How I Sold a $25K AI Agent Offer (Not N8N or Make or Vibe)

Hey there, I'm Parker Rex. I led tech for a startup that scaled to $73 million, went down to $47 million after we were fighting with Door Dash every single day, and then sold for $23 million bucks. After that, I did a lot of different things. And on this channel, we're today going to be talking about the $25,000 agent and how I'm going to be packaging up AI agents to hit 100K a month in profit while also building a SAS startup. Sounds crazy. Everyone's be like, "Wow, you're you're crazy." But these are possible with the right agents in place. So, I got some feedback on the videos and I'm supposed to talk a little bit more about why you should listen to me. So, I've been in the startup game for a long time. I'm 32 and I've done pretty much everything. I've been a little nerd forever, but yeah, I did the drop shipping thing while I was working at the startup. At the startup, it was called Delivery Dudes. At first, I was the guy who' like buy all the things for the food delivery service. Then I started working with the engineers. There's kale. We were figuring out an AI algorithm to speed up the delivery of the food. Big problem to solve. Um, and then just some other interesting things like how to negotiate a deal with Verizon so that you can get thousands of tablets and then how do you build systems so that you don't get them stolen and how do you recoup them and you just learn a lot of stuff along the way. And then I did an AI or sorry a uh music startup that was a marketplace. And here's me in Vegas. I was chasing down Bill Tai who was one of the earliest investors in Canada. If you've ever heard of it, that's me and two engineers, Gary and Dne. And then yeah, I went and I like tried to go all out on that idea. Went and pitched on stages in front of VCs. And uh after that didn't work, I just went all in on AI, basically using AI every single day to learn it because this is bigger than the mobile revolution, bigger than the cloud revolution. It's the biggest thing that you could be a part of. So this is a daily update where I go through some questions and answers that come in and then I am changing it a little bit. I go through some quick news that should be somewhat evergreen. If you do watch this at a later date, I try to keep that short. And then I go into strategy on building to 100k in profit a month while doing a lot of different things as a one-person team with multiple agents. We're talking dozens and dozens of them. And then we go into the build pipeline. So that's stuff that I'm building for the SAS. That's stuff that I'm building for Vibe with AI, which is this thing right here. See the VTV. This is a media company. So I talk about the things I'm going to be building for that in our inside of our community. And then I I think there's a third thing that I talk about in the builds. Oh, it's the content generation system that I'm building. So, this video that you're watching right now is long form because I'm going to have tons of trading data to use to make something that kicks off 20 steps and then turns this long form into short form across all platforms, rehashes it, finds the best clips on my behalf, and then posts it on my behalf, as well as text content, both short, medium, and long form across all platforms. So, that'll be something that I'm building inside of Vibe. And uh luckily for you, you don't have to pay for any of like the strategy side of it. So if you want to go build it yourself, you can do that. So let's jump into questions. Uh this channel is the best AI news out there. Thank you. That's Basil over there. She's the number one cheerleader. Thank you for that. Uh do you have a YouTube strategist? Says Marlin. No, I don't. I do not. What I do is I stay up way too late reading Twitter and finding out what's going on. But that brings us into something that I'm going to be building inside of the community, which is an automation that basically does that for me. So, I have a bunch of curated resources that I use across different websites, forums, uh, newsletters, and most importantly, Twitter because that's where it all happens. And I need to have that process automated. So, I don't let me just say no, I don't. What does one do as a YT strategist? Cool. I want the fisherman prompt. Also interested in any favorite Obsidian plugins. Oh, you saw I used it. Yeah, the fisherman prompt. I'll post it in here. Let me snag it. So, it's inside the school thing. I just added 51 prompts, 27 videos. There's more than 27. And I kicked everyone out because I ran at an alpha with uh 40 people and you Anyways, I I had to kind of clean house and then reinvite everyone. So, there should be 40 people joining today. But let me grab this. I'm not gatekeeping little prompts. Uh where the heck did I put it? I put it in here. This is use this prompt to do this. So, for each prompt, I try to put in like the context of it, but I'll put it in this board and then you can just uh paste or actually no, I'll just paste it in here. And yeah, you are famous. You have a great uh thing. You have a great profile pick. Maybe one of the best NFTts I've ever seen. that thing funible. Cool. Thanks for the questions. Now, let's get into the meat and potatoes of this. A lot of stuff's happening. If you didn't know, GPT, what Sam Alman did was he launched super super viral uh image generation right before he was closing around, which is awesome. We did the same thing when we were selling our company. When we went to market and we had an offer for over nine figures, we would know that we were going to the table. So, we'd pull back on profitability and pump revenue because in that business, GMV matters. In this business, having users matters. That's what they're using to negotiate. So, he pumped pumped pumped. Um, they added a million users in one hour. That's kind of nuts. They're talking about doing an open weight language model. I don't need to talk about that much because I think it's just a BS like PR play. And then this is just a really interesting thing that I'm going to be implementing into the content generation system that I'm building where this guy, go follow him, Apollinator 3000, he's got mega alpha, he's got the juice on AI, but he basically broke down all the things that you get graded on and 50 insights to help your tweets go viral. So, I'll be using these to help inform the tweets that are generated on my behalf. human in a loop always like I want to see everything especially Twitter it matters most to me maybe LinkedIn I'll let it fly after a little while but I want to see all the drafts and then last thing I wanted to cover which has to do with content generation at scale which even if you don't generate content I don't care you're going to want to in the future every company is going to have to like everything's getting commoditized of abundance coming 300 million jobs changing by 2030 time to start being more human 56 million views for three apps one creator Bible chat couple joy 15. Like these numbers are ridiculous and it's just one check. So I'm going to go and study what she's doing. Yank all those scripts and then basically remix them. So that'll be part of one of the automations, one of the agents where it's like when it gets to the Tik Tok one, it'll be informed on that. And I just think it's crazy. like distribution has always been a thing but even bigger premium on it. Now let's talk about the 25k per agent and uh I don't have this 100% figured out so that's part of what we're doing together. So when you build something you want to get leverage on it, right? Like Naval always talks about code and media are the highest forms of leverage because code you build the app once and then you make updates to it, right? But the thing is is like then you can sell that a million times plus and then same thing with media. You're watching this video. This will live on the internet so long as I want it to. If anyone rips it, they can put it wherever, right? So it's here and it can be watched and people while I'm sleeping will watch it a thousand times. You can see the same quantity of banker's hours during a week in watch time off one and decent video. So yeah, you can get millions of views and it goes. So this combo matters more than ever. As the cost of coding comes down, then the cost of media in my head is going up and up and up. And the combo of the two is what's going to be magical. So to that point, I think that there's still like leverage like I I started to get into the AI services business as a need for cash flow to fund my SAS product called MAP because I couldn't raise money for it. I went and wasted much time doing that. I don't have like a Stanford badge, you know, I'm not like super pedigreed, but I know how to build stuff. I've done it for literally my whole life. And so I said, "All right, well, f it. I'm gonna take these high lever skills and I'm gonna go to work. So, what I did was I went into a company that was a real estate company, real estate tech or prop tech. We just saw a military plane. That's badass. Um, went into a prop tech company and the problem that we solved for, this is how you should always think of products is like what problem are you solving for? So, the problem was real estate agents have a lot of things that they're doing. It's mostly follow-up. It's mostly nurturing. It's things in sales. It's closing. It's contracts. A lot of stuff on the phone. Inputs of phone are much slower than voice. Like that input like your average typing speed on a computer is double that of a phone like from typing. So that interface kind of sucks. We all are used to it. But it's all about data compression. So, one way of compressing something is a great example of an article I read where if you're at a table with your grandparents and they knew one another so well that after years of living together, the grandmother when she goes to ask for the butter, the grandpa already goes to hand it to her cuz he already knows what's coming. He has context of what's coming. He doesn't need to even hear the words. She doesn't even need to utter the words. There's a lot of that. Like that's an example of a data compression because getting the words out is actually slower. Same thing with like hand gestures and little like facial gestures. The point is is like there's different modalities for getting a message across and doing text and doing all these things like email and SMS, MMS, they're not that efficient. And so what we did for this was we have the agent and they're in the middle and then they have like text, sorry, uh let's use proper terms here, SMS, MMS, email, and phone. And this is just for comms. We have all these things coming in, and that's both inbound and outbound. So, they're using this to get leads. They're using this to bring in new leads. This is used all across their funnel from top to bottom from the first impression to the close. There's a physical aspect to it. So, when someone sees a property, that's how they're going to notice. So, how do you get QR tied into this? But with this agent, it required a bunch of like it went well, right? But it required a bunch of custom work because I had not set SOPs. I had not disqualified what I would not do. So, when I'm planning this out, my next move is to put SOPs or standards around what my qualifiers are. So, for me, it's going to be picking a text stack and sticking to it. Now, when I did this one, it had a reals part, it had a Python part, it had a create react app part, which was deprecated sunset a long time ago, and it had all this dependency stuff going on, and it was super uh just not a system that I was used to. It's not nothing wrong with it, but it wasn't a system that I was accustomed to. So for me it's okay to do this well for the client to be an expert in it get the job done save them time money whatever the use case is it needs for me to be using NexJS it needs to be using Tailwind three not four that's important for LLM uses it needs to be using obviously TypeScript on the deploy side it needs to be using GCP or some sort of like container which should be really easy. it should be anyone but like defining what these tech stacks are and then defining what the um objective like basically does my goal for these things that I'm building align with their goal like is this going to be a lead genen multi-agentic like workflow is this going to be a growth agent that'll find ways to go and interact with people proactively go and talk to them which is slightly different lead genen could be outreach reach with automated email, but that's not interesting to me. I think that's just like spam at scale. I don't think that works. Um, so I would then define like, and you would do this too if you take the same strategy, define exactly what it is that you enjoy doing that you're really good at doing, and you build around that. You'd figure out like, is this going to include voice? Is this going to include uh text? What about email? Another one that I would really care about is like their obviously their budget and then is it done um like basically are you going to have them on your shoulder or not which is like done for you which would be like I do the whole thing or is it done with them? Um let me just scooch this stuff. So, I'd be putting this together and I'll refine this more, but it's, you know, that's kind of what it would look like where each uh part in the process from the proposal to the discovery call to actually like figuring out the delivery dates, like it's all defined, right? So, in this proposal and then like the way that I'd be thinking about it just has a couple things in it. It's like here's the stack, here's the minimum, here's the goal, and then the follow on option. Like, am I building this so that I'm saying goodbye? Would they need more work after this? Could this lead to building another $25,000 agent? Would this lead into um being able to further develop my own core attack, which would be the agents server, which is a cluster of them. It's a swarm. So, this is like how I'm thinking about basically boxing this up. And you get a much higher multiple on the cost per transaction than you would if you just jumped into NADN and you were a spammer. Not all are spammers. I was mean, but there's just so much content out there about these automations and I've done my research on it and there's a place for them. But it's connecting this together with custom software where I think like the real alpha is. And then when you get done with the project, if you have this as a set sort of template, then you could come in and there's probably like 75% reusability, which is the metric you'd be optimizing for, which is like when you finish the project, it's like you're asking yourself how much is reusable, right? And so you might go build one and just specialize in one goal that a business has. you might do a couple of them. That's for me to figure out and for you to figure out. So, that's the 25K agent plan. I think it's going to work really, really well. And as so long as I do the the qualification properly. What else are we doing in strategy? Oh, this tweet just reminded me of whenever you're building something, this is like really well put. Gary Tan has a succinct way of explaining topics and in the case of building anything, how do you get it to product market fit? How do you feel like people are going to want it? You talk to the users, you iterate, you do 10 things in the next week, they even tried to get users or improve the product. Did any of them work? If they did, double down on the ones that worked and repeat. And so that's what I just did with this new media company that I'm using. And again, like the media company plays into a bigger master plan, which is to attract people that are optimistic about the future, that enjoy learning about what's possible, that want to get more leverage on their time using AI. And so when I look at this, it's like I launched this first version and it was called Troublefree AI. And you can go to the URL. It's not very good. But I stood this up and I was like, "Okay, like this is dec. I have some prompts and some workflows." And I was like, "Okay, let me have it promote back to a paid community and I'll let people in for free." I was off. I'll let people in for free and get to talk to them because then there's no barrier to entry. And so after a couple well it was a month couple weeks maybe a month of doing that and actually getting on the phone with people that were in it I learned exactly what I needed to do to give me confidence to then go and launch the proper one that was paid. So that process while it may you know you might eat some costs up front by doing like basically like trial slash in my case it was the free beta you get direct access and that's the best way. So, I ended up doing a call probably I probably did like four or five hours worth of calls with customers and you then take each call and you iterate on it. They're not going to tell you exactly what to do. That's your job as the product strategist to figure out. And I made a whole course on this actually, but it's so important that you guys figure out like it's on you to take what they're telling you and then form it and sculpt it into actionable stuff that you implement. Let's see what else. Um, I think this gets into the build stuff. So, let's talk builds and I'll kind of run through what I'm going to be doing inside of the community. You get all this stuff for free of how I'm like thinking about it. But the actual builds I'm not going to share with the world because it is kind of a thing. It's kind of a lot of work. Let's see. So on map, if you're not familiar, this is mapthemap.com. Unlock your potential, become your best self. So the idea is it's an AI health tech company that builds tools to help you live smarter and happier. And it has all the daily drivers built in. So notes, tasks, health, chats, and calendar. Idea being that you want to become the brightest version of yourself in the future. And AI can help you do that by becoming not just like this like coach, but actually being like a representation of what you could be if you accomplished your goals. So, it's training you. It's it's basically like if you had a mirror image of yourself in the future that was at 100% potential, what would that look like? And it does that by talking to you and figuring out what your goals are across all areas of life. I was really focused on health at first because I went down the rabbit hole of well for a couple reasons. One, the rabbit hole of health is just really depressing. Um, I made a website called crazyhealthstats.com that live plots the cost of obesity not only in dollars but in like just like look at that. Like it it shows you the number of people in each country and it's in real time. And so this is something that's like going whack-a-ole style like crazy. And all the data is in here. But it made me want to go big on the health part because it's such a a problem and then also because health is more objective. So you can pull data and you can run it against some set of rules. So most of the medical journal or a good portion of it rather can be put into a set of if else statements like oh is this hypertension? Well, did your heart rate exceed some level of uh BPM over some amount of time? And so that was more objective and would be pulling data off there to then use the calendar to plan what your goals would be. So if I wanted to run a half marathon every single month, which I did using map, which is really cool, is it would pull data off of Straa's API, it would store it in Superbase. It would use that as the grounding to then figure out where is this guy at? Um, and the goal planning was done at the top of the funnel. So, you enter the product, the first time user experience is you basically just seeing like what the heck is going on. Um, I can show you in here, but I'm going to get into an explanation of like what I plan to do with this product now because I had to pause development on it to spin up these other income sources and now things are are brightening up where I can get back to it and also because the agents SDK came out. So, that's really helpful. Well, crap, I can't seem to find it. Where's the slides? Fundraise docs. Here we go. So, this should better explain the product and I'll talk through my strategy for building this. So, what I was doing was I had this health dashboard and so this is pulling in from Whoop and Aura and Apple Health. So, there's also an iOS companion app. I also write Swift. And then there's a calendar component to it or a calendar app within it. If you've used notion calendar, I rebuilt that from scratch. Rolled the whole thing. A lot of client side stuff, pretty annoying. Took me forever. And then the coach would write your goal plan into that and check for basically availability. And then, you know, when you show up, it needs to know your time zone. There's always nuances, but then you can see on the top right there's your actual sleep information. Workouts would look different. So before you do the workout, it's going to look way different than after the workout where you actually get the insights of the thing, but having live time auditing on you can be really annoying. So it's solving for a design problem, but I just needed to get all of it in there. Um, and then originally I was doing this where there's like multiple agents based on the area of life that you're in, but that's a UI like miser uh like nightmare. You don't want to do that. So, I'm thinking different here. Uh, there's tasks else and then there's like me. Here's all the stuff I had built, right? And so, and then the scale of the problem. But what I'm thinking with this in order to set it up, what I had done before was I had all these agents that basically just have a set of calls that they can do. Like an agent at the end of the day, there's a couple different definitions for it, but it's like it's basically like they have access to tools. And so those are just Python functions. And if you've pulled up a really basic agent like the ones in the Verscell templates, which if you're trying to learn how to build these, go to Verscell templates and just build one from scratch. Just pull up one and then have it next to you and see how it works. But if you clicked on AI, I rebuilt most of these the day they came out because I wanted to figure out how they work. So I made this one when Next, not the uh like framework, but it's funny when Next.js JS the framework presented at next which is the Google conference the day this came out I rebuilt it because I was like whoa this is cool you can stream in components and this is an early version of what's to come so what was the pain is you'd use what's called Zod validation to figure out what is the tool that needs to be called and that was kind of a nightmare because if you had like this one agent is like more of a bar trick than anything but if you had it doing operations on your behalf within the software then it got a little iffy And if you had, call it nine agents in the calendar because you need them to do all these different things, then it got to be a nightmare. But luckily, the agents SDK came out and it at least makes handling swarms, which is the word that they use for uh having like dozens and dozens of agents, it makes it a lot easier to work with. So, what I'm doing is in this case, I'm going to be setting up a what's called a turbo mono repo. A lot of words, right? But if you have your folder structure and I got to reload because Excaliraw is not in love with Brave browser. But if you have your mono repo, let's say like it's not a mono repo to begin with. What you typically have is you have a bunch of these like folders. And so you might have let's get rid of this. You might have like app oops app. And then inside of app, you might have components. And then you'd have your page.tsx. And then you'd have, you know, your like layout.tsx. Maybe you'd have like a utils folder or whatever. So this would be like potentially the actual dashboard, like stuff that you see and all the middleware and whatnot. But in the monor repo turbo repo which is just a monor repo but next does what I plan to do is basically have this architecture where then I'd have just another one but this will be like the python thing and I'm not going to name it properly because I'm going to totally botch it on the spot but this would be like where all the logic for the agents are. So you might have agents and then inside there there'd be like calendaruler.py pi. And then actually what I'd probably do is I'd probably have this and then I'd have uh calendar and there'd be a bunch of those, right? So you'd have calendar and then you'd have like six of these ones and then you'd have um you know like tasks and you'd have health and you have all these broken out in this sort of setup which mirrors the way that I have it set up when I work with integration code which looks like this. So if I just pull up cursor I go to map and then I go to uh let's just do a get. So, the way I set it up in here is I have all the API stuff for external integrations just mapped out in really small modular folders so that it's easier for me to work with because there's so many of them. And then it's easier for other developers to work in and then it's also really easy for obviously AI to work in. So, I have the Google calendar one and then each one of these represents a method. So it's like, oh, I have ACL, I have calendar list, I have all those things. And if I just pop into one, I can see the TS docs, I can see everything. It's all there. And then there's one service at the top that handles the access tokens, the actual it's a singleton pattern, right? So that's the plan. I'm going to end up like turning this I'm basically gutting a lot of this stuff because I tried to run it last night. One of the kind fellows, Hari, shout out Hari, that watches his channel's like, "Yo, dude, this sounds kind of sick. Uh, can I jump in?" I'm like, "Yeah, sure." And then I'm like, I need to go fix this thing because I haven't run it in like four months. That's not that long. It wasn't four months, but you know, when you bump like major versions on things, it takes a second. So, I'll be gutting a lot of that. And that's the strategy there. And then we'll just get it running and then it's all UI from there. Figure out how to get the UI to to do to do, right? Um, the way that I'll do that is I'll paste in like some examples. So I'll like start by going through the docs, right? And then I'll see a couple of the examples and then I'll take what I like to do, which is like, oh, like grab uh an example like this little import statement. Then I'll do a search operator in here with Python as a language, find examples where other people have done it, and then remix just to learn it quicker on the strategy for vibe with AI. So, I'm trying to figure out like how to provide the most value, how to get people super stoked so that we can build stuff together. And I think it's just building automations. I think it's like there's little automations you can do inside of communities where it's like, okay, anytime that uh I post new content like it needs to be listening somehow. I'm not sure how to do this. Maybe I can create an admin user. I can take Basil's account and then hijack it so that it can go and like do a browser agent to figure out like when some action happens, make sure that it gets posted back. When some person joins, make sure you do this or that. Um, and then that way like everyone's kind of always up to date on what's going on. And I know they have web hooks, but I am curious like how this can work and how you can like set up a bunch of like basically like a nurture sequence that like cool here's all this content, but then how do you get people to actually go and do it? How do you get them to be reminded about things? Like how do you work outside the bounds of this? And then we'll be standing up the site that's just called vibewithai.com. Um yeah, so we'll stand that up and we'll probably use AI a bunch for that cuz it's going to be simple like directory style stuff with um SEO SEO pipelines set up and those are trained on SEM Rush's blog templates. So, there's five of them of like the best way to write uh different blog posts and then there's a bunch of parasite tactics that we're going to use and I'll be like posting that basically like in the in the program like hey this is how I plan on doing this, how I plan on doing that. Um, and I think that's for the most part it. Yeah, the content generation engine is actively getting developed so that I can take all this long form stuff and then have it like be posted back and it it should go really really well. I'm excited about that. If you guys want to get a couple of the like code snippets or if you're interested in any of the stuff that's going on in map, then just comment the word map below and I can share whatever part of the project that you are like curious about. Um, and then of course always ask questions. I think that, you know, it it would be worth my time obviously to answer them and I love hearing from you guys. And I'll see you in the video tomorrow. Go check out all the stuff that I talked about. I hope this is helpful. Oh, and like and subscribe. Yeah, yeah, yeah. Like and subscribe. Okay, fine.