Start learning this AI Coding Toolset WAY Before It Blows Up (task-master + cursor)

I'm Parker Rex. I led tech for a startup that sold for 23 million bucks. After that, I tried Airbnb for music. It didn't work out, but what it did do was light a huge fire underneath me to learn how to program everything across the stack. So, I went all in full-time learning how to do that. And then ChatGpt came out. Of course, this was a game changer. Immediately recognized that this is going to be bigger than the iPhone, bigger than cloud compute, maybe as big as the internet. So, I said, I'm going to spend all of my time, literally all of it, learning how to leverage AI across programming, marketing, and business while also building my own SAS product and an AI media company. On this channel, I do a daily upload sharing the strategy, things that are going on in the news of AI. No BS, no fluff. First, we start out a little questions and answers. Shout out to another animal 01. Instantly, the winner of the comment belt. I thought I was getting comment belt at 500 subs where you get these little badges, but apparently I need to get a lot more watch hours. So, another animal. You're the proud winner of the the comment belt. Oh crap, because you said a couple awesome things in here around Well, first of all, you said I should wear a suit. So, you know, I just I got real dled up for you guys today. Maybe there's one girl watching besides my fiance, but let's Anyway, so he gets comment belt. So he he said something good about the suit. Sure. But beyond that, there was someone that was saying, "Hey, your videos are not add proof enough." And I was like, "Okay." He says a whole bit about the Gary Veaynerchuk strategy. Now, love him or hate him, the guy's made a lot of good calls. He calls all these platform shifts, maybe not NFTTS, but it's, you know, still not out on whether or not this thing's going to work. But this is more or less what I am doing with this channel is there will be a ton of automation and I mentioned this in today's upload which is the 8th of April if you're watching this at a later date. I talk about my five marketing strategy and using AI automation and orchestration to take long form content and then break it across every platform to the tune of thousands and thousands of posts across everywhere across text across well that's short form text mid form text long form text same thing with video same thing with generating carousels all that so shout out to him for the belt And when I do get badges, you'll be a proud badge holder. And then there were a bunch of other comments that were just asking for the giveaway that I did yesterday around the starter kit. So, turbo turbo starter kit, the development treasure chest. I will show these things in a second, give you a quick update on them, and then reply to all the comments. Let's see. Oh, someone else said that I too long didn't watch easiest way to make 5K is selling courses in community access. Join my community and buy classes. Thanks Idol Observer for the free ad. Now I'm just going to make every video. It'll just be six seconds long and I'll just say go buy my thing. So, thank you so much for the free ad. You're great. Now, quick update on the turbo starter kit. I was up till three last night trying to get something done in a cheap fast way. Actually, I was just frankly just exploring. I know how to do sometimes you know how to do the thing the right way or the the proper way, but then you decide, hey, you know what? I actually want to try something new. And so this turbo kit, it is open right now. Like you could go to this route. It's just Parker Rack Vibe with AI turbo kit. This will be housed on the ivibewithai.com. Shout out Basil for great comment on using that as the URL. But this will be housed on there. And I was just trying to take all of the LLM's.txts, txts, which if you're not familiar, this is something that a lot of people do as a pattern to take all of their documentation and put it in a big text file for ingestion of LLM. And so I was going to try to do this without doing rag or retrieved retrieval augmented generation, which I've rolled myself several times. I've also used Azure. I've also used AWS. I've also used GCP. I've also I've done it everywhere. Edge functions base. But I was trying something different where I was curious if instead you could just do flash all. So flash being Gemini and they have a million context window. And I was running up on that to begin with. I was like, "Oh, I'll just take a bunch of these LMTXT files out and then I'll just prompt inject it, which means when you send a message off instead of it just being like, hey, and then well, I guess that would be you, right?" So, this is let's say user, then this is assistant. So, if you said hey, and you had a question about the Superbase.js JS SDK instead of it just jamming all of those in there, which I was trying to do because I thought it'd be cool and I thought I could cache it so it'd be quicker. Instead of doing that, you actually need to do some form of rag. It would be better with that amount of tokens. So instead of it doing that and then just like coming back, you'd run out of this really fast, but it was late. Instead of doing that, you'd go and it would have lb with vectors and then you'd have a system prompt that would say something along the lines of, "Hey, you're a super helpful research developer assistant and you're responsible for answering questions about XYZ topics. Prefer you the use of these documents." And then there's a lot of other stuff that you can configure. Rack is kind of a rabbit hole, but that'd be an easy way to do it. So, that will be coming out. If you want to, you can go and check out the link now, but I'm not going to reply to comments until I put in a video that is just a quick loom explaining how to use the turbo kit. It's got a bunch of good stuff in there. I'm adding tests and something that I came across that is the purpose of this video, which is just the goated AI coding stack. It's absurd. I need to add that because I would feel remiss if I didn't. So, let's get into some of the AI news. This will be evergreen because I make takes on it. So if you're watching this at a later date, you can call me an idiot or you can call me a genius based on what I tell people. Let's see what old says. I think it's the degree of automation that matters to get to 100 truly 100% automation. You know, you can make a complex thing and never touch code. That that's one thing, but I'm less interested in that question than when a coder becomes 10 times more productive. Tada. That's it. I talk about this all the time on the channel, but competency is an all-time premium because the amount of leverage you get, the multiple you get on what you're capable of doing, obviously much higher if you're competent. So, that's interesting. Another one, the godfather of of vibe coding, Andre Carpathy. And that's downplaying him. He's much more than the guy that just said the word vibe marketing, which is a pretty popular trend. Let's just take a look see real quick before we get into it. But he wrote an article yesterday that came out and it's really interesting because what he does is he's just like, "Hey, this whole the future is here but it's not evenly distributed thing." Is a good pathy statement, but he brings up the fact that it is distributed. You just have to use it and it's multiplying everyone's output, all these great things. So, I'm going to own these keywords. That's the point. That's why I bought like 10 different domains. That's why you see Vibe coding and Vibe marketing and AI growth agents and all these things in every single title. It's what I talk about. It's what I'm interested. It's what I believe in. I believe strongly that there will be a one person billionaire. I'm not saying it's me, but I know that that's going to be a thing. A oneman team. It's going to be maybe like a 100 person like a 100 agent squad that's helping. But yeah, these are taking off. So then we get into some of the things that Facebook's done. Facebook aka Meta released Llama 4 and it's not doing so hot. There's a lot of rumors that they gified their arena score. They trained it specifically to win on a leaderboard. Then the leader who runs the whole program stepped down. Whenever you see stepped down, it probably means they were fired or something went down. But that's like the PC way of saying it. And so on here you can see, yeah, they have a 10 million context window, but it does the worst on retrieval. It's horrible at coding at a 16% on the coding polyglot test from Ader. If you're not familiar, Ader is a fantastic CLI tool. Just teaching people how to use Whisper in school. So that's why that's But if I go to my workflows, you can see Ader is on here. And that API key is burnt. So good luck trying to take it, you freak. But Ader is a CLI tool for coding. Pretty good one. And you can see that Llama is not doing so hot on there. On the opposite side, you can see that Gemini 2.5 absolutely crushing it. Kind of wild. Really, really good to see. And you can see there's more evidence that people are finding that they just kind of gained this. It's like, okay, you did really good on this thing, but you're 16 on this other one, and you're It's just all over the place. Literally, the only one that was good was the one that it may have gamed. So, announcement on GPT25 Pro from the guy at Google, they just extended their usage limits on Vertex. And the way that you set that up in short, I go into this in like every little detail in school, but in short, you will, if you want to get this as a hosted instance, I guess you call it, it's not totally hosted, but vertex is the most control you can get over Google's platform and their AI offering. And Vertex has the model garden in it. So you enable the model that you'd like. So in our case, obviously, Gemini 2.5 Pro. And then you can go in and you can actually set up access to it, right? You need your API key obviously, but then you do something called a G-Cloud O. So it's in the CLI and you'd authorize with it. And then because it'd be coming from an app, then you do just Google up like Google Cloud O application default login. If you put all those words in, it will show you all the steps to do this. It gives you a JSON credential tied to your CLI and your cursor or whatever ID you're using. So if those requests are coming directly from your CLI, maybe it's coming from client or something, then it'll be okay. It'll work. Let's talk about this. Johnny IV and OpenAI are making this her device. Kind of a wild thing. They're considering spending 500 mil on some sort of device. So that's interesting. If you didn't know, Johnny is the goat designer who made iPhone. Now on to this. Everyone's freaking out because the CEO of Shopify is running his motor about the use of AI. He should be. And there's a lot of doomers in here that are like so scared. But at the end of the day, he's talking about how in order to stay ahead as a company at their scale, then you have to be growing, I think, something like 20% a year, which sounds daunting, but it's not. If you have leverage, which is what I talk about all the time. Yeah. In a company growing 20 to 40% a year. It's similar to the Red Queen Race, which is based on Alice in Wonderland story where you have to keep running just to stay still. True. It does sound daunting, but given the nature of the tools, this doesn't even sound terribly ambitious to me anymore. It's also exactly the kind of environment that our top performers tell us they want. learning together surrounded by people who are also on their own journey of personal growth and working on worthwhile meaningful hard problems. Yeah, that's the whole thing. That's why I'm like really like super bullish on community is if you put it in place and you put the effort in, it's a ton of work up front. I definitely underestimated that just to be totally honest. All these little things around it like oh you got to make like a video sales letter and then you got to like seed it with content. But then now I'm seeing the purpose of it because if you have someone in there that's working on a QuickBooks integration, they don't have to like go to Reddit and dig around if someone else in there's already done it. Just like, "Yo, what's up? Can you teach me this?" And that's already starting to happen, which is really cool. So he talks a bit about using it effectively. Yeah. And then basically saying you got to be using this thing all the time or you're going to get left behind, which is more or less true. You just get leverage, that's all. So every every time you're thinking of adding headcount, then seek AI first. That's not a doomer quote. That is a smart one. And then what we're going to be talking about is is the goat setup. This is a product that I used all day yesterday and we'll be demoing for you guys today on starter kit which is Claude Taskmaster. It is absurd. It is everything I've been looking for. Every problem that I've been running into gets solved with this which is great. Other things OpenAI launching their responses API. So go check that out. The reasoning tools. I'm not gonna talk about this one. Sorry bud. I haven't used it so I don't I don't know enough about it to talk about it. And then Cloudflare's launched auto rag in open beta. Now, as someone who's only used Cloudflare as a CDN and for buying domains, hosting and stuff. Well, I guess I've done Cloudflare Wranglers. I've done Cloudflare Workers. I've done R2. I've used it a fair amount, but I haven't gone all in. So, I'm really curious because they just keep releasing all this stuff. So, the fact that they can just spin up a pipeline. I really like that. And the offering of having everything in one place is really pretty enticing because it's I've done this in Azure, but then you're in Azure and it's awful. Makes you want to cry. It's so ugly. Just like convoluted. This is crazy. This guy made Tom and Jerry with a prompt. Like literally a minute long scene. So there's going to be movies written by AI, which is wild. Look how long that prompt is. But that's that's where we're going, ladies and gents. Then what else? There's a lot of news today. The prompt watch. I just wanted to show this because I'm gonna be exploring it. There's a whole service around boosting your brand visibility in these search engines or these which are search engines, but it's basically just SEO. And that's a big part of why I make media. That's why I do the long form stuff and that's why I'm rolling out this vibe marketing strategy is at the end of the day, the shift in market share for search and discoverability will go away, dissipate, slowly atrophy from Google to these, but they're still using similar ways of finding stuff. They want to deliver good answers. That's what they optimize for. They're working really well and growing really fast because they don't have a bunch of blue paid ads in front of it. But Google's also fighting against that. But it doesn't change anyone's strategy. I think this is just SEO with a new flare on it. So, it is cool to see that people are building them. Like I said, Cloudflare is building a lot of stuff. They have this agents website, which I'm going to do a deep dive on in Vibe with AI. I think it's really cool. Anytime I see something like this, I just go build a little throwaway app. Try to time box it to an hour. It typically goes way longer than that because if it's interesting, then you naturally will go and do it. But if you say you're going to time box it, then you go and start and then I'd probably be using something like this that lets you convert your Versell next.js project automatically to Cloudflare. So that seemed pretty interesting to me. There's the auto rag thing. What else? Oh, this may be the best thing I've ever seen on the internet in recent times from John Carmarmac. He was a huge inspiration for me when I was learning to program. I literally had his name on my calendar every single day where I just pretend like I'm going to become him. He's the goat of programming. But someone was complaining saying, "The way that Microsoft recreated your game and it's AI generated is just disgusting and spits on the work of every developer everywhere." Well, go cry river. No one cares. He immediately says, "What? This is impressive research work." And then he gets a, you know, reply saying, "I respect your work all this, but you can go read it on your own time, but I will pull out the best part, which is building power tools is central to all progress in computers. Everything that John built was for someone else to then build upon. The world moves forward when you take these primitives and you make them better. That's a power tool. And then on the idea of people having less jobs in the space, well, just get better. It gives you a multiple. And he just says, don't use power tools because they take people's jobs. It's not a winning strategy, which I think is fantastic and a really strong truth. Uh, other things. I was going to watch this, but it's too long. It's just on DPS stuff. So, now let's get into Taskmaster AI. This is the bulk of the video which is the goated cursor stack. So what we're going to do in this is solve the problem that we had before and I'm gonna just go through and show you how the tooling works and then I can go through and actually execute some of the tasks so you can see how it works. But we're going to uh swap this massive prompt injection for doing rag. We have a couple of options. So we can use a Dino edge function. If you're not familiar with Dino, it's the word node remixed and the developer of it now works in Go, but doesn't matter. Dino, it it lets you basically make function calls outside of its own environment. So you can use they use them at superbase all the time where I can make a call to open AAI. It's not reliant upon XJS routers, route handlers can do that. I can and it would go and call make the vectors and then store them and do it automatically when it's use doc somewhere I can do cloudflare I can do Azure so I'm gonna just say that I want to do it in adino edge function and what I'll do is I'll open up the turbo kit you guys if you comment below turbo kit please then you guys will get a comment when it's totally ready if you guys are antsy and you want to check it out now before it is ready then it is available So what you do with this taskmaster is it keeps control of your task just what it sounds like but it's a really nice CLI tool that also has MCPS. So the point of it is you can install it globally and then when you do an init the initialize function it's going to add a few things to your project and you're going to need an enthropic API key. I don't start crying because I I use it for the entire day yesterday and I think I spent a quarter. So, pretty good. And the way that it works is it's going to add in two folders. It's going to add in this tasks folder and it's going to have all the different tasks in it. And the tasks are classified by a bunch of different things. So, it has the dependencies, very important because you can set up a dependency tree and then it knows where to go. the priority, the description, the details, the test strategy, status obviously, and then it has a JSON file that has all of them in here where you can then break them into subtasks. So, this is me last night trying to figure out how to do Gemini 2 flash in a cheapo way, messing with cash and next didn't like that because next cache are not friends until cash comes out. It was messing with the build somehow. And you can add subtasks, you can expand things, it figures out the complexity of the different tasks that you want. So then you're driving from here because it has MCPs because it's added this tasks and it's added this script that's running it where you can put in your PRD and then it creates the tasks. It also, like I said, has the complexity analysis. So if you had tasks and you're trying to figure out which one to go after, it has a few different parameters to look at. You can look at the complexity store score as an input. It'll look at what's dependent. So it knows what to do first because you can't go in the case of something simple like we're making a very fancy header for a client on their website. It's got a lot of animations. Well, we don't want to do the animations before the header is actually there. So that's an obvious one. But it's going to be looking at that and it has the PRD example in here. I use a specific prompt for writing my PRDS ahead of this and then I reformat it into that and I bind it to a key. So I do that. If you want that prompt then type the word PRD prompt below. You'll get this one as the expert technical product manager for feature development. That's what I was before I was a full-time developer and research guy. But I know that job really well obviously and I worked with a lot of architects. So it includes all the things that I did. And then you have the prompt variable which is PRD instructions. This is where I'd go and I highly recommend using Whisper Flow for that. We're going to have the CEO on to our community to kind of teach us the tricks of the trade. But he beat like a three trillion dollar company aka Apple on voice. But by using Whisflow by running your motor inside of PRD instructions, you will give it all the details of it. And you want to explain the classic product manager method of like, okay, narrow down the problem, try to come up with a solution, provide context of examples if there are one. If you're doing something heavy like a reverse engine on an existing product or an existing feature, then I'd go to deep research before I do this. So I'd run a deep research. I'd run the reverse engineering prompt that I've written to then get all the context for that to include under just a context header in the PRD instructions. So then you've narrowed the problem. You've come up with a solution. You've provided context on the reverse edge. You've provided links to in this case it' be like dino edge function with superbase. And then you provide rabbit holes areas that could blow up and then no goes. So areas you shouldn't touch because you need to provide rails for the LLMs. So I run that inside of here with Gemini 25 in ask mode. It's going to give me a dope PRD. Then I'll take it and I'll say this is a great first draft. Now reformat it using this format. I paste the format and then below it I say and then write to prd.txt. I change it to agent so it can do a write. Now I have my thing and then I run taskmaster task master space uh I think it's parse. I haven't memorized every single one of them. You basically run a command and then it will parse out and that's when it does the call to claude. So yeah, I do taskmaster list is one of them. I can this parse prd. So by default it will break into 10 tasks and that's awesome. But the point of prd is to capture the entire essence and detail about your idea. It'll take the PRD and turn it into structured dependency aware list of tasks. means it'll look at your entire PRD and figure out which tasks are foundational which are built upon that foundation and so on. So you end up with a linear path to completion simply by completing tasks in order of dependency. It also has a table. So you'll notice if you went into your package.json that you'll see it added a CLI tool which is the cle table 3. So it's using that to then show you the tasks, right? And other thing that it adds is in your MCPS. So under cursor, it has the cursor rules, the dev workflow that explains how it's working. So it gives you these global CLI commands if you weren't using MCPS and it will go and do that thing. Now, it's funny. I'm not going to demo this because I'm running up on time. We only have 30 minutes together. But you know exactly what to do now. You go to the Taskmaster AI. You download it. You run a couple of commands. You get your API key. You run Taskmaster in it. It's going to ask you a couple questions about your project. You write out the PRD as I explained, which he wouldn't explain on that GitHub. So that's the value you gained here. You get a couple of the prompts if you want them. Make the comment. And then you're off to the races where you would just say, "Okay, let's get started. Which task should we start on?" It's going to go and it'll call one of the CLI commands like taskmaster list. It'll show you them. It's going to say, "Hey, we need to get started on number one because it's, you know, two through 10 or dependent." If you get to one that is way too broad, and this will happen when you give it a very beefy PRD is you'll say expand on this with research. That's when it does a perplexity call to get additional information if you didn't provide enough research. And then if it doesn't need additional research. You just need it to be expanded without a perplexity call. Then you just say expand on this one. It seems tough. And then it'll it'll go through that task and come up with more subtasks, more research and it's just absolutely ridiculous and I really really like this workflow. So definitely check that out. And off camerara I will be doing this. I just think that it'll be too boring for most people watch. And check out Aayal Teladano. That's the guy who's made it. And lastly, just to wrap, we can do this together real quick. But the way that I can make like a quick writing agent is I can look at someone who writes a bunch of stuff and like does this full-time for a living and then find their kind of rules. You'll see a lot of copywriters do this, but I'd look at his, maybe I'd use Tweet Hunter X to find one that's popular, maybe that the market thinks is really smart. But all I'd be doing is I'd be taking this and I'd be turning those into rules and I'd use them as the custom instructions inside of a cloud project or inside of a chatbtd project. And then I'd go to every writer who's had things pop and I'd pick those and provide them as context in the examples. So custom instructions and examples are the way to go. and you'd make the writer specific to the platform you're making content for. So, if I'm doing tweets, then obviously I'd get tweets. If I'm doing Medium post, I go look for the ones with lots of claps. But it goes back to just how to write good prompts. You know, that's the most important thing is giving it a role, giving it a task, giving it examples, then let it go. And then getting those back, reading them yourself, and making it better. And you do that a few times and then you go from, you know, like the 80% one to the 90% one and so on. That's what I'm going to be talking a lot about on this channel. You'll also see this in action. Tuesday, April 8th, I'm shooting to roll out three more videos on my main channel. You can go check that out. And I give all the behind thescenes stuff on the school. You can check that out, too. But yeah, it's going to go from the daily video and then I'm going to be rolling out the main channel videos. And then I'll be using that process I just explained to then go into short form and then go into written and then go into image plus written and then go into voice. All these all these things. So if you enjoy the video, make sure you like it. That's your way to pay it forward and I appreciate your time today. Have a great one. See you in the next one. Aka tomorrow.