Planning an Agentic Organic Content Pipeline (step by step no slop)

Hey there, I'm Parker Rex. I led technology for a startup that sold for 23 million bucks. After that, I tried Airbnb for music. Did not work out. Got really good at the technology part of that though and went full stack. Then chatbt came out. Everybody knows it changed everything. So I went all in turning away all their opportunities. Spend every single day working doubles to know how to leverage it across code, media, and business. It's changed everything that I do. It's crazy. So on this channel, this is a daily upload channel. This is not the main channel. The main channel is where we go deeper on the actual builds and implementation of the things that we talk on here about. But oh, we can see we just finished something. I'm just trying to run this agent while I'm doing the call. So yes, it's going to run into a migration issue. We do questions. We do five questions if we have them. We do some news snippets if it's worth talking about. So sometimes we won't talk about it. And then we do AI strategy. So let's talk about the questions. First of all, I need to fix this migration thing. It's going to repair and revert those. If this works, I'm going to be really surprised. Okay, it worked. Awesome. This is Taskmaster. So my whole thinking behind this is I just want to have this going while I am running this. If you're not familiar with Taskmaster, check out the video that I put out yesterday about it. and the founder of it is actually in our Discord for the paid community and yeah, he's really really helpful. So, it's doing all sorts of stuff right now. But on the question side, we have a bunch of them. And let's just go in. So, DTECO Dan I think is his name. He was really helpful in asking a question about Convex versus Superbase. Which ones to use? I answered him. He came from the main channel and I said that I would prefer to use superbase but it's all preference. I know convex does a lot of I think they do like compute and they they have like a lot of different things that they've added to it. So I try to bias towards the primitives but we had a good chat about this and I think a lot of it just has to do with what you're used to. So I'm used to superbase. I'm used to using it for everything. So, I really like it, but you should follow him because I looked into him and he has I think he works at Google first of all. Awesome. And then he has a podcast. He has some interesting folks on. So, check out him. Next question is, I almost built Quilip like app. It's a great idea. MC MCP should get the code context and return the documentation related only to that context. We talked about this yesterday, but Quillipe is an MCP that you would reach out to get fresh documentation for. So, pretty cool. This guy says, "Yet another douchetuber riding the trend to train using the moronic term ever created in vibe coding. You're an idiot." Thank you, stupid internet people. One, if you actually watch the video, you'd know that I'm kind of making fun of vibe coding. I think that that's a big thing is this is a trend that's going on and it means a lot of different things to different people, but person who coined it is the goat at coding and at Python. So, Andre Carpathy. But if we type this in, this is what the market finds interesting is this phrase. And because they found it culturally interesting, it brought in a lot of new coders and it's changed the way that a lot of developers work. I implore that everybody gets better at coding. They get more technical. It's the easiest time ever to learn. When I first started learning when I was 18, I'm 32 now. I was just doing treehouse classes and it was so hard, so boring. And then when I tried again when I was 23, I was taking linear algebra classes from a Brown University course that I got access to, that was impossible. Now you just have this 247 always on tutor that you don't have to feel, you know, like you have a lack of confidence or you don't feel like a asking a lot of questions. So I think it's the best time ever and vibe coding is the gateway drug to it. So it's really cool. Another one is I stumbled upon this from the other side of the world. Thanks. This is very eye opening. Wondering if you could have a checklist of tools to look into like superb base. It is super but close. And this is something that I'm actually going to be doing is I'm going to create different assets for each video because a lot of people think that I just talk too fast. I'm leaving them behind. And I'm not going to change the way that I talk. But I'm going to add more content around that. And then it'll all be hosted on iviwithai.com which is what this is being like this is what's happening here is we're building iviwithai.com so it can host all the free content the paid content but for each video there will be an automation that goes and creates a one-pager as well as a one second a one pager oh I'll come back to that a one pager and a summary and basically just more stuff for people that are having issues. So, that was nice. These comments change the way that I work, by the way. So, make sure you do make comments and I'll try to answer as many as I can. I know we have time for maybe one more. Let's go back. YouTube Studio. For instance, one of these that I need to build is an automation that pulls these ones. So, you just spent months writing these Python things. You didn't you haven't seen what's out there. There's probably a lot of stuff you didn't need to do. 100%. 100%. And this is in reference to me talking about how you can handcode a lot of things that don't need to be handcoded. There's a lot of people that overengineer solutions. So, you just at the end of the day, everyone wants to move forward. I'm not hating on the people that make these types of content. It happens both on the technical side where people overengineer things that don't need to be overengineered and then it happens in the no code space where you have a lot of people that are making really hypy content about faceless stuff about just oh this sounds like this British accent assistant from James Bond and it's going to update my calendar tell me the weather and it's going to light tokens on fire and take longer than just looking at the weather. So I think it it happens across the board. And then here's just a bunch of comments for people that want the PRD prompt. They want the turbo starter kit. Turbo starter kit is gonna be dope. I have made three different versions of it and was unhappy with the first two. This third one's going to be good. And me and Hari are going to be working on it tonight. So that's not working. This is an example. When you're using Taskmaster, you shouldn't be making a YouTube video while you're doing it, especially when you're doing SQL stuff because I know what it's happening and I should just take the time to actually stop it rather than let it light money on fire or these are free requests, but you get it. So, that's on the comment side. On the news side, everybody knows that 40 came out, and if you want a really in-depth video about that, you should go watch Theo T3TG. But let me pull up the couple of notes that I put in to our channel on the vibe with AI which comes with our community. I changed the price because I don't want a bajillion people in there and I also want to make it a business that I can then hire people to come and work for because this won't be just a school community. This will be its own standalone brand. But the price going up tomorrow and I just changed it back because someone was like, "Hey, I checked and it was 37. Now it's 50." So I moved it back. But in here you will see all the big news. And this will be another automation, but on the 40 side, you can see that it's basically a 40 replacement. It's a workhorse. That's cool. It's cheaper. It's faster. It's better. It does better on the coding only for Python. So, if you're using 40 for Python, then go to 41. But besides that, it's not that great against other languages. So, I always go to the AERS polyglot benchmark. It's really really nice for that. And then Windsurf and Curser are both doing free unlimited 4.1 requests this week. Other things, Grock dropped workspaces. So that's pretty cool. It's their version of projects and I was playing around with it just to take the PRD that I was writing for iviwithai.com and trying to make it better. But the prompt was just river trash absolute dog water. So I didn't get the great outputs. But what it does is it just basically lets you work in a project as you would in Chad GPT and you can put in the instructions. What I don't like is that you don't have prompt variables. What I love about vertex is prompt variables. So, I hope that they make it more like this. And always use pass keys, by the way. I've never typed in a password in my life since I started using these. They're more secure. They're easy to set up, and you don't have to type anything, which is just the best. Sometimes you get those dubs, and you just hang on to them. Let's go into the prompt gallery. You can see all these things in here. But when I'm looking to make a new one, the prompt template's really great. So what I do is I hit help me write and then I do F1 and this opens that if you want a dope PRD and you drop the word PRD below and it will actually reply with the IV vibe with AAI website later tonight when it's rolled. But then I'll hit generate prompt and what this is going to do is going to take that and it's going to plus it up and optimize it for this workspace. So that's awesome. Now what I don't know is I didn't change flash so it might not be as good. Let's actually let's see what what's different about it. So the easiest way is you just spin up sheets new open up a spreadsheet. We're just I want to know does changing a model in here change the model that's used for the prompt intent creation. So what I'll do is I'll just say output and we'll open this up and we'll open that up. Close this. And we know that this one was for Flash 2.0 Flash, which is a total total workhorse. Let's open that one up. And then let's just see if anything changes. I doubt it. I bet underneath my bet is that they want to make the best one possible. So probably going to be the same, but let's just see. I'm curious, so I'm sure you are, too. Let's move that up. That's system prompt. And then that is or system instructions and system prompt. Now, let's do the same thing. Let's cancel out of that. Let's switch the model to 2.5 Prox. And is this the latest one? Yeah. Cool. So, let's go back and help me write. And then let's do F1. Boom. Generate. Let's see if there's anything different. Move this way down. Move this to the top. Cool. That's going to be two flash. And then let's [Music] grab that and stick it there. And then 25. 2.5. Cool. Wow. 2.5. Good job, Parker. You're so talented. Okay, so I'll grab this. Oh, it is better. Look at that. It could just be random. It could be just that it happen to be better on that one. H. Definitely longer. Suggested prompts up wrap. You are tasked with creating a comprehensive prompt. You will be provided the following. So it looks like it put more in the front of this one on 25 steps outlined. honestly looks really really similar. It's probably the same. But anyways, I would do that. And then this is what I hope to see out of Grock is that they have these prompt variables. So when I hit insert, it's going to take that and you'll see these little outlines. What's really cool is this can map to a cloudr run app. So it will go and containerize it and then you can hook that up in a front end. So you could literally take this and it' be like make meme adopprd.com and it's just three fields that people are putting in here and that' be pretty cool or you can open it as a notebook or you can get the code. You can do all of them. It's pretty nice. So that's on that front. I wanted to call this out. So, Superbase, I've said this before, but they just added another one for declarative database schemas, but you run a command with Shad CNN that then brings in all the rules from Superbase about how to use their products. So, if you're to go and read this one, it's an outline of how to do best practices for using declarative database schemas. if you're not familiar with what those are, it makes your things when you're writing migrations be native to the code that you are already writing in. Whoops. So, instead of it being in SQL, if you like writing TypeScript, then you can just stay there. So, if I go to docs and then let's see query data, you can see that one, it's the way that you're doing this. This would be on the SDK side. So, I guess it's actually going to be on the migrations. Let's see an example. You'd se select your dialect all these things and then you actually define oh user table. So instead of create table and writing it in SQL you're writing it in TypeScript. So they followed suit with drizzle because drizzle is really popular. You can see I'm using similar web and they get 300,000 people to the site a month. Another good way to see popularity of something is go to GitHub and then put in a TypeScript operator. So I do language TypeScript and then this is a search operator. You can see it now has that little thing there. And then I can just grab this and let's see like I don't know that let's see how many people are using this for Postgress do search and you're going to see a big number 31,000 examples that are open source. So pretty popular project. This same tactic is something I use whenever I'm building stuff whenever I'm trying to learn something new. I'll just jump in there and find examples and there's usually a lot which goes into just tech choice in general. If you see someone say on X like, "Hey, this thing just came out. It's going to kill that." That's usually a red flag. So, don't follow that. Next up, let's talk some AI strategy things that I'm doing. So, I added, if you've been following this channel for a little bit, then you'll know that I have this content pipeline and I always am trying to do the things manually before I add more stuff and or sorry, before I try to automate anything so that the quality is good and I know the steps, it follows Elon Musk's five-step algo. If you're not familiar with that, let's just pull that really quickly. Canva, I made a graphic for this and it's actually the background of my computer. So I remember don't automate prematurely. But the idea being where is it? Not in there. Oh yeah. So question every requirement and this goes for any workflow. Question every requirement. Delete any part or process that you can simplify and optimize. Accelerate the cycle time. So in my case with this, it's YouTube videos. And then automate. So that's the part that I'm at. And I'm doing the same thing with YouTube videos, the same thing with doing code commits, the same thing with working on my community and my brand with Vibe with AI. All those things I'm doing manually now. And I've reached the bandwidth. I've reached the cap of what I'm capable of doing manually, but I've done it so many times that I know all of these. And now I'm going into the question every requirement. So, in my pipeline for YouTube videos, I used to have this flow, which was, let me make sure that this is still on screen. It is. Great. I used to have this flow where I'd record the long form. I'd have an AI transcription edit done, which is Premiere Pro. Haven't deleted that yet because I don't have an MCP that'll do that for me. But if someone writes an MCP for Adobe Premiere Pro, let your boy know. Then I would export it to GC cloud, not drive, G-Cloud. And then that's using Google's publish and subscribe or pub sub or, you know, if you're from Florida, it's not like a pub sub sandwich. It's publish and subscribe. Then that's being emitted to let Aonic know that, hey, this new thing is here. Let's run their ASR, which does a couple things. It pluses up the audio. It transcribes with whisper. It creates the subtitles. It creates an HTML file with summaries, both short, long, and medium, timestamps, and all these things. And then that would be put back into a database. So, you're going, you're doing something, it's coming back, it's returning it, and then it kicks off the next thing. There's really nothing in here that's too fancy. It's just a bunch of steps. But that's what I'm doing now is cool. I've done long form seven videos a week on daily. still optimizing that where oh I need to do it in the morning because there's people that reach out and they say I watch in the morning and I'm a night owl so I might as well just do the morning the night before. Shout out Basil for reminding me of that. And then now as I'm adding more videos because the main channel is popping then I realize oh crap like I need to make these processes easier. So I know that the leverage on this channel is much higher. like I get 10,000 I almost had 10,000 views in two days which is really cool and I know that I need to be pumping content into here but having 10 videos a week with zero automation that ain't it. So that's what I'm doing now and kind of ideating through is okay cool if I want to get to phase three which involves shorts because shorts are bananas in terms of top offunnel virality the interest graph the algo pushing stuff I need to automate some things so that's what I'm thinking a lot about right now and let's start to like think through it in terms of the actual tactics Cool. Is this the right one? No, this is not the right one. I was going to switch away from Excaliraw to go to Canva, but the thing is is Canva doesn't have hot keys. So, I don't know why they wouldn't. But as I think this through, there's a few things that I go to, which is what are the things that won't change on your dayto-day? And like how do you set systems in place such that everything that you do plays into the next thing and anytime that you add new output then you need to change the way you're working. So when I look at this this is like a tactic this is like oh browser agent for school. All right, like home server. These are the things that you would end up doing for one particular use case. But I like to abstract it and say, okay, well, what are the things that I need to do every day in order to accomplish my output goals of the 10 videos a week on YouTube, as well as getting to shorts ASAP. So, let's talk about that. I think of it in terms of a power list. So, a power list, this is a term coined by this guy Andy Fcella. Very polarizing figure. Don't watch him if you're a snowflake or if you like can't listen to someone yap about being aggressive. He's just an aggressive person. But he also did 75 hard. It was transform transformational for me. And then he had the power list which is what are the five tasks that you need to do in order to call the day a win so you can get a dub. The W. And those things are non-negotiable. And then as soon as it becomes a habit, then you drop one off and you add a new one. So I had a habit on my power list a month ago to do a daily video. Now it's just happening. Yes, I'm still optimizing it where I'm going to shift it to the night before. But now I'm changing this and I'm keep looking up here because visuals are really helpful. Like if you want to get a new habit, then you want to change your environment. So, I changed my environment for these YouTube videos by having this arm here with the microphone, having a box light here, having a camera there, having this little drawing thing that I took the concept from the guy Nick Save. I'm sure if you watch me, you know who he is. But I had tried launching this channel a ton of times and I just didn't have the right format. The content was good, but the framing wasn't and the production stream, like the production, the way that I can just sit down and just dial it in for a half hour just wasn't there. So, shout out to him because that was helpful. But it all plays into this power list where it's like, "Okay, each thing on here needs to be optimized. Do it manually." And so it'll change your outputs. And so when I look at the power list, I recommend that you make a power list. I just recorded this and apparently still had my face on so I had to go back. But this is where I landed where it's these are the things that I'm going to push myself to do every day. If I don't hit them, then it's technically an L for it for the whole day. Now I go really hard on the different things like almost unachievable but then it that constraint forces me to think of how to systematize everything. So if it takes me I want to put half an hour of every day into the daily video. That means I probably need to get it down to about 20 minutes. So that means I need to have some automations around getting the questions into there and then automations around the news and then the rest can be about strategy and that'll go into the production as well. I have five school replies. So five school replies is not within my own school but other schools that are free because I realize that that's a growth channel that I can leverage where I can go and provide value and actually spend the time answering real questions for people and then they might click on me and learn about me and I'm happy to spend an hour doing that split between other people's and then mine. So on mine, I will be doing real posts, right? Like things that I've thought about. They're an attempt to be as helpful as possible for every member in there. And then I'll do it manually at first, but it's in the daily tab. And later on, once I've done it manually enough times, then I can have a browser agent that uses my O token to then read through these and understand the sentiment. And there's all sorts of other things in the automation world that I would like to be doing where if someone has not done the things that I want them to do, like to go and actually take action on the content in there, then they might be overwhelmed. So I can send them a message after 48 hours of inactivity or 72 or something. I don't want to be annoying, but just to check on them and make it really, really personal. So I'll do that manually and add that. And then automations. So, I need to be doing an hour and a half worth of automations each day. And I'm going to map this to my calendar. So, when I get this power list, I map it to my calendar so that it actually looks like what my day looks like, right? So, I have that and then that'll go in there. And then I have the different types that automations that I can choose from. So, NAN, Zapier, make that'll be hosted on an agent cluster. So, the agent cluster is hosted on Netcup. It's basically three times as much compute and storage space just specs on the hardware for it's 3x that for the same price. It's like $12 for something that would cost like maybe even like 50 or 60 bucks on Digital Ocean. And so I need to set that up and that has Docker Swarm on it for failovers and load shedding as well as uh load balancing, right? It'll have the ivibeai.com. It'll have the nadn running locally. It'll have nomic as an embedding model which is better than text A data from OpenAI and it's open source. And it'll have all the other apps that I have. So, it'll be great. And what else? I'm going to be doing that. I'm going to be doing OSS contributions. So, I have a bunch of projects that I'm going to be open sourcing. They're going to go in phases. So right now they're all closed, but I could say that this is for any project. If I want to make something added to Taskmaster, for instance, I want to do that so I can help out that team. I have a lot of thoughts. But the phases for these projects, it's basically everything always goes alpha. And this is the MVP phase. Or maybe you're just using it. Maybe you have a friend or two. And then it's always performance based for me. So, I'm I never really have deadlines. And what I mean by that is the deadline is now. It's just go as fast as possible to get it out. And you want deadlines for release dates, but more so I think of it for like a marketing event. So, you know that as you're getting closer to GA or general admission, basically anyone in, then you need to start yapping about it. But I'm going to be yapping about it the whole time. So then beta would be a polished version of the alpha. It probably has most of the functionality. This would be like almost a V1, but it's like let's call like V.9. But it's basically like if I I do it by users. So if I had five here and I want to get to maybe like 50 here and then as I go out of beta, which is just looping back through everything that comes in from users, then I can go to GA and GA is when I'll be OSS. So beta, the way I'm going to be framing this and why I love the community is everyone that is in there has an interest in working on real projects and most of them are developers that want to learn how to leverage AI or they're people that just want to know how to use AI in general. So I cover both those bases. It requires more work and I can only do that because of automation, but it'll make my offering superior to everyone's because the price will be similar to those who just focus on one and it pushes me to have more and more leverage on the AI side so I can get both. But because people in here want to use the projects because that's part of the package, then they get access to all the beta stuff and then they want to use it, they want to contribute, they want to give feedback and then we're doing PRs on a loop. So, it's like as we go use the thing, we're just making it better, better, better, and then we say, "Okay, uh I don't know, let's say April 30th would be the external date." Then you're rushing towards that, you're posting on things. And then when it comes to the distribution side of getting your first 100 users, then I'm looking at channels. And it's like, yeah, I have the YT stuff, but we want to go outside of that because people that know me will likely be more lenient. So, I always go to Reddit. Reddit, they just will Cleveland steam you if it's bad, which is great. And I have done Product Hunt, so for GPTs for devs, one of these projects that got 100,000 hits, but it was free. So, I should have made it paid, but I couldn't. It was illegal because I was using uh Apple's IP and then just wrapping them in GBTS. But I did Product Hunt, Reddit. Reddit was the the most successful one. And so that's how I think about the OSS side. So putting a lot of time towards that will be worth it. And then office hours. I do office hours every day in our Discord. So we had one today. There's like five people on. It was dope. But I just hang out in there and then we just run the motor together which is great. And then on Wednesdays I don't do an office hour. We do like a a G meet. So I'll do that. And then the last one, this will be the trickiest, but it's like two hours dedicated to trying to make a great main video. So, this won't mean the output of the video because these are going to be higher quality. So, my goal with this is seven vids on main or sorry, on daily and then three vids on main. And the question might be, well, Parker, these ones are higher leverage for audience building, but it doesn't bring in the fact that all these second order effects of doing the daily that I get like these things that are like squishier, which is like I get the hard skill improvement. So, it's like the classic saying of sets and reps, you get better at the thing by doing the thing multiple times. If I'm doing seven a day or seven a week, then I get better at being on camera, it makes me know about the AI news because I'm looking at it every day. It gets me better at automation. It gets me better at the programming, not actually coding, but programming the content. So you have, you know, in production you have a program manager that comes up with like, okay, this is what we're going to do. We're going to do five questions, five snippets, five this, five that. So it'll give me better at all these things. And I also think it serves a different audience. So that's how I'm thinking about it. And then I'll map that to my calendar. And I actually want to just see what this would look like. Let's just do it together. So let's look at the calendar. And I used to be the craziest person with my calendar because I was doing so much stuff that relied on others. And then I didn't do calendar for a long time because I just was sitting there in growth mode learning how to code. First it was getting really good at just like backend as a service which isn't true backend dev and then it was backend dev and then it was learning how to use all the SDKs and building those and then it was client work. So I didn't use calendar. So, coming back to this, it's like, well, the daily update on Tuesday night will be filmed from 9 to 10 because I'm a night owl. And then I'll categorize that to content. And I think that's fine. I actually prefer if it was purple because I think color coordinating your calendar is something I've like obsessed over and people are going to hate on me, but it's nice at a glance to see like where things are at, but we'll leave it blue. We have the daily office hours, which is school. We have other stuff we need to add. I go to the day ahead to know like what's the perfect day look like. So I'm going to change in the mornings. Easy stuff. Baze, what do you think? What should I do in the mornings? School communication. School communication and engagement. So 30 minutes school, 30 minutes other schools. And that'll be from realistically I've been staying up so late. So I'm going to set this this will this will be better. This will be good for me. I'm g change this to 8 to 9 every day till the end of time. And then what else? That bundles those. I need the half hour for the daily, which would be great if I can get it down a half hour. So I'm gonna shoot for that tonight, which means 20-minute video and then 10 minute post. It takes a little while for it to render, so that might be inaccurate. I'll put it to 45 minutes because you got to worry about the rendering. We need an hour and a half of automation work. So, let's plan the ideal day for that. Let's look at it. Um, that's content. Oops. So, automation stuff's pretty lightweight, so I'd probably start the morning out with that one. Automation. I'll put nad zap slashmake slash aagentcluster because it'll involve that. So you will technically be doing a little bit of coding. It's more like devops and then that's every day till actually I take Saturday mornings off. I know it's crazy. So I'll do that and I do Sunday mornings off. So maybe let's just try that because weekends are going to look different for me. I'll do that at a different time. And then I can just add a new one. This will be let's change content to purple. Let's change this. This one will then be automation. Actually, I want it to be yellow. Sorry people. Yellow automations because the color lightning. Then we have code. We have I'm going to just call this media school automations. What other ones do we have? That's code. That one covers that. Office hours is school. Main is media. Yeah, those are good. Cool. Let's change these. Okay, that's good. Automation is going to be that all cool. School. We have the weekly Q&A, which is cool. We also have a half hour block there. I'll leave it open. I want to do video planning in the afternoon because it's a little lighter. One main or main channel planning slashproduction filming. And as you actually start to map this out, then you recognize like maybe not. Let's see. Repeat every weekday. Oops. Every weekday. Great. And then this would push all client work which I have just been doing nights and weekends then I'm not going to even map that because I just know inherently um main channel media all what else we did main we did office hours we would then put coding around lunchtime so oss contribution And it's mapping all the things that I love to do. Like I was just telling Basil about this. It's like really like how do you it's a classic like icky guy the Japanese philosophy around like how do you find something that's fulfilling that delivers values to other that you like doing and that's magic. So we'll change that to code all events and then we need to change this to every weekday. I'll worry about weekends later. So this feels pretty good. We've got that. We've got that. We've got this. We've got the automation. Yeah. So, all this is totally doable. It's just going to force me to be better, which is what we all want. One conflict there. But that's enough for today. I'm running long. Hope you guys found this video helpful. Oh, I already filmed this part. Kitchen Pal's cool. Map's cool. Agent cluster is going to be dope. I got to go build these things. So, that's it for the video today. If you found any of this stuff helpful, then obviously leave a comment because I want to talk to you. It actually is really fun being able to talk to people and get their comments. And then what else? The main channel has really good videos on Taskmaster, which is what I was using at the beginning of the video. Make sure you like the video because then it gets out to more people and we'll all at the end of the day end up hopefully becoming friends. That's it. I already said comments six times. Baze is really good. She's my YouTube coach. So, if you ever see her in the comments, if you're mean to her, you're getting autobanned. I'm booting you. So, everyone should actually comment a flower for Basil. So, all right. See you.