Cohort 16 - Session 2 - 7/17/2025

have, you know. All right, cool. All right, go ahead, George. So I've been in Replic a lot today too, and I've— I'm within the preview window and in the agent, the message agent section. Yeah, it keeps bouncing back and forth at the bottom of my screen between working and reviewing your message. Yet I don't recall ever uploading a message, and I don't know why it's doing that. It's been stuck that way for a while. I don't think it's actually doing that. Why don't you try this? Just try refreshing your browser really quick, like the whole browser, and just, and just see if it updates that or if it still says that. But just in the last maybe week or so, it started doing that to me too, right? It's clearly not doing anything, but it's like, they are reviewing your message, and I'm like, oh, well, what are you doing? And then I give it something and then it just gets to work, right? So don't always trust that little status bar, that little status indicator. It's having a couple— okay, it's having a couple bugs right now. Um, so yeah, um, but you can always refresh it and just see if that works. And at the very least, you can be like, hey, tell me something, you know, and see if it responds to you, you know. And, uh, you know, do one of those don't write any code, tell me what we're working on, what should we work on next, what's the best way to do this. So, um, very cool, guys. So, um, yeah, how many of you have spent more than 3 hours in Replit today. George has. Sam. Yeah. Ashton, maybe.

I wish I had the time. I didn't.

All right. If you don't have the time, you don't have the time. It's okay.

I went back and I was trying to re-engineer the prompt on the Replit so that I could put it into Replit again because Everything I did with Replit didn't look at all like what I wanted it to look like, and I've already built the app before. Yeah, so I'm like going back and creating a user design guide and a couple things to plug in. But ChatGPT was having some issues for me.

Like I was saying, what kind?

Yeah, like I was saying, like, hey, because you just say build a canvas with this, right? And then it would say, oh, here's your stuff, and it wouldn't put it in the canvas thing. And I was like, I don't know what it's doing.

So you just have to say, no, you didn't do that. Use the canvas tool and put it into a document for me, right? Yeah, sometimes you just got to remind it what it's supposed to do. And sometimes you got to start a new chat too, right? You can start a second chat inside of your project, right? Just on the left, click on your project and then just start talking to it fresh again and say, hey, I want you to review everything in the project and let's keep working on this. I wanna build a design guide. Let's work through this problem, right? And so, yeah, you can, what happens usually, and we're gonna talk about this right now about kind of optimizing some of Replit, but you can overload your AIs, right? The context, right? The context limits that they have are real, right? And you know, I'm not sure exactly what Replit's is. There's some debate about it online. Is it 8,000 tokens? Is it 40,000 tokens? Is it a million tokens? I've seen all three of those options, you know, suggested out there, and it's not very clear. But same is going to be true of ChatGPT. Even inside a project, after you've been talking to it for a few hours and it's, you know, it's cranking and building stuff, you got to kind of let it rest and give it a new— just kind of let it reset itself and say, all right, fresh brain, let's go, right? So that will probably solve that issue.

Um, George, to kind of say it to tell it to go review the other things in the project and then—

Yeah, yeah, I— yeah, so theoretically it's supposed to know everything that's in the project already when you ask it, but sometimes it's like, oh yeah, let's build you a new thing. I'm like, dude, no, this isn't a project. Like, look at everything that we've already discussed. Oh, okay, thanks, now I've reviewed it, now let's do this, you know. So sometimes you almost have to remind it what it's capable of and, uh, You know, we're still dealing, guys. I, a lot of people are like, you know, I, I got in into some fun debates, you know, on LinkedIn over the last, you know, day or two just for the post I put up. And a lot of people, you know, I, I, the post was about, you know, AI deniers are a thing, like people that just say like AI sucks, it's a fad, it's, you know, same way that people were about the internet back then, right? But, um, but yeah, a lot of those people are like, yeah, you know, we've had, we're kind of at the end of this run, right? Like LLMs have reached their capacity and their capability levels and we're done. And those people just, I'm sorry, they're wrong. They're just, they have no idea what's going on. Even very smart developers that are way more technical than me that are like, here are all the technical reasons why. No, you don't understand that there's new ways that are being worked on right now. But the reality is, guys, I think we're in the second inning, right? We got a long ways to go here, guys. We are right at the beginning. The vast majority of people don't know it. Like I saw a report not too long ago that was talking about all the different states and, you know, what percentage of businesses are actively using AI in those states. And Utah was ranked number 3 at 12.5%, right? And okay, so almost nobody's using it. And by the way, I bet most of those businesses, I bet 90% of that 12.5% that are actually using it are only using ChatGPT, right? And they're using it to like make their emails better. They're not building complex strategies and going between different AIs and using 7 different systems for the right things. And we're going to get into way more advanced stuff. The stuff that we do today, guys, is going to be way more fun in my— you know, this is the fun day, right? Yesterday is like, okay, we got to kind of learn how to think like this a little bit better. We're going to have some fun discussion as well. But, but yeah, today it's going to be, you know, agent systems doing crazy things, images, video. We might play a little bit more with ElevenLabs and voice stuff if you want. We actually will for sure, because I'm going to show you how to implement it actually inside a Replit and how to really put something like that together really quickly. George, did you have a question before we move on? It says that your hand is raised on the platform here. Maybe I can just lower your hand. There we go. Cool.

Just show that real quick.

Yeah, please go ahead.

It's just weird to me.

Yeah, I'm happy to look at it and see if we can. Whip it into shape. Just got to call it an idiot.

Oh, okay. I think it might— okay, it did put it in the thing. Sorry.

Got it. Now again, guys, using strong language, right, whether you want to use profanity or not is up to you.

But, but yeah, I had to basically reset. Get out. Can you run multiple? Do you mess up ChatGPT if you try to run like multiple tabs with—

maybe. I don't actually know the answer to that because I've had it work sometimes and I've had it not work sometimes. Like, yeah, I've had to request timeout while I start to do some other things with it, right? So yeah, um, I think that's what I was doing to it. Yeah, I think, I think you've got one agent at a time available to you. Uh, now there are some systems like we're going to get into Manus right now. We're going to play with Manus, and Manus you can have at least two, and if you get into the pro version, I think, you know, 5 or 10 like systems of agents working for you simultaneously on different projects. So that's pretty cool. Anybody here use Manus? Anybody? We played with it. Yeah, that was pretty cool. We're going to— you're going to like this. Sam messaged me earlier. He's like, dude, I built a ton of stuff. Now I got to figure out how to sell it. How do I make money with it? How do we do that? So we'll be talking about that today, of course. In fact, we'll dive right into it with some of what we're going to do with Manus. So Manus, guys, is kind of this next level. We're taking it another level up here in terms of complexity and what it's able to do. You know, in a business, you know, with or without AI, right? You've got employees, right? And you have employees usually, at least when you grow, when you're a solopreneur, right? You kind of, you have to do it all, right? But you usually have employees that are dedicated to different types of projects, right? You're going to have a sales team, you're going to have a marketing team. Maybe if you're a startup, that person, you know, is the same person in the beginning, but then it eventually grows and you've got a customer service team and you've got a product team and you've got a, a design team and you've got a development team, you got all these different teams, right? And so, you know, I was saying yesterday, when I'm developing with AI, I'm not just sitting here waiting for one agent to finish, right? I'm using, you know, I'll have a conversation with one team and then I'm going to move to another team and get them on something and I'm going to move to another team, get them on something. So I always have 3 or 4 things and I'll even oftentimes set Manus or something like that to work on a 4-hour project. And then I go to bed, you know, or I go, you know, watch TV and fall asleep or go hang out with my wife or whatever it is that I'm doing. I'm— I'll set it off to do stuff. And so I like to— I like to have my agents doing, doing a lot of work. Sometimes I'll put Replit on a really complex thing, you know, right before I'm, you know, instead of sitting there and finishing it and making sure it works, I'll just be like, okay, now work on this. I've created a great plan with it. Okay, now work on this thing and make sure you don't mess this up and do this and all right, go. Okay, implement this, implement this new, you know, SEO across the entire site or whatever it is. So yeah, so anyway, with that being said, we want to make sure that, that we dive into this madness here. And by the way, to answer your question, Sam, on Lovable, I have. Yeah, so Lovable was one of the ones I pulled up yesterday. There's Lovable, there's Bolt, there's, you know, Cursor, there's, there's what is it, something Base44. There's a lot of different ones out there. That, you know, are what I would say competitive to Replit. I actually did some research, some market research today on it. And, you know, there's just this pocket in Utah at least. And it was— I was curious because there's this pocket in Utah of a bunch of people that are like, you know, really kind of high up in the business ecosystem, right? I've, you know, worked with them a lot and they do, you know, they're the VCs and, you know, some big tech companies that they're owners and all that stuff. And, and they never ever bring up Replit. They're like, they're just using Bolt and they're using Lovable. And I just think those programs are vastly inferior. And so I did some market research. It turns out Replit's worth about 3.5 times the next closest one, right? It really is the biggest in terms of this style of platform. Um, they're currently worth probably close to about $4 billion right now. And the others, you know, one of them's worth $80 million. One of them's worth, you know, $1.1 billion. One of them's worth about maybe $800 million. So. You know, they're all good and you can play with them. What I have found with Lovable and Bolt and these other ones is it really does require a lot more third-party integrations, right? So you can't do everything inside of that. Like just as a small example, like I said yesterday, Replit now allows you to connect your domain or to buy a domain and connect a domain and do all that right inside of Replit without even having to go to GoDaddy, right? Now it couldn't do that a week ago, but now it can. And so Replit really is focusing on saying, okay, when you're on replit.com, you never have to leave, right? And that's cool. Now, if you want to build really cool stuff, yeah, you're going to leave, you're going to grab APIs, you're going to connect Stripe, you're going to connect different things like that. And, you know, you want to know how to do all of those things. Okay. But we're going to dive into—

we're going to dive in and throw in a quick plug for Base44 just for getting your front end looking really good. Like it was so easy to use and it just made it really pretty. And so like I might use that for screenshots to move over into Replit. I don't think it has much strength on the backend, but I liked what it was doing on the frontend a lot.

Yeah. One of the things that, that you can do just, just while we're on that, on that note, because I think it's really good, import code or design right here. Okay. So you don't have to always start from scratch. You can also start right here. Bold or lovable or bold or lovable. You can— that's actually kind of a slap in the face a little bit. Oh, yeah. See, that's the cool thing, Ashton. That's what I was coming in to show you because I was like, I bet you have Figma designs already.

Yeah. Yep, exactly.

So yeah, you just push that and then connect to Figma. You just log in with your Figma and it'll just be like, okay, what do you want to do? So pretty cool. Pretty cool. And they're going to keep getting better, right? Pulling those things together. Obviously you can, you can import from GitHub. Who here does not know what GitHub is? Raise your hand. GitHub, guys, is really critical. When you build whatever it is, anytime anyone's coding, what you're really coding, it's not all this stuff here and it's not even all this stuff here. This and your browser, those are what's called a rendering environment. It's rendering what all of this says. It's all this code. This is what makes your website come to life and do stuff. All of this is files and text basically. Right? And maybe some images in there if you have some images stored for your logos and whatever, all those different things like that. Okay. And so what GitHub does, GitHub is a code repository, right? And you can sign up for a GitHub account for free at the levels you guys are working at right now. It's definitely going to be free. Let's just click on pricing here. But basically, yeah, you're going to unlimited public-private repositories like you can do all kinds of crazy stuff. Now, if you want to pay $4 per month, right? Then you can get all kinds of other stuff and even code. There's even some AI coding stuff you can do if you start to dive in, but you don't really need to. But what GitHub does is it's a repository, a place to put your code. And if you do hire a real developer— so Ashton, I know you've, you know, maybe got another developer that's working on something or another team, and, you know, maybe they don't want to work in Replit. Okay, that's fine. You give them what you've built and then just push it to GitHub. And the way that you do that, guys, is you just go here to this little plus thing inside of Replit and you just scroll down to Git. This is your version control for your app, right? So you can, you can do that and all you do is go into settings and connect to GitHub. Boom. And so that allows you to push into GitHub and then GitHub is connected to, you know, it's a, you know, 2 decades old, you know, developer repository and it has thousands of plugins and all kinds of things you can do with your with your app to make it stronger and beefier and whatever else you want to do to it. To tell you what I'm doing, guys, with like GenAI API and other things that I'm doing that are actually production and ready and going, I'm still pushing everything from Replit. I'm just telling you, you can build some big stuff in Replit. But I'm sure there's limits that I just haven't run into. And I'm not a developer, right? And so maybe developers just love pushing into GitHub and pushing other ways and using Vercel and other— there's other systems you have to use to push it. And by the way, you can use ChatGPT or Groq or anything and be like, hey, I want to push my Replit project into GitHub and then I want to deploy it outside of Replit. How do I do that? Oh yeah, you can use this, this, or this and just listen to it and follow it, right? You guys are all capable of doing that at this point. But, but the reality is, you know, I think Replit's moving and continuing to push towards this, this idea of, you know, you don't have to leave the platform, right? That's, that's what they're trying to do. Which makes sense. I think it's a good strategy. Okay, so Manus here, guys, is a new agent system. Uh, it's been out for, I don't know, 4 months maybe now. So it's pretty new, pretty new. Um, it does— when it first came out, it was hard to get in, uh, get into. But there's all kinds of things you can do, right? Like video, audio. Man, audio. I want to play with audio here. So cool. Oh man, okay, I want to see some— I want to see some examples of this. Why is it— why is it doing that? It's kind of a weird functionality. I'm like clicking this down. Oh, hide use cases. I want to see like the audio examples. But anyway, um, let's see if we can find something cool here. Yeah, this is just like explaining stuff. I want something that's like really useful. Productivity. Let's see what we got here. Uh, explore more use cases. Okay, come on, productivity. I'll show you guys a couple things that I've done, but it's kind of fun to go through and see 3D scan service. What's that? Interesting. That's a car.

Did you click, click into that?

How did I, how did I find this? Yeah, this is manas.im and you just start scrolling down. Um, let's see, we're gonna find Yeah, this is a, this is a pretty cool one actually. We'll just, we'll just click on this one. So this is like a job search, right? Please search the following websites for any job openings, compile the results into an Excel spreadsheet with the following column. These are the things that I'm looking for, right? So maybe you want help finding a job, or can you can see here every once in a while it'll come up with a little list of like X's next to right there, right? So it's like, going through and doing these tasks. It's doing that, that chain of thought reasoning and showing you this. And so it's actually running a replay of what it did here. And so we'll skip to the results here. But by the end, hey, here are the job openings right here, the different companies or different job openings in those companies, job title, you know, location. Okay, that's kind of cool. Here's all the stuff. And I think that's kind of a mildly, you know, efficient thing. What I would have asked it to do is you know, pick the top 10 or 20 and, you know, rate them, or I want you to rank them by, you know, salary, and then get all— make sure all of the application links are there and fill out all of the applications for me. Like, you can have it do that. You can actually have it like type things for you and write things for you. Um, let me, let me pull up, um, before we do a lot more with, um with Manus. I actually want to go into operator. So operator is part of ChatGPT if you are on the Pro platform. Okay. So the Pro package, right, which is $200 a month. So it's a little bit more. It's still early access, but I believe, yeah, let me, let me say the next thing I'm going to say after I show you what it does here. But you can basically say, you know, book me. Come on, work. Book me 2 flights to Cabo for Saturday from— come on— from SLC on Delta. What it's going to do is it's going to take that request, it's going to break it down and say, all right, what do I need to do here? Then it's actually going to open up a browser for me and it's going to do things. Not every website, for example, has an API. Right? Not every database has an API, right? And so sometimes you need to do that. Oh, you know what, it's, um, this is actually interesting. It's saying this site can't be reached. Delta.com. That's super interesting. Okay, fine, just do it through Expedia. It's interesting, some sites have bot protection and it detects operator as a bot, right? And so you have to, you have to kind of deal with that. There's a couple things it can't do for you. Like, I wanted to go in and do a bunch of stuff on my Shopify, um, platform the other day and, uh, like update a bunch of stuff for me and even process some refunds for me, things like that. I had this stupid thing where I had hundreds of people— man, operator just— operator struggling today. Um, but I had, I had literally 106 purchases made on my website that were all fraud. They just identified like a $3 product on my site and used 106 different credit cards, you know, that they'd stolen somehow, right? And they're just like testing them on my site to use it, right? So I had to go through and refund all of them. It's taking me forever. I was like, hey operator, can you do this? And it's like, no, I can't, you know. But anyway, what it's doing here is it's literally, you know, it's using Expedia. Clearly Expedia is allowing this, but it's acting like a human. It's operating the browser like a human. And so it's just typing things in. I said this Saturday, so it should pick it for the 19th. Now I didn't, I didn't tell it when I want to come back. Maybe we're just going one way. Yeah, it did choose one way. Because I didn't specify. So obviously a bad prompt is going to get a bad result. But to adults, it's going to have to assume two adults probably, and it's going to go find it for me, right? So I could have it book it. Now I can even give it if I wanted to, and I wasn't sitting on here with 7 people watching me, I could give it my username and password and I can give it my credit card and I could give it all that stuff. Now that sounds scary, but this is an encrypted session that it's opened on a computer. And if on another computer, that's a headless browser. Basically, it's a headless system, meaning there's no— nobody's watching it. There's no screen connected to it. Right. It's just, you know, it's just running on a computer and, and getting this up. It's all virtual, though. But, but the AI can see it. And the way that AI can see it— and this is, this is pretty cool— is by— it's going to YouTube and it's called Omni. Yeah, let's see. Omni Parser V2 right here. So basically what it's doing is it's breaking up that website into a bunch of different little, you know, it labels them, right? It says, okay, this thing is this and this, you know, 375, this is this right here. And okay, it looks like I can click on budget, right? 71. It looks like I can click on organize number 91. And so it knows where it can move its cursor to click, right? So it's actually moving physically a cursor around, typing things into different fields and doing stuff. But that's what it's seen behind the scenes. That's how it knows how to browse the website. So pretty cool. Pretty cool stuff there. Operator again is struggling. I have a hypothesis and I'll take you guys to the end of what I was going to say here. So anyway, it selected a flight. That's cool. It's ready to go. ChatGPT— or rumors got out that ChatGPT is about to launch a browser. OpenAI is about to launch a new browser to compete with Chrome. Right. And by the way, Anybody know how many users ChatGPT has? I don't know, 200 million, 100 million, 500 million. It's a lot. Isn't that crazy? Like, they are just, you know, other people come out like, there's other things, and, or ChatGPT sucks now. No one— no, ChatGPT dominates. Now Chrome has 3 billion users, okay? So they got some stiff competition. But my belief is that the new browser will have operator-style control. You'll be able to get in and then log in all your sessions, everything, but just like I have here, I'm not afraid to log in to Expedia or Delta on my Chrome browser. Once people feel comfortable logging in to all of their stuff and they know it's secure and it's encrypted sessions and all of this stuff, then getting agents to just do stuff for you on anything. Because I could have said one of the cool things I can do here, is I can actually take control, right? And we'll minimize that for now. And now I have control. If it would stop crashing on me. But I have control now of the browser where I can come in here and I could, I could go and open up maybe. Come on. I don't know why it's not letting me do it. It might just crash again. It might just not be working right now. But theoretically, I've done it before. I can click here and open browsers and tabs and I could open up like my Google Drive. I could open up a spreadsheet or whatever and say, "Hey, I want you to find all this information and throw it all in here for me." Operators being a bad demo today, but whatever, it is what it is. But you saw it moving around and working and operating. I just wanted to show that. Sorry, let's close this thing right out here.

Hey, John, quick question on the operator.

Yeah.

I've found use cases like that where it's like, oh cool, I'll try this. But this in by no way, shape, or form is saving me time on anything so far from what I've seen operator can do. But maybe you have examples that it's like actually is saving you time and it's not just like, oh, this is kind of cool in theory on the operator side.

Yeah, the operator side, you're absolutely right. You're absolutely right, Kyle. Um, I, I can book that flight way faster than I can babysit it to get it done, right? And so, you know, I want to be able to just tell it something and then I go shower and I come back and it's done. Right. But in order to do that, you have to give it everything it needs all the way through.

Right.

Like, what seats do you want? You know, I don't want to have to tell you every step along the way. Just, just use your brain, do the best outcome possible. And then, you know, if I want to adjust my seats later, I can do that. But, you know, I like Delta Comfort Plus, put me in there, whatever. I like aisles. But you'd have to go through and be pretty specific. Right? And in the amount of time it would take you to write that prompt, you might be able to just book those flights to Cabo, right? And so, so that's why it's a demo, right? That's why it's early access. It's just like, hey, here's what's happening, right? I think that's really the idea here is just so you can understand that. But Manus, on the other hand, is crazy. Okay? I'm not finding a ton of great prompts here. My prompts are freaking amazing. So we're gonna use those. We're gonna look at those prompts. Okay? The things that I'm doing with them are actually saving me, you know, 100 hours or 50 hours or 10 hours even, right? With different things like this. So here's a cool example right here. So I had somebody reach out to me that was interested actually in taking this course. And they said, hey, you know, I want to take your course because I work with HOAs and, you know, I need, you know, there's these HOA registries and there's like thousands of HOAs. Just in Utah alone. And I want to, you know, I want to be able to like run email campaigns to them, right? But going in and extracting all of that data just is taking me forever. And so can I build an AI that can go out and do that for me? And I was like, of course you can, right? Come take the course, we'll teach you how to do it. But then I was like, that sounds like a fun game. You know, like I was saying in the beginning, like fun challenge. It's like a video game to me. So I'm like, okay, hey, challenge accepted. I'm going to try this. So I came in here and I just like typed A, clicked search, and I was like, okay, yeah, there's a ton of stuff here. And you click into one of these here and you can see, okay, here's the HOA number and the address, city, state, zip address. Here's the email, here's the manager's name. Like, hey, it looks like all the information's here. And so if you could, if you could automate something that could actually, you know, reach out to all these people and first, first get all the data and then reach out to people, like that would be really, really cool. And I was like, I want to know if I can do this, right? And so, you know, I just turned to Manus and I said, hey, Manus. And I literally copied his prompt or not his prompt. That's funny. His message to me on LinkedIn, right? He texted me or he messaged me this right here, right? He said, look, I want you to just go there. I want it all into a spreadsheet. And so then I copied his prompt from LinkedIn, pasted it to here, and then said eliminate duplicates. Provide a long and detailed CSV of all of this information. Boom, go. Right. And so it started working on it and got all— it worked forever. Like, this is a very, very long thread, okay? Because there's so much data here. Okay. And then what happens in Manus is it said the current task context has reached its limit and cannot continue. But Manus is very smart. Manus can inherit the compressed context and all files into a new task to proceed. So it can take all its progress and all you do is click inherit and continue. And then over here on the, on the left, you just— here's the next one, right? And it did it again. Inherit and continue. And this is huge, right? Just massive, massive amount of data, right? Did it again and did it again. Each of those runs took about 40 minutes, right? So I just a couple of times I'd check and be like, oh, I have to do it again. Check. And here's what it gave me at the end, right? I've got this final CSV. I've got, you know, all the data, address, primary email, chair email, like everything I need. It's, it's all there ready to go. And it extracted everything. It extracted 82,000 total rows because, you know, every name has a couple different, a couple different, you know, entries, right? If you're looking for Brookhaven, if you type A, it's going to come up. Type B, it's going to come up. Type C, it's going to come up, right? So anyway, that's what happened there. And so it went in, but it found all of them, removed All the exact duplicate entries, all this stuff. Boom. 2035 unique HOA entries. He's like, this is exactly what I need. This is crazy. And I just messaged him the file and I was like, here, you know, you can take my class, but here's the file. Go have fun. You know? And I, he's like, what do you, you know, what do you want to charge? I was like, dude, it's on me. That was, that was like just a challenge, you know, for me. And, but it wasn't really a challenge. It was a prompt. And then I pushed a few buttons. Like the total input for me was like 3 minutes. Right now, you know what, it might have been a tiny bit more than that in the beginning. I might have asked it a few times to like try some different ways to do this. I don't remember exactly. Man, I hate this thing. This is so long. Let's see. Let's see what happens here. Yeah, like I had to tell it. Yeah, I didn't have to tell it anything in this one. I did something recently though with all of my— yeah, this is pretty cool. So basically I said, okay, look, I have these recordings of all these cohorts that I've done with the class that you're in right now. I was like, I've said a lot and the students have asked lots of questions, you know, can you compile this into complete transcripts for every single video? Right? And there's, there's literally most of my courses are 3, 3 sessions or 2 sessions, right? So there are a total of 41 videos. That I had over, you know, 15 different cohorts, right? And I said, okay, I want you to, I want you to just, you know, pull this out. And it tried to do it. It tried to like use the browser to go and do it and it couldn't figure it out. And then I was like, dude, don't, don't do it that way. Do it a different way. Right? And then anyway, it figured it out and, and was able to very quickly use an efficient way to just kind of go behind the scenes and grab it all. But now I have, I have all of the all of the different sessions. You know, this is just a summary, but like, here's, I don't know, cohort 2 complete transcript. This is a massive file, right? But the whole thing, everything that was said during that entire thing, all the way down, all the way through. So now I can take this obviously and I turn it into a book, How to Build a Business in a Weekend, right? I could turn it into a guide that I give to you guys, right? I could sell it as a, as an extra thing on my website, as a, as a lead, a lead-in or something like that with some HighLevel, whatever, right? But I've got all these different, you know, things that it's done now, all these different cohorts. And so pretty, pretty cool, right? Pretty cool to be able to do that. So that saved me huge amounts of time. Now, I could have manually done it. I could just open it up and, you know, Premiere and, you know, do the transcript and, you know, copy the whole thing and paste it over. But to do that with 41 videos and then compile it all like this and organize it all, that's just a ton of busy work, man. Like, I don't want to do that. I want to put an employee on that, right? I want to get someone else to do that for me.

What are the costs on, on this one? This is wild what you're doing.

Yeah, it's crazy. So it's actually, first of all, if you want to jump in, you know, I'll share my little link with you guys. Can you know, even if you sign up for just a free account, you can do that. I think you get 500 free credits by clicking on that link right there when you share. So go ahead and do that. But, but the cost here is very, very little. Let's see, account. Yeah, upgrade, whatever. I'm currently on— what am I on right now? Which one am I on? I might be on Plus. Oh yeah, current plan right here. Yeah, this is what I'm doing right now, $39 a month, right? So you get a certain amount of credits, right? And if you run out, you can buy more. You can upgrade if you're really using it a ton. But $199 for like 10 different, you know, 10 concurrent tasks, 10 employees working on projects for you at the same time, like Now, one of the things I was looking at today and I was kind of reading through a thread and commenting on it, thinking about it, is a lot of AI today really does require babysitting, right? Even if it is doing good stuff, like you still got to manage it. Like even with Manus, it's doing lots of great stuff. I still have to come back and watch it. And so not a lot has been done to create great workflows to manage multiple agents. So that's, that's an opportunity. You could build that solution inside of Replit, by the way. So one of the things I wanted to do, I've actually started working on it. Um, it's in ChatGPT here in one of my projects, AI to-do. But the idea here is— I called it TAD, um, just anyway, but TAD is, TAD is my, um, yeah, it's my to-do. Basically what I wanted to do is, is say, hey, can we create an agentic system that's a to-do list? And so I can literally just give— I, I just add a to-do item to it. Like, for example, you know, book me those tickets to Cabo or, or, you know, publish a new song to CD Baby or whatever, you know, so that it goes live on because, you know, whatever, any, any task that I have, right? Like pay my, my water bill, you know, whatever it is, I can just give it the to-do things, right? And then anytime it needs my input, it just asks me for it. It sends me a text and it's just like, hey, I need, I need this. Or it sends me an email or I can just get in the system and I can just answer all its questions really quick. And then it gets back to work until it needs me again. Right. And so, so basically I'm just giving it a to-do list and it just completes the to-do list while it's, while it's, you know, it's working anyway. So I'm working on that. I think it's kind of fun. I'm calling it—

you basically give it access to the different sites and things that you're trying to use. Like if I was like, I want you to make a list of all my Home Depot purchases and break down what they were for, like go to this website. Yeah, you just need to be giving it—

you just have to be specific and, you know, give it, give it access, right? If I need it to go pay my, you know, my SESD bill for my power, I got to give it my username and password, right? I got to have it do that. My inner— or my battery here is like dying on me. You know, I'm plugged in, it's going down. Maybe I need to close some things. I don't know what's going on here. We're gonna actually, just for fun, we're gonna close out a bunch of this stuff. I don't need all of these Shopify things open. My website here, I'll do— we're going to use GoDaddy later, but yeah, I'll just leave that open because we'll do that. We'll close that and that and that and that and that. We'll just give it a little bit more a little bit more juice. We'll keep using that AI Manus. We don't need this open anymore. Browsers take up a lot of— yeah, it went from 8% to 9%, so that's already improvement. I switched my power cord, but that was weird. I don't want this to just turn off all of a sudden. If it does for whatever reason, then, you know, I'll switch it over to my desktop that I have right here and we'll, we'll get you guys back in. All right, so Manus is not that expensive, obviously. It's, it's great, but the types of things you can do here are incredible. I'm going to share a couple more because I just want you— I share these examples and show what I'm doing here because I, I hope it just sparks an idea in your mind. Um, so this was, yeah, this was the Shopify, like, hey, I need you to refund 90 customers, and it just couldn't do it. It couldn't get past the bot protection on Shopify. So whatever. Even when I took control and tried to like complete CAPTCHAs and things like that, which by the way, it can complete CAPTCHAs on its own, but it, you know, there was a bot protection. It just would not complete. It would not let me get past it. So couldn't do it. Um, so I've got a, uh I've got a neighbor here in West Mountain that owns a decent-sized, you know, construction company. And basically he, you know, he's— anybody here ever work construction? One of the things that's really tough in construction is the bidding process, right? And because you get— what you'll receive is you receive something kind of like this. You know, it'd actually be easier if I just go like this, paste the text. Just open this document here. So you'll get a document like this. This is like, okay, hey, we need— this is the design. These are the, you know, the plans, and they're very, they're very technical, like detailed documents, tons and tons of information, site plan notes. And we need, you know, this much of this and this much of this. And you've got a scale here, but you have to go through as a construction company and say, okay, well, we want to give you a bid on it. Well, in order to give you a bid, we need to know exactly how many linear feet of curb there are and how much, you know, 5-inch slab there is and how much, you know, what do we need for all of these different things where we're building a big complex site, right? And it's like a 20-hour to 30-hour task. Like, it's a long task and there's specialized software out there that helps you with it. But even with that, it's very manual and hard. And he's like, dude, can you— and, and not only that, you can spend that 30 hours, put together this beautiful bid, you know, and send it over and get rejected. Right? And now you've done a lot of, a lot of work as a construction company, usually as the owner or somebody high up. It's usually the higher-up people that are doing this kind of stuff. And, and it's a waste of time, right? And then you might not get any money, get paid for it, right? So it's a really kind of risky thing to do as, as a, you know, an owner of a construction company because there's no guarantee that all of your effort is going to even, you know, lean into anything. And so what I did He said, can you have it do what's called a takeoff? That's called a takeoff. That's, that's all the information that you need in order to create that bid, right? And then from that takeoff, you can create the bid itself, right? And so I basically, at first I just gave it to ChatGPT. I was like, hey, do a takeoff. I just tried it and I was like, yeah, it didn't work. Okay. Then I gave it to Manus. I said, hey, do it. And I was like, yeah, these, these numbers are way off, right? It tried. It was this one right here, right? It was one of these guys right here. And maybe it was this one. Maybe it was this one. Yeah, it didn't even work. It kind of, it couldn't figure it out, right? Maybe it was this one right here. This was maybe the first one I had it do. Yeah, here we go. Here we go. So yeah, it came up with all these, you know, site areas and all this stuff and, and none of the numbers were even close. Like it was just picking things. Like I don't know how it was figuring out what it was doing. And you obviously, you know, could go back and read it and try to figure out exactly what it did. But scrolling through is looking at things, creating files. As I was watching it do it, I was like, it's actually going to work pretty well. I'm just so excited to send it to my neighbor. I send it over. He's like, yeah, not even close, man. And I was like, okay, well, I have another idea. And so what I did is I basically gave it a task to— I said, here's, here's the, the, the construction plans. And then here are the, the documents, 1, 2, 3, 4 documents that were created manually by This guy that knows what he's doing. These are correct documents that were done. So what I want you to do is I want you to build software that pulls all this information off and then, and then, you know, test it on, on the plans, right? See what, see what output you get. If you get the right answer, right? And I'm sorry, then I want you to take those answers and I want you to compare it against the actual answers, right? And then I want you to keep iterating until you get it right. I want you to design a system that keeps iterating and changing it until you get it right. And it took it a couple hours, actually. And so it went through and so it created a basic framework at first. And then when it, when it ran it the first time and said, hey, I'm 90% off, I'm 96% off, I'm 99% off now, I'm going to improve it by doing this, this, this, this, and this, whatever. So I'm checking, I'm adding validation checks against known measurements and trying to figure out how to do all this stuff. So anyway, long story short, it finally finishes and what it gave me was a bunch of— sorry, let me— oh, you know what? I think I actually have this, this thing here. Construction takeoff automation. So it gave me all these Python tools, right?.py. Python is a code, a code base, and, and a Python, a Python script basically does a bunch of stuff for you, right? But it does all these cool things. And then not only that, it created a README and documentation for it, a user guide of how to use all these Python things, right? Introduction, installation, getting started, interactive mode, batch mode, calibration, measurement tools. Here's all this stuff, right? Everything that you need to understand how to do it. And I was like, okay, well, I don't want to do it. So I took this entire folder and I opened up Replit and I just dropped it in. And I said, all right, here's what we're trying to do. I need you to build using all these Python scripts that have already been built. I need you to build me a user interface or a UI here so that I can, I can then test this and use it and wait for it. So I've got— I was on with Replit 2 days ago and I was like, guys, I need you to give me a search functionality for my, for my tools here because I have so many of them. But it built it. It built the construction takeoff app. And here we won't even get it running. We'll just click on the domain here. And so it built this thing that now my neighbor can just jump in, upload some documents. It's taking a second to spin up the server, the autoscale server, because he's not using it right now apparently. But here it comes and wait for it. This one is pretty complex one. So it's actually going to— it's not optimized. I didn't optimize it all. It's kind of an internal tool for him. But, but I did publish it and I want you to see it. It'll be so cool. All right, here we go. So basically you come in here, you create a new project, saves it to the database, project description, you create the project, and then you come to the PDF upload stage. You'll upload that, that document. Then you'll have to go through and you actually have to do about— it's about 20 minutes of scale calibration. Each page that has scale, you have to go in and like go through and just make sure that you label a couple of things. And then you export the results, right? It does the whole thing. Boom. Now all of a sudden he can whip off a bid in 20 minutes, right? Not 20 minutes plus letting this run for about 10 minutes, but it's pretty amazing, right? So that's, that's using all parts of the AI brains out there, right? This is saying, all right, we're going to have Manus figure out how to solve this problem and iterate until it gets it right. Then We're going to take its, its results and we're going to drop it into another Agentic system that figures out how to get it all running. Right. And it took me maybe 2 hours to get it running inside of Replit.

So anyway, any questions on that?

Yeah.

Can I, can I ask a couple of questions there?

Yeah.

So with that, that's what we do. We do construction and the estimating process is like one of our most time-consuming things that we do. If I were to upload, let's say, like 2 or 3 years of documents where it says, here's the plans and then here's our takeoff. We do 3D takeoffs and then we, we also do 2D takeoffs depending on what the, what the project is. Is that going to help my accuracy or kind of what comes to mind is, is there a different scope of work where it's like you mentioned concrete flatwork, this is how we do those takeoffs. Versus curb and gutter or a different type of site work or site utilities. Like, is it better to be specific with it? Always improves accuracy.

Yeah, always. So I don't think you need 2 to 3 years of results, but if you give it 10 or 20 right examples of like, hey, here is the project, here is the takeoff, this is what we did, these are all the 2D ones, these are all the 3D ones, be very specific as possible, right? The software really struggles with what's called the edge, edge cases, right? Where something isn't standard or it's not, oh, I wasn't ready for that. And when it runs into something that it's not ready for, you know, it's, you're flipping a coin, right? You don't know what's going to actually come out of the other end, right? It might get it right and it might not, right? And so anything you can do to train it as close to what you actually need as possible. So it might, it might be something as simple as, hey, yeah, we're going to train two different models. And then in our UI that we create, we're going to say 2D and then we're going to say 3D, right? We're going to have just a toggle like, hey, this is what I'm uploading, right? This is what I'm— this is what I'm needing out of this. But, but yeah, man. Yeah, I didn't— I realized as you were talking that, yeah, you're doing that stuff. So this could obviously save you guys tons of time, right? And, you know, build your own system. Follow my thing. This took me a day, right? One full day of work to get this done, probably about 8 hours in total to make all of this happen, pull it all together. Now, it might take you 2 or 3 days if you're uploading lots and lots of data, right? And so what you would do here, here's how I would do it. I would say basically the same thing I said up here at the beginning. By the way, I'm more than happy to share this prompt with you, whatever, you know, and you can add more to it. Yeah, but I might say, I might say, hey, Hey, you know, here's 2 projects that we've done. They're manual, you know, but I want you to iterate. I want you to build something that will help us do this in an automated fashion. Then when it's done, you say, you give it a new one, a 3rd one, and you say, test it against this, see if it works. Okay, then a 4th one, test against this, see if it works. And then a 5th one, test again. Oh, I need to do this right. And eventually it should be able to kind of fill in those edge cases. Right, where all of a sudden it's now able to, to produce really awesome stuff for you. And then putting it together into a UI, just like creating buttons you can click, Replit can do that in its sleep. That's easy. That's just a matter of getting, you know, the black text on a black background to look good, you know.

And will, will Replit produce like a PDF that can be downloaded?

Yeah, totally. Okay. Yeah. Yeah. You can have it, you can have it produce all kinds of stuff. You can have it produce HTML. You can have it send an email. You can have it send a text. Whatever you want to do. Now, in order for it to do that, it has to install different libraries, right? So for example, with, uh, with GenAI API, I have a pretty complex email structure. Um, I wonder if I have that up here somewhere. Let's go to the assistant here. Yeah, so, so I have, um, in order— so guys, you're gonna have to integrate an email service. I recommend— and it's an API, right? It's just another API. I recommend SendGrid. SendGrid is free for up to 100 emails per month. And then it's like, I don't know, a fraction of a penny per email after that. It's so cheap. Right? So if you have any number of real users, it's paid for, right? And then it's like, I don't know, $7 a month. It's very, very cheap. Okay. But SendGrid, you guys might've heard of Twilio. Twilio is a texting service, right? If you ever get one of those shortcode things, it's probably powered by Twilio. Like that verification, 2FA verification is probably coming from Twilio. SendGrid was purchased by Twilio, right? So it's actually the same company, but basically, yeah, Twilio, SendGrid, right? SendGrid does email and it's going to come from here, right? The default is no reply, whatever, right? So I've got all these different things, no reply, info, enterprise. Those are really kind of the main ones I'm using right now. Then Replit setup all of the email templates for me. All the HTML is just saved in the code, right? And so I didn't have to go and design all those things. It just built these nice, beautiful things. And so if I click, you know, password reset or whatever, if I just try that right here, I go log in and I forgot my password and I say joncheney7@gmail.com, click send reset link. We'll let that send and I'll look at it in a second. Okay, check your email. Boom. All right. Obviously, this is another one of those pathways you got to follow. You got to check all these things, make sure they work. Okay, but certificate emails, when you, you know, when you, when you do a registration or start, when you do a certification on that, that's going to happen there. So, so anyway, here's the, here's the quotas, emails per day. You know, this actually needs to be different. I actually disabled a couple of things because it wasn't working right. But, but this is the full email setup and I just kind of worked through it and, and built all these different things and I'm paying, you know, I think probably like $20 a month for my one email address with, you know, I'm through Google Workspace and Google Workspace gives me access to all kinds of other things, obviously Google Drive and big limits and, you know, all that stuff and all the, all the documents I need there. So I already need, you know, that for GenAI API. And then, and then I'm paying for SendGrid, right? And SendGrid just has its APIs and, and I, I set up my senders and I'm good, right? Then I can go to my dashboard or activity. Let's see if there's any activity. I'm sure there's activity today because I just sent an email and I know a couple people have purchased courses today because I got notifications on them, right? Yeah. Anyway, boom, there we go. 35 requests today. So people have done things today. I don't know what they're doing, but stuff's happening. That's cool. And so I go to my email. I should theoretically be able to go to Jon Cheney 7, and there's my password reset right there. So it created this email, it set up this reset password here. If I were to click that button, it should just pull up, create new password, boom, new password, confirm new password, reset. It updates that in the database, takes me back to the sign-in, and I'm rolling again, right? Really awesome, right? I mean, that— I, I know you've all experienced a password reset. There's nothing like crazy about the functionality, but it's cool to set it up manually and be like, I built this system that works exactly how you would think it would work, right? And again, all I did was act as a user. Okay, well, if I can't log in, I need a button. Okay, well, it's probably going to send me an email. Great. That email needs to look good and be simple, and I need to push a button and then fill in my new password, click reset, and then I sign in, right? Like, I'm not a genius. And, you know, any— but anybody can figure that out. You just maybe have to write it down on a notepad and figure out, okay, here's the steps that I want to do. And then tell Replit what you want it to do. Or you really can just say, hey, I need a password reset system, and it'll build it right now. It might ask for something else. It might ask for, you know, a different API than SendGrid. It might choose a different system like that, or it might say, hey, yeah, we can set this up with Google Workspace. And I actually, when I first did it, when I first set up my email, I set it all up with a, with like crazy settings behind the scenes inside of the Google, inside of the Google Workspace where you had to go. I had to go into all these crazy things inside of, you know, I don't know, rules and accounts and settings and all these things and create this special sending system and give it permissions and do all this stuff. And it took me, took me some time. It took me some ChatGPT talking to figure it out. And then later on I was like, Wait, SendGrid's a thing. This is, you know, I had never done this before. I'd never set up the email before. And then I was like, actually, I've got a better idea. And so I built the aliases, use SendGrid, set it up. That is the way to do it. That's the best way to do it. And I've, you know, continued to do research on it and I'm, I am correct that it's the best way to do it. One thing I did in here, guys, you might not have, uh, or you may have noticed, but you have the agent, which you've been using a lot of.

Okay.