The 30% Apple Tax Is DEAD — Build Agents to Operate Your Apps

They finally cracked Apple's walled garden. Huge deal. Like a 30% on your app. Big deal. Or 15% if you're doing less than a million a year. But that's crazy. This is a daily upload. My name is Parker Rex. I led tech for a startup that sold for 23 million bucks. After that, I went all in on AI. I built a lot of agents and I talk about it on this channel. So, like I said, the Apple stuff, I just want to go through it so everyone understands because I personally had to deal with this when we were operating on iOS. And just a quick rundown on what happened was basically they've been in hot water ever since Epic, the founder or the company behind Fortnite came after them. So kudos to Epic because what they did pushed this where they implemented direct payments in Fortnite on iOS on August 13th. Then they removed it literally the same day and then the next day they filed an antirust suit. This has been going on for 5 years almost and finally things flipped where they basically ripped the band-aid on commissions. And so just quick highlights, the first one that they tried to do was, oh, we're going to make it so that you can have an external purchase link and it's instead of 30, it's 27. But then they added these like scary looking screens where it's like, you better not leave, you better not go. You're going to die. You might get a virus or something. So it was kind of crazy. And then they had like really strict ruling around it. If anyone's built on iOS, you know that dealing with App Store is a nightmare. It's so tough. Um, we operated a food delivery app that did not have guest checkout for about a year and a half and we were doing at that point probably 30 million a year with it. And then all of a sudden they say, "Oh, you don't have guest checkout. We can't release your new update." And we're like, "Uh, hello. Where you been the last like year and a half?" So it's super inconsistent, super tough. So the walled garden is only cracked. It's not broken open. But that was their first step was doing that. But now it has basically, you know, they basically said, hey, you cannot impede developers ability to communicate with users about external purchasing options. So that's awesome. And that would be the scare screens. You cannot impose restrictions on the placement, formatting, style, or language of the buttons because if it's some rule where it's like, oh, you got to go to some scary looking menu or something. And what else was interesting about this? I also found in a a chat there was slacks from the iOS team just saying like they're just like talking about how bad they could make it look. So the impact on profitability for businesses previously paying 15% they now retain it and those that were getting 30% haircuts now retain it. Like that is a bananas of return. And let's see, now we can own the customer relationship. So transactions processed directly allow businesses to capture the email addresses and other contact info. Basically, you could throw a pixel on them. You can start to bring them into your funnel in a more effective way. And I know that there's some differences because they've been in hot water with EU. But beforehand, big tech was trying to dodge the fees. I don't know if you've ever had the Kindle app, but you couldn't buy a book on there cuz they didn't want to get the 30% haircut. It's more like a satellite application, which is awesome that it's changed now because you can invest further in your iOS app without fear. So before companies like Netflix, Spotify, and Amazon with Kindle app primarily offer access to content purchased outside of the app to avoid that 30%. So very very cool. and the impact on them. I mean, they basically Apple just got estimates say like a 25 to $32 billion deletion. So, I'm actually curious. Let's look at uh Google stocks. Let's see what Apple did. So, let's type in Apple. Not much after the hearing, huh? I mean, most people don't care, but they did process a staggering 1.1 trillion in developer billings in 2022. So, you can only imagine that's larger. So, that's just a big deal. Shout out to Tim Sweeney, the leader over at Epic Games because that must have been a very expensive battle to win. Very, very cool. So, that one is done. Now I want to talk about let's pick maybe a couple more. Let's do how about this oneshot analytics. So this guy's name is Yohai and I met him when this must have been three years ago where he was building something called Baby AGI and you can see he follows me because we met and we chatted and he was liking what I was doing. I was trying to build agents in the like just general productivity space at the OS, not the OS level, but like something that could control all your different apps, but it was before MCP existed. So, we we talked a fair bit about that. But what he has done in replet is actually really interesting. There's a lot of different ways to make agents. And what he did here was he says it's time to add Google Analyt. And he made a oneshot analytics dashboard which I think is pretty cool where he just types that in and then he got this. Now if you go and you look at his page, he also built this thing called VCedia. So it's like an encyclopedia style thing with all the information for what's going on in the VC world. So I vibe code to startup intelligence platform introducing VCPedia startup funding discussions captured on X analyzed with open AAI enriched with Exa got the wrong one with Exa and then a daily newsletter using Reend built with the replet agent. The only thing that I've seen that's negative about this is the performance but I think that's just by nature going to happen if you're not going and optimizing things. You don't optimize first, right? So when you get into it, basically if you just want the automated daily newsletter, you can go here. Sample image attached. So this is pretty cool, right? Where it has a bunch of the stuff in here. It may not be perfect, but I mean the guy said he built it in a week and that's really impressive and very cool to see. I haven't used Replet in probably two years. So if anyone is using Reddit, uh, sorry, Replet, can you comment below just with your raw feedback on it? whether it's awesome or if it's dog water. You could comment uh replet is dog water or replet is awesome. And here's the thing I built with it. But yeah, this wasn't even something I plan on building, but the idea came to me weeks ago and I put other builds on hold. And so you can see that he basically gets these daily summaries of the exposts and it's not super UI friendly but it works where he just has these keywords and this is a search operator in Twitter or X and then it comes back and he says that it was occasionally picking up funding rounds. So he started imagining combining this with a graph-based approach and you can see this pretty cool. So if you're familiar with graphs, if you I mean the simplest way to think about it is um like your brain has topics and they connect on different edges. So an edge would be the line. So if I say the word edge, now I have an edge that's thinking of edge networking or the edge of this table. And that's kind of a really really dumbed down watered down way of thinking of how this works. But it's used with clustering. It's used with traversing databases. It's uh a nice visual representation of information on the screen. And so by the end of it, you have complete profile pages for every entity. Very cool. This is kind of like he rolls his own I don't even know what it's called. It's like pitchbook. Pay like $40,000 a year to get this information and he just self rolls it in a week. So I've been making a lot of jokes about, "Oh, I'm going to roll my own neural link." Someone comments and they're like, "What's up with the headphones?" I'm like, "Oh, I just rolled my own neural link. Oh, I just rolled my own payment processor. Oh, I just rolled my own air conditioning. It's a meme. I don't think people catch on. All the neck beards get angry, but I'm going to lean into it. So, does it work? Yes. In the last week, he caught key funding rounds. He's catching about 200 of them on the top end. Yes, it makes mistakes. Did I really vibe code this? Yep. Literally just the replet agent and some docs and he did five versions in one week. So, very cool. You can see that it comes back with, hey, I need to optimize. Then it comes back and it tries to figure it out. Now, the coolest part to me is I saw this and I was like, oh, is that mermaid? No. Then I realized this is how Claude outputs things. So, if we look at this, let's actually grab this image because I want to walk through this. This might take a minute, but this is really, really cool because it's kind of a combination of things that you could do in a make.com or an NAN or GCP, but he just did it in here. So, I in my head, this is probably a bunch of Python because I know he's a Python guy and a bunch of um SQL. So, let's see. He has the Oh, and then a basic front end. And I know they use vit for um replet. So yeah, you have the admin interface and then a query setup. So it has all the stuff that it needs to store for the queries and then the search query database with auler system. So not sure how that's set up, but sounds cool. I don't know what a celery task is. What is a celery task in this context? Also, how's this work? If you were to guess, let's skip this. So then it calls the Twitter API. It retrieves tweet data, stores it in the database, processes it to do the entity extraction. I think it's with 40. Does the identification across is it a startup, an investor, a founder, a funding round? And then it separates them. So it's kind of like he's doing ETL. He stores it. He does some transformation on it or classification on it and then he's loading it into a new database in a new way. In this diagram, a celery task is a background job that handles in the execution. Okay, so it's like a cron. Got it. Once triggered input and scheduling. Okay, that's fine. So once those are all stored, then the new ones are sent to Exa. So, Exa, as I recall, is kind of like, let's just see, like a SER Exa API. Yeah, web search web researcher. Cool. Yeah. So, oh my gosh, it's a nice looking animation. I haven't used Exa much. If you've used Exa, I'd love to hear from you guys below on what you're using it for, like straight up, cuz I'm going to have to go use this. This looks cool. And then from Exa gets the web search results and then he processes them with OpenAI to get structured info. And then structured info goes back to enrich the database. Over here it looks like he analyzes entity resolution identifies candidates with name similarity scoring. Whoa. Vector embeddings candidates. So I guess in this case are the candidates people or I think it's deals confirmed duplicates. So he's doing dupe coverage. Then you have the newsletter. So this comes off of entity database. Once all this is enriched and updated, you got the daily trigger that fires, prepares funding round data, processes it with 40, generates the content, generates the subject headline, combines it into the newsletter, stores it in the newsletter database, newsletter DB fires the resend, resend, updates the delivery status, and then goes back and writes. This is so cool. So so so cool. Yeah, what a cool thing. And that is going to be it for the video actually because I want to end with some fire. So no more stories. What I wanted to just talk about really quickly is how do you delineate? And I'll make a whole video about this because it's kind of a big topic, but it's like I see a lot of questions about like when do you choose make versus n versus GCP versus let's say replet for aentic workflows because that's what these are and I actually it depends on the use case, right? So in my case, if it's something that I want to sell as a service, then it's one of these, right? But if it's like if I'm selling a service, then I use that. If I'm building a business with it, totally different thing like a SAS where I need this thing to have infinite customizability, infinite scalability, and I can ultimately put a front end on it with O and sessions and all these things, then that's when you go for something like GCP. But then the question is is like okay well if it's just for you for you with interface like with UI then probably replet if this is just single serve sort of deal like I'm thinking more and more about this and I see a lot of people on the NAN subreddits trying to figure out O and figure out sessions and figuring out all that stuff and you can do it but again you're going going to be limited. Like I haven't seen I say you can do it cuz like one person on Reddit figured it out, but I just don't think that you'd want to have that as your dedicated back end. Maybe if it's just a small app, but yeah, that's how I'm thinking about it. If you guys found this helpful, make sure you like the video. I have a lot more stuff like this coming. I think the Apple stuff's bananas. I think this replet agent flow is really, really cool. I have more stuff that I'm going to be covering tomorrow, which is on the MCP plus A2A, which is agent to agent. I'm going to be talking about a bunch of the AI influencers, the top people to follow, about codegen with taste and this speedy OS agent, as well as some hot takes from the founder of Super Maven, who basically made the Cursor tab a lot better when Cursor bought their company. If you want behind thescenes stuff of me building, then you can also check out our school. We have a really active Discord. As you can see, people are always trying to figure out the next edge on different development workflows, marketing workflows, all of that. Right now, we're running a special. It's half off because we're going to be moving off of school just to provide an elevated member experience. So, can't wait to see you guys tomorrow.