plenty of choice. Usually with you take a look at perplexity. For example, you can choose Germani, you can use GP T five point five, you can use opus. And the same is true for Notion AI, for example, which even lets you use some of the Chinese models. And Raycast is following the same approach. And some of these also let you use local on device models , most of them don't . There are plenty of these services out there, plenty of apps on the app store that some of them are quite scammy to be honest. Like they make you pay an expensive subscription to give you like some credits for GPT five point five. Yeah, there are a lot of different business models which I guess we can get to in a second, but you can, I mean as I think like perplexity and Notion AI are each have maybe you know half a dozen or so different options. Whereas things like Raycast, I was looking at their website today, they advertise that they've got over thirty two different models, which to me is too many, but I mean we can get to that in due time. That is maybe also the downside of these products, right? The yeah, too much choice can be. I mean, it's really confusing, right? If you are an average user and you don't know the fine differences between what's GPT five point five, I guess it's better than five point four because it's a bigger number. But what's the difference between four point six sonnet and four point eight Opus and what's fable five rest in peace if it ever comes back . Like it's already quite challenging to keep track of these models on a monthly basis . And when this when these aggregators present you with the menu with thirty or fifteen models , if you're not tuned into the scene, it can be confusing . However, if you're a power user of these tools, it's pretty great because like you just pay a scene like what's the reason why these companies and these aggregators keep popping up? Because for most people, it's a pretty good deal. Like instead of signing up for ChGPT and for Claude and for Gemini, you just sign up for the one aggreg ator and you can use all of those models . The difference from my perspective and maybe the biggest downside of these aggregators is when you're using these models through the aggregator, you're not using the official hardness or the official ecosystem of tools that are available for those models. When you use GPT five point five through Raycast , you are going through the Raycast prompt. You are going through the Raycast harness. You're not going through open AI, which also means when you're using GPT five point five in Raycast or Perplexity, you don't have access to the, I don't know, advanced voice mode of ChiGPT. So it's not just the model. And I think this is may arguably the most important factor to consider here. It's not just the model that you're using. It's the surrounding corollary of tools and integrations and features and design that is completely locked into the individual experiences, right? Right, right. And I think that's a really important thing to keep in mind because over time , you know, one of the things there was, I think there have been a lot of predictions at times that models would become a commodity and that one would be as good as another . And while that may be true to a degree with the underlying model, if it's a frontier model , it's not true necessarily when it comes to all the other pieces of the puzzle that are integrated into tools like cl clawedawed code , you know, GPT five point five and Codex and all of those things. And so if you're using a tool through one of these other services, you're not going to have the same access to those tools that you might want. Now I think that just I think that as a result it helps to know what kind of user you are before going into these tools because I think that these tools that aggregate lots of different models are good for lighter use and more of a chat bot type of experience more than anything else. That's not to say that they don't integrate with your system to a degree. Some of them, like we'll mention Alter , try it works with a lot of the Mac apps themselves and uses some of the accessibility tricks that like Codex uses in order to use things like calendar or reminders or something like that , but they're also more a more generalized approach to models as kind of a commodity and layering on their own set of tools that you have to then kind of decide, well, are is what? This aggregator is laying on top. Is that better than what I can get from one of the large frontier models themselves. Yeah, most of these, most of these apps are chatbot style apps. Perplexity is maybe the only one that is also considering the desktop agent trend with their personal computer that we're going to talk about right in the Notion Tutor Group with its Notion , but that's very locked into the Notion ecosystem, right? It's not a desktop agent. I guess what I'm trying to say is that we haven't seen yet from the aggregators, with the exception of perplexity , the equivalent of something like the Codex app , but for productivity, that lets you jump between models. There are plenty of aggregators in the coding space, like in the programming space. I mean, take a look at open code, for example. That's essentially it's a terminal UI is like clog code or Codex , but it lets you it lets you use any model you want. But that's very like it's not the type of consumer app that to a degree Codex can be, right? Right. Exactly, or something like coworker from something like cowork, right? Yes, yes. Yeah, absolutely. And I want to touch again real briefly on the pricing models here too, because one of the downsides I think, of aggregators is it's a little bit it can be confusing. I mean, there are some apps , some apps that are literally just pay once apps. And you might say, well, how can that be if these various agents have API costs and monthly subscriptions and that sort of thing. Well, it's because those are bring your own key, which is, you know, means you go and you sign up for something like open AI's subscription or anthropic subscription , you get an API key and you dump that into this app and then you have it available. But here's the thing is that those API costs tend to be higher. You get less for a higher price than you would from one of their consumer subscriptions . So there's a lot of variables there . Others like Raycast will charge you a higher monthly fee to bring in a certain amount of usage of each of these models, but it's not always clear how much you're getting for that added price . And then there's also credit systems that exist out there , which I personally am not a fan of where you're where you're using up credits that you buy in chunks . And again, that's something that Notion does, but it makes it hard to, I think, makes it hard to kind of figure out how until you've used it for a while, how much it's actually going to cost you kind of on a weekly or a monthly basis. So there's a lot more complexity I think in terms of the pricing models than simply saying, Oh, I'm going to sign up for OpenAI's twenty dollars subscript ion. Yeah . And the challenge for all these services is that when you are signing up for a Ch BPT subscription or a cloud subscription , those subscriptions are heavily subsid ized to the point where even if you're paying like the highest tier of a Ch apy subscription, the two hundred dollars a month plan , you can actually potentially run up to ten thousand dollars of inference on that two hundred dollars subscription, right? Yes. So in other words, if you were to use the API instead of your subscription , you'd hit ten thousand dollars instead of two hundred . And you can just run some tools on your Mac to count the token usage of Codex and it'll tell you where if well if this were an API subscription you would have spent ten K this month . Instead instead you spent two hundred . So that's the challenge for these companies that they cannot subsidize anything here. They can give you a base level of a subscription. For example, with Raycast, which we're going to talk about later , I quickly ran out of my allotted usage of fifty messages for DPT five point five and fifty messages for Claudopus four point eight. And that's it. Like it doesn't matter. And obviously like it gets complicated because then open AI it actually lets you connect your subscription to other tools, right? But anthropic doesn't because anthropic really wants you to stay in the anthropic ecosystem. So it's a tough market out there for these aggregators. And I think some companies are doing their best to give users a good deal. But I think especially if you're a power user and you think, oh, well , I'm just gonna sign up for Perplexity or I'm going to sign up for Raycast and I guess I'm going to have unlimited messages with GPT five point five or Opus. That is not the case. That's not going to happen. Right, right, absolutely. This episode of App Stories is brought to you by Mercury Weather. Mercury Weather is the thoughtfully designed weather app that shows all of the essential weather details at a glance. It has a gorgeous colorful interface that dynamically adapts to the conditions , with a warm orange palette on a sunny day, icy tones on a cold day, and deep blue for a rainy night. Mercury uses a glanceable chart layout to present the hourly and daily forecast in a way that feels intuitive right away . A really cool feature for when you're traveling is Mercury'sp T friorecast . It's a feature that automatically shows the weather at your destination right in your daily forecast timeline. That way you always see the weather where you're going to be not where you are. It's summer here in North Carolina and that means thunderstorms. And what I really like about Mercury Weather is that it has great radar that lets me see where a storm is going and where it's likely to head. Of course, they also have severe weather warnings and all that sort of thing too, so if I go out for a long walk, I know whether I'm likely to get stuck in a downpour. Mercury weather handles other serious weather too. Mercury offers fire maps, storm and hurricane tracking with live positions, forecast baths, cones, and intensity , plus there are widgets so you can keep up to date on a specific storm or the closest one right from your home screen. Mercury Weather's gorgeous interface makes it a delight to check the weather every day , even on gray and rainy ones. The app's business model is simple. No ads and no selling user data. Mercury is available on the iPhone, iPad, Apple Watch and Mac, and you can download it, use the standard features for free, and upgrade to Mercury Premium to unlock all the features. Just go to mercuryweather. App slash app stories to download Mercury Weather Now. Use that link to let them know that you came from App Stories. That's mercuryweather. App slash app stories. Try it out, get all the standard features for free. Our thanks to Mercury Weather for their support of the show . All right, Frederico, let's get into talking about some of these apps because there are a lot of them out there , and I think one of the ones that we want to talk about is Raycast because I know that you've been diving back in again, and I think Raycast is one of the most interesting because, you know, it started off its life as a very developer oriented launcher app for launching both applications on your Mac as well as just doing various launch ing scripts and doing other things with APIs . And they added on an AI layer, a couple of different options, I think, in their pricing schema that you can sign up for. And each gives you a slightly different amount of tokens and various models and different things that you can do with them. How has been using Raycast so far? So I've been checking out Raycast again on IOS to prepare for this episode and what's especially interesting is that a few days ago on IOS they added support for system integrations as well as MCP directly on IOS , which means it's Raycaston IOS is now one of the few experiences that let you use system integrations for Apple reminders and Apple Calendar and Health in addition to connecting any custom MCP you want ? I mean you can literally count these tools that have mobile access on one hand. I think it's Raycast, clawed notion and perplexity, and that's probably it of the big players. Yeah, those I know of yeah. Yeah . And so they have two layers of the subscription. I think they have the nine dollars ninety nine cents a month and the twenty dollar a month called Advanced AI. And Advanced AI is the one that lets you use GPT five point five OPUS four point eight, like those big heavy frontier models . Although as I mentioned, you will run into limits quickly. I think it's fifty messages seven days for those models. So in just a couple of days, I've already run out of GPT five point five. First you have and four point eight messages. However, it's a really nice experience. I think they've done a really nice job . They have a settings page where you can go in and you can see all supported models organized by company, by provider . So you will see all these models for Google, for open AI, for anthropic and so forth. They support some of the Chinese models as well, including GLM five point two and Kimik K two six and Kimi Code two point seven . You can go in, you can choose which ones you want to see in the model selector. So if you're not interested in the meta models or in the Gemini models, you can just turn them off in settings . They have a really nice UI for the model picker and they have a really nice UI for the thinking selector. There's a sl ider that lets you choose no thinking, low, extra high , all those things. What's especially nice here is that they have a collection of built in popular MCP servers, but you can just go in and enter any custom MCP URL you want. And actually what's even nicer is that they train that the in their system prompt I guess they tell the model how to use Raycast itself . So you can go in , open a new chat and say, hey, can you set up the Notion MCP for me? Paste the URL of the Notion page and the model running Raycast will bring up a permission prompt saying I'm going to set up this MCP for you. Just click through the various permission prompts and you're good to go. So you don't have to deal with the settings at all. It's just doing it for you. Exactly. You can go in and deal with the settings manually if you want to, but you can just ask any model . And I guess that's part of the system prompt that Raycast gives to the models . There's a settings page where you can approve the classic like do you want to always be prompted for tool use? Do you want to always allow like that kind of thing? I think it's nice that Reycast I think it's interesting that Reycast has built this ecosystem of features that are mostly consistent across iOS and MacOS, right? You create an account, you sign up for a subscription that syncs across IOS and MechOS. You have these chats, you can pin them, you can access them from the home screen with widgets. They integrate with the live activity. So if you start a chat, but it's taking a while, you can close the chat, it goes into the dynamic island and it keeps working . This tool usage on IOS is really nicely done. You can inspect to what's going on every time you call an external tool like notion . And it integrates just like Claude does on IOS with reminders calendar and health kit. So it's a really nice way , especially if you use a small model that is, however, intelligent like Sonnet, for example. You can say, Hey, can you give me an overview of my health stats today? And it's going to call on healthcare and it's going to tell you, well, okay, here's how much you walked today. Here's your resting heartbeat and so forth. Sure . I think Raycas can be a little confusing because of the many models they support. Yeah. To me, I think it's a little too much is kind of where I'm coming back because I think I don't know I think there's diminishing returns after you get beyond kind of the bigger names that we've already mentioned throughout the episode . Yeah . Yeah, it can it can be it can be a little too much especially I wish there was like I don't know some kind of like get started bundle right that tells you here's the best ones your which is a nice generalized coverage of different topic areas and that kind of thing. Yeah. However, I forgot to mention , I gotta say , you can also plug in your own API key . So if you want to keep using GPT or five point five or Opus or any model that because you ran into limits, you can just put in your API key and then it's you're off to the races with your own API billing, with your own API account. Right. Make sense, makes sense. And I mean that's a nice option if you're a heavier user and need that kind of thing. I think too, one of the advantages of these aggregators that we didn't mention that you reminded me of in talking about Raycast is that one advantage is if you have the same set of prompts that you want to run on a lot of different models, this is a way to do it because these aggregation apps will let you save prompts and projects and things within them. And then you can just switch the model, switch the thinking and use the same kind of thing against different models to see, you know, maybe you want to see what the differences are and come figure out which is the best for what you're trying to accomplish . Yeah . All right, what do you want to cover next? So let's talk a little bit about Bolt because I think Bolt AI is a very good example of a native Mac and IOS app is the IOS app yet Federica or is that I wasn't in test flight a while back. I think it's been in test flight for a while but I think I think if you just email the developer they're going to send you a test flight link. Yeah, but you know, focusing on the Mac , it's a very native Mac app. And so I think for people who the, you know, the electron of it all really bothers for some of these other models that are out there , that this is a nice alternative because it's a native Mac UI. It's got a left sidebar, a right sidebar. On the left, you've got things like your projects and your prompts and that sort of thing. On the right, you've got all the parameters you can set for each one of them. And in the center is just your chat interface where you do all that kind of thing. Yeah . Yeah, I think the developer has spent a lot of time really fine tuning the MacOS side of things in terms of UI and performance. And they have this like system wide hotkey for invoking chats from any where. I think it's really nicely done. Right. And it works too with like things like screenshot and dictation. So there's a lot of different ways to kind of get into the app and get going no matter where you happen to be in the system. Yeah , yeah. I'm a fan. I really like this one. It's hard to come by native Mac stuff. Yeah, it really is. And this is a pay once app. This is just, you know, you pay for it , you put in your API tokens, and you're good to go. Yeah . We should probably cover perplexity next . This is arguably the most popular aggregator, probably one of the one of the first companies to figure this out . They started as a search engine replacement that used different LLMs along with their own search index to let you chat with we b search results. And over time, they have expanded, they have their own models, although they're not particularly good. They have their own search API , and they're still integrating with all of the frontier models from Google to Anthropic to Chad GPT, they're really fast in adding support for the latest model that just came out like usually yeah, whenever OpenAI announces GPT five point six, you can rest assured that within within twelve hours, you will be able to find that in perplexity. Yeah . They have on desktop, on the web , they have this model counsel thing where you can ask this is what you were talking about before, John, you can ask one question and you can compare results across three different LMs , which is an interesting way to see how different models from different companies deal with the same search results and give you an answer. They have expanded into agent territory with personal computer , which started as a sort of virtualized sandbox in the cloud, but obviously and famous ly they rolled out personal computer on the Mac , which is take on open claw essentially that connects to your Perplexity account. You install the Perplexity app on the Mac and it lives in the menu bar, you grant a bunch of permissions and then you're able to have an agent on your Mac that has access to Mac apps and the Mac file system . All of this while still choosing between different models as the orchestrators behind the scenes that coordinate with the desktop agent and the file system and apps and accessibility and Apple Script, all these things. And you can also do this from OiS and iPad OS. So you install, I don't know, perplexity on a Mac mini or a Mac studio, you live it running all the time, and you can chat with it and you can do things from your iPhone while still picking between different models. I believe they now use GPT five point five as the default orchestrator, but you can also use OPU so you can use Jam and I got forbid if you want to . I think I really like the design of all things perplexity. I think they have really good designers . I don't like their CEO as much or he should be better at interviews and statements in general . He likes to disappeared a lot lately, which is probably for the best. Probably because of all the lawsuits that were filed against Riplexity. I think they have a solid product with personal computer . I think it can be a little finicky sometimes , especially if you want to do things like via personal computer, call command line utilities that you have on the Mac. I had to build a whole wrapper in personal computer to get access to some of my own stuff. Using like Apple script or something, isn't that what you were doing? Using using like a private framework or something. I don't recall. I don't use personal computer as much these days, especially since Codex for IOS came out in the CharjPT app . But I think overall it',s the best designed and arguably the best harness for multiple models , especially on desktop. The model counsel thing, I think is really well done. It is really well done. And I think especially if you do a lot of research, it's actually quite good for that. I mean, just kind of the leveraging the web search alongside what it can do from an LLM standpoint is quite good and and they have got ten I mean early days perplexity was not very good about sourcing. They are good at sourcing now, I think in general . So that's a pretty good experience. And they have a new feature called the Brain which just came out about a week ago, which is a memory system . It's a little bit it's a little bit beyond what is kind of the default for a lot of LMs you might use. And it kind of ties in with the whole personal computer thing and I haven't spent a lot of time with it yet, but it's very intriguing to me too. Yeah . All right, what do you want to cover next? I want to talk a little bit about Notion. I think notion is worth mentioning again because I think one of the things that these aggregators and this is where I think perplexity and notion kind of set themselves apart is that the apps we talked about at the outset are more focused bringing a lot of models together to do the same kind of things you could do with those models on their own . The thing about especially notion is that Notion is a useful product on its own without Notion AI . And I think that that to me and we'll talk about where the future of these things is going, but I think that that really is worth keeping an eye on because I think that's, you know, notion started out. I know when I first used Notion AI, it was really good at helping you set up notion pages, you know, setting up a database and doing relationships between different database fields and things like that, which was a big help because Notion can do a lot of things . And by using Notion AI, you can just have it do it for you or give you instructions to do it. But it's gotten a lot further from that now. I mean, now you can, you know, it has an API , it has a CLI tool , all kinds of things that allow you to hook your other agents that you might be using elsewhere into Notion itself. And then when you're in Notion , you can then query a big database, for instance. And for instance, I'll say the other day , I was working on Amazon Prime Deals. And I have a really big database of deals and I wanted to narrow it down and I just started asking Notion AI questions like give me narrow these down, filter them down to just the Apple products and the Apple accessory products. And boom, you know, like that, I had that. I was focusing that. I could do that myself. I could, you know, set up the filter and do all the things, but this was a lot easier. I was working on something else. It's there querying my database. And then so I started using that to just kind of find things and ask other things like what's, you know, what are the things that are what are the deals that are available today and are the biggest discounts of all? You know, putting together multiple parameters to come up with things that I thought would be worth considering for some of the stuff we published about that. And that's just incredibly useful. And I know you've gone a lot further with some of the worker aspects that are built into this to do even more with Notion AI . Yeah . Yeah, I live in Notion myself . I think I've built a really nice system with custom agents combined with workers and the Notion CLI on desktop . The idea being that I for types of content that I write, whether it's our Mac Stories or my upcoming I IOOSS andP twenty seven review or the newsletter , I've tried to identify any repeatable routine or things that I do often . What can be offloaded to a dedicated custom agent . And for example, I have one whose whose job is to only organize my bookmarks, things that I clip from the web . And I have another one just for my iOS twenty seven review and I have another one just for proofreading my stories. So I've tried to identify things that I do on a regular basis and turn those things those routines into custom agents . Then what's really so an custom agent has instructions. The instruction is basically an option page that you can go in and tweak. You can actually ask the agent to tweak its own instructions. So it's got this sort of compounding effect after a while. And each agent can use a specific model, right? Notion has a drop down that lets you choose between different models . And each agent can have its own set of permissions. So maybe my proofreading agent doesn't need access to the Fire Crawl MCP , but maybe the IOS twenty seven review does and the one that deals with scraping the web for my book marks does . So you can set tools per agent and permissions per tool and per agent. So it's really fine grained. Then the final step was the workers. Now workers are basically a code that runs in Notion Cloud . It's a secure sandbox that runs some type script, I believe, so version of JavaScript that's similar toou Cdlfarel workers like, it's this isolated instance that runs in the Notion servers . And what's nice about the Notion workers is that they have tools to talk to Notion agents and to talk to databases . So to give you an example, I have an OSHA worker that sends any new article that we publish on Mac Stories to an Ocean database just as a backup . However , for my IOS twenty seven review database where I clip all kinds of things. And there's a whole thing that I built that we'll talk about in September that I actually created with Fable for those few days that we had access to Fable . There's a worker there that deals with all kinds of tasks, like for example running through my database of features that I'm supposed to cover. Each feature is tied to a chapter and to an app and the worker has some code to duplicate features. Make sure that the same feature does not end up twice in the database . And all those things I deployed into my Notion account via the Notion CLI on desktop using Codex . So that sort of gives you the idea of the ecosystem that you have with Notion. Obviously , it can be expensive , right? The Notion AI subscription is its own tier on top of a Notion subscription and they have a credit system now. So keep an eye on those agents and make sure they don't go alright because otherwise you're going to have to recharge your wallet in Notion a bunch of times I am fascinated by and I think I mentioned this a few episodes ago. I'm fascinated by this idea of notion. Notion has always been at the forefront this like the theory of malleable software. Like it's something that can be anything . But now they're doing that with AI on top and it's easy to get carried away with it. So like just like anything in Notion . Just like anything in Ocean, just like anything with AI. So that's why I try to keep myself in check so that I don't waste hours just creating a project system. Yeah, coming up with systems that ultimately never lead to any actual work that gets published anymore. So that's what I try to do. Yeah. Yeah, I think to me what's interesting about Notion is what they're doing is not so different from what you'd see in other tools, but they're doing it in a more user friendly way because essentially what you're working with are things like skills and MCPs and servers and things like that. But they put a nice face on it and I think it makes it a little bit more approachable for a lot of people. The credit system to me is going to be fascinating to see how it plays out because I think I get why they're doing it. It makes a lot of sense. On the other hand a lot of what you can do with those workers and what you're doing you could also do simply by using codex outright or clawed code with the Notion CLI tool and have it dump the results into Notion . And so what you're effectively doing is paying for those credits to supplement what you're already paying for for an agent like Kodax . And that I think is a difficult business model potentially for Notion because it's on top it's more basically more tokens on top of tokens for a lot of people, which I think is a hard one to swallow for some people. But in any event, it is a very good tool, so I'm pretty happy with it myself. This episode of App Stories is brought to you by Claude from Anthropic. One of the things I've been doing a lot lately in Claude is working in Co andor Ik have, a specific project where I have combined Sparkmail , Notion , and the Apple Reminders app as a way to get information out of my email and somewhere where I can actually use it. Sometimes that's turning into a reminder, sometimes that means taking information out of an email message and putting it in notion. But with the connectors and co work , it's easy to do that and to have a project that's consistent and works the way I want it to day after day. It's a fantastic way to deal with email . Claude is the AI for minds that don't stop good enough. It's the collaborator that actually understands your entire workflow and thinks with you. Whether you're debugging code at midnight or strategizing your next business move, Claude extends your thinking to tackle the problems that matter. If you're a developer spending half your day on tasks that you wish you could just hand off , clawed code runs in your terminal, reads your code base , and can take things like writing tests, refactoring or debugging without you hand holding it all the way through every single step . Claude Code is at the frontier of egentic coding. It's not just another coding assistant. It'll handle project wide refactors for you while preserving your coding style and showing its work. And of course you stay fully in control too. I've used clog code for a bunch of different projects. One of my favor ites web One of my favorites is a web app that I use for posting to Mastodon and Bluesky using their APIs. It's a great way to handle both services at the same time . For problems worth solving , get started with Claude today at Claud . AI slash app stories that's CLAUDE AI slash app stories, and check out Claude Pro, which includes access to all of the features mentioned on today's episode. That's Claude. AI slash app stories. Our thanks to Claude for their support of the show. We should talk though, Federico now about where we think things are heading. And I think the two tools that we finished with perplexity and notion are, I think, to me at least very good examples of where this is heading because I do think that aggregators are going to continue to feel the pinch of a lot of the most popular frontier models wanting to keep people as their customers and not aggregator customers. So I think it's going to become harder and harder for the aggregators to continue to aggregate except for when they're willing to charge things like API pricing, which is generally higher than subscription pricing. And as a result, I think it's the companies that can offer something beyond just bringing a bunch of models together that are going to succeed, the Notions, which has its own useful tool completely outside of AI, and Perplexity too, which I think has just done a nice job of integrating things in a way that a lot of aggregators haven't . My theory is that notion will be fine . Perplexity will eventually be acquired. I think so too. I mean, I think if I were if, you ask me to put a bet on it , I will say perplexity gets acquired either by Microsoft or Samsung . That would be that would be my bet. I think Notion will be fine. They're probably too big for any company to swallow, except maybe open AI . I could really see Notion as an open , especially if open AI , let me rephrase, if open AI is serious about building or oper ating system , I could see a really expensive option acquisition . And also as a way of getting further into the enterprise since they 're a little behind in that area, I think yeah , but I think I think you're right. I think these aggregators will continue to feel the pressure API costs and and it's going to be especially challenging for these aggregators to match feature set and the lock in of the bigger companies. Like the moment that you have a Codex from OpenAI a clawed from Anthropic, supposedly Anthropic is also working on new agent feature features with this project Conway that's been seen in recent leaks for a while now . It's going to be hard to justify going from an ecosystem of tools and integrations to something like Perplexity or Raycast that only lets you chat with a chat bot, right? So I don't know, these are interesting products right now. I'm just not sure what the end goal will be. Yeah, I mean they're pressured on two sides. They're pressured by the features that you just mentioned and also by the fact that there are just a lot of aggregator apps out there. There are a ton that we have not a lot of mentioned. Oh yeah, there's so many . There's a I could mention Type in Mind Yeah, that's the one that's a web elephantus that one I don't know that one. There's a few of 'em in setup, I know that are 's a shira . AI that's SCI R A which is like a xity clone. Very or. Po is a popular one. There's so many alter alter, 's another one. Alter on the Mac . There's so many. It would be like a five hour episode. But I think the takeaway is true for all of them. Yeah, absolutely. And that's I mean, I think it's it's good to remember that things like Codex , things like Cowork , those have only been around for a few months. I mean, those are like since the beginning of the year type products. And those are the kind of things that are going to shift the landscape for the aggregators because there is absolutely lock in once you've got yourself a bunch of product projects in one of those tools that hook into your email, maybe hook into your notion, hook into your calendar, your reminders. That's where I think the lock in is going to be, and that's being adopted by the frontier model companies in a way that I don't think the aggregators will ever be able to match. Yeah . All right, Federica, well we are going to do a little update during the post show for App Stories plus subscribers all about kind of where we're at with a lot of these tools and what we're using these days because while we've talked about it before , it changes quite a bit. I mean, I think Federico's probably switched teams four times in the last yes but, I been've st for a few months now, which is why we want to talk about it. I know, I know. All right, well, we'll talk about that in the post show. In the meantime, you can find the two of us over at maxstories. net and we're on social media where Federico is at Fitchi. That's VITIC I, and I'm at John Voriz. JOHN VW R HWS. Talk to you next week, Federico. Tao John