Avoid this AI Agent Trap, Build Workflows FIRST(Free template + 5 step algorithm to win)

there. I'm Parker Rex. I led technology first startup that sold for 23 million bucks. After that, I tried to do Airbnb for music. It didn't work, but it lit a fire underneath me to go all in on programming. A year into that, Chat GPT came out, changed everything. This is the biggest thing ever. So, I've been using it non-stop for the last three years to learn how to leverage it in coding, marketing, general business, productivity. On this channel, I share a daily update on my road to 100K in profit a month across AI services, SAS, and a community that I'm building. It's like a a media company and allin-one. Sounds like a lot. It's only possible with AI. So, the way that we're doing these videos, I've tried a couple different formats and the one that makes the most sense to me is we do five questions that you guys put in. So, any questions that you guys have, please drop them below. And it seems like my mic is a little loud. I got some feedback that my mic should be louder. So, hopefully this sounds okay. But, we go through five questions that anybody puts in. and make sure you ask any questions you have. There is no bad question. We go through five pieces of news that are relevant if you're building in the AI space, some takes on them, and then general strategy of the stuff that I'm doing. And it typically goes like this where I'll do the strategy and then that turns into a main channel video. So, I'm covering the thing that I'm actually going to build and put out into the world and then I put it out into the world and I package it up and I put it onto the main channel. The one that you're watching right now is the daily And the main is just my name, Parker X. So, let's get into it. On the question side, PRD prompt, thank you for fixing the mic volume in this video. Nap, no problem. Of course. Thanks. Said, "Great content, mic volume, PRD prompt, and turbo kit, please." Okay, we have a purity prompt that I'm going to be sharing and it's part of the strategy, right, is put out stuff that people want and then have a way to be able to communicate with those people in the future. So, instead of just posting it as a like notion link where I don't have the ability to communicate with them moving forward, I need to set up a system for that. But, if you guys do want the PRD prompt, this is what it looks like. I'll open up a code editor and just show you guys real quick. And I got feedback that I should make this larger. So that should be big enough. Let me know if it's too small. Again, you guys are the oxygen for this channel. So, it's really helpful when you guys say stuff. It's so funny that you can't do it in there. How about let's go to text edit. New document. F1. Cool. So this is the PRD and it is really nice because it actually asks you to fill out any open questions that you have afterwards. PRD is the key. It is the key people to making these agentic systems work really well. And we're going to be talking about agentic workflows versus agents. That's kind of the theme of this video. So each video should have a theme, I believe. But yeah, so I'll be building out a landing page and then I'll be replying to every single one of these about those topics. What else? This video showed in my recommendations. I've done a lot of serless functions. I answered this in yesterday's video, but he basically is talking about how if you just have aic systems that are sequential, so it's like a waterfall effect, then you don't get the most value out of them. Totally agree with that. And I'm going to be doing a deep dive of this on our Discord and then I'll probably share it on some of the channels that we have across YouTube. But if you didn't know, we also have a Discord and I think I went into this on here on the Aentic side or maybe I think it was actually in the school. But if you're curious about going deeper on Agentic systems and how to make the most out of them, there's this video called Agentic Patterns. And I'll just open it up so you can see. But this creator is really cool. He actually follows this channel which is nice. But yeah, go follow code with stav and just watch the first two minutes. Even if you're not technical, it's worth it. So you'll understand agentic systems a lot better on the new side. So I had mentioned this turbo kit. So turbo kit is something that we're putting out through vibe with. We have a lot of different open source projects that we work on at Vibe with AI. So one of them is turbo kit. So we have a channel for each one of these and whenever there's an update there's just a quick zap that launches and says hey this thing got updated. The idea being that a lot of the agentic sorry a lot of the web apps out there that use AI have the same primitives. They have the same sorts of things but setting them up over and over and over again is really bad. You can think of this as like a wrapper almost like a wrapper template. So we're putting one of those together and I originally just rolled a fresh one but another one of our open source projects map which is a fiveinone productivity system. It's multi-aggentic platform that allows you to train a model on what your future self looks like. So it lets you kind of figure out those goals and then break them into tasks, break them into your calendar, all of that. And instead of doing a separate turbo kit, what I realized is I'm doing double work. So, I jumped in last night and just kind of went ham on all the stuff required that I need to get this project running. And it has all the documentation in it. So, let's see. Let me pull that up. Where is map? Is this it? Yeah, this is it. But this will be what the turbo kit is forked from. So, let me increase the size of this. But in here it has a full readme that explains all the things that you need to do to get up and running. So the database, how to use Docker, some basics on how to run things. So build scripts, the MCP setups. If you're not familiar, this allows you to communicate with a service using AI and natural language. And so this one has a superbase in here. So you can use superbase both locally with an MCP as well as remotely. It has the Taskmaster MCP, which if you're not familiar, check that video out on my main channel. But Taskmaster is a product manager built in to make sure that your AI is staying on track. It is by far and away the best thing that I've come across in the space in terms of keeping your agent online and actually pulling out more value and an idea. It's absolutely nuts. But it's going to have that in there. It tells you all the steps on what to do. So I worked really hard to make this as easy as possible. It also has cursor superbase rules. So you can see over here in the cursor rules if I see like database functions. These are official superbase documents on how to get the best out of it. So, I used these and was able to take 5,000 lines of SQL code across multiple files instead of just doing a squash, which means to take all the migrations and then slam them together without optimization. I was able to take this and then optimize it using those rules. And it's great because now all the queries will be much quicker. Databases are powerful. Most apps are just really thin wrappers on top of a database. So, this really matters especially for the newer vibe coders out there. Think of it that the database is not only a spreadsheet. A lot of people just pitch it as that. This can have its own functions. This can have its own triggers and automations and calculations running on it. It does compute for you. So the more that you can put on the database, the better in a lot of cases when it comes to transactional stuff like making sure that numbers are correct and all of that rather than putting it over the wire and doing it on the client, which would be the front end. So it includes all of that. It includes all of these. What else? The branching strategy. This goes into actually the open source side of map. So, we're open sourcing this all the people in the community and it explains more so like how it has a homepage that has your daily summary with your tasks, your local weather, the option to plan the ideal day for you, chat with voice support, a unified message feed that includes Slack, email, iMes, Twitter, WhatsApp, social media. So, you can think of like the texts app. It's an aggregator for that. So all your coms, your health map that has a dashboard with health metrics, all this stuff is going to be open source and available for anyone that's in the community. And I'll just show you a quick demo of it so that you really understand kind of how this product works. But to the point of the turbo kit, it's going to be forked out of that project. And I'll just sort of redact a lot. I'll be pulling out all the specific things for map and just use the primitives of you know the documentation and all sort of those things. So, I made this demo back in June and it was when I was trying to raise venture capital. It didn't work, but the product's fantastic. I really So, you can see this was the original homepage and on the homepage you can see that I had set a goal with them saying I want to run a marathon by this date. and it had a really basic view of what your day looked like, the tasks that you needed to do, your goals for the quarter, some stuff around calorie tracking, and then on the right hand side, you can see the back end of how it's actually creating goal plans. So, if it didn't have enough information about the goal plan, then it's going to break it out. So, it might say, if you just scroll up a little bit, you can see that it actually breaks it into what type of goal is it, and then you pick one, and then it's going to bucket things and have different logic based on the decision that you made there. Then it breaks it out into the separate tasks and the separate calendar events looking for calendar conflicts in your existing calendar. So like I mentioned earlier with the database, it's doing the calculations, it's doing the batch extraction, transformation and loading or ETL from Google calendar into our superbase database for 36 months of calendar data. And that's inclusive of six different methods within Google calendar. Just so you guys know, Google calendar has a lot of different things. So if I looked at the map project and I just typed in calendar and I looked at the actions we have creating action events, this is the bridge between our front end and the logic behind the scenes with Google. So there's a a few thing things that you learn. Um and these patterns will be really nice to sort of mimic when you're doing your own things. The way that I do this is I have a library and then within library each one of these folders represents the integration code that we're writing. So if I have Google calendar you can see all the different methods in here. We have access control list calendar list channels colors customs events free busy and settings. Customs is the only one that is not mimicking the exact function from what Google provides. So adding a sleep event obviously that wouldn't be in Google but it maps perfectly. If I typed in Google Calendar API node, I'll go and I'll see here. And then when I go and I actually see the reference for it. Let's see if I can't find it. Um, it's already enabled. Maybe I'm on the wrong account. If I go into creating events, you'll see that we have all these different things in here. I'm just quickly trying to find I can't find the page just now, but on the lefthand side there would be a method for each one. So that's what it maps to in short and I just JavaScript counter API reference. Here we go. So you'll notice that under V3 you have all of these different things and so each one of these represents what's in the pattern that we're using. really easy to work with. So you have deletes, you have gets, inserts, lists, patches, and you'll notice that we have, if I went to calendars, you have clears, deletes, gets, index, patches, right? And then in here under patch, you'll see I have all the TS docs. So it says this partially updates a Google calendar with new properties. It's using a Google calendar service above it. It's all type safe. Very nice. So you get the kind of the squiggly lines to let you know if you've missed something or not. And then if I clicked into with Google calendar, you can see that this is where it's authenticating. And this is a singleton pattern for the ooth client. Meaning that this is the one-stop shop where every method that will be called when we're talking to Google will be running through rather than having multiple instances. It's just a really nice pattern. So that is a huge huge news on the open source front. And let's see. I guess it's like semi closed actually. No, it'll be it'll be fully open. It's just if you want to collaborate with the people, that's where I kind of gate it because I want to have our community have like a big incentive to work together on it. So that's on the turbo front. Then let's see, did I grab the video from school? Yeah, I did. That was about the Gentic thing. Other news stuff, Canva launched a competitor in the no code space. Not really no code, but like if you looked at Lovable, it's a competitor for Lovable. So Lovable, you guys probably already know because you're watching this video, but you can build some stuff with it. They've added the ability to actually see the code and modify it. Same thing with bolt new. You guys know this one. What do you want to build? The problem with a lot of these is when you get out of them. So you can do all this stuff in here, but then leaving it, it leaves a lot of room for error. like it will use the wrong libraries, it will use the wrong configs because it needs to run within a browser. And so a lot of the folks that go, they try to leave this and they really really struggle to get what they want out of it once it's in an IDE. So that's what we also do is we find a lot of the people, we only have like 15 people in there, but a lot of them are making that that jump and it can be really tricky. My recommendation for you is if you're doing it on your own, then read documentation. look at what version numbers are newer that the LMS are probably not trained on because if Superbase has a new version that came out after October or after March, then the LLM is going to struggle because the base of its knowledge was not trained on it. So, I think Canva is going to have a big leg up against these folks because Canva has 250 million active users and these are newer startups. It's just going to be up to the delivery of it and the quality of it. So, they have the distribution. We'll see how it goes. It's TBD on the quality because Google launched theirs and it wasn't so great. Lastly on the news is Sam Alman had a talk with TED and sort of my takeaway was that it's growing like gang busters. They have 500 million weekly activives. So I'm not sure if you just want to extrapolate that and say that there's two billion people using it a month. That's maybe true. I'm not sure. but they just doubled their usage through that image gen thing that went super super viral. And my recommendation for everyone in our school is when you're developing on these things. If you are, then pick one and run down that lane. I because I do this full-time. I don't have a job. I literally just build SAS and I build community and I build all these things. I use two. I use Google and then I use OpenAI. But it's because I've used I went all in on OpenAI since their first thing came out. When 35 and the AI SDK came out, I went all in and now I do two. So, on to the strategy. The name of the video was about agent versus agent. And this is something a lot of people confuse. You don't go agent first. Agent means that it's standalone, that it can make its own decisions, and it can lead to a lot of lowquality experiences. I always reference the Elon Musk five-step algorithm for this because it's incredibly pertinent. It's incredibly relevant rather where I think I had it where was it? But it's a fivestep algorithm that basically says the last step is to automate and that is totally true where you want to question every requirement that's in the thing that you're trying to optimize first. So if you've watched any of my videos before, if you haven't, go watch some of the previous ones that are popular. But whenever I'm automating something, I do it manually first because I need to know the steps because I actually need to go through the thing. I can ideulate on what the steps will be and that's a good starting point, like a nice exercise, but then I go manually do it. For instance, with this YouTube channel and my main channel, I do seven videos a week on this channel and then I do three on the main. I did it manually and I tried to do it manually until it broke and then I figure out all the steps. I question, well, does this step really need to exist? For instance, having a third party that does audio optimization. If it takes an extra 15 minutes each time that I do that across the week, then after 4 days, it's already at an hour. After seven days, it's at what, hour and 45, then that's just too much. So, I question it. Maybe it doesn't need to exist. Then once I've deleted a few of those things, I try to simplify and optimize the process. then accelerate the cycle time, the amount of time that I can get through it. So in this case with the daily video, I was struggling to figure out what's something that can be easy for me that keeps me on the ball that delivers as much value as quickly as possible. Then I come up with a checklist. So that's an example here. Shout out Basil for helping me with this. But it's figuring out like, okay, there are certain things I need to do. I need to write the title first so I know what I'm doing. I need to have a hook so I know that I'm actually saying something interesting and delivering. I want to plug and plug is a tough word, but I want to mention the fact that I have this community because I know the people in it are having an awesome time in there and I know that they will gain value. I just know it and I know that it's going to work. I need to mention that I have this other channel. I want to mention things that I can give away for free. Not just being like, hey, what's a thing that I can say like comment this word for nothing. This one forces me to then have something worth mentioning. So it makes me build and then obviously there's post-processing things. So it comes back to checklists and then the last step is to automate right. So that's the kind of guideline for these. Now when I say agentic workflow, what I mean is there's humans in the loop. It means that I've come up with the checklist. I'm doing all these manually. I'm not just saying to an AI like go and write the title because the title could be bad or go and write the script that includes all these things because these could be really cheesy. I'll do it manually a bunch of times and then step two would try to it would be actually building the Gentic workflow whether that's in make whether that's writing that in handwritten code but then I'm chaining these together and each time that one goes to the next I'm reviewing this output. So if something happens and something's generated and there's an artifact that's created, there's something outputed like a title, I need to jump in and be like, well, how does that look? Because I don't want to be a bot. And so this is an agentic workflow where these things might have tools in them. It might have, you know, for title written, it might be a research tool that goes and finds things that are working. It goes and finds the news that's interesting and then it might summarize them. That's two tools. So it might be that and then it does the output. So they still have tools in them. They're still AI nodes if you will, but they're not agentic in the fact that they can just they're not like a full agent. Agent means that this would just go all the way through. And if you did that prematurely, back to what we just spoke about about automating, you're going to end up with bad outputs. So that's the big difference. So you always go agentic first and then you go to agent. So you're seeing it in real time. I'm fully transparent on everything that I'm doing. If you guys have any questions about this again, please comment below because I'll be going through those as you see with my new process. Five questions, five news, and then strategy. But that's that's the gist of it. And I'll be sharing the actual code behind the scenes inside the school obviously. But then you see bad examples where if you do agent first, then you end up with this. Like there's this guy who comments the same thing over and over and over again. and I noticed it and I'm like this is like maybe funny the first couple times but this guy comments he's a competitor against Alex Hermoszi. Alex Herozi owns school. This guy is a worker at and they're different products but they compete in the same space. So he has an automation that is full agent mode where it has rules that say always have this gap between always use lowercase. use this hook that I think is really good. Only use these emojis and make it maybe up to five things with something inspirational and friendly at the end. If I look at that and then I search his page, you'll see, wow, he uses the same thing every every time. And he might think it's working because like, okay, he got two comments, he got two likes, but this goes into the fact that like to me this looks like, you know, shout out Joseph Henry, like good job for setting something up. I think it's garbage. And maybe that's a little aggressive, but it's just not the way that I would take it and you don't want to be a bot. But versus if you did this manually and you you found things that work and you still had it go back where it's like you get the maybe it's even a text message through Twilio that says like here's what I'm going to say. Text back Y if you want it to go and then it gets the check. Like to me that's a lot better. And I don't think that there are things yet that are full agent mode that fully represent you because your thoughts will change over time. That's your moat is that you if you have something that's programmed to be you, it'll only be so good for so long. Otherwise, you're just stale. So, it's going to require a lot of tweaking, a lot of that. And then I wanted to talk about this. This is something that I believe very, very deeply where we have people that are joining the school and we put them into two buckets. And it's really code and media, right? And so code involves beginners and that's why I named it vibe. It involves experts. You have the beginners, you got the experts, you got the mid mid-level. Then on the media side, you have sales, you have marketing. And this covers a lot of areas of business. And a lot of people are wondering, they're like, "Well, these two have the most leverage." I always like this is to me like the juice and I think there's massive crossover and there's massive ability now a day with AI knowing how to talk to it knowing how to unleash the PhDs that live in your computer to do both because this is so true the old path was school job savings retirement the new path is skills audience assets it gives you optionality I think that everyone should get more technical it is so easy you just have to trick yourself. Sounds crazy. So that's a message to media folks. But a message to the code folks is everyone should learn how to adopt this media set of skills because when I'm making a video, it is an always on thing that has leverage. If it's good content, then it will get a lot of views. And the way that you make the good content is you put the checklist in in place. Like I think that this video will outperform a lot of the videos that I had before it because it had a system and that system still requires creativity within it but has structure so people know what to do. So moving forward this is kind of the thing that I'm going to do where I have these topics that I cover and then towards the end of it I have the videos that are coming next. So on the main channel it will be about the agentic workflows. So actually going in deep on them the five-step algorithm talking about Elon how it is so important to follow this actually building an agentic content workflow this will be something that I'm going to be using and then I'll share for as a freebie I'm going to be covering an agent hub. So this is a place where I have my media scorecard. So a scorecard looks like this. This is taken from when I was doing leadership at delivery dudes the company that we scaled to 73 million and sold. We had a scorecard that you look at every week. I look at it every day and it's falling behind because I've done everything manually so far. It is breaking. So that's now my wakeup call to be like, "Okay, time to optimize. How do we make a media scorecard that not only just fills itself out, but then has all the context of the performance of the videos so that I can analyze that, know what is working and what is not working, you can bring in other things like current trends that are going on. So I can see across even little things like if I typed in vibe coding or vibe marketing, is this still something that's on the rise? Is this something that I should be focusing on or is it not? But there's a lot of different things you can do when you start thinking about it where it's like, oh, it did have a spike. It seems to be gaining popularity. Maybe we do a video like that, maybe not. But it's really like wrapping all of those things into one scorecard with the video tracking. Another video will be about the YouTube agent. So having a scorecard like I mentioned obviously but then the flows of the actual automations. So when I list out all those steps where it's like you know 20 different steps for the YouTube video what does it look like when I actually build out the workflow versus doing it manually. Same thing with this concept is going to print. I think that that's something we talk a lot about in the school. We have people that run their own businesses in the media space and we're realizing that one agentic workflow is actually a skew and it actually if packaged properly is much better to offer as a software as a service or something usage based than to go and set up a bespoke NADN or make thing because when you have a bespoke NAND or make thing it is not levered. you can get high ticket stuff and then you're like praying for retention. I don't think that that's the future. I think the future in order to build enterprise value in order to actually have repeatable leverage, you need to take one of these, make it so good for one specific workflow, then build a brand around it, build a tool around it, and then start charging. One example just to close is I have a prompt library. And this prompt library is the antithesis of the make an end thing because it's not no code. But what's so cool about it is these are essentially just XML files that go and are chained together. So if I have this brainstorming prompt, it has all the XML things. XML is what the internet is comprised of. It's what the GPTs are trained on. It's this flavor of XML, which is just straight up XML. And it has tags around it. So what would happen is you run one of these and I have what's called a Mimo prompt library. So, it's not running right now. I need to fix it and update it. But, MIMO notebook, what this does is it allows you to actually run the thing that is in the XML. So, I have a playground that has a bunch of different prompts in it and then I can just select them and then have it run across all the different providers. So, I can see the different outputs. So I can really quickly iterate on all the different prompts in here to understand. Well, this one is a writing one. So maybe I should use entropic. But I don't actually know. So let me see that up against 40. Maybe I can prompt around 40. Reduce the cost by 90%. Maybe I can go open source. What if I went open source and it's totally free? What does Gemini 2.5 look like if we do this? So this would be the the kind of the coding way. And I actually prefer this. you get higher quality stuff and it's a lot easier to see. I've tested running and optimizing prompts with make, but you still like there's something about like clicking on a really laggy node and like waiting five seconds versus just bam bam bam there they all are. So I will be making this a product as well. Like this meta idea is like you could make a product that's just called the prompt library that takes this library all of your best ones and then sells that as a product. This one is technical writer for doc rocking. What it does is it goes out given a parame URL it will write out a perfect document for that library with usage with all the things that you care about in it that takes away all the pain of the the research and then you hand that out as context. So there was a lot of stuff that I covered in this video. I hope it was really really helpful. I hope that some of the improvements that I've made shine through in this and the quality of the video. Of course, I have to ask you to like and subscribe. That's my way of knowing if this video was good. This will go out to more people. If you want behind the scenes content of all the stuff that we're doing in Vibe with AAI, go ahead and join. We're about to raise the price because that's just how it works. I would rather have a really high quality community that people are willing to pay like a dinner's worth of money to be in it for 30 days. We have calls every week. We have two different calls based on the time zone and we basically just like run the motor on different topics and it's been awesome so far. So, enjoy the rest of your Sunday and I'll see you tomorrow.