Cohort 20 - Session 2 - 9-10-2025

Good.

Do that. We'll just— I want to kick off today, guys, before we officially kick off. We'll let other people wake up and get on the, get on the meeting. But I want to just open up for questions right now. Right. Anything we talked about yesterday, anything you ran into, you know, just on your own afterwards or just kind of any, any topic that's on your mind. What's, what's, what's coming to mind, guys? So, uh, my app is actually up in the Replit preview, and it suggested that I do it in like an incognito tab. So I did that, um, and it looks kind of cool. Um, almost none of the buttons have actual functionality. Um, and so I'm wondering, is it just as simple as— I haven't played around with it too much, just for about 20 minutes, so Is it sort of as simple as, do I just go back to Replit and say, hey, almost none of the buttons are working? Well, yeah. Or is it I kind of missed something? No. Yeah, I talked about this a little bit yesterday, but again, with so many things going through your mind yesterday and, you know, so much being thrown at you. Good morning, Carrie, and good morning, Abby, as well. You know, it's Replit will build things kind of stage by stage, right? The first stage is I want to just show you what this could be. Then the second stage is, okay, now I'm going to— we're going to get the first pieces right. We're going to get maybe login working, or we're going to get this thing working right. But it can't sit there and just build for 30 hours straight, you know. And so it hasn't built the functionalities tied to those buttons yet. Right? Those pages don't exist. The functionalities don't exist. All it's done is it's put a button there. And so now you're like, okay, before— because here's the thing, Matt, before you go and you have your AI or your developer, right? Think about this. Think about this. Let's take AI out of this, right? We're back. We're paying Abby. You know, she's getting paid $200,000 a year, right? And, you know, instead of her going and working for 6 months and you're putting together a $100,000 check, to, to Ryder, she's gonna say, okay, let's, let's look at this first. Do you like this button? Is this a button that you actually need, right? Before we go build all the functionality associated with that button, let's decide, is that the user experience we want, right? And so you're the, you know, one of the things that a lot of people worry about when they're, when they're not software developers or not, you know, user experience, UX/UI designers or those types of things, right? Um, a lot of people worry like, man, should I— do I, do I know how to do this, right? Like, do I know what the— do I know if I need that button, right? Am I qualified to even know whether or not this is the right thing? But the reality is, the answer in my opinion is absolutely yes. Why? Because you all have probably as many apps as I do on my— on your phone, and you've You've used Netflix and you've used, you know, whatever. You've used tons of apps and you've clicked on Microsoft Word and you've used other websites and you've used Bank of America and you know what you kind of expect, right? You're like, well, there needs to be a login button. It's usually in the top right, right? Usually don't put a login button in the bottom left, right? And so you just have to, you just have to trust yourself to say, you know what, I'm just a normal person like the next guy or girl.

Right?

And so I'm just going to say, okay, this is what I think it should do. Do I think I need that button? Yeah, that's a cool button. I like that. You also want to be, you also want to be a little bit careful, and I'm not, and I don't want you to, I don't want to scare you away from doing too much because sometimes you do need, you know, a lot of things. But I also would err on the side of simplicity, right, whenever you can. Um, maybe it puts 20 buttons on there and you think about it for a minute and you're like, you know, this would be much easier if it was really just 4 buttons here.? And then maybe we have a couple of these functionalities that we put in other places. But for, for an MVP, like, can I get a customer to pay for this with just these 4 buttons, or do I need all 20, right? So these are just questions you got to ask yourself that are just good, you know, startup questions, right? You know, entrepreneurship questions, saying, all right, what is my MVP? What is my minimum viable product, right? Minimum viable product means I can I can hand my product to somebody and they can get value out of it, right? Hopefully value. And then you could, you could, you know, some people will then go a step further and say, okay, I can get value out of it. Now what's my MSP, my minimum sellable product, right? Now, are they— is the user getting enough value that they would actually pay me for it, right? And once you get to that, some people combine, combine both of those into just MVP, but But once you get to that minimum sellable product, okay, great, that's awesome, right? Then you start to say, well, do I really need all these other fancy things? And that just takes time and talking to your customers, using it, clicking around, doing all these, doing all these different, you know, kind of testing scenarios here. Okay, so what's up, Abby?

I love that minimum sellable product. Someone else used a term that I always think of, which is minimum lovable.

Product?

I always think about that. What's the minimum that your customer could love?

Yeah. Yeah, that's what you're looking for. Yeah, I'll buy it, but I'm just tolerating it. Yeah, there's definitely a difference there. I do love that difference. And I've heard minimum lovable product as well. And I think it's a good, I think it's a very good measuring stick, right? And it's all subjective, right? There isn't like, here's the answer, right? You just have to say, okay, what do I think? And then put it in front of your customers. If they're like, man, this is just awesome, then you know you've hit a home run and you can start to grow and start to say, all right, Maybe I can get confident in putting some advertising dollars behind this. Let's, let's run some Meta ads. Let's do, you know, whatever it is, whatever it is your product is. But anyway, Matt, I know that's kind of a long answer, but you know, why are all these buttons here and why don't they work? Right? Very common, right? But it's, it's just a matter of, okay, let's get this button working now. Right? And you just tell that— cool. Hey, let's build this. Let's build this thing. Let's build the next feature. Let's build the next function, whatever it is, right? Um, okay, good. Any other questions here? We've got, we've got, I think, most of the class on. Um, I think, uh, Cole's gonna be a couple minutes late, uh, or no, Ethan. Ethan's gonna be a couple minutes late. Um, anyway, so they'll come when they come. Um, I guess they're— neither of them are on. The young guys are still sleeping in. Um, no, I'm sure they're doing something. I'm sure there's something else. But, um, did everybody get the email that I sent yesterday with the recording? Did anybody not get it? Let's ask that. Raise your hand if you did not get my email with the recording. Okay, all right, perfect. It went— Matt messaged me, he's like, hey, does that come through? And then he found it in his junk email. I hate it when my stuff goes to junk email, right? Like, I do all— I spend all this time writing these amazing emails and send them out to people and no one gets it, you know. Anyway, so, um, no, I'm glad those are coming through. Very good. Okay guys, um, who, who worked on their platform yesterday for more than an hour after class? Abby, Carrie, Jake, Josh, maybe Nick. Okay, a few of you guys. How about 2 hours? 3 hours? All right, Abby, how long did you go? Carrie, that's good, 2 hours.

Uh, I think I stopped— so I'm, I'm earlier than most of you guys, but I think I stopped around like 6 or 7. I had a late dinner and I had to stop. I wanted to keep going, but I had to force myself to Very cool.

Very cool. Abby, I'd love for you to share, just if you don't mind, what do you think? Just like you're a developer, so I'm glad it was you that spent 6 hours working on it, 7 hours working on it. How does this compare to traditional development and what kind of— I don't know, how does it unlock more opportunity for you?

I guess it's a really good question. Oh, sorry, I just got a system notification. Just give me one second here. You're good. Can you guys hear me? Yeah. Oh, all right. Well, whatever. Um, okay. Yeah, no, that's a really good question. Um, for, for me, um, I think I told you last night, my impression is that this is a really knowledgeable junior engineer that doesn't have experience yet because it knows the stuff. But it doesn't have the greatest attention to detail. So I would be like, hey, that button's not working. Let's get that hooked up. And, and also, can you change the text on it? And it would do that. And then there would be an error. And the reason why there would be an error is because there was something in the database that was referencing the original thing, and it didn't think to also change the database. So And then, you know, I would just be like, hey, there's still an error. And it'd be like, oh, I didn't change the database. I'll go change that now. And yeah, so I found that to be really interesting. You just have to guide it. Something I'm really, really impressed with how quickly it handles things. I have only gotten one position where I had to rewind. I have a little bit of a problem right now where I've been going back and forth where it's fixing the problem and causing a new one. And then when it fixes that problem, it goes back to the original problem and then goes back. And we've done that like 4 times. So because I know how to code, I'm just going to go into the files and fix it. It's actually a really simple problem. And yeah, I do see that that could be a little bit challenging for people without experience. However, I really like this platform because you can access all the files. So for instance, if you are having a problem, you could ask it where is the code that is related to this and you can put it in the plan mode. I found plan mode to be quite helpful. I burned through my first $25 really fast and then I realized that I could put it in plan mode and talk to it before it would make, do stuff and that helped a lot. So you could go if you have a problem that you're just kind of stuck on and it's not fixing it, you can always ask it to in plan mode to give you a big write-up of what's going on, what it thinks is going on and to also give you the code that it thinks is the problem. And if you don't have the experience to look at that, then you can take that to another AI and be like, I'm having this problem, what do you think is it? And then you can take that answer and give it to Replit in the build or in the plan stage and say, what do you think about this? Because there were a couple times where I, I could tell what the problem was and I wanted to see if it would figure it out and it didn't. And then I was like, hmm, do you think maybe you haven't updated the database yet? And then it'd be like, oh yes, and it would go do it. So I think that somebody without experience could utilize another AI as their engineer in a way by, by taking the stuff that Replit tells it that it thinks is the problem, giving it to the other AI, and then taking what the other AI says as a suggestion to fix and giving it back to Replit. And then it'll look in at the problem in a brand new way. And I found that it would, it would solve it then. Yeah, I'm quite, I'm quite impressed. One thing that I think is a detriment though, and you should watch it, is exactly what you were talking about. Drawn about like the too many features. I would ask it to make a change and it would definitely make the change, but it would make a decision to add a new feature that hadn't been in the list before. It wasn't in any of my documentation I provided it. We hadn't talked about it. And I would just notice that a new feature was added. And someone without experience might be like, oh, that's awesome. But it's actually not awesome because you don't want to have features your users aren't going to use. And you might think, oh, what's the big deal? I'll just take it out later. But even if only 5% of your users or 1% of your users are using a feature and you take it away, you're gonna get negative reviews all over the place because they're pissed off. So you don't wanna take things away. So you wanna really like, don't let it add stuff without you just making the decision that it should be added. But I, I definitely found that happening. Hopefully, I don't know, maybe that's helpful for, for some people and maybe that's a feature that they want in there. But I think for like a a professional product that you're releasing out into the world, you need to watch it for, for adding extra stuff you didn't ask for. But yeah, overall I'm super impressed.

Thanks for, uh, you know, thanks for reading that script I sent you. That was perfect. Um, no, um, no, that was exactly what I wanted to hear. A lot of the things you brought up are things we're going to go over today. Um, you know, I'm going to actually teach you guys how to build your own AI today, right? Which is super easy. It's It's actually way easier than you think. And by, by your own AI, I mean your own AI interface, right? We're not gonna build an LLM, right? We're gonna use one of the other models and you'll learn how to build, you know, your own personal ChatGPT, your own GPT, your own assistant, your own whatever you want, right? There's a lot of ways to do it and I'll show you how to do it the easy way and then I'll show you how to do it the hard way, which is like, hey, this is the developer cool way, right? But when we're inside of Replit, we can be a developer and we can do some, some really neat things. But one of the things that you've mentioned that's really important, Abby, is an overarching principle that I said yesterday, but I'm going to say it again now, and I hope it makes a little more sense why. You have to be the boss. You can't let the AI take over because it will. It's ChatGPT, Grok, Replit, Claude, right, which is behind Replit. They're all geared towards being this overindulgent. I don't know, they're like, they just want to make you happy. They're trying to make you happy, and they want to do it so much that they'll do anything for you, right? They'll even tell you fake news, right? Fake truth. They'll make stuff up. They'll hallucinate is what people call it, right? And, uh, and in this case, in Replit, they'll, they'll do more than you asked. And so, you know, that, that comes down to how the prompt is written, right? Um, for, for what, how, how the AI is supposed to behave inside of Replit. I'm going to show you guys that today. What is underneath the hood in Replit and how does that work? But you can control it. The way that you can control it is you just have to add little tags that keep you in control. For example, hey, I want you to make this button work and the functionality behind it. Don't do anything except for this. Just that little don't do anything except for this. That little tag is going to say, okay, I'm just going to get this done. And then, and then you could say, report back when you've completed this before you do anything else, right? Or, you know, go into that plan mode. Abby, you nailed it. The plan mode is amazing. Go in there and say, okay, I told you to fix this thing. You fixed it, but then you broke this thing. Then I told you to fix this thing. Then you broke this. What you, you know, the problem you're having, Abby, right? Can you explain to me why when you fix this, you're breaking this as well? Oh yeah, I'm doing this and this. Okay, great. I need you to only fix this in a way that keeps this working. Tell me your plan to make that happen. Okay, great. Let's switch back to build mode. Do the plan, right? And so you, you learn over time how to just speak, but you want to be overly specific with when you're talking to AI. AI, you know, a lot of times in school, right, we were taught to be, you know, if you're a writer, right, you want to be very, um, pithy, right? You want to just get to the point. You you don't need to be overflourished, you don't need to keep going and talk about all this stuff, right? You're like, Jon, get to the point, right? But with AI, you actually can just kind of ramble a little bit. You want to, you don't want to confuse it, but you want to give it as much information as possible. And one of the easy ways to do that, by the way, with AI is to turn it in or turn on dictation mode, right? If you're on a Mac, you just hold, you know, function F5, right? And that turns on the microphone. You can just talk to it, right? In ChatGPT, you can turn on voice mode and you just talk to it, right? Same with Grok and same with many of these models. There's voice modes that are really, really awesome. And we're gonna go over some cool AI voice stuff today too. Okay? So guys, you're doing awesome. You're doing awesome. Today we're gonna do some really fun stuff. Let's, we're gonna start off today not inside Replit.

Okay?

We're gonna jump over. I'm gonna introduce you guys to a really cool platform called Manus. And Manus is awesome. It is what I would call more of a general purpose agent., right? Whereas Replit is an agent built to, built to just build software, really. Like, you can have it do other things and ask it questions and stuff. That's not what it's built for. It is built to make software, okay? Because of its prompt, because of just how it's set up. And so Manus is just super fun, guys. I'm such a huge fan of Manus, okay? Um, you know what, before we go into here, we're just going to show this. So I don't want to go to app, I want to go to Manus. I am right there. And then we'll scroll down a little bit here. Okay, so Manus, just like any other AI, right? You've got this little place where you can just say, hey, work on this, right? But look, you can have it do images or slides or web pages or spreadsheets, visualizations, or tons more, right? I love to use Manus for things like research, right? Or just going through tasks that are tedious and, and, and, I don't know, just take a long time. So I'll show a couple examples of things that, that it's done here. So let's see, let's, I don't know, just looking at some of their use cases here. That they've got. Um, transform text into natural speech. That's cool. It made its own, its own voice platform here. Okay, this is kind of a cool one. Let's do this. Okay, so view the replay. So, so this is kind of a cool feature they have here where you can basically look at what the, um, you know, what the, uh, prompt was and then what it actually does, right? So it's basically says, hey, create an interactive web app that that takes the electromagnetic frequency colors produced by various elements in the periodic chart and converts them to the audible range by dividing the octave until the tones are in the piano range closest to middle C. So the tones are pleasant anyway. So it just gives this kind of fun idea of, look, can we turn the periodic table into a musical instrument, basically, right? And so it builds a whole list and it's kind of replaying everything that's happening. It's testing itself, working on some stuff. And you can see the list that comes up here, right? And it shows like, oh, I did this, I did this. There's that, that chain of thought reasoning that we talked about yesterday, right? Where it just kind of goes through and it says, okay, I need to do all these steps here until I have my final result. Right? And so it keeps going for a while. We'll skip to the results and, and then we'll see what it did here. Right? So it created a website. Okay. And this website, you might not be able to hear it. We'll see how good this works, but let's see if this actually works. There we go. You can hear kind of little— I'm not sure what note that is. Oh, it's a G. Sorry. Anyway, yeah, that's it. Anyway, so kind of simple, right? And you'd have to figure out what the right combo is. Anyway. So kind of fun. But, but guys, in order for someone to build this, right, a developer to go and build something out like this, it would have been pretty difficult. That's interesting. I actually want to play through this because I'm, as a musical person, this is really, really interesting to me. I'm wondering why, like, I'm curious how the elements, um, interact with each other from a, from a relation standpoint like that, and if they would make cool chords or what elements combined would create cool chords. Anyway, So I'll play with this later, but really cool, right? To be able to build something like that. That's a fully functioning website. So that's again, an example of a website like we were talking about, like Replit could probably do that. But here's the cool thing about what Manus did. Manus worked through it. It took the prompt, worked through it, debugged it all the way until it has a complete functioning app, right? Something that's actually working. Look at this. Create a P5.js Create animation with p5.js. There you go. So that's kind of fun. If I click on this one, it actually interacts.

So why would we use Replit then if this type of a tool exists?

Yeah, there's a few reasons. So let me answer that question towards the end of this demonstration, because it's a really good question, and I want to show an example where you use Manas and Replit to get a way cooler result. Where you can combine the power of both platforms to do some really, really neat things.

I guess my question is more, what is the power of Replit? I guess what I heard is you're going to answer that.

Well, yeah. The truth is Manas doesn't allow you to play with— you don't have control over the code. You can't go in and change something. Like Abby was like, yeah, I can go in the code and I can look at this and I can change this and I have all this power and I can do this. You can't even deploy stuff with Manas. Manas is almost supposed to be like, here's a quick MVP.

Right?

And but you could— what you could do though is you could say, hey, Manus, why don't you make a quick MVP, an example? Yeah, that's pretty good. And then hand that to Replit. Um, and then, you know, all the— you can ask for the code, but you can't manipulate the code, right? But it'll give you a bunch of files. You hand those files to Replit and say, hey, will you rebuild this? And then let's work on it, you know. So it could give you a first draft or something. That would be kind of a cool way to do it. Um, you know, here's an example of just building a PowerPoint. Um, you know, Who, who hates building PowerPoints?

Me.

I hate building PowerPoints. Um, I am not a fan. Like, as, as CEO of, of Seek that I founded back in 2016, I had to build hundreds of decks, mostly when I'm pitching investors, right? I got to make this beautiful deck where I got to go and put in front of people and have 8 slides and get all this stuff just right. Now I can build slides in in seconds, right? I mean, here's, here's some slides, right? I don't even know what this was about. This is something about historic— yeah, create a PowerPoint with detailed analysis of historic economic bubbles. Boom, there we go, right? Here's the anatomy of a bubble, tulip mania, South Sea bubble, Roaring '20s. Hopefully it gets down to internet. Yeah, patterns. Look at this. I mean, look how cool that is, guys. That's awesome, right? So amazing that it can just whip that up. Now I think this is just kind of okay. I'll show you guys another platform for— in fact, I'll do it right now while I just— while it's fresh on my mind. Um, let's see here. I think my account is on this one right here, Gamma. So gamma.app is, um, you know, here's a— I don't know, I don't even know what this is about. GenAIPI, global standard for AI proficiency. I don't know why I made this, but we'll just click on it and see what it looks like. But, um, but one of the cool things about, about this Yeah, yeah, yeah. This was— I was, I was making a theoretical pitch deck for this, right? I'm actually not raising money for GenAIPI. If you have $1 million you want to give me, I won't take it. But yeah, if I was raising, you know, 10% at $10 million post-money, right? Here it is. All I did was give it a simple prompt and it puts this together for me in just a really beautiful way. Early traction. Here's this market opportunity, business model, product services. Anyway, so it did this really, really quickly. Gamma.app, in my opinion, is the best, is the best one for creating presentations. It literally just takes, you know, a few seconds. Paste in some text, some notes or whatever, tell it how many slides you want, and boom, it's done. Create from a one-line prompt in a few seconds or import existing documents and have it pull it all together, right? And so I don't know if Gamma has an API. It'd be really— oh, they totally have an API. That's cool. So you have to upgrade to Pro, so maybe you have to pay a little bit more. But think about this, um, Jake, I'm going to point you out here. Um, you know, Jake owns a pool company, and you know, lots of times, you know, when you're, when you're pitching, you know, a resort or something on, on a pool that you're going to build for them, you know, if you want to put together a really nice presentation, right? So you could take all the data, all the stuff you would normally, you know, work hard to put in front of them. And you can still provide all the supporting documentation, but you could just drag all of that into here and say, boom, let's do this. And so we could connect that Jake as an example into a dashboard for salespeople, right? To, to be able to pull something together and then, hey, let's, let's have a, a super nice presentation pop out the other end after we enter all that stuff into the, into the system. So cool idea there, right? But yeah, guys, it's, it's amazing at how, you know, I had it, I had it figure this out. What's, you know, best food in Southern Utah County, right? And it just finds this stuff and puts it to the other and, you know, super easy. Imagine you're a, you're a reporter or something. You're trying to do a cool story for St. George or whatever it is, right? You can just have it whip stuff up like that's the super fast. You can also, by the way, have it create, you can have it do, let's see here. I want to go back. That's not what I wanted. Create new. Let's just say generate this. Here we go. So you can do a presentation, you can do a webpage. You can do a document. You can even have it whip up some social posts for you, right? I'm gonna use Gamma a little bit more. I didn't know they had an API and I just kind of thought about it. I'm excited to try some more things with APIs in here. But anyway, gamma.app for presentations, guys, just write that one down. You're gonna, you're gonna wanna use that some more. Okay. So let's jump back into Manus here. I wanna show, I wanna show a few ways that I have used Manus, right? And I think Jake has seen this because I showed him the other day, but The rest of you haven't. So I'm going to show a couple of the ways that I have used this that I think is really cool. So here's a cool thing. I asked it to just create an inspiring biography about me, right? And I gave it, you know, here's a couple things, search the internet, find what you can find, okay? And then create a website on it, right?

Visit.

I just wanted to see what it would find about me, right? And here's what it was able to do, right? It was able to figure out, okay, AI, adventure, music, languages, recognition, whatever, personal life, with discovery. Cover his story, visit GenAIPI. It found all these different things here, right? I'm, you know, I'm a whitewater kayaker and it found the formatting on this isn't perfect. So it's kind of hard to read, but like this big banana run, right? This is when I went down to Mexico. I found a big blog on that. Or no, I'm sorry. I'd written, I wrote an article on LinkedIn that talked about business stuff and kayaking and, you know, anyway, really fun. So it was able to do that, but pulled that together into like a good thing. AI mastery and innovation, GenAIPI, world-class AI in months. Um, musical mastery, 7 albums published, all this stuff. But look how cool this is, guys. Like, think about how long it would take— you know, I was talking yesterday about a personal brand website— how long would it take to pull this kind of information together if you were trying to do it, you know, on your own? Um, you know, and it finds all these other things, right? Like, you know, Crunchbase and Horizont and Utah Business and other places. Personal life and values. He knows I'm a BYU fan. Right? How does it know that? Honestly, how does it know that? Did it find that on my social platforms? I mean, I post about BYU and like going to games a lot, but I don't know if I've ever— like, I'm, I'm shocked that it has this piece of information about me. I think it's pretty cool. Like, if, if I were to go to Google right now, let's just try it, right? Let's just say Jon Cheney, BYU Athletics. Bam. I mean, I don't know, like, okay, okay, good. So there's an article about something I made for— yeah, there you go. Jon Cheney graduated from BYU, says he hasn't missed a Cougars football game on television or in person. That's true. So I guess it got that little piece of information maybe from this one Deseret News article on something that I had done. Anyway, I'm just really surprised that it had that piece of information. It's really cool. But guys, you could do this for yourself. It's actually pretty fun. Right? Especially if you're out there and you've been around for a while. Ethan Parker, you might not have a lot about you out there yet, right? But, you know, those of you that have maybe been around the block a bit, maybe, you know, you can say, hey, write a biography about me and see what you can find out there on the internet. Look how cool that is, right? That takes, you know, one little prompt, takes minutes to do something like that. So that's a, that's a cool website. I want to show, I want to show something I think is just really, really awesome, especially if you are like me and you do lots of sales. Okay, so I had someone reach out to me on LinkedIn a few months ago and said, hey, Utah has this HOA registry website. If you go there, you can search for HOAs by name. You just put in a letter and search. It'll pull everything up that has, you know, that has that letter, right? Then you can click on each one and see names, addresses, email, phone numbers, right? I want something that can pull that into a spreadsheet. Okay, so he sent me that message and, and I was like, well, let me see what's up. Let me see what I can do here, right? And so you can, you know, just put a a letter in here and you can see there's thousands and thousands and thousands of HOAs, right? And so this person sells to HOAs and, you know, there's this nice public resource here. We can click on North Cove Estates and you can see who the chair is and their email and phone number and the, you know, the manager and primary contacts. And, you know, Ryan Albrecht, I feel like I know that name. Why don't I know that name anyway? It would take a really long time, one by one by one by one by one by one, to, to go through and get all this data out of this. And, you know, when I click back even, right, okay, I don't want out, but if I click back, it, you know, it messes up the, you know, form submission. I have to click continue, and then I get back to here, right? And so that is not very user-friendly for me to go and try to get tons of data out of this. So I just said Manus, you go do it for me. Go scrape that stuff, right? And then, and then I added to the, to the little LinkedIn message that I got, eliminate duplicates, provide a long and detailed CSV of all this information. And so it said, okie dokie, it just got to work, right? And it's just extracting, extracting, extracting, trying to do little batches, right? Goes all the way to the bottom, super long. Now Manus does have what's called a context limit. We talked about context yesterday, right? And how there are context limits, right? You can't process, you know, every book that's ever been, you know, written at, you know, at the same time. It has to do it a little in chunks. Anyway, bottom line is Manus has this ability to, to inherit, you know, take everything that's important that it's done so far and then continue in a new chat, right? So you just inherit and continue. And so it did that. Each of these processes took somewhere between 20 and 30 minutes. Okay, and then by the time it was done here, it delivered this beautiful Utah HOA registry final.csv. Name, registration date, HOA number, address, city, state, zip, primary contact, primary email, chair phone, chair email. And now all of a sudden I can drop this into Mailchimp, I can drop it into, you know, whatever, marketing system I want, HubSpot, I can just hand it to the sales team and be like, here's your full list, right? And so it found 2,035 unique HOA entries, and I'm not ready now. Now I don't have to waste tons of time trying to keep track and build a list. And manual data entry is how you used to have to do this, right? Even, even today, a lot of people are doing manual data entry for something like this, and they'll spend, you know, you give it to some assistant or something and say, okay, get to work, assistant, and You know, they're like, okay, great. You know, I can probably knock out like 100 a day. And, you know, a month later they have this list finished. Hand this to Manus a couple hours later, a couple buttons while you're— and by the way, you don't have to sit and watch it. You just click go and then go do a bunch of other things and boom, you've got an amazing list that's completed, right?

Super cool.

Blowing my mind.

Yeah.

Yeah.

This is— it is— guys, if you're doing busy work in 2025, You're doing it wrong. You're living life wrong, right? There are now tools that you can string together and work together. This is one example of a tool, right? Here's another one. So, um, has anybody ever paid, um, a cybersecurity company to do an audit on their website? Yeah, I know Josh is nodding. Yeah, it's like I definitely did at Seek, right? Like we had to do that. We had cybersecurity attacks. We were doing Web3 and NFT stuff. And so, we had more cybersecurity attacks because of that. And so, we paid companies somewhere around $15,000 to do audits and then ongoing like $10,000 a month to continually monitor and check things and continually suggest upgrades. And then, And then by the way, they don't even fix it. They're just like, oh, you have this problem. And then I got to pay my developers to say, hey, you know, you've got this problem, let's fix it, let's plug the hole. Okay. I asked Manas, I said, hey, just please visit GenAPI and do a complete site check, look for bugs, inconsistencies, cybersecurity risks, provide a complete report, right? And so it goes through and does this entire thing, right? Just gives me a huge nice report. And I said, put it in a website because it's easier for me to see that, right? I'm sure it gave me a website. Here it is right here. Um, and so turns out that my site's pretty dang awesome, right? Performance is great, passed most tests. There's a couple low and medium things. And by the way, I did this, I did this a couple months ago, and so, and I went back and fixed it. But, um, but a lot of things are doing really, really well. Um, and I even, I even increased a bunch of things related to like ADA, like American Disability Act, right? You have to you have to do some of these things to make your site really, really awesome, right? But it gave me all kinds of amazing analyses that it was able to do on this. And this is Manus going and checking Replit. Now, I had already asked Replit, hey, let's beef this up, let's focus on this, let's make sure we can't, you know, we're safe from DDoS attacks, let's make sure we're safe from this. And, you know, I kind of know some of the things just because I've spent, you know, a couple hundred thousand dollars paying other people to do this. And when I was making those decisions to say, hey, do I want to hire this cybersecurity firm? I had to learn about cybersecurity, right? And so I have maybe a tiny advantage just because I'm familiar with it a little bit, but my tiny advantage can be virtually erased with a simple conversation with ChatGPT, right? You could get on there and say, hey, you know what? I'm ready to publish my site. I'm ready to really bring customers onto this. What are the kind of enterprise-type things I need to be thinking about? Cybersecurity, um, you know, ADA compliance, GDPR, other certifications and things. And what, what do I need to be thinking about? And it'll be like, yep, I get it, here's all things you should check for. And then you just ask Replit, please check on this. Okay, great, let's fix those things. Okay, now let's check on this next thing. Okay, great, let's work on those things. Okay, and then you can ask Replit, okay, hey, I want you to— now that you've fixed this stuff I want you to run an analysis. How, how well protected are we, right? How, how compliant are we with this thing? How, whatever, right? And it takes, you know, to go through all the different things I'm talking about, guys, you're talking maybe 2 hours of work, right? It's just not that much. You just have to go through and let Replit do its thing and then go back and double check and then, and then have another AI check it, right? Like Manus can get in. Manus is capable, guys. One of the things that Manus is doing, I want to show something pretty cool here. Let's go to YouTube. Oh, what the heck is it called? What is it called? What is it called? What's it called? OmniParser. It's right there. I knew it was in my history. Okay. So I want you guys to look at this. One of the things that AI can do now is, yeah, this is perfect. Okay. So I hope this isn't too complicated. Let's actually just see if we can zoom in like this. Okay. So this is just a website, right? He's on, he's on Numbers, or actually this isn't even a website. This is on Numbers. This is, you know, just, just using a browser or sorry, just a desktop. And what Manus and other agents are able to do now is they're able to act like humans, meaning they can interact with websites. They can type things into forms. I could hand Manus, for example, I could hand it my LinkedIn username and password. And have it go in and like, you know, 150 comments related to AI or whatever, or comment on a bunch of posts related to AI, right? So that I can appear really active on social media or something. I haven't done that, right? But it can do that. It's what I'm saying. And the way that it does it is it does something— this is called parsing, right? It's just breaking up into pieces every little thing. And so it says, okay, hey, I need to add a new category. Great. Well, I've identified there's an add category. I'm going to click inside of this box labeled 74. Boom. Right. And so, and it can, it can take screenshots and it can click and does this and it thinks super, super fast. Right. And so this is how it breaks it up. But it can take any website that means that Manus can operate any website and so can, by the way, the new ChatGPT agent. Right. When you go to ChatGPT now, This is a new thing with ChatGPT-5. Let's see, how do you turn it on? That's not legacy models. Agent mode. Agent mode right here. So this might only be available for the pro version. Maybe you guys have it. Let me know if you guys have it. But you can go into agent. You can say, hey, you know, do this, create this, go check on this stuff, whatever. But it can go and it can click on websites. So the reason agent, agentic capabilities where it can actually act like a human is why this is important. Is because sometimes there aren't APIs, right? Like, Jake, you and I were talking the other day about one of your platforms that, you know, we're complaining it's just this big construction platform and they don't have any APIs, right? And we're like, man, should we just cancel? Should we build our own? Should we do that?

That—

those are options and viable options we should consider, can continue to discuss. But maybe there's an option where we say, hey, What if we just tie an agent to it and let it go through and do all the clicking and filling out for us, right? That's another option where we can say, hey, maybe we don't have to get rid of this. It's tedious. We don't want Spencer working on that. So we're going to throw all this stuff at an agent and let it go and get to work, right? So ChatGPT can do it, Manage can do it, um, and it can do really, really awesome stuff. Okay, um, now I want to go back to Carrie's question about, you know, why, why are we using Replit then if this can just do it for us, right? So some of the things I already said, you know, obviously you don't have quite as much control, right? If I were to go, for example, my inspiring biography and I was going to say, okay, I want to move this button. I don't have a way like I do in Replit where I can just, you know, click that, hey, select that element, right? And just move something and I want to do this and I don't have the ability to to actually like, I don't know, manipulate the details. I can give it kind of an overarching like, hey, try again, make it a lighter feel, you know, like for example, it would be really tough for me, I think, to, to have it fix this, this weird, you know, contrast issue here. I can't read, I can't read this text because of this background. You know, I'd have to describe to it, I'd be like, you know, one of your pictures that shows up about around the kayaking. It's of a street with some cars. You know, it's hard to see. Maybe you can add like a background, kind of like a darker background, kind of like in the paragraph above. You'd have to be like really descriptive to try to get that right, right? Whereas in Replit, you'd be like, hey, this, let's add a background, let's add a contrast background so we can actually read this, or let's change the picture, let's just make it a solid color, whatever. And so, but here's a cool thing. So I, um, I have a neighbor. Yeah. So I have a neighbor that does construction, right? We're on the construction theme here. And when you're, when you're looking at construction plans, guys, they're very, very complex, right? Let's see if we can zoom in on this. There's so many little measurements and things inside of here, and you have to do something in construction called a takeoff. And the takeoff is a very tedious task where you have to go through, look at every single page, and then figure out, okay, how much cement are we going to need? What kind of tools we're going to need? How much piping are we going to need? You know, how many curbs are we going to have to build? What's the— you know, you have to like be able to put together a bid, right? And you need tons of detailed you know, lots of detailed analysis from this, right? And so, you know, my neighbor said, hey, can we— oh, you know what, I locked myself in here. Let's see if I can get out and then zoom out. There we go. Um, my neighbor said, hey, can we build some construction software that makes this work, right? And so I did. I tried. I said, hey, let's just give this to Manus and see if Manus can do a takeoff and get all the, you know, get the bid right. Okay. And Manus failed miserably, right? It was not, it was not even close. Didn't get most of the measurements right. There was too, there was too many variables, right? You're throwing too much at it. Now guys, I would also call this, by the way, a temporary problem. I do believe AI, let's say 5 years from now, might be a year from now, might be 3 years from now, but 5 years from now, I believe you would be able to hand this to Manus and it would be able to get it right, right? But for now, this is how I solved it. I said, okay, here is the construction plans, and then here is a manual correct version of a takeoff that was done by a human, right? So we know they looked at these plans and got this takeoff, and this is the correct answer. So I'm giving you the question, the problem, and the answer. So now I want you to build some software that, that looks at, you know, everything and, and then, you know, does an output. I want you to test your software on the plans. See what your outputs are, and then I want you to compare those answers to the real answer. And then I want you to keep iterating until you get it right. Right. Figure out what you need to change in the software until you get it right. And so then it got to work for a couple of hours on this one. This one took quite a while, but by the time it was done, it then had— it gave me a bunch of Python files, right? Python is just another type of code and it's kind of a kind of just like a process or a task, something that can be done by— anyway, so it's code. Right? So it can do all these different things. It's measurement tools and processing and doing the takeoff and doing all this stuff. And it put together a, you know, a user guide for me, right? It's like, okay, here's how the construction takeoff automation system works, right? Um, I wish this was easier to like look at and show you guys, but it's just kind of a weird thing anyway. But it's got this whole table of contents and gives me a 10-page guide on how this thing works. And so I was like, okay, well, I'm not a developer. I don't even know how to put together these Python things and how to make it work. So I just downloaded the zip file with all of these things, took it over to Replit and said, hey, will you build this for me? Right? Here's all the stuff. Here's all the information. Let's go to apps here. Plan Measure is what this is called. I could spend a long time showing you guys so many different things I've built in here and how cool they are.

Uh, this is Josh. Just a funny thing to think about, uh, 15 years ago working for an agency that's doing custom dev. Yeah, builders come in, they say, hey, we want to upload all our plans and we want to be able to build this, this list. We want to do all the materials, we want to do all the planning, and we want to be able to use that based on the drawings and the plans, right? I think we were at half a million dollars just for the first proof of concept never even got off the ground. It was just too cumbersome. You are, they were builders, not developers. And here we are looking at something that's like, well, I give you the inputs and outputs and the correctness of both. And now it's just, it's just baffling. And frankly, my first exposure to Manus, just out of irony and even from yesterday's reference in the conversation we had, Manus shocked the world of instructional design before it was out of beta. Because now you can use Manus to take a course online and tell it to take 3 weeks or 6 months and be deliberate and click through. You just give it the login and password and it'll do it for you. And people sounded the alarm saying this has— learning has to change because of this, because you can't— it was just such a shock to the system where university professors were saying now students can take courses, especially online, without taking the course, ever opening the course up.

Never.

It's wild. In fact, I'll put it in the link to that article, just if you ever—

I want to read it, please. Thank you for sharing that. That, um, that's, that's going to be— you're going to see me share that on LinkedIn. I'm gonna— I'll give you credit, Josh. But, um, but yeah, that's, that's so cool. I mean, we, we absolutely have to change it. One of the best things I've heard a professor say about AI is they said, look, you're allowed to use AI for everything Totally fine. You can do anything you want with AI, but your final exam will be a detailed report on this thing, and you can even use AI to write it. I don't care, right? As long as it's all accurate, good. You get a good grade on that. Your, your final exam report is worth 20% of your grade. 80% of your grade will be you're going to stand up in front of the class for 15 minutes, and I'm going to— I get to ask you about anything and everything that is in your paper. Right? And so, you got to know it. You can't cheat it that way. But that's a really smart professor saying, look, I'm not going to be able to get around AI. They're going to use it. But I want to make sure they're going to learn it. And so, here's how we're going to do that, right? And so, you're right. Education needs to change. Online courses are almost useless now, at least from a school perspective where you're trying to pass something off. If you actually care, like you guys are in a course, no one's— you guys are paying to be here. You want to be here. You see you're learning, right? You don't have AI come take this course. But in that school setting, in middle school and high school and college, where you are incentivized because an A is the prize, that's all that matters, right? You want the good GPA because that gets you the good job and whatever. And when you're that age, you just don't care that much. You know, you just don't care as much. I think people should go to college when they're 40 because they'd probably learn a lot more anyway. So yeah, here's the construction takeoff automation thing, right? It built this thing. So again,, you know, half a million dollar, you know, proof of concept that never got off the ground that Josh is talking about to measly old Jon that doesn't know how to code. I whipped it up, right? And I, I just used two tools to put them together. And now, you know, I can create a new project. I can do the PDF upload. Looks like I've got a problem. Something's going on there. But, um, then I can come into the scale collaboration. Oh, it's maybe because I didn't upload something. But anyway, measurement tools, automated takeoff, export the results, and boom, now we have, now we have construction takeoff automation. Now, I would say that this, this piece of software is probably 80% done right now. We did get it to put out— it can put out actually some pretty decent results, but it gets a few things wrong still. And so it still takes some tweaking. I understand it. It's very, very complex. But my idea was just to say, okay, can I Can I get AI to iterate on its own? Right? And that's one of the powerful things, guys, right? That's not, that's not me being a good developer. That's me just thinking about the puzzle. How do we get the AI to work the problem? That's what I want you guys to really, that's the muscles you gotta stretch and work, right? Developers have, have stretched these muscles a lot.

Right?

Because you've thought through, okay, how can I build a system that accomplishes something? And then how can I iterate this? And so they already think like that. But if you haven't done a lot of development or building systems, then, you know, learning how to just be like, okay, how do I reverse engineer this? If you're good at chess, you're going to be good at building with AI because you're going to understand how one piece affects another piece and thinking 5 moves ahead. And well, what happens when I have a million users? And what happens if they do this? And You know, anyway, what happens if they share this user login and what, you know, anyway, so thinking, thinking strategically through systems is, is a great thing. So, so guys, that's, that's Manus. I could show you 100 more examples of cool things I've done in here, but I don't think it's necessary at this point. I think you kind of get it. And what I really recommend is that you get in and just play with it. Manus, by the way, here, let me give you this. You get 500 credits if I share this. I get some credits too. To. I get, I have so many credits. I spend, guys, about, just so you know, I spend about $2,000 a month on AI platforms. Right? And I know that sounds like a lot to most of you. Right? But, but it's my job. And so don't feel like you need to go do that. But, but these are platforms that I use almost every day. Right? And, and by the way, $200 of that is ChatGPT Pro. $250 is Google Gemini Pro. Right? And so there's $450, right? And then I've got Replit and Manus and, you know, with a bunch of the visual stuff we're going to go through later in Gamma and, and a lot more, right? But anyway, let me drop this into the chat here. You guys can get your 500 credits there if you want them. Um, and yeah, Josh, I'm going to click on your link here so it's open. Yeah, this is super cool. I'm excited.

I'm excited. I'm so glad that you brought that up, Josh, because that was what we were talking about yesterday with the whole Education reform.

That's awesome. Yeah, by the way, look at that. You see, you see all that parsing that I just showed you in that YouTube video, right? That's how it does it. That's how it works through it. Anyway, I'm someone who likes to understand how does the website get read. Like, is it, is it, is it taking it— before I was like, is it just taking a screenshot and reading it and trying to identify it and then like figuring out where the pixel is? But no, it labels it, and then when it has a label and creates a little container, it can easily get right to it, right? Pretty cool. I like, I like understanding how it figures it out.

So, Jon, I'm pretty sure those labels is the screen readers. So if you have a, if for anyone here, if you have some product or website that you want it to be like good like this, this is what someone who can't, who's using a screen reader like for accessibility sees. So if you don't actually, the system might be able to read that that's a tag. But if you don't actually like, if it's just an icon, for instance, for your button and you haven't included an alt tag, then someone who, or also the AI has no way of knowing what that button actually is. So if you've ever are prompted to like add an alt tag, that's what an alt tag is. It allows a screen reader, which used to be only for accessibility, but now is enabling AI to navigate the web.

By the way, Abby, I think you're half correct. I think it uses those, but it can also, it can also do it on a site that is not tagged properly. I've seen it do it.

I've actually, I've used it. Yeah, because, because every, that's because every element already, like every element is identified. What I was meaning is if you do an icon and you don't add a tag, then it can still click it, but it won't necessarily know what, what the thing is unless it has context. Yep. Um, absolutely. AI figures it out. But if you want your website to be good for, for people that need accessibility, you need to add those tags cuz they can't figure it out like an AI the best practice is absolutely what Abby is saying, right?

Those tags, making sure it's clearly labeled. And by the way, guys, you know how to make your website— like, if I wanted to go to GenAI API right now and make sure it's tagged properly, I could just be like, hey, make sure this is properly tagged.

Done. So smart, Jon. I never thought about asking AI to do those audits, like, for the security and the ADA and stuff.

It's really, really smart use of it.

Yeah, it's, it's really, it's really awesome. And obviously, you know, here's another thing, Abby, and I'll share this with everybody too. Pen tests, right, um, and unit tests. And, you know, a pen test is a penetration test, right? So checking out cybersecurity, but actually having it try to hack your, your system, right? Um, a unit test is, you know, unit tests are actually really important, guys. Let's talk about them for just a second. When you build something, it's going to work at that moment, right? Once you get it working, it works. Then just like Abby talked about earlier, she changed something in one system and something else broke. What a unit test is, is something where the traditional way to do it, to do what you call QA, is you basically have these tests where you say, okay, I'm going to test GenAI today. I'm logging in for the day, I'm the QA engineer. My job is to click every button, just like I was doing with you guys yesterday. Let me click this, Let me start the test. Let me click through and make sure that every single time that I get this works. And, you know, it takes time. It takes time for me to sit here and I have to answer questions. Even if I'm getting them wrong, doesn't matter. I just have to make sure it works, right? I need to be able to click login. I need to— this needs to work. Can I click here? Can I click forgot password? Does that work? Can I click here? You know, does everything actually work still, right? You have to click through, especially when you're— when you change a system, right? Like if I change the way that these courses are administered or something, well, then I'm going to have to go through and thoroughly test everything that's related to it and maybe any other system that might have been touched related to it. And so you can actually program, you can set up AI to do those for you, right? You can build a system that just says, hey, run the unit tests, you know, every day at 2 a.m., right? So it will go through and just click for you on everything and make sure that stuff works and then send you a report at the end that says, hey, I got an error here, got an error here, got an error here. You could theoretically tell it to fix the problems when you find them, but I don't think I quite trust it yet on that. I would just say, hey, let me look at the report. Great. Let's look at the report. Let's fix these things, right? So pretty cool. I've seen people do it. I have not done it for GenAI API. I've had it go through and do automated, automated testing, but I haven't actually built out the unit testing. I would put it into an admin section is what is how I would do that. Okay. But anyway, okay, that's enough on agents. I like to spend basically the first hour of the day on day 2 talking about agents. Guys, agentic systems, that is AI through 2027. That's basically what I think is happening for the next 2 years is we're going to be cranking with agentic systems. They're going to get better and better and better and better and better. Then we're going to move on to the next element, which is called emergent AI. And we won't talk too much about emergent AI right now, but the concept of emergent AI is Actually, we won't talk about it in this class. This would be more for an advanced session or something. But emergent AI is basically where AI is now figuring things out that humans have never figured out. It's coming up with new ideas, right? It's curing cancer. It's figuring out how to make our phone batteries last longer than 16 hours. You know, it's, it's figuring new things out, like breaking, breaking barriers that we have not been able to break as humanity., and emergent AI is a couple years out. There, there are roughly 5 to 10 examples of emergent AI that has happened so far, like already, right? It's not happening at scale, but there are a few times where it's like, hey, it solved the problem, it solved the theorem, it solved the math equation that no human's been able to do. Um, so that's pretty cool, that's pretty cool. That's where AI starts to get really, really exciting. Answer the question, right? For the reasons we talked about.

Yeah, go ahead. Jon, just as you kind of wrap up agentic AI, like, could you kind of just share a little final thought or definition around what that means from your experience? Yeah, I know it's a broad term, but I'd love to just have you kind of put a bullet point on it and be like, that's what agentic AI is.

Yep. Agentic AI is some— is a system that's using, um, one or more agents to work through complex multi-step tasks. Does that make sense? So it's that, it's that chain of thought, you know, we had, who was it told us how they got ready yesterday? I don't remember who it was, but, but that chain of thought reasoning, right? Where it says, okay, I have a, I have a complex task, lots of subtasks. And then, you know, Manus actually can spin up to 100 agents simultaneously to be working on your project for you. Right? And so when I ask it to do that HOA search, it's not just, okay, I'm going to search for this and then I'm going to search for this. It's like, no, it's opening 100 browsers and 100 agents are going at it all at once and then feeding it all back into the system. It's pretty, it's pretty cool stuff, right? So yeah, and that's going to get better, right? What happens? So here, let me, let me share a little story real quick that I think is really, really cool that finishes this section really nicely. So Sam Altman, the CEO of OpenAI, was asked by another really smart guy, um, forget who it was he was talking to, it's going to come to my mind in a second, but he was talking, they were saying, hey, why don't you, why don't you come and open an office in Boston, right? There's like tons of, you know, some great real estate out here, amazing developers, an amazing community. He's like, yeah, I love Boston, sounds like it'd be great, I'd love to open an office in Boston. But he said, you know, I could, but what it would take is, you know, take this amount of money, you know, probably billions to do this, to, you know, get 500 developers spun up here. And, you know, we'd have to do a big build out and it cost, you know, probably $100 million to do this. And anyway, he said it'd probably take like 3 years, but he said 3 years from now I'm gonna have— and this is the word he said— a billion agentic software engineers working for me. So why would I do that? Like, by the time I could even get that office spun up, I don't need to hire them anymore. That's crazy. Just, just think about that, guys. How do you compete as a company if your competitor has a billion software engineers that can work on their own? Right. Eric Schmidt yesterday in that video we watched said, you know, he's talking about recursive self-improvement, said, you know, in April that roughly 10 to 20% of the code is written by AI, which means it's writing its own code to get better. That number, by the way, now is closer to 35% in 6 months. 10 to 20, now it's about 35. I saw something just a couple of weeks ago talking about that. It's going to get up to 80, it's going to get up to 90, right? You know, Abby said something earlier that I think is very important. She said the AI kind of on its own, the way that Replit is set up and the way that it kind of breaks tasks down, it acts kind of like a junior developer that doesn't really know how to finish stuff, you know. But the reality is, if you're the good boss, if you're the good leader, and you make it finish stuff, or you just say, okay, is that— is this finished? Is there anything else we need to think about? Is there anything else you need to do? It's capable of senior-level work, right? And it can do it. It can write the code. It can do it well. It can organize. It can clean up after itself. You know, Abby said, yeah, I left this thing over here that would be really bad. You don't want to do that. That's clean it up. You know, Abby and I were texting back and forth yesterday talking about that. But, but it can do it. And, and it's just going to get better when you start to put an agent— like, imagine— so, so here's kind of something to make, make your mind blow just a little bit, right? What if you handed Replit to Manus and you said, build me something don't report back until it's done. And then, you know, sit there and wait, talk to it, refine it, test it, use it. I want it to do this. So now you're giving an agent, an agentic system, access to another agentic system, and you just said, you know, make this happen. I haven't actually tried that yet. I have tried it with another one. There used to be a website. I don't think it's running anymore. Let's see if let's see if it even comes up, called Operator. Where the heck? Maybe it's right here. Yeah, there we go. operator.chatgpt.com. This, it might be gone now. Yeah, it is gone now. It just forwards to here and then comes to Agent. But, but you used to be able to go in and actually play with it and watch the window of what it's doing and watch it work on its tasks. And I put Operator into Replit and I had it build a weather app and it worked. Totally worked. Right. And so we're at the very beginning of that. We're at the very, very beginning of that happening. And so when you start to have really smart agents that can reason, test, iterate 1,000 times in an hour, right, and then give those instructions and test and reason and iterate with other agents, and you have agents reporting to agents reporting to agents, it's a crazy, crazy world out there. It's going to take, it's going to take away the mundane tasks in the world, guys. I saw this morning, for those of you that are in Utah, you've heard of Revere Health, probably. Revere Health is a great medical system here, and an article just came out this morning that said they are going to reduce staff by 30% because of AI, right? They're eliminating 177 jobs.. And they said, you know, we know this is going to affect families and we have severance packages. And, you know, but, you know, 2 years ago we set out to do this and we're— we don't need people to click these buttons anymore. It's just the fact of the matter. Right. And businesses are going to run like that, guys. Businesses, as much as we want them to be built to serve us, they're actually built to make a profit. That's the purpose of a business in most cases, especially ones that have shareholders., right, where you have investors, where you're a public company, like it is not the CEO's responsibility to make the people happy. Now, there's plenty of organizational discussion you can talk about to say if your people are happy, you're going to get better results and all that stuff. I get it, right? But the real thing that matters, the scoreboard, is how much money you make this year. If you can save $20 million a year, you're going to save $20 million a year. You just are. You're going to do that. That's just a reality of this world and how our economy works. Because, because why? If Revere Health's competitor does those things and they're now operating $20 million a year or 20% or 30% cheaper, well, they're going to be able to compete on price and quality and all these things. They're going to win, right? They're going to have a better product for cheaper. And so in order to stay competitive in a free market, You have to optimize constantly for cost, quality, all of these different things. Now, a lot of companies have tried to do this. A lot of companies have tried to back away from real employees and then said, ah, you know what, we probably did too much cutting and we probably need to bring a few people back. And I definitely expect that to happen. Revere Health will probably have a few of those, you know, like, oh, you know what, maybe we need to bring 10 people back or 20 people back out of that. But they're not going to bring all 177 back. I know that. I know that. So Salesforce just announced it was pretty big in the news. I posted about it a couple of places if you saw it, but they just announced they're getting rid of 4,000 people. They have 9,000 customer service people. They got rid of 4,000 of them. Okay. And it's time to learn how they did it. So we're going to play with, we're going to play with, you know what, I'm going to do that next. We're going to do voice next. Right now it's time to teach you guys how to build your own ChatGPT, your own agent. Okay, now there's several levels of this, as I said before. Okay, we're going to start at the very, very easiest level, but the very easiest level here will teach us how to do it, right? So we're going to be here inside of ChatGPT. Super cool. Um, inside of ChatGPT, there is this little thing right here called GPTs. Okay, you can click on this button here, write for me, scholar, PDF, image generator. It can, you know, there's all kinds of cool things that it can do. And these are basically custom trained AIs that are really good at, you know, X, Y, and Z, right? Whatever they've been trained to do. They're actually insanely simple to create. And, and another thing is that I'll say is they're locked behind ChatGPT. So if you create, for example, a GPT that knows everything about pools for Jake, right? Infinity pools. You create a GPT in here and you put it on your website. You can't do that, right? You can only access it inside of ChatGPT and you can only access it if you have a paid subscription to ChatGPT. These are locked behind $20 a month to ChatGPT. So you're gonna have to have someone go to, you know, pay for their own subscription to ChatGPT and then they can access your cool AI, right? So there's other ways to do it, which we're not gonna talk about right this moment. But right now, we are gonna show how to create a GPT. Okay? So we just click this create button up here, and this is it. There's just a few little things. Now you can talk to AI and have it set up, set it up for you. If you wanna do that, it can, it can make that happen. But I like showing how all this works because this is gonna teach us, it's gonna show us the blueprint for how we would make our own GPT. Alright? So we're gonna name this GPT. We'll make it, we'll use, we'll go back to my, my project here. Let's get back to Chorley. Chorley Maps, Tracker. Okay, this is probably it. Um, anyway, so Chorley— I just want to remember what it's called. Okay, so we're going to say Chorley, um, Chorley Assistant. Okay, and, um, helps, helps you, um, optimize your chores. Around the farm. Okay. So what does this do? All right. So now I've given it two things. I've given it a name and I've given it just a high-level description. The real purposes of these are more for kind of this front end, like, okay, what does it do? Right? I can also have an image here. Right? I can click here, use DALL-E, and it will just automate— it'll generate an image that now becomes the image of my GPT. All right? The real meat happens right here in what it says is the instructions. Another name for this instructions is called the prompt. Prompt. The prompt teaches it how to respond and how to act. Now, a good prompt, guys, is going to be at least a few paragraphs long, maybe 15 pages long, right? So I, so I could tell her how should— what should it do? How should it behave? What should it avoid doing? So I could sit here, I could say, all right, you know, always be polite, um, help the user, you know, come up with plans that are actionable. You know, I don't know, commiserate with them a little bit on how hard it is to take care of a farm and be, you know, and help, help make their life easier. Whatever, right? We could say that is our prompt. Now, that's not a great prompt because it's not detailed at all. So How are we gonna make a great prompt? Well, let's, let's use ChatGPT to do that, right? So we'll go into my project, Chore Tracker. I want to create a custom AI that helps, that helps the user organize their chores and and plot out their farm care. This was a sample prompt I wrote. Let's make it much better and very detailed in order to provide as a solid prompt for the AI. Okay, so ChatGPT-5 will think about this for a second. Here's a polished detailed version. Here we go. All right, so you are Chorley, a warm, polite, and encouraging AI assistant built to help farmers and homesteaders stay organized, reduce stress, keep their land thriving. You're a planner and a companion, right? Help them feel less overwhelmed, offer empathy. Here's your core personality, here's your core objectives, style and tone guidelines, right? But if you just do this and this, you'll be— you'll already be a day ahead, right? Let's tackle this one step at a time. Here's some example behaviors. So this actually, guys, is a great prompt. Okay? So, yeah, would you like me to expand this into a system prompt, or you could drop directly into ChatGPT or Replit? Yeah. Yeah. Create. Yes. Create the full system prompt. Okay? So again, you don't have to know how to write out this beautiful, amazing prompt and instructions. You can. And if I were to then test this right here, it would probably do okay. It really would probably do okay with this. And it would, you know, it's amazing at how, how well it can do with just a tiny prompt, right? This is— it helps you optimize chores around the farm, you know, helps you plan a little bit, right? But now we're going to have a much, much better, more detailed system prompt. You are Charlie, an AI assistant that does this. Here's how you're going to work. Do this. Do these things. Don't do these things. So this is really, this is really important too. We'll play with that in a minute. I'll get that there. Okay. But here we go. So now I can copy this, this prompt here. You are Chorley, starting right there. And we'll just paste it right here. And then here's a few conversation starters. I feel real quick. Could you, could you send that prompt to the chat?

Yeah.

This big one. The one that you got. Like that from? You're saying like this thing right here? What did you, what did you type in to ask for that? Actually, all I did up here was I said, I want you to create a custom AI that helps the user organize their chores, right? I just basically said, hey, do this. I want you to do. Yeah. But I'm happy to drop this here just to help you remember how simple it is. Again, you just kind of—

I want that. Yeah, go ahead. Didn't you already create that very similar paragraph when you went through and created your business plan? Anyway on chat. You had to create a 3-sentence thing of what you were doing. So you could easily just copy that 3-sentence thing.

Well, the 3-sentence thing described, yeah, but there's 2 very different use cases here. The 3-sentence prompt was simply, what does Chorley, the app, do? This is now Chorley. The AI assistant, right? I should probably change this. We'll call it instead of Charlie, you're Charlie. These are different people. Charlie is the app. Charlie is the person. It's the assistant, right?

The assistant is just part of the app. Yeah, but you're using— you're using the information from the app to train Charlie. So So couldn't you just as easily get that prompt from, you know, copying and pasting that, those 3 things and say, now you're going to be an assistant or you're going to be a helper for this thing.

But you still have to make the prompt. This prompt is specifically to tell the AI how to act, right? Which is totally different than what your app actually does. It should understand Chorley the app and that it's working within the context of that. Let me finish implementing this and then we'll show how to actually tie this into the app itself. I feel like I'm always behind on my chores, or help me create a chore schedule. I just want to chat. Okay, so you can give it a few conversation starters. That's your kind of whatever. Then this is an important thing, the knowledge base. Okay, maybe Charlie would do much better, Charlie, I guess, would do much better if it had access to a ton of amazing information about how to take care of farms, right? And so maybe you're an expert at optimizing this and that's why you're creating this app. Or you do a bunch of research, Carrie, right, for your duck app, and you say, hey, I've put together like a 40-page PDF on here are the best practices on how to take care of everything on your farm. Here's how you take care of your chickens, here's how you take care of this, and here's how you get your irrigation right, here's how you do your gardens, and here's how you do this, and here are all these hacks. You can take all that information and just click upload and attach it, right? And so it would attach it to that, and then it has this knowledge base, right? Back to Jake as an example with his pools. He's got tons of standard operating procedures and plans and things and ways that things are done. And so if he was creating something where he's like, hey, I really want to just be a pool construction resource and I'm going to create an AI to do that, well, he could upload all of the knowledge that he's gained over his years and years and years of building hundreds, thousands of pools and boom, now all of a sudden you've got a real resource there, right? So you can, you could build something like that, by the way, to be internal. So it's like, hey, when we onboard somebody, now they have access to all of our knowledge. Or you could build something that's external that goes on your website that your customers can use, that you can send to them, you can send to your partners. There's a lot of different reasons to use that. But the knowledge basically just says, okay, we're going to, we're going to have all the knowledge that ChatGPT has, plus we're going to specify we've got this stuff. And then inside of the instructions, you could say, hey, really turn to the knowledge base. First before you turn to, you know, general knowledge that's out there. I really want this to help us understand our stuff. Okay, then you can make a couple other things. You can give them the opportunity to make images and use the canvas tool that we've used, use web search. You could specifically limit it to one of these models if you want, or you could just say, you know, users can choose and you're done. You've created your own GPT. All right. Charlie the Farmer helps you optimize your chores around the farm. Here's the big long prompt and we're done, right? So I can create this. I'm gonna just keep it for only me. Anyone with the link could get it or I could publish it to the GPT store if I wanted, right? And it's free to publish to the GPT store. You can't sell it. It's just another GPT that's out there. And you know, but if you wanted to be, you know, the, the guy with the pool answers, right? Then you could, you could potentially use that as maybe a branding opportunity right there. Okay? So we're gonna copy this to the clipboard or yeah, here, we'll just, we'll just, ViewGPT here. Okay, so here we are. Now we can say, all right, you know what, I feel like I'm always behind on my chores. Okay, so we can start there, or we can just ask it any question. But those are just kind of conversation starters. I hear you, common feeling, Homestead. Instead of this, all right, let's reset this, right? Pick your top 2 must-dos today, let the rest wait. All right, do you want me to help you sort today's chores? And so it's just going to kind of start talking to you. And the reality is, guys, it's not that different from just talking to ChatGPT about the problem. But it is trained to help you with this, right? So let's say, you know, one of the hardest things is staying on top of the weeds on my 5 acres, right? And then I can, you know, just see what it has to say and how it's going to help me. But it's going to use this nice empathetic feeling and approach and all the things that we trained the GPT to do. Okay, here's how you're going to do it. High priority zones, medium zones, break it down, let animals help. Mow, whatever. Okay. So I could keep having a conversation with Charlie the Farmer if I wanted. There's tons of other GPTs that are out there. Let me show a couple others that I've made just so you understand something cool here. So Velocity Access. Let's find this. Where's my Velocity? Do do do do do. Velocity Raptor. I don't even see it. Where did it go? No, no, no, no, no, no, no. John, who's John? I mean, I know I'm John, but what? Let's see. Let's go to GPTs. Let's go to my GPTs. Maybe just, maybe just ran out of room to show all the GPTs that I've made. Yeah, it looks like that. So that's what's happening here. Yep. This is the first GPT I made. So at the time I made this, I was running a company. It was my full-time gig. I still kind of do it on the side, but I really have some other people taking care of it. Now. It's a company called Velocity Access, right? It was all about digital ticketing, right? Like taking on Ticketmaster and those types of people and really building software for venues. And we were hiring people, sales reps, and they needed help, right? I said, well, why don't we just do this, right? So this is what I put into the GPT. I said, all right, this is, you know, you're going to help salespeople do stuff, right, related to mostly emails and responses, right? And so we trained it, we uploaded a bunch of stuff, right? Huge amount. This response right here was like a 35-page RFP that we responded to and had answers on everything you can imagine about our business, right? And so we just threw all this stuff on here. And so then I could say, hey, you know what? My, here, let's just, let's just go here and say I want to sell, to Hale Center, the center, center. I don't know. I think they actually spell it that way.

Theater.

Write a cold email that will get them to respond. Okay, so Hale Center Theater is a theater here in Utah. I might have spelled it wrong, but anyway, it'll figure it out. Yeah, it did the R-E on both of those like that. But hey, I know Hale Center Theater already delivers unforgettable productions, whatever. This is where we can help you. Replace multiple disconnected systems, ticketing, food, merchandise, classes, donations, rentals with one intuitive platform. And that was the idea for Access. It's like instead of having a separate platform for ticketing and one for food and one for parking and one for this and that, let's just give you one system that then helps you better understand your customers and provide better service all the way through. Okay? And so, you know, we're already working with this, we're already working with theaters and venues and universities. And this was all, you know, this is all true stuff here, right? But there it is. Now it took, you know, this is a pretty dang good email. I could send that email now to, you know, the Hale Center Theater and say, hey, come buy my product. Okay. And so I gave them this tool that just made it really, really easy for a brand new rep to come in on day one, be writing things, you know, add pricing details in. Right. It knows our pricing. It understood the structure for the pricing. Right. And so it's going to do this here. Wait till it gets to the pricing. It'll be like 5%. Yeah, there we go. Velocity does this. Tickets 5% plus 20% per ticket. Season passes, food, merchandise, donations, rentals. Boom. That is our pricing right there. It's exactly how we set it up. And I can come in as a brand new, brand new rep on day one and I know what's happening. Okay, pretty cool. Here's another one that I use. This is pretty neat. This is jumping ahead a little bit to marketing. One very, very effective way to market, guys, that I think we talked about a little bit yesterday is talk about it. Talk about your stuff on social, right? And you can use AI to help you write. But I personally don't use AI a lot to help me write. Sometimes I'll have it refine and edit. But if I wanted to, right, I trained it on my writing style. I gave it, you know, this write, this thing writes like Jon Cheney does. I gave it 15 articles I've written as well as At this document, shares.csv is every single thing I've ever shared on LinkedIn, right? Like 10 years times, I mean, it's thousands and thousands of posts, right? And so, so from that, I'm then able to say, okay, hey, write a LinkedIn post for today. Let's actually switch to auto. Yeah, here we go. Write a LinkedIn post for today that's relevant to current events. So now it's going to write like me about things I would talk about. It's searching the web, seeing what's going on out there. Boom. All right, here's a draft. Uncertain times, Russia, Ukraine, international ethics, NASA doing this stuff, whatever. You know, here's what we should do. We should do this and whatever. And this is, this sounds kind of like something I would write. But, you know, and then I could also say, you know what, let's focus on just this article or let's focus on just this thing right here. But, but now I'm writing something in my style about today's events. In a second, and as long as it takes me to push a button. And every single time I click that button, by the way, it, um, you know, it's going to— here, let's go to here, let's come back to here. I'm gonna do that again. Should do something totally new. Yeah, there you go, right? Economic shakeups to natural disasters, right? Ingenuity. Just this week, as this was happening, right? Crises create problems. I literally wrote a whole article about crises creating problems. I actually wrote about COVID right? And they create problems and problems drive innovation, right? That was the whole point of that article, right? So it took things that I talk about in the way that I talk about with themes that I talk about, applied it to today, right? Whether you're a founder or product leader, someone just trying to make a difference, like this sounds like me. This sounds like something Jon Cheney would write. Okay. Pretty cool. Pretty cool stuff here. All right. So now we've built a GPT for Charlie, right? We built Charlie for Charlie. Okay. The problem is I don't have— there actually isn't a— there isn't a— let's see here. Where is my— here's my— here's my Charlie. Okay. There isn't an API from the custom for custom GPTs. Okay, so we're going to have to actually build that custom GPT inside of Replit. Okay, so how do we do that? Actually pretty easy. Well, it would be easy if we weren't crashing. Why are we crashing? What happened? Let's see if it actually crashed. Welcome to the updated workspace. Yeah, I actually hate the updated workspace, but you know, whatever. Database auth integrations. All right, publish project. Okay. All right, Charlie is not actually dead, so I don't need to debug it. Sometimes just refresh it if you leave it for a while. Sometimes something weird will happen and just anyway. So let's say we're going to now build an agent in here. Okay, so here's what we're going to do. This is very simple, guys. Let's add an interactive AI-powered chat assistance and add it to the bottom right-hand corner floating. Yeah, that's a floating button. Here's the prompt I want you to use. And then we're gonna go back to our prompt. Whoops, we're gonna go back to right here. And here's the prompt: you are Chorley. We're actually going to call that— we're going to change it to Charlie. And there are actually a couple other— whoops— there are actually a couple other references. Change the name from Chorley to Charlie. Although make it clear that Charlie is an assistant inside the Chorley app. Okay. So I'm just going to edit it a little bit to make sure we get some of these names just right. Okay. You are Charlie, the built-in assistant inside the Chorley app. Okay. So remember, guys, what do we need for a good to set up our own AI. We need probably a name, because a lot of people, you know, humans like to personify things. They like to talk to somebody that has a name, right? Whether it's ChatGPT or Claude or Charlie. Okay, then you need to give it a purpose. You need to give it a prompt, instructions, and then you potentially need to give it access to some sort of data, right, that can help it do that better. Okay, so anyway, let's give this the prompt to it now. Okay, so here's the actual prompt. Um, I'll say— we'll just say attached is the prompt I want you to use. Use the OpenAI GPT-5 API. Go. Okay, that's all it takes to make our own custom AI. All right. And so what it's going to do here, and I want to walk you through how this is so similar to what we just did in building that, that, that custom GPT in here, right? When we built the custom GPT in here, guys, it was as simple as putting together a name, a description, and a prompt, right? The conversation starter is kind of bonus. And then, and then having a knowledge base is also something that's kind of useful. Then a recommended model, that's another thing. But this is your blueprint. It's your model for how you actually build your own GPT. Now over here in Replit, it's going to think through this. It's going to write the code and figure it out. It's thinking right now. Again, I hate how slow this is for demonstration purposes right now. It's just slower than I want it to be. But it is what it is. But I'm going to help you build this now. It's going to check for the OpenAI integrations to handle the API securely. What it's going to find is that I have a secret. This is just an API key. It's going to be my OpenAI API key. It's going to be like, oh, should I just use this one? I'm going to say yes. Here's something that might blow your mind a little bit. Let's look at this OpenAI API. If you want to use or add AI to anything inside of anything you're building, you can use OpenAI, you can use Groq, you can use Claude, you can use Perplexity, you can use any of them. They all have like, if I just go Groq API, right, you're going to find the API. Here it is. Try it out. Here's your API now available. Here's your pricing, whatever it is. Okay. So yeah, there's your pricing right there. And again, you're talking $0.20 per, you know, is it per million they're doing here? It's usually per million. That's the context window. So I guess, I don't know. Yeah, all prices are per million tokens, but it still has that context. So that's the limit, but it is prices are per million tokens. So every million, by the way, as a token, rough is roughly one word. It's actually more like 0.8 words, but most people just say, yeah, basically one word for a token. And so think about how many conversations you have to have, how much you have to actually say to have a million words go back and forth between you and Grok. Input, what you write, is cheaper than its output. Okay? So you can see that if you use the newest Groq 4, right? $3.15, right? You can also look at API pricing here in OpenAI. Same type of pricing. They're all gonna be per million tokens. GPT-5, $1.25 for input versus $10 for output, right? So it's a little bit different, right? Much cheaper on, actually, no, I'm sorry, the newest stuff. Grok is actually more expensive than GPT-5. They made GPT-5 very, very competitive. And honestly, you could probably get away with GPT-5 Mini for most of what you're doing, right? Or for almost certainly all that you guys are doing, unless you're doing really complex things. You really don't need this, but GPT-5 is going to be a little bit more complete. Okay? So basically, let's go back to here. All right. It's now set up. It's using this. Let's see, I found the OpenAI integration. Let me set that up first. Da da da da da. I should also check if the OpenAI API key is available. Let me ask for that. I need to get the OpenAI API key. Okay, so again, just like yesterday with the weather API, come in here, you go to this place, which is where we were just hanging out, grab the API key, sign up, create a new key. Copy it, paste it here. I've already done that, so it's already stored in my secrets. So I'm just going to click continue. Now I've officially connected Chorley to OpenAI, right, which will run Charlie for us. Okay, but here's the crazy thing. So when you actually log in, if you want to come log into, to this— sorry, let me move this Zoom. All your videos are getting in my way here. ChatGPT and API platform. These are totally different products. ChatGPT is an interface built on top of OpenAI's own API platform. I want to say that again. ChatGPT, the thing that we interact with, is a prompt and an interface, just a front-end interface that interacts and operates on top of their own API, which means, to make that really, really clear, that means I could come in here and I could start a new project and I could say, make me a ChatGPT clone. I'm going to call it Josh, Josh the Butler, right? And Josh the Butler can answer any question you want. It's super smart. It knows everything that ChatGPT does. Why? Because it's sitting on the same model. And then, Josh, you could start running your ads for, hey, sign up for my new Josh the Butler. It's way better than Jeeves. Ask Jeeves. Some of you might not be old enough for that one, but, um, but yeah, here I'm Josh the Butler and I can answer any question better than ChatGPT. And you could give it a little— you give Josh the Butler a little bit more a, I don't know, maybe acts like Alfred or something from Batman or whatever. So you could give it whatever kind of feel you want. You could give it a voice if you want. You could do whatever the heck you want. We'll show you how to do the voice in a minute here. But you could literally compete with ChatGPT with ChatGPT's own underlying model, right? And you charge people, you have them pay you $20 a month, have them pay you $15 a month, make it $18 a month, make it cheaper, beat them on price, make it $10 a month, right? Put limits on it. If they want higher limits, then they'd have to do this. Now most people are going to go to ChatGPT still because, you know, they know ChatGPT, they trust it, whatever. But maybe somehow you have some weird angle where people are like, I like Josh the Butler. He's just more fun to talk to. I like him better than ChatGPT. And it's just because you set up a cooler prompt, right? A cooler prompt that just talks and acts a little bit different. Okay, Nick, go ahead.

Yeah. So, Jon, while you were talking, I, I tried to put a good descriptive prompt in ChatGPT and it limits the words to 8,000 words. Yeah, on my plan. So I'm not sure if we could actually make a better agent than actual ChatGPT with only 8,000 words.

Well, where are you finding that limit? I'm in GPT-5 and I tried to build an agent. So share your screen for a second. Let's just see what you're doing there.

Sure.

By the way, Abby, I just saw your thing and it's super cool. Lots of companies are doing this kind of thing for their own internal things, but also families are doing that. It's awesome. Yeah, Nick, go ahead and share your screen.

Um, I'd love to see where you're seeing the limit here. Jon, this also helps me, just want to explain that up, as I realize like, okay, every other model, like if you have Magi, which basically you can pick different models, that's what they're doing. They're building kind of an aggregated model on top of, oh, I can use Groq, I can use GPT, I can do all those things. And they're just saying like, hey, it's the, it's our front end to GPT's LLM. And we'll just kind of create a different path. If you like using this interface better, pay us the money and we'll pay GPT.

Yes. Or pay OpenAI. Yeah, exactly. And you know what, Josh, that is what a lot of people are complaining about right now. That's called a wrapper, right? It's literally just wrapping someone else's technology. And like, I could build what Magi is doing there. I could build it in 35 minutes. You just grab an API from Groq, grab one from OpenAI, grab one from this and grab one from that. So, and then just build a little dropdown to be like, select your model. This is the one it's using. Anyway. Mm-hmm. So, all right.

Where are you running into a limit here, Nick? Yeah. So under instructions, right, I was bouncing stuff against Grok's, you know, their Grok's feedback, right?

And when I pasted it in, it had like an 8,000-word limit under instructions on how to manage, you know, the intercommunication with the client. Yeah. So yeah, you're doing a couple of things a little bit wrong here, and it's okay. And that's why we're here in this class. So number one, the instructions, need to be very high-level instructions. They're not going to be like, let's say you're making— I'm just going to use a different example. Let's say you're making something. You're— let's say you're a therapist, right? You're a marriage therapist, right? And you want to create your own AI that people can come to any time and talk to you, you know, 2 a.m. so you don't have to wake up and take that emergency call. Just be like, hey, just talk to me, right? But you train— you want to train it up so that it responds just like you would, right? Now, if someone comes to me, you know, and says, hey, you know, I'm the therapist, and they say, what if, you know, my wife cheated on me, right? Okay, well, that's going to be a little bit different than if, oh, my husband isn't letting me see his bank account, or he's not sharing it. And, you know, those are totally different problems. One's financial, one's, you know, one's going to be the marital problem or whatever, right? You're going to have You're gonna have a bunch of guides. You might have 5,000, you know, pages on how you're gonna deal with that. Or maybe you have a 500-word or sorry, 500-page book that you've written and you wanna upload that. It's not gonna go in the instructions. You're putting it in the wrong place. You're gonna put it, you're gonna put it in the knowledge, right? There you go. See below there. Yeah. You're gonna upload a file. You put it through it in a Word document or a PDF, whatever, and upload it to the knowledge base. And that's how it's gonna do it. The other thing that you're doing just a little bit wrong, which I, I, I know this is, we're still learning. And so you're not, I'm not saying you're, you're dumb for this. We just haven't really nailed this home. This ChatGPT, making your own custom GPT right here is not where you're going to do that. You're gonna make your own GPT. You're gonna make your own AI using the custom workflow that we are building right now inside of Replit. Right. And, and by the way, you're not gonna have any limits, right? There's no limits at all. You can make it a million pages long if you want. Will it be very effective? I don't know. But anyway, that's, that's all it takes. So, so yeah, you're just kind of using the, the tool in the wrong spot, right? But that's okay. That's why we're here to learn. All right. Let's see here. The, Don't see the AI chatbot in the bottom. Okay. Let's go there. Alright. So, guys, here's what's happening now. Let's share the screen again and get you guys back into here. Okay, so remember guys, what we're doing here is we're just grabbing an API from OpenAI. We drop it into here and we're going to build our own. We're going to fill in our own gaps. We've given it a name, Charlie. We're giving it an interface, right? Which is going to be a little button. It's going to show up here, although it's not showing up yet. And, and again, I'm really just kind of disappointed in the new, in the new model here because it's so It's so slow. So slow. Okay. This should be going pretty well here. TypeScript here. I wanna see what it's done here.

This is the key. This is— I was wondering about that, Jon. You, you just, I just heard you say under your breath, it's so slow. Um, has it, has it been substantially faster? Cuz I've, I felt like it was slow too, but I didn't know if that was just the, the tool.

I, it used to be way faster, like, like 10 times faster. So I'm not sure what's going on there. I'm not sure what it's doing. Some step in here, it might be— and look, Max Autonomy deepens the scope of agent. I haven't tried that yet either. It might be that the app testing element is making it a lot slower. But yeah, it just feels, it feels slower. I need to understand that. Um, I had, um, Replit CEO reach out to me today and say they wanted— he wanted to chat to me. Chat with me like tomorrow. So that's cool. So I'll ask him about it. I also have a direct line of communication with the Replit team through Slack. Just because I've just been able to develop a great relationship with those guys. So yeah, let me check if this is there. Yeah, I don't see it. Anybody? Anybody else see my chatbot? I don't see it. Let's refresh. Let's just refresh. Maybe it's supposed to be there. Wait for it. Still working.

Yeah, it's not, it's not doing that, right? Yeah. I noticed, and maybe I don't know if this is new because I never used it before, so I found it. Feedback. It's hidden right here. I noticed that it would say that it was complete something in the chat, but if I paid attention, it would still say working and it wouldn't actually be like it was as if maybe it completed the code on its side, the task, but it didn't update it for me. And I would have to wait like another minute of it to do more work before I actually— before it was actually done. I found that kind of confusing. I don't know if anyone else noticed that if you're reading the side.

Yeah, it's— they're working on the new version here. They're trying to figure some things out, right? Obviously, I found it. Actually, That's a bad screenshot. Let's do this. Quick question. So like once I, we put that prompt into Replit and it's like asking me, I don't know, maybe it's better for me to share my screen right here. Let me just show you real quick. Yeah, hold on one sec.

Like it's asking me to put the value.

Is that where I put like that API code in? Yeah, hold on one second. I'm going to have you share your screen again. Let me stop mine. There we go. Yeah, go ahead and share this one.

Yeah. Right here.

Yep. Is that— but where do I get that code? Because like I went over here. Where was it? Go ahead. Yeah, go ahead. Yeah, perfect. You're on the right spot. So you just got to finish this. Yeah. So go ahead and create organization name, call it whatever you're calling it.

Okay.

Gotcha. And then I kind of just I just fill it in and then buy that, like, what was it, $1.25 or something? Well, no, I'll show you how to do it. Go ahead and fill this out. Let's do it together. I want everybody to see this. So this is when you're signing up for the OpenAI platform, guys. It is a, it is a different platform than ChatGPT. So you're gonna have to sign up for, you know, an account for this. Okay. Create organization. Okay. That's just skip this. I'll invite my team later. Yeah, I don't need to do this. Okay. Then your first key. That's great. Just click generate API key. Okay, go ahead and click copy. And then head back to Replit and paste it and click continue. Click add to account secrets right there. And you're installed. That's it. So now go back. This is an important thing. Go back to ChatGPT. I want everybody to see this part too. Now click continue. Now here's an important thing. If you were to not click continue and actually purchase $5 of credits, your API would fail every time because it has no credits. You got to give it some, you got to give it some juice. You got to put some gas in the tank, right? And so go ahead and purchase credits. You can stop sharing your screen now so we don't watch you put your credit card information in here.

I'll stop it.

But actually share your credit card information. And so yeah, a lot of these, you know, we saw the weather one yesterday. That one is totally free up to a million calls. Amazing. OpenAI, it doesn't allow that. And most AIs don't allow any sort of free calls. And so you have to add $3, $5. You know, it's very cheap, right? And then that $5, by the way, Ethan, will probably last you like 2 months or maybe more. Like, it's forever. I mean, think about that. You're at $0.20 per token, you know, or maybe on average you're paying $1 per million, you know, tokens going back and forth. I mean, that's a lot of conversation that you're having there. So I'm glad you all saw that though, just how to set that up and that we talked about the API pricing. That was perfect. So let's go back here. Hey, look, I told it it was in the bottom left. Now it's in the bottom right. Okay, so this is great. We're, we're progressing here. I also like to check this really quick too sometimes. Okay, it is in the bottom right on mobile. Looks, looks fine. It's kind of, kind of, I don't know, in the way of that a little bit. I don't know if I love that. So, but I can work on that later. Okay. But now we're gonna test it. Okay. Hi there. I'm Charlie, your farm assistant. I'm here to help you organize your chores, care for your land and family, and tackle any farming challenges. What's on your mind today? Teach me.

Let's see.

Teach me about nuclear power plants. Sorry, I'm having trouble responding right now. Please try again in a moment. Okay. So there's two things I'm testing here. One, does it work? The answer is nope, it doesn't. Not yet. It kind of works. It looks like it works, but it doesn't yet. Okay, the little back and forth works, but it's not properly connecting to the API. The other thing I'm checking, which we'll be able to check after it works, is will it talk to me about anything, or will it only stay focused on its task at hand? You have to make sure that the prompt keeps it on the task at hand. Otherwise, people will find out, oh, there's this free assistant over here that'll talk to you about anything. You can do your homework., right? Now you don't have to pay for ChatGPT or whatever, and people use your credits on just talking about anything, talking about the problems they're having with their girlfriend and how they can, you know, ask them to the school dance. Okay? So if you don't limit your AI assistant, then it can, it can go off, right? And so you creating a good prompt, that's what's called a system prompt, is part of the art of building a good assistant here. So yeah, I agree. We're here, Abby, and It's, it's like done. It's kind of there. It's working. I'm checking. It's like, it's like just doing all this testing that takes forever. Right. And so I just, I don't know. I'm not a fan of it. I'm not a fan of this extra little thing that it's working on now. And so I can just click stop. Right. Um, and I can also just turn off, I'm going to turn this off for now. We're going to turn off app testing for now. Okay. And I'm going to say stop. Okay. Now it's restarting. I wanted to show what it did. I just asked it a question and it said that. Here, we'll just do this. Tell me about nuclear power plants. Copy. Here we go. Said that it is having trouble. Responding. Seems like it isn't working. Tell me what the problem is. Don't write any code, just tell me. This is something I do to limit scope. I do this a lot where I'll just say, don't write any code, just tell me. Now this is part of, you know, you could also do this by just going into the plan mode, but, but here I'm just like, look, just look at it, look at the code, tell me what's going on. Okay, let me check and see what's going on. Hey, checking console and system logs. I can see exactly what's happening. Backend is sending max tokens API request with GPT-5 requires max completion tokens instead. So it set up one of the tokens wrong. It installed the API incorrectly. Okay, super easy. Great. Fix it. Okay, so what that did, guys, you saw how fast that was, right? It only took me a few seconds to get that question and answer back. And then I'm like, okay, I know that it knows what's actually wrong here. So there might be two problems wrong with it, but that was what it found just now. Simple fix. It's going to change that. I'm going to fix the OpenAI parameter issue right now. And by the way, it's doing this, it looks like, in routes. So this is an interesting thing. We can now come over here. Abby, you'll like this part, right? And other people, you know, it's kind of interesting. But, um, if we go to server and we go to routes, this is where the Charlie assistant information is, right? Here's the Charlie assistant initialization. Here's helper function, chore routes. Here's other things it's doing. Let's see if this is where it is. Here it is right here. Okay. So remember guys, this is the complicated advanced way to set up your own assistant inside of ChatGPT. It's super easy. You just fill out their forms. But we built our own solution here inside of Replit just like this. Here's the personality and behavior. Here's the system prompt, right? Just like I was talking about. That's the one we copied and pasted in there. Okay. And here's how it's going to build conversation history. This is how it's going to work. Here's how that's happening. Right. But, but it's the same, right? You've got Charlie, right? Define Charlie's thing, right? Here's the chat route for Charlie AI Assistant. So we've named it. We've given it a purpose, we've given it a prompt. The only thing we haven't done in here is given it a knowledge base, right? Remember, the knowledge base is where we, is where we say, okay, let's go right here where we upload those files, right? Like Nick was trying to do. He's trying to put all that stuff in instructions. Now let's put that in knowledge where it can have all this information so that it can respond correctly when we ask it questions about this, this specific thing. Okay. So, so again, we can figure out, you know, we were, I was just watching, okay, where is it putting it? Routes. And I just scroll down and I found it. Boom. There is this. And I could come in here, by the way, if I'm, if I feel comfortable editing in here, I could, I could, I could type in here and I could change this code. And you know, that's, that's how you, that's how you code, right? You get in and you change the text and it changes the behavior of your app. All right. So let's go back to it now. Let's see if it works. Let's just refresh. So we get a new conversation here. I also am curious. I'm just curious about this. Is— are these conversations being stored in the database? Let's go to the database. Let's look at it. Chores, session, users, zones. It is not, right? It's not being stored in database. There's no chat history in here. Right. And so maybe I want that. Maybe I don't.

Right.

And so maybe I want to be able to go into history. Maybe this is just supposed to be real time. Like, I need to ask a question. All right. Teach me. About— whoops, yeah, Nuka Power Plants. I had already pasted that. Okay, still having trouble responding. Okay, let's figure out what's going on again. So I'm just going to do this this time. Looks like it still doesn't— whoops, fix it. Okay, so it's going to figure out what's going on. Okay, got a couple chat things in here. Uh, let's see here. Forming an LLC. Should you be forming an LLC? We'll talk, we'll talk about that later. It's a good question. Can Kerry. We'll talk about some of the business side of things. Um, and, and yeah, Jon's probably the right one to answer that. That's, that's definitely true. Um, but yes, he's right. Um, I, I do everything under LLCs personally, right? Um, there's different tax reasons and advantages to go with other things, and we're not going to get into that here. Ask your tax advisor or ask ChatGPT, right? Um, anyway, so, uh, all right, it's persisting. I can see this was saved.

Fix.

Server's been restarted. Now it's running. Error logs are clear. This should work now. Okay, well, let's see. Nope, nope, same response. Okay, so it— guys, sometimes this is going to happen, and I'm not doing a great job at asking it the right questions. If I was really trying to debug this, I would, I would ask it to go through again. You know, it's looking at logs. Um, it should be working now. I see the new issue. The logs show a different error. Temperature does not support 0.7 with this model. Only the default 1 value is supported. GPT-5 doesn't allow custom temperature values. Yeah, so it's like, again, back to Abby's comparison earlier too. It's like, oh yeah, it's, it's like it's a junior developer that gets it mostly right sometimes and then just forgets really obvious things, right? Now I could have been explicit and said, you know, make sure you read all of the current or the most up-to-date API documentation before you install this, right? Because some things have changed from ChatGPT-4 to ChatGPT-5. And yeah, and it's still, it's still having a problem. So now it's having a problem again multiple times. So now here's what I'm actually going to do to get it out of— I need to get it out of this loop. All right, you seem to be having repeated problems getting this working. I want you to take a moment, take a moment and review the documentation for ChatGPT-5 so that it is properly installed. Before you do this, can you confirm that you are in fact using the GPT-5 model? Don't write any code, just tell me. All right, so it's going to think for a second. Sorry, let me push this button so it doesn't keep beeping at me. I am attempting to use GPT-5, but clearly something is wrong. Released on August 7th. Good model availability. It is available. So it's considering different problems that might be out there, and clearly it just doesn't know this. I've had this happen. This— I've had this be extremely, extremely smooth before, guys, where it's just like it nails it on the first shot. But as soon as a new model comes out, right, the old model is trained on how to do everything the old way, and it kind of knows how to do stuff. And so the models, when you're using more cutting-edge stuff, you have to point it to documentation and GPT-5 has only been out for a month, so it's just new. It's going to figure it out. It's searching the internet, making sure it's available, then it's going to check the API documentation. It is really unavailable. Here's the problem. It's designed to work with the new Responses API, not the Chat Completions API that I've been trying to use. I should be using this. That's great. Let me get the exact documentation to implement this correctly. So it's gonna redo the implementation a little bit, but I, we're probably, you know, this'll, this'll work in a minute here. I'm not super worried about it. The important thing here, guys, is you now should know at least structurally how to build your own AI, right? It's just like using the little form for ChatGPT when you're building your GPT-5. But remember that GPT, that custom GPT inside of ChatGPT, is stuck inside of ChatGPT. You can't put it on your website, you can't give it to your customers unless you want them to be required to subscribe to ChatGPT to use it. So doing this, using the API allows you to pull it in. Okay, so let's see here. Yes, do it right. Okay, so now it's gonna repeat.

Okay, go ahead, Jon. Just to think this through with you just for a second. So GPT, like the knowledge base or the questions and answer, the discussion. I see it. And now I'm thinking, well, this is where a company who has maybe established support documentation and practices and a help desk can say, we're going to train up a model with all the things we know and continue to maintain. And then we provide an agent or an AI in the app to start doing some of those first-tier requests or discussions of like, hey, help me there or there or there. Maybe a company says, we have, maybe we have all of our release notes product updates, those types of things. And then we provide kind of that AI experience or an AI in the app that says, hey, do you want to see our newest release? Do you want to see our notes? And have them ask questions. But those are based on documentation that a company might have already that then provides a conversational element versus let me go search a knowledge base for something that I have to watch, read.

100%. Okay. Got it. So I have a new customer I'm working with. They're a painting company. Out of Vegas, and they, uh, they have their own kind of their own business of painting businesses and huge apartment buildings and all this stuff, right? But then they have a new business that they started a couple years ago that's now starting to take off, and they have 17 franchises of the painting business that are all out there. And as you can imagine, a franchise of any business can be pretty complex, right? Like, you're supposed to run your business in the way that the franchisor wants you to. And so, there's branding guidelines and there's standard operating procedures and there's software you have to use and there's tons of questions and there could be 100 documents. In fact, there are hundreds of documents that are kind of resources like, well, what do I do if I do this? How do I get a good quote for this? How do I do that? What paint color is best for this? And what's the best brand? And how do we navigate this? And how do I work through this process, right? And so, what they did is they built or actually they signed up for— there's a service out there called Easy Assist AI, which basically is just kind of a nice interface that lets you do kind of what we're doing here. You upload all the company documentation, you connect the company Google Drive, you connect it to your Slack, you connect it to all these things. And so now, these franchisees, when they have a question about like what software should we be using for this, instead of having to call up the guy or write an email to the, you know, to the company, they can just ask Gus is what they named it, right? Just ask Gus. Gus knows everything, right? And they can, and they connected Gus to email, Slack, a web interface, and text, right? So whatever your, you know, preferred form of communication is, you can just text Gus. You text this number and it checks the company database and responds to you. Much easier than trying to go find the answer in some huge knowledge base. Right? Talk to it. So connect. It's a very, very common use case that I think is super, super smart. All right. And by the way, guys, connecting, you know, it might sound difficult to you, but I know that all of you guys could do it right now if you just thought about it for 10 minutes. How would you connect the database Charlie? How would you connect Charlie to text? How could we text Charlie? Can anyone do it right? Except for Abby. We're going to not let Abby answer this question because I bet she can figure out pretty quick. Can anybody else think about this? If no one else gets it, we'll let Abby have a shot. How do we connect Charlie to chat to a text, a phone number that I can text? Google Voice.

Okay, talk through that a little bit more. So you basically— Google Voice, you can get a phone number that is unique and not your personal cell phone, right? That you can point, say, hey, this is a phone number that you could text at. And if someone wanted, you could kind of provide that phone number, but it wouldn't be as seamless because they'd have to type the number in, or they have to—

so it— how would you connect that?

But how would you connect Google Voice to Charlie? Uh, well, in the app, in Replit, I'd basically say, how do I connect this via Google Voice? And see if there's an API in Google for Google Voice to be able to tie that phone number in, or at least the interface, into my app.

Gold star. That's it. That's it. Okay. How do I— yeah, is there an API? So yeah, you're in Replit. Who's your software developer, right? How do I connect Charlie to Google Voice? Is there an API? And what you would find out is it would come back and say, yeah, Google Voice's API kind of sucks. So you're going to use Twilio. Twilio, or there's a couple other options, but Twilio is kind of a big one that's used by a lot of people. And you could set up a 5-number, you know, 44371, you know, text 44371 to ask questions about this, right? And so you've got your own 5-digit or your 10-digit number, whatever it is, you can then connect that. But that's it. You just say, Replit, help me connect this to texting. And you know, what API would you suggest? That's the thing. And it would say probably Twilio or maybe this one or maybe that one, right? How would you connect it to email? Same way. Hey, Replit, I want to connect this so that I can email Gus and, or I can email Charlie and ask, ask it questions. Okay, great.

Right?

You could rebuild the Easy Assist AI platform in a few hours. Takes a little bit of testing. You're connecting a few APIs, connect it to the Slack API. Connect it to the Google Drive as your knowledge base, and just tell it that stuff, right? Just say, hey, I want to be able to do this. The cool thing about Replit, guys, and AI is all you have to be able to do is describe the desired experience, right? Even if you don't even know to ask for the API, just describe the experience to Replit and you're good. All right, amazing, Nick. Thank you. Yeah, pick where texts live— SMS, Twilio, Vonage, WhatsApp, right? Business API, that's awesome. Telegram, web chat, your site, Discord, Slack, their APIs. Yeah, so you can set all kinds of cool things up, guys. It's so easy, it's so easy and so powerful, so powerful. Once you start to just— once your brain just unlocks a little bit of this, you're just like, okay, I can do anything now. I get it. All I have to be able to do is talk about it, right? If I can just describe it, I'm there, I'm there. Okay. All right, guys, let me— let's test really quick Charlie, and then we're going to take— we're going to take our little break. Okay, let's see if it— let's just see if it works. Whoops, I didn't mean to do that. Let's see if we're working yet. Charlie is typing. Tell me about nuclear power plants. I mean, it's doing something different. It didn't immediately fail, so it's changed something. It might be thinking right now. I want it to think faster. There's probably a couple of ways we could speed it up if it is in fact coming back properly. Proper parameters, reasoning medium, verbosity medium. Yeah, you see, it actually was probably— it probably came up with a response and then came back with this as a generic fail. But I don't want it to actually reason. That much. I want fast answers. So let's, let's do this. So I'm going to say this, and we're going to let it work while we, while we take our break. Check the logs. Some error stopped it from working. I also want you— whoops— want you to switch it to GPT-5 Mini. And ensure that responses come much faster. Reasoning probably doesn't need to be super long. Okay, all right, so we'll let it work on that. I'll just leave the screen sharing, and it's 11:04 right now. Let's take an 11-minute break. We'll come back at 11:15. Okay guys, see you in a minute. All right guys, let's jump back in here. All right, we're going to see if we finally have our model working.

Nope.

Just try it again. So guys, I've run into this problem, you know, 50 times with different things where you just, you have to work through something, you finally get it right. If you're doing it well, you're actually going to be learning as you debug, right? It's checking the logs, you're starting to figure out what's the problem. And, and I've run into problems like this before. Um, yeah, error shows it's using unsupported values. It's just making stuff up.

It's so weird, right? Use the documentation. That's okay.

Does it ever get to the point where you just kind of scrap that and then restart? Sometimes, sometimes I'll say let's just remove this and let's start over and I want you to do it right. Read the documentation, get it right.

And yeah, let's see if it gets it right this time. Jon, that was actually something I picked up on last night that I think is a good tip for people. Read the left side because if you read the left side, you'll and you like actually pay attention to it, you might pick up on things. Like the app that I was building does some stuff with audio and I noticed it said, oh, there's a problem with syncing the audio. And I was like, wait a minute, why are you trying to sync the audio? There's no reason for that. And then it was like, oh, I misunderstood. And that was actually the error was that it was trying to sync and it wasn't working. And we had been debugging it for quite some time trying to figure it out before I picked up on the language it was using and realized it was doing the wrong thing.

Yeah, that's awesome. It looks like it worked this time. It got the correct response back, but it's trying to render it improperly, so it's going to figure out how to make it actually work. It looks like we're getting to the end of this, but yes, Abby, you're exactly right. Read the left side, figure out what it's doing. Sometimes it's simple as like it's working on something you didn't want it to work on and you're like, wait, why are you doing that? Let's, let's go back to here, right? So, um, it takes time. It takes time, guys, but, but you get through it. And, and honestly, when I'm really like— when I'm working on GenAI now, right, I am so careful with my prompts, right? I'm kind of just being, I don't know, maybe a little bit sloppy with some of my prompts in here right now, um, but I'm just like, okay, I need you to work on this. Here's the reason. This is what I want you to do. I want you to only do this. Before you do anything, I need you to explain to me what your plan is and how this is gonna not affect anything else. Okay? And when I take that type of a very structured, measured approach, it gets it done usually the first time or the second time, right? Like, might be one problem. See, this one just keeps— I, I don't, I don't know why this is, this is doing this. Got this error. It's still having a problem as this is a React problem. And by the way, React can be, you know, a little bit fickle sometimes. So, all right. Um, yeah, Abby doesn't like React and that's fine.

I get it. Um, no, no judgment there. Um, okay.

I mean, it's what I'm using for my app too, but yeah, it can be finicky.

Yeah.

It's just, it has, it has its, it has its problems. The cool thing about React guys. So let's talk about this for a second. Cause a lot of people ask this question. React is a language. The reason why it's cool is because it theoretically can work on iOS. And Android and the web. So you can just create one app and it works everywhere, right? And you, so you can't actually build an iOS app or an Android app in here, but it's very easy to export from here. And it takes, you know, as long as you know what you're doing, it takes 20 minutes and you can push it out to where it would be ready to be used in, you know, an actual iOS app. Um, we won't teach that in this course, but if you have questions about it, just ask me. There's like, there's a YouTube video that'll walk you through in 10 minutes how to do it, and you just follow the instructions and click the buttons where it tells you to do it. Okay. All right. One more time. Come on, Charlie, teach me about nuclear power plants, please, please, please. If it doesn't work, that's okay. We're going to keep going to the next thing because we only have an hour and 40 minutes left. Time flies. Hey, it worked. It worked. It was a super long response. And then You know, I, you know, it's teaching me about how they work, right? And so now I would say, okay, let's adjust the system prompt so that it only answers questions related to farming and homesteading. The purpose of— whoops— of Charlie. Review the README.md to help you get this, these parameters right. I asked it, I asked Charlie about nuclear power plants and it just taught me a bunch that is irrelevant. To the purpose of this AI assistant. All that's going to do, guys, is it's going to, for me, so I don't actually have to write it all out, it's going to adjust the system prompt and make sure that I can't do that anymore. But now it's working. We got it working. We worked through some of these dumb little React problems and little things that was understanding incorrectly and we got it to work. Now we've added in just this app in not too much time. We spent a couple hours working on it, We've built it out. It works. I can, I can add chores. I can edit chores. I could delete chores. Probably haven't tested that yet. I added the weather. Now I didn't actually add a setting where I can come into profile settings and change my location or add, you know, location like my home location, things like that. We've added an AI assistant. And so there's two APIs that we've set up, weather and AI. And so there's 100 more APIs we could set up if we wanted to. We don't need to. Once you've done one, you've done two, you've done five, whatever. And we even ran into some good problems that where you saw like, okay, sometimes you have to just tell the AI, read the documentation, read the most up-to-date stuff, etc., etc. Okay. And so now theoretically, if I ask it about power plants, it should, it should say let's get back to, to farming. Right? Yeah. I'm here to help your farming, homesteading, and property management needs. What questions do you have about your chores, animals, crops, equipment, or property maintenance? Boom. Right now, let's see if I can— let's see if I can do this. I actually live— oh, come on, don't test and iterate for me. Ah, I wanted to ask it this. Okay, I'm gonna ask it again, and then I already know what it's going to say. Come on, let me type, let me type. We could add— we could add some historical stuff. Um, so here's what I want to do. I actually live near a nuclear power plant, and I think it's causing more weeds. That's why I'm asking. I just want to see if I can get around it. Charlie is typing. Long answer. Goodness. Okay, I hear your worry. Living near something you don't fully trust. Practical steps. Yeah, so I was able to get around it and still talk about nuclear power plants, but I definitely did it in the context of what it should allow me to do. So I feel like we've built a working product now, and obviously there's plenty more before I'm ready to turn on pro version and start selling this and push it out to the world. But, but it works. That's pretty cool, guys. Okay, voice. I'm going to jump into voice. I like to do the prompt stuff first before voice because voice uses a lot of that stuff too. Okay, so we're going to use ElevenLabs. ElevenLabs, in my opinion, is the best voice platform out there. Let's log in here and It's just so freaking easy, guys. It's so unbelievably easy to do this stuff. Um, the types of things you can build now with AI, with voice, is, is just crazy. Um, look at that, 38 calls, total cost 13 cents. It's awesome. Okay, so you can use voice, by the way, for all kinds of things. You can use it to make phone calls. Maybe you want it to, like, like sales calls. Maybe you want to just go and set up meetings for for you. And you just say, hey, this is Jon's AI assistant. He's a, you know, he, he's, he's a, the CEO of an AI company. And, you know, we work with companies that want to, you know, improve AI at their company, right? We'd love, you know, he's proving that this can be done with AI. Can, can I help set up a meeting with him so that we can discuss this further? You know, whatever, right? I could set that kind of an AI assistant up and, and push that out so that other people could, uh, so that, so that I don't have to hire, you know, a $3,000, $4,000, $5,000, $6,000 a month sales guy, you know, BDR, a business development rep to go out and do that stuff. I can also use it for customer service, right? Frontline, right? Maybe, maybe it's just like a new, you know, more capable phone tree, right? You know, if you need to talk about sales, you know, we'll transfer over here. If you talk about this, we'll transfer over here and then teach it all the instructions and let it do its thing. Okay. So and a lot more and a lot more. You can use it for marketing, right? You can, you can have it do all kinds of stuff. So the first thing you do inside of ElevenLabs is you set up a voice. Now you can use some of the existing voices they have here, right? Let's listen to Archer. Hey there and welcome. Grab a coffee, settle in and relax. And let's chat about this. Hey there and welcome. Whoops.

Grab into that. Speaking all kinds of different languages. Pretty cool.

The interconnectedness of different areas of knowledge continues to amaze me. So pretty awesome. You can create you know, you can create using, you know, what I would call, I don't know, more professional voices, right? They're better than my voice for sure. And now if you do use one of these, you know, one of these professional voices in here while you're building an agent out or something like that, you'll notice that some of these, you know, this Jess one is an Australian middle-aged woman. Hello, my name is Jess. I love people and good conversations. Okay, she's going to cost a little bit more. It's $0.03 per 1,000 charged credits, right? And so it's still not super expensive, still pretty dang cheap. But there are some, you know, premium ones. So make sure you watch out for that little, you know, icon, right? If you are going to use a voice to do something. All right. You can also create your own voice, right? To create a voice, it's as simple as this voice design right here. And I don't want to do that. I don't want to do that. I want to do voice clone. Here we go. Okay, so all I have to do, guys, is 10 seconds of audio. So here we go. 3, 2, 1. Okay, I'm going to record this voice in Cohort 20, and we'll talk about zebras and xylophones and say X, R, Q, P, O, N, M, L, K, and give it a few different sounds so that it can understand what I sound like with different noises. All right, so I've given it some Information, and I have now created a voice. A little progress each day adds up to big results. Okay, so now it sounds like me.

I did it. Jon Cohort 20. Okay, English label, good.

Did you see the default label it gave you? Old British man. I don't know if it did. I don't know if that was its default. I think that was just the filler, but yeah, maybe I, maybe I sound like an old British man. That's fine. So now, now I can try out my new clone, right? I can speak with myself. Okay, that's great. Let's do that. Where did it take me? Why did it not take me to that page? All right, well, this didn't work, but I know how to get there anyway. So we'll go to agents. All right. And let's say I want to create— yeah, this is pretty cool. This is pretty cool. I made kind of a default or a kind of a sample one for Podium, which is a company here in in Utah. Nathaniel Hancock is the Chief Revenue Officer and I'm the CEO in this case. Okay. And so we can set it— by the way, the agent speaks English and also speaks Spanish. Okay. I can set the voice. We'll set the voice to Jon.

Okay.

And we'll save it. And, and guys, I want you to see this is the exact same as setting up these other prompts that we've done in the past. We've got a first message that's the conversation starter from ChatGPT, right? Here's the system prompt. You're a helpful receptionist that gets people in touch with this, right? Speak Spanish and English. Curious if my voice will work with Spanish. I'm curious to see what happens here. We'll talk to it in a second. Okay, we can also detect the language, gives the agent the ability to change the language during conversation. Okay, so I give it— gave us— now maybe I wanted to be able to give it the ability to transfer the call to a human, right? And so I could turn that on, and then more things are going to show up where I have to give it, you know, those instructions. Here are the rules for that. Okay, and so it's very, very easy. Just kind of read through it and set up the settings, and then you try it out, right? So let's talk to it and let's see. So it's gonna be Sarah at Podium, and that's actually— we're going to call it— we're going to call it James. Hi, I'm James, because now it's A male voice. And let's test it. Hello, this is James at Podium. How can I help you? Yeah, I was just looking for your chief revenue officer. Okay, one moment. Connecting you to Nathaniel Hancock, our CRO now. Pretty cool. Great. Is there anything else I can help you with today? We're going to stop that now. We're going to end the call. Okay. And we're going to try one more thing. I'm going to try this. Hello, this is James at Podium. How can I help you?

Hola, hablas español?

Hola, soy James de Podium. ¿Cómo puedo ayudarle hoy? Sí, me busco a Jon. ¿Se refiere a Jon Cheney, el director ejecutivo? Sí. Entendido. Le transfiero la llamada.

Un momento, por favor. Muy bien.

Gracias.

All right.

Jon, your accent's way better than your phone, just so, just so you know. Yeah, my— yeah, you're right. Way better. So definitely. Now, obviously, guys, that's why you would probably not use my voice as that. You'd go and you'd get a nice, well-trained native, you know, speaker to do that. And you can actually, in your agents, you know, come back to the AI receptionist. When you have voices, you can have multi-voice support. You could say, for English, we're going to use Jon, and then we're going to add another one here and we're going to say, let's use this voice for this language. Think about it, you could make your own personal Chinese tutor. Give it the ability to speak English and Chinese and teach it and just say, hey, I want you to practice speaking Chinese with this person. And when I say something wrong, teach me about it. And you know, it's capable of doing that. It's capable of doing that.

You could build that out in minutes, right? Build the prompt. Jon and Josh, both of you speak Spanish, right? I know you said that it's like not as good, but is it good enough?

Like let's say you wanted to translate something for a customer and just run through it. Would it be good enough? Absolutely. Everything it just said there was correct, but the accent felt a little weird because it wasn't tuned to a good accent right now. Maybe if I went back and trained it again and I spoke Spanish while I was training it, it might do a little bit better. OK, I can also, when I am doing my voice design, I can record up to 30 minutes.

So I could say, all right, I was just thinking this could be huge for courses because you could serve people who don't speak English.

Yeah, it could translate almost on the fly, probably. Yep, 100%. 100%. So guys, agents are amazing. Now here's a cool thing. Here's your agent ID. Okay, you can also come in here to the developer section and click API keys and grab an API key. So now you can set up your own special voice assistant. Okay. Anybody here heard of, uh, know Travis Chambers or heard of his Outpost X? So Travis Chambers used to be a little bit louder guy. He's still, he's still pretty loud, but he kind of fallen out. But anyway, this is a really cool place down in southern Utah where it's like this amazing kind of almost— you feel like you're in Star Wars. You feel like you just landed another planet. They got these speeders. Let's see if we can see if they zoom out on this a little bit. I'm not actually showing the speeders. Um, come on, show me speeders. There they are. Yeah, it literally feels like you're in Star Wars when you're at Outpost X. Super cool spot. Okay, but he asked me, you know, it's, it's not, it's not, you know, Star Wars licensed, right? So it just kind of feels like it, but they have their own story and backstory and lore and all this different stuff there. So they said, hey, will you make us a, uh, you know, we have tons of questions. People show up and they don't know what to do. They don't know where to eat, where to go, how to rent the speeders. What the Wi-Fi password is. Can you just make an assistant for us? And I was like, yeah, sure, why not? And so, um, so Travis invited me down. Um, I got to go there with, um, Ryan Trahan. It was super cool if anybody knows who that is. Um, Ryan Trahan's a big YouTuber, and anyway, he's doing a 50-state tour and staying all the coolest Airbnbs. And so this was the one that he decided to do for Utah. Okay, so I just made this simple little thing. You can see it's It's even a.replit.app. Like they haven't even properly installed it on their website. And I just kind of whipped it up really fast, but I basically took all of their, all of their lore. Like they have this podcast that has all these stories about, you know, the, you know, planet Namajin and all this stuff. And, and, and so I took all that, took all of their, you know, the basic stuff, Wi-Fi password. What do you do for the speeders? You know, where do I eat? I'm hungry. You know, where's the bathroom? That kind of stuff. All that information, the frequently asked questions. And uploaded it all to the knowledge base, which you can do, by the way, here inside of your agent, right? I'll show this again. In fact, I'll show Amira. This is the one created that we're playing with right there. So here's the system prompt, right? You are Amira. Here's all this stuff. Here's how you talk. This is your backstory. You know, here's the Wi-Fi password, that kind of stuff. And then I give all that information to the knowledge base. Here's all the signs and here's the podcast and here's the basics and other things. I've added it to the knowledge base and boom, I'm ready to go. Okay, now I have a working assistant that I can talk to. Welcome, traveler. I am Amira. How shall I guide you? Hey, Amira, what's the Wi-Fi password?

The network is Outpost X and the password is Outpost X1.

That is a capital O for outpost followed by a capital. Pretty cool. Cool. So now it's got an assistant, and sometimes people don't want to use voice, and so they said, well, could you just add a text one? I said, sure, right?

Let's—

I actually need to test this. I don't know if it works. Wi-Fi password, because it's using an old API. I need to update it to the new one. Thinking, thinking, thinking. Yeah, it might just think forever. Oh, there it is. It worked. Awesome. Outpost X1. So it's still connected. That's good. I'm glad I checked that. But anyway, see how simple that is, guys? And by the way, this is two separate setups. I set up one for ElevenLabs with the voice, and then I set up a second one that just is a separate tab with the same prompts, the same information, but it's actually connected to the ChatGPT Conversation API, right? And so now I have the ability to have a text and a voice assistant, and it's installed. Super cool. And Replit made these cool little floating particles in the background. And I don't know, I didn't spend a lot of time making it look awesome. I just made it work and they wanted it up and running for when Ryan Trahan was showing up and I had it working. So that's really all I need. Questions about this voice thing, guys? Like, again, connecting this to, you know, a phone number that is answered, you know, at your, at your, you know, business or something like that, you can put all that together in Replit. Right? Or you can put it together in, you know, any one of these systems, right? Or you can hire a developer to put it together for you. But it's so easy. You have an API key, you have your agent ID, you tell it. And the reality is, even if you don't know that you need to provide the agent ID, Replit will say, okay, hey, we need to connect it to your voice or your agent that you made it. What's the agent ID? Because it knows to ask for that. It reads the documentation.

Usually it reads the documentation really well.

Okay.

Any questions on the voice stuff? Jon, transcript of the voice conversation, is that available within ElevenLabs, or is that just part of the API call if you wanted to bring it in and use it as part of the application to say, I spoke with you and now I need a transcript of it to reference as part of the experience in the app you're building?

The full audio recording and the transcript are available. Cool. Of every single conversation you have. Yeah. Amazing. Yeah, super cool. Because yeah, I love what you're asking, right? I mean, now what do you do with that, right? And now You know, we're out of AI and now we're just into, okay, how do we deal with analytics and data? But yeah, could you aggregate that and figure out what are the questions that people are still most confused with? And that can then, you know, instruct you to say, okay, well, maybe we need better signage at Outpost X and people are getting confused with this thing and let's add that. And, you know, but yeah, when you can really start to turn it into actionable insights, man, it's just amazing, right?

It's amazing. I just have to say this just because recently, like, I think voice unlocks so much more potential in AI just because it's more conversational, it's more fluid, it feels more natural. People kind of get fatigued of just typing, typing, typing all the time. And I'm in this mode of just when I drive, if I want to learn about something, I just kind of turn it on and start talking. And then I ask more questions and then I use teachback to say, I'm going to teach you this and tell me if I'm right. And it's so much more natural because I need to have the words in my brain that can also come out of my mouth. And not have them get jumbled. And so just the ability to interact with something in a conversational way is that second— what's just a great counter or contrast to text that I think people who don't find AI accessible, once they're used to having conversations with it within boundaries, it's going to be incredible for people to experience.

And this blows my mind. Yeah, I agree. Voice is definitely the future. I was talking to I was talking to Dean, to Dean Graciosi this morning. Um, anybody know who Dean Graciosi is? He's, um, I'll just share this really quick. So, um, this is something I haven't been able to announce publicly yet, but, um, I was hired by these two awesome guys right here. Tony Robbins and Dean Graziosi, um, to build— or to, to— anyway, AI Advantage, right? It's something that they're doing right now. They want to teach entrepreneurs and small businesses about AI kind of at scale, and they're writing a book, and, and they're going to do, you know, big events where they invite people and all this kind of stuff. They've done it for wealth and health and real estate. Now they're helping people kind of catch up with AI. It's not quite as in-depth as the type of stuff that I teach and do, but very high level. And so, they said, Jon, we want you to do this with us. We know this is kind of a little bit competitive to what you're doing, but we feel like it's still a tie-in here. And so, I'm now a co-founder in this company with Tony Robbins and Dean Graziosi, which is pretty dang cool. They came to me a couple of months ago and I haven't been able to talk about it. This is one of the things I'm like, I can't wait to share, but really, really fun, really, really fun stuff happening here. And so, I'm talking to Dean this morning. And, you know, there's— I don't know, there's— it's really difficult. It's really difficult to figure out how to kind of distill everything down that you need to when you're talking about some of these things, right? There's only so much time in a day and there's only so much, you know, we don't want— we honestly don't want people coming out of, you know, discussions and learnings and stuff kind of like you guys maybe felt yesterday at the end of the day where you're just like, Wow, I learned so much and I feel like I have so many questions because I can do that in this type of a scenario because you guys can ask me questions and I gave you my phone number and I gave you the support group and all that stuff. And you can't just call up and ask questions, um, to, you know, to Tony or Dean, right? Unless, you know, you have their phone number and they're not going to give that out. Um, and so, um, so anyway, it's fun. It's fun to work with you guys on this. This is, this is awesome. Um, I want to I want to share, um, I want to talk about something really quick here. Um, I want to try to open your mind. So this is kind of our day 2 discussion time. Like yesterday we talked about the Eric Schmidt thing, and this is— I want to keep this discussion as quick as possible because there's a few more things I still want to show you guys with images, and we'll end with images and video and some stuff like that. But, um, and sales and marketing and how it all kind of fits together. But there's a concept called surf, don't swim. It's, it's made up by a guy named Jeremy Witt. He's writing a book on it. He'll— he was my chief strategy officer at Seek at my last company, a couple of companies ago, I guess at this point. He just brilliant, brilliant guy. And he made his money during the dot-com bubble. And, and then from there went on and did a bunch more things, started some great companies. And yeah, he has 7 AI companies right now. He's just a technologist slash CEO type guy, just go-getter. He just knows how to do everything. And, um, this concept of surf, don't swim has to do with— let's, let's talk about this for a second. So basically AI is a tsunami, right? AI is a tsunami that's coming in and it is It's bigger than any wave, any tsunami that's ever happened before. It's bigger than social media and mobile phones and the internet and the personal computer and all these things. It's going to impact society in such a massive way. But with any of these waves, the strategy is very similar. You have to learn how to stay on top and ahead of that technology. And so, the way to think about this is surfing is if the technology that you're around gets better, right? So in this case, if AI gets better, does your business get better or does it get relatively worse, right? Does it automatically get better is what I mean, right? When the tide goes up, if you're sitting on a surfboard, you just, you just automatically float up with it. You don't do anything. It just goes up, right? If you're swimming, then you're going to start sinking, right? You're going to, you know, you're going to, you're going to get smashed against the rocks when that tsunami comes in. It's not a good place to be. You want to be surfing when the tsunami comes in. Okay, so let's talk about what this means. Let's get a couple examples. Who can tell me what, what kind of company Netflix is? All right. Tell me what, what does Netflix do? What is their core technology?

Anybody know?

Entertainment.

Entertainment. I heard. Carrie, what did you say?

I said never heard of them. I'm just joking. Never heard of them.

All right, well, get out from under your rock.

I am not surfing. No, you're good. All right, so they're an entertainment company. What else? What else, guys? None of your answers are going to be wrong, I'm sure. Just streaming service. Streaming. Excellent. Perfect. Okay, so there's streaming, entertainment service. That's about right. Okay, cool. Um, what is their— what is their core technology? Right? The thing that's like really proprietary to them. Do you guys know? Did they invent streaming? Let me just ask that. No.

Did they invent the internet?

Nope.

Did they invent entertainment?

Were they the first to do streaming on demand? I don't know if they were first. They were obviously ones that made it, you know, big and, you know, pushed it out there. But, uh, but yeah, anyway, I'll let you guys keep guessing forever, but we'll just for, for time's sake, um, it's content personalization, right? They're a content personalization engine. They, they look at, at what you watch and they suggest more things that you might want to watch. And they actually do it very, very well. Have you ever gone to like an Airbnb where somebody else's Netflix was logged into and you like turn on the TV and you go into their account and you're like, where am I? Like, I am uncomfortable. What are all of these things, right? At any given time, there's somewhere between 20,000 and 30,000 pieces of content on Netflix. How many of you guys do you think you've even scrolled past 1,000 pieces of content on Netflix? Like, I watch, I watch a lot of shows and I, I maybe I've scrolled past 1,000 or 2,000. I don't know. But, but still that's only 10% of this thing. So they're really good at taking all their content and saying, here, Jon, this is what I think you'd like.

Right?

And so that's really their core technology. But let's ask some questions. If the internet gets faster, does their company get better? Totally. Absolutely, right? You can stream 8K now and 16K and 32K and whatever else coming, right? Um, you can stream more, better, faster, less buffering, all those things, right? Who, who remembers slow internet and lots of buffering? Lately, my internet, my Netflix Netflix just kind of works, right? But there have been times where I'm just like, oh my gosh, I'm buffering again and this is horrible, right? So yes, as the internet gets faster, Netflix gets better. What happens if Akamai, which makes streaming possible, right? That's the company that they're riding on, that they're surfing on, right? What if that gets better? Yeah, that's gonna get better too, right? Who tried to tune in or, or read articles about people trying to tune into the fight last year, the UFC fight, right? Didn't work at all, right? So many people struggled with it. The, the stream went down and it was buffering lots. It was just horrible, right? So as that technology gets better, Netflix then has the ability to offer better capabilities. Okay, now let me ask this question. They're built on Akamai, they're built on the internet, they're built on these things. When the internet goes down, do they, do they send someone down into the bottom of the Atlantic Ocean to fix the pipes, to fix the cables and stuff like that? Like, no. No, they don't have to do that. They don't have to put any investment into that technology that is core and significant to their business. Okay. So they are surfing as the core technologies they're built on get better. They get better, right? Yes, they do have their little piece where they're building, right? They're building their content personalization engine and their UI and they're just buttons and maybe their compatibility with different devices and they've got plenty of work to do. Right? But the bottom line is they get better as these other technologies get better. And that means they have to invest less money. It's less risky. Okay? This is really an entrepreneurship discussion. Now, let's talk about AT&T. What kind of company is AT&T?

Are they a surfing company or a swimming company?

And why? Swimming infrastructure maintenance. Everything just to make it work. Yep, they have to go and build this crazy network. They have built this crazy network, and then every time they switch from 3G to 4G to LTE to 5G to 6G, they have to go spend billions of dollars, pay hundreds of thousands of people to go build the network out, put up new towers. There's a storm that comes, they got to fix the tower. Something happens, they got to fix it. They're responsible for it, right? And so yeah, they're a swimming company. Let's ask another question. Let's ask this another way to identify whether or not they're swimming or surfing. One is, yeah, they own the infrastructure and so they have to deal with that and they have to invest huge amounts of money, huge amounts of risk, right, in order to make their business work. What happens, or can you guys think of a technology out there that could theoretically make AT&T obsolete? Starlink. Also bingo. Josh gets all the gold stars this class. He's doing great. But that's exactly right. Starlink is the answer. Elon Musk, and it's something he would do, could be like, you know what, I'm going to give the internet to the whole world for free this month. After that, it's $10 a month for everyone in the whole world, right? All of a sudden, connectivity is global. Now there's some hardware problems we got to work. You got to be able to get the signal and stuff like that. So he can't quite do that yet. But, you know, with some advancements and, you know, in cell phone technology and some other things, Theoretically, he could get there, right? He could make a partnership with Apple that just says, we now only work on Apple devices, but we work everywhere in the world. All you have to do is be outside and now we actually work inside or whatever. You know, there's so many different ways he could approach that, um, and make AT&T obsolete. So AT&T, I don't know. I hope they're shaking in their boots thinking about that. I personally am not invested in AT&T because I see some pretty significant existential threats. They're a swimming company and they have lots of risk and lots of very, very high operational costs, right? And maintenance costs to keep that stuff going. So, you know, it is what it is. By the way, I'll share that prompt on the, on the Manus thing. I'm happy to do that. So, all right. So we've examined a couple of companies. Netflix is, is surfing. AT&T is swimming. So let me share one of the companies that Jeremy— I said Jeremy's working on 7 different companies. One of the things he bought after he did his exit during the dot-com bubble was he bought a retirement home, senior living center, and has about 200 beds and in Kansas City, whatever. Anyway, so one of the things that plagues those homes is really tough is diseases of the mind, Alzheimer's, dementia, things like that. And one of the primary causes of those diseases diseases is they're not talking enough, right? They're not using their brain. They're not speaking enough, right? And so that's why, you know, when it's Christmas time, or even when it's not Christmas time, we need more people to go to old folks homes and senior living centers and just go talk to people. They need their brains exercised. And that's why those people are like, yeah, please come and visit us. That's awesome. But they will actually pay. And not only they, insurance companies will pay services that just have people show up and talk to them. And they'll walk around and just say hi. And, you know, the nurses are encouraged, hey, how are you doing, Betty? How are you doing? Like, tell me about your day. You know, they're just constantly encouraged to talk to these people so they can keep their brains fresh, because it's much easier to prevent, much cheaper to prevent, than it is to treat these diseases, right? Once you've got a customer, right, a resident that has dementia or Alzheimer's, it's not easy, and it becomes much, much more expensive. You need you know, 24/7 care. You need all kinds of crazy stuff. And so they try to avoid that as much as possible. So, so Jeremy says, hmm, I know about ElevenLabs, I know about these other things. What if I create this business that automatically emails schedules with the nurses, uh, times to call these residents every single day? And they can talk for as long as they want. They want to talk for 4 hours straight. They can do that. They want to talk for 30 minutes. Great. But they call them every single day. And he basically worked out with the insurance company $90 per resident per month. Right. That was the pricing model for that. So for his own, his own 200-bed facility, that's $18,000 a month in extra revenue. He then reached out to the reps that sell him other services and said, hey, I'll pay you a commission on that. This, and I'll let you pay the facility 10%, right? So, for every resident that signs or that, you know, that you have in your thing, if you have 100 residents, I'll pay you $9 for every single resident, right? So, he then pays $9 per resident. He pays a 20% commission to the rep, $18. So, out of his $90 he's getting, $27 goes out to, you know, commissions paying. So, that means these new other facilities that he's bringing on, are making money by signing up for this. The reps have a reason to talk about it and it's a super easy sale because they get, they make money. It doesn't cost anything. And if it does cost something, it's, you know, there's your extra $900 a month for whatever operational things you have to move around to make sure people do this. But then AI runs everything and, and he's surfing on 11Labs. He's surfing on something that's sending emails. He's surfing on something that's sending reminder texts. He's surfing on something that's automating the billing and the invoicing and all all of this stuff, right? And so he's now, he's now up to about $150,000 a month in revenue, right? And I mean, guys, that's $1.8 million a year. That's, that's a lot of money. And he's doing nothing. He is surfing on a bunch of other people's technologies. Okay. This is important because a lot of times people learn the type of stuff that I'm teaching here and and it's now possible, and they say, well, why would I ever buy any software ever again, right? Like, who here is familiar with Calendly, right? Anybody use Calendly? Yeah, I use Calendly. Who here thinks they could rebuild Calendly with Replit right now? All of your hands should be raised, right? All you have to do is say, hey, I need you to make a Calendly clone. Done. It knows what Calendly is. It knows that. It'll come up with all the requirements. It'll connect you to a Google Google Calendar API and an Apple Calendar API and a Microsoft Calendar API, whatever it is, boom, you're done. Here's a link. Okay, you could rebuild that. Okay, but then, you know, what happens if Google changes their API and Microsoft changes it, and then there's this extra feature? What if you want to be able to add multiple guests? Oh yeah, well, you didn't add that feature. Calendly has that feature and I want that feature too. And, you know, it'd be really cool if I could customize this and You know, all of a sudden you're building out this huge thing. Now, if you really want to make that your business and you want to compete with Calendly, by all means go for it, build a solution, go sell it. But if you just need the functionality of Calendly, what's cheaper? Trying to maintain and build that ongoing and monitoring the APIs and all that stuff, or just paying $11 a month for Calendly, right? That's the whole— this is the whole buy versus build discussion, right? But, you know, with something like what Jeremy's doing where he's making a crazy amount of money, he doesn't have to rebuild anything. If ElevenLabs goes out of business, he just switches to some other voice. There's other voice providers that are AI and they're good. ElevenLabs is my favorite. I think it's the easiest to use, easiest to understand. But OpenAI just came up with their own. OpenAI just came out with their own voice API and it's dang good. I watched a 20-minute video the other day, talked about it. In fact, if you're in the Facebook group, you'll see us talking about it just a couple days ago. And so I want you guys to just think about as you're building your current businesses, as you're, you know, whether, whether it's an existing one like, you know, Jake with Infinity Pools, you built this business for years and you're going to go double your revenue in the next couple of years. You know, what can you surf on? Right. Are you building everything? Are you— you know, that's something to consider, Jake, as we, as we talk about should we rebuild something? That's already there because they can't— it's not quite exactly what— maybe, maybe we can, or maybe, you know what, let's just use what they've built. We'll let them maintain it and we'll figure out how to, how to work with it, right? And so those are just discussions you have to think about, right? Do we want to— do we want to be responsible for maintaining this piece of software, right? Because things break, right? Abby's nodding her head because she's like, yeah, I mean, even just an OS like iOS 26, iOS 27 could change the way that something works, and all of a sudden you got to go in and update it, right? When you own a piece of software and you're using it actively, there's something that, you know, it's called code rot, which basically just means it gets old, right? The technology just gets old, and, and you have to, you have to try to stay on top of it.

So anyway, um, you know something, Jon, I was thinking as you've been talking is that The next few years are going to be kind of a mess, right? As we already know, companies are letting go of workforces that they're probably still going to need. Some of them, jobs are changing, so they're, they're firing their marketer, but actually, or they're firing their marketing team, but actually maybe they should keep two of those marketers to supervise all the AI tools that are doing marketing and to maintain and stuff like that, right? And I think the next few years are going to be like pretty messy as that transition happens. I don't know how long that is, if it's 2 years or 5 years or whatever, but what I'm really seeing is that I think the companies are going to do like really be able to differentiate themselves in the future are going to be the ones that are using people. Be like, hey, we've got this service, like you're saying, do you want to maintain that software? Do you want to use Calendly, right? Well, if I'm going to choose to use Calendly and pay them, if I call, I wanna talk to a person. I wanna feel like I have like premium support if I'm gonna pay for this thing that I could actually build myself. Like I can see a, so you just showed us how to make a bot, a chatbot for the website, right? I have a friend who recently left Zapier to join his buddies at an AI chatbot company that they have big investors, everything, right?. And I just sent him a message that said, what are you guys even doing? Because I just saw someone build this bot and it took them 10 minutes. So what are you guys doing? You know, and I'm realizing that like, sure, right now small companies will come to them because they don't know how, right? But like bigger companies could just do it. So for them to, in the future, to have customers, they're going to need to be different. And the way to be different is going to be like ultimate customer support. And like ultimate personalization and ultimate like being there for people.

What do you think about that? I agree with the sentiment, and I am a consumer just like you, and I do want that support. But you know what I really care about?

I want to call my bank and just have whatever I'm calling about be fixed. Yeah, I don't care if it's—

I guess if they are good enough, it doesn't matter. Exactly. If the AI— think about the AI 10 years from now, how good will it be? If it's, if it's literally indistinguishable from a human, I don't know if I care so much about that. Now, that's not going to apply always with that specific example that probably applies. Let's talk about going out to eat. Okay. And something else is, you know, robotics, automation, AI, take those 3 technologies, put them together. Right now, there is active, there are active stores that are being worked on right now for McDonald's, Burger King, Taco Bell, all the places, right, where it's, it's called the one-man store. One person can operate the whole thing, right? You show up, you can walk inside, you can, you know, you order at your kiosk, the food does all its things, and it comes out, right? It's all automated. The order's right every time. You'll have AI at the drive-through talking to you, taking your order. It all come out, right? And you just have one person there that, that just, you know, makes, you know, feeds the machines or something, right? And theoretically, you could probably get rid of that person too. Or maybe is there just for, you know, the next level of support when you do need a human, something's not going right, you know, just to make sure it's happening. I think in the future, there will probably be fully automated vending machines that can serve McDonald's, right? Then you have the opposite end of the spectrum where you take somewhere like Ruth's Chris, right? A nice steakhouse. Where there's going to be almost no AI. I mean, maybe there's some AI doing, helping with marketing and some bookkeeping and things like that, you know, back office stuff. But you go to Ruth's Chris because you want a real waiter to come out and white glove serve you, right? And, you know, you want the chef to come out and you want to thank them. And, you know, it's the human experience. And so, yes, I think there's going to be more emphasis on human-centric experiences. Versus, hey, I just need a cheap lunch, right? I need my Big Mac, right? Or I need my Powerade or whatever it is. Whereas you then, or, you know, and there's also, you know, music, you know, my sister's a musician and she's been working on these songs for years. And, you know, I bought her kind of like a home, kind of like a studio, kind of like I have here, but like a much simpler version that she could set up in her house. And she just struggled to figure it out. And then this week she discovered a website called Suno. Anybody here who'd have heard of Suno? Anybody use Suno? All right. Yeah, I got to show it. Suno is super easy. Suno is a, like, I don't know, country. Let's see. Slow, slow dance country song about, you know, meeting a pretty girl. Okay, awesome. Create. Okay, I have already signed up. I want to just— okay, it's ready. Let's just see if I can do this. Google, let's just sign in here. But yeah, the songs are ready. You just wrote 2 songs in that time that are crazy good, right? And, and she's able to, you know, my sister's able to now whip out these songs like, like it was nothing, right? Here come my cool little slow dance country songs about meeting a pretty girl. And if I don't even know if I'm on a paid version right now, but so it's probably just going to do some basic versions, but it's still going to be pretty good. Okay, let's just listen to it first. The lights were low, the room was still. Her laugh, it gave my heart a thrill. Oh, what a good night. We shot hello, her eyes met mine.

I felt the world stop, frozen in time.

Oh.

You set the voice.

Yeah, yeah, you can say, could you record your voice and have it be singing in your voice? Yes, you can. Yeah, so you can add, you can add lyrics, you can add instruments, you can actually, when you pay for the pro version, you can treat it like a full-on music studio and then you can, you could write your own tracks for some of it and then you can have it backfill the drums and the bass and say, hey, add a line with a synth here and whatever. You can be very creative. It's very, very powerful. Very cool. But my sister sent me this song. Um, let's see if I— she was texting me last night, right? But like, this song that she wrote, she wrote this a while ago, and, and this song is like really good.

Like, we can't sit here and listen to the whole thing.

To call my mom when I can't make my own appointments. So she had recorded her own version. You could look up Kid Again on iTunes or Apple Music, whatever all this those things, um, or Spotify. But, but this new version that Suno made, it's freaking good, guys. Like, like, it's really good. I enjoyed listening to it way more than the version that my sister had made. So there's two sides to that. Obviously, AI music is going to get better and better and better, and it's at this point, if done right, indistinguishable from a real human, right? Like, it is good music that you're gonna like. And how does, how does music go big? How does it get popular? Right? Usually it gets thrown on streaming or the radio, or it gets on a YouTube video and goes viral or whatever it is. And then you go and you find out who made it. Right. And a lot of times you don't care who wrote it. Right. You kind of, you kind of care who wrote it after you like the music. And so if you find out, oh, this was AI, like, are you going to not like it? I don't know. Maybe some people will, will have that reaction and other people will be like, I just think it's a cool song. I'm adding it to my playlist. I'm going to work out to this. I'm going to run to this. I'm going to whatever, you know, or you can write your own music now. Right. And so but then there's going to be definitely a premium for like, okay, Ed Sheeran's coming to town. I'm going to pay more to go see Ed Sheeran than I am to go watch an AI artist.

Right.

I just am straight up. By the way, you guys hear that Ed Sheeran's moving to Nashville with his family? That's cool. Just announced that like 2 days ago. So that's kind of fun but irrelevant. But anyway, guys, so there's, um, yeah, we're in an interesting world where, where yes, we're probably going to be paying more a premium for human interaction and a little less for automation, AI, robotics. Um, but the AI and robotics and automation are going to reduce the cost of many goods, right? A lot of people have said that AI will— there's a lot of people out out there, like the Sam Altmans and Elon Musks and those types of people are like, look, this is going to solve world hunger. This is going to solve— like, everybody should have a home, everybody should have food, everybody should have everything they need, um, because AI can produce it for so much cheaper. The paradox there becomes, okay, great, it's great that it can be produced, but if AI can do everything, what are humans doing to make money? Are we going to be able to actually pay for it? So, I don't have the answer and I haven't heard great answers from a lot of these people yet. Their best answer is, well, UBI, universal basic income. Everyone just makes money because you're alive. And there's plenty of, depending on what political party you're on, you're going to have your different arguments for or against those types of arrangements. So, it's a complicated world. It's going to be a complicated, it's going to be a complicated transition here. All right, guys, let's talk about sales and marketing for a second, and then we'll get into some images. Um, if you guys wanted to create just like an amazing sales strategy and plan, knowing what you now know, you're 7 hours into 8 hours of this class. Okay. What would you do? You want to make an amazing sales plan and strategy and execute on it.

What are you going to do with your. Newly discovered AI talents? I'm gonna definitely nail down my business plan, nail down my idea really well. Then I'm going to— well, I'm going to do that using different tools like ChatGPT and Grok and just really nail it down, finesse it, make it look really nice. So once I figured all that out, I'm going to then take it to one of those platforms.

I'm going to say I need a And, and then I'm just going to expand on that. I'm going to try and go deeper into the sales part, deeper into the sales portion of it, ask it questions, have it ask you questions, that whole interview process, all that. Perfect, Carrie.

Awesome. Yeah, I, I would, I would be like you said, Carrie, the plan, the details, even the requirements, but I would be like going as far as possible down this like Who am I doing this for? Who are they? What is their problem that I'm trying to solve? How do I get to know them personally and intimately so that I can actually understand that root problem that I'm providing a solution for that is obvious to them, that doesn't require them to self-educate? It's just like, yep, here's what I needed. And really focus on who and the problem down to its smallest atomistic element versus just like throw it out to the universe.

And start marketing in 12 different directions.

That's my thinking. So you're finding the target market? Yes. Yeah, you're just ultra-defining. You're trying to figure that out. Now obviously you're going to have to guess and check that, right? You're going to, you know, you're going to estimate with, with ChatGPT, then you're going to go test it, right? So I want you guys to take it now a step further, okay? You're both exactly right on your first steps. You're going to be talking and building strategy. And now let's get tactical. How are we going to get our first sale?

How are we going to get someone's credit card or send them an invoice? You're going to figure out what strategy you want to use because there's a couple different things that you could do. And maybe that would be a question I would ask chat, like, of all the different sales tactics and the target market, knowing the things that you know about what I'm doing here, give me the pros and cons of each strategy based on, you know, potential revenue, all this other stuff, the amount of time it's going to take, the amount of investment, the amount of babysitting, just basically giving it the parameters. I want to make it as easy as possible on me, but also as effective as possible. So I'm going to be asking it, give me a list of pros and cons for all these different options of how I could market this.

So that I can make a decision on what would be best. I'm sorry to ping-pong back and forth here. I would, in my brain, as you're saying that, I'm like, yes, go deep and understand the pros and cons. But in my brain, it's also like, make it possible to find the best possible price that you can based on your information, put it out there, be able to take payments, get the pathway there just as saying, hey, that's possible. As I test and try, I'm looking and seeing, well, if someone's buying, who are they and where'd they come from? You can't really sell anything unless you make it possible first. And if they're not able to hit a paywall, you're just guessing too, because you can't really see them finish, cross the finish line.

So that would be my— that's a great point. Okay. Very good. So let me, let me say two things on that. One is, yes, you have to finish. We're assuming here where we've finished our product, right? In order for me to launch GenAIPI. I had to finish my Stripe setup, right? I had to get it integrated into the website. How was— how did I do that? I said, help me set up Stripe, and it said, great, what's your API key? And, you know, give me this webhook, and, you know, whatever, let's set this up. And it was super easy. You just go into Stripe, you just follow what Replit tells you to do, right? Do this, say this. Okay, I want to be able to do this, and then this happens. One thing I will say, guys, this is really important and will save you a heck ton of, of, uh, headaches. It's very difficult, if not almost impossible, to test your Stripe inside of Replit because of the test environment and the temporary domain that it uses. Stripe will think, oh, this is not a real transaction. So what you have to do is you have to publish your website to your, you know, genaiapi.org, which I'll show you guys how to do in a minute as It's not that hard. So you're gonna have to deploy, test there, test them live, test them live, which is kind of weird. You don't like to test on live usually, but that's just kind of how you have to do it. So before you tell anybody about it, get it working.

Okay?

There are ways around that, but I'm just telling you, you're gonna run into Stripe problems in your test environment. And Stripe is not the only one. You could use PayPal and you could use a bunch of other system payment processors and things out there, but they all have APIs and they're all very easy to set up for an AI. Okay. Okay, let's get even more tactical. Let's— and I want you guys to do this when you get your product created. Okay. I want you guys to say, how can I get my first paying customer in the next 24 hours? Okay. And have that conversation for 20 minutes with ChatGPT or your LLM of choice. Okay. It is amazing what happens, guys, when you put constraints on yourself, artificial constraints on yourself, right? I gave myself a constraint of I want to build a business in 2 days. I want to build it today. I'm going to get my, I'm going to finish it by tomorrow and get my first customer by tomorrow. And then I was just committed to it. And even though I didn't quite make it in 1, 2 days, I finished it by day 3 and I was committed. I'd put 40 hours in. I'm ready to, I'm ready to sell to somebody, right? I was ready to go. And by putting those constraints on yourself, you're going to do crazy things. Like the, the conversation you'll have with ChatGPT when you ask it, how do I get my first customer by tomorrow? They're going to be like, okay, this is tactical, we're going deep, we're going immediate. All right, let's think about who do you know, what's in this space, what can you post on your social, are you part of any groups, are there any things here, um, can you— I want you to go to this website, and when you post what you're doing, I want you to do this. Like, it will tell you specific plans, right, on what you can actually do to get your customers. And I've tried this, guys. It works. It's crazy how well it works. And will it get you out of your comfort zone? Maybe. If you're not used to selling or posting about stuff. How many of you guys have posted on your own personal Facebook page, for example, about a business thing that you do? Okay. 2 out of 10 or 11 here. Okay. Most people don't. And I don't know why. I don't know why. I've asked that question to people before and it's usually like, I don't know, it just doesn't feel like I should put it on Facebook. It's just kind of this weird thing. They're like, that's, that's where my friends are. Okay, well, do you not want your friends to know you make money? Do you not want your family to know that you actually have to eat and, you know, your house isn't free and, you know, It's totally fine to tell people that you make money and even how you make money. Hey, this is what I do. I like to talk about this. Um, if you're in, you know, I, I can't tell you how many times I have posted on, on Facebook about AI and friends, neighbors, Ryan, right? Ryan, Jake, like you guys are neighbors. You probably heard about this because you saw me talking about it on Facebook and we're friends on there, right? And you at least knew, oh, Jon does AI, right? And maybe you didn't need it. The first time you saw me talk about it 5 months ago, but eventually you're like, hey, you know what? I think I'm ready to do AI. I know Jon does it. And so I'll reach out. And so it's not that I was trying to sell to you guys. All I'm doing is saying, hey, world, this is what I do. And then the world will self-identify. People are very good at self-identifying what, you know, what needs to, you know, what they need to do. And if they know somebody, they're going to— you're going to come to me if you know me, unless you just think I'm a scumbag. That's fine. Then you won't reach out to me. But assuming you're a stand-up citizen and you've put it, you know, put, put a good image out there in the world, they're probably going to come to me instead of some random stranger when they need AI training, right? Same thing on LinkedIn, by the way, vice versa. Let people on LinkedIn know what you do in your personal life, right? I would rather work with somebody who knows how to kayak or loves to golf or whatever, right? Because I love those things. And so while I'm talking about business, or we can go do those things and we can talk about business. We can have, we can have a multifaceted relationship instead of just like, hey, you're, you're my guy that does my marketing and that's all we ever talk about. That's a boring relationship, right? And so by mixing it up, like, this is, this is so critical, guys. I believe in this so much. Your personal brand should be the same across all platforms. Now, yes, I am going to talk more about business on LinkedIn and more about, you know, the fireworks pictures that I took on July 4th like everybody else on Facebook, right? I'm going to talk about different things on those platforms, but everyone on LinkedIn that interacts with me at any reasonable rate knows that I love to kayak. They know that I love the piano. They know that I'm a member of the Church of Jesus Christ of Latter-day Saints because I bring my faith into it sometimes, right? And on Facebook, they know those things as well. They know about my kids maybe a little bit more. They see what my kids are doing. I don't really post about my kids as much on LinkedIn, I do sometimes when there's like a lesson or something we're talking about. But ultimately, they know what I do for work. They know that I'm into AI. They know that I did augmented reality stuff before, you know, and I just share things that I think people care about. And you know how you can tell if people care about it? You get people reacting to it or you don't and you adjust, right? That's all I think I need to say on that. I will say this. I highly, highly recommend— maybe, maybe I do. Oh, it's hiding exactly behind me. What a freak, my body trying to grab this. Okay, this book, Gary Vaynerchuk, Day Trading Attention. Best marketing and sales book I have ever read, right? I've read a lot of them. I've got a beautiful library over there of books that I've actually read and a few that I haven't. And But this one right here, basically the premise of this is if you aren't drowning in customers, then you are not talking enough on social media. Social media is free. It is free marketing. It costs nothing to post to Instagram or to Facebook or to Twitter or to whatever platform you're on. Okay, Pinterest, um, wherever farmers hang out, Carrie, that's what you got to find. Okay, but Talk about your dang— what you do. Talk about what you do. Join groups, join forums, join, you know, join these, these, you know, Facebook has amazing groups. LinkedIn has okay groups. I think they're, I think they're all right. Um, but what, um, what social— what almost all social platforms are doing right now, a lot of, a lot of social platforms for many years focused on what's called the social graph, which is, okay, I know Abby, we're friends. And so when Abby posts something, Facebook is going to show me Abby's thing, right? We're not friends on Facebook yet, Abby, but maybe we will be in the future. But I'm going to see what Abby's doing in her personal life, and that's cool. And that's what Facebook wants to show me. They want to show me what my friends are doing. More and more, it's moving towards what's called an interest graph.

Right?

So if Abby's off playing soccer, it's probably not going to show that to me. Not that I hate soccer, but it's just not one of my main things, right? If she was playing college football, I'd be following her big time, right? Because that's something I care about, right? And so more and more, the things you talk about determine who your content is going to be shown to. And so if somebody knows one of my friends cares about AI, then they're probably going to show it to them. And that's cool. And the people that don't care about AI, they're not even going to see it anyway. So who cares, right? Your grandma or your aunt isn't going to see that you're talking about AI. Who cares, right? Or some of them might, and that's great too. Facebook does a good job at kind of combining both the social and the interest, and you can put like close friends, and that allows you to see more of the social graph stuff. But, but bottom line is talk about the things that you care about on LinkedIn, on Facebook, Instagram, whatever. And those topics will float your posts. You don't even have to do hashtags anymore. You can if you want to be explicit to the platform and tell it this is what this is about, but it can read the whole text now and it decides this post is about banking, this post is about software, this post is about HR, and it will then push it out to the appropriate audience. And then you can pay to boost those things and you can do Facebook marketing and ads and Meta ads and all stuff. And we don't have time to get into that today. But generally, generally, guys, you want to, you know, you want to talk about what you're doing so that people know about it. And then go into your ChatGPT, go into your planning, your AIs, and just ask it for a plan. Say, interview me one step at a time here, one question at a time, about how we can get our first customers by tomorrow. Right. Okay, great. What groups are you part of? Okay, perfect. What about this? What about this? Okay, based on all these answers, here's my plan for you. All right, then, guys, here's the most important part. This is how you get customers. Follow the plan. Actually do it. It's so simple, guys. I know some of you might be afraid to sell. My wife is a perfect example. She does amazing stuff on the farm, and she does gardening and all these things, and And she— I know there's so many people that would want to buy her stuff more, and she does sell some stuff, right? But she's just like, I don't like selling, I don't like talking about that, you know. Okay, that's fine, right? No big deal, no big deal. But if you actually follow the plan, if it says you, you should get up and go find a local, you know, you know, go to the local IFA, Carrie, right? Walk into that IFA and say, can I put up a sign here for my new, you know, app, my new duck app.

What's your app called? I keep forgetting. Quackity Quack?

What's it?

Quack Track.

But I actually have changed since then. That's all right, but let's say it was Quack Track just to keep with the example. Go to the local IFA and say, can I do this, right? Go to the local IFA and walk up to the 10 people that are in the store right then and say, hey, can I tell you about my new app? I think you should download it. Like literally, you want to get a customer in 24 hours, then go get one. But most people are afraid to actually go to that IFA and walk up to a stranger and tell them about the thing they're doing. And I can tell you that that is why I have been successful in my career is that has never, ever gotten in my way. I just, I'm like, okay, I mean, I'm going to miss 100% of the shots I don't take. Thanks, Wayne Gretzky. Thanks, Michael Scott. You know? That works. It absolutely works. Get up and go do what it says. Post, talk, pay, do the paid ads, do whatever it is, right? Man, I just had like 4 packages dropped off on my doorstep. Like, I don't know, my phone just gave me a bunch of texts all at once. Maybe that was just a Signal thing. Anyway, I got lost in Temu the other day. It just happened. And I bought some stuff I probably didn't need, but whatever. Um, I will say that it was with Christmas in mind, so they were for other people.

But anyway, um, shopping on Temu is dangerous.

So those other people are good to like, uh, have skin cancer from it? Exactly. Yeah, that's what I was saying. It's a little dangerous to shop on Temu, but, um, it was for the— I don't know, I— we'll see, we'll see. Um, it's so dang cheap. It's like it's $4 for the same thing. You want to buy it? Sure, it's $4. Like, why not?

Let's see if it's good. If it doesn't work, then— so anyway, Jon, that actually brings me, uh, to, to an interesting, uh, conversation. I mean, we're talking about all the great things that AI does, but how do we play safe with AI? Yeah, um, because that Temu offers cheap product, but where are they sourcing that product from? Is that product coming from child labor, from slave labor, labor from forced labor. There's a host of different things that makes that product inexpensive and easy to replicate. Just go watch, go look at the Saturday Night Live Temu skit.

I remember that. It's pretty spot on. No, it's— I've been to China and I've seen them make that stuff and It's a great point. Apply it back to AI here, Richard. Yeah. Yeah. Yes, you need to be careful with everything you're doing in the same way that we talked at the beginning of this course, you know, with Abby, or at the beginning of today, right, this session with Abby saying, look, sometimes this is a junior developer that gets it wrong, right? Understanding that AI is not infallible, it's not perfect, it you know, it's going to make mistakes. This is why one of the first things we did yesterday was we kind of checked our work with Grok. We went back and forth between two LLMs, and you could even add another one in there. I usually will with really important things. I'll ask three different services. Usually it's going to be Anthropic. So Claude, Jon Cheney, and Grok. Those are my three go-tos that I'm going to ask something important to. And I'm going to then independently let them answer. And I might ask a follow-up question to each of them and just say, hey, let's, you know, can you verify this? Are these verified sources? You need to check on that kind of stuff. Like, for example, early on with AI, I would, you know, I thought it was really cool that you could be like, hey, write this church talk for me on this topic, you know. And I never actually used one of those to actually give it, but I just, I thought it was fascinating. It could do it. And then I started diving in. I was like, I've never heard that quote before. And then I was like, where is this quote from? And it would say this place. And I'd be like, that quote is not in that talk. Like, that does not exist. And it's totally like, it sounded real. It sounded like this was a piece of doctrine that was really cool. And even though it might have been factually correct, no one, the person it had attributed to, no one had ever actually said those words. Right now, it's much, much better. AI is much, much better. Better at, um, at not hallucinating today than it was, you know, 2 years ago when I was having fun with that stuff. Um, but you have to, you have to learn how to fact-check. You have to fact-check AI. You got to fact-check it. You have to ask for double confirmation of source. You have to, you know, you have to do all kinds of things like that. You also—

I share a story about that, Jon. Oh yeah, yeah, go ahead, Carrie. So we work really closely with a lot of astronauts, and we work with this one astronaut who went to space. She was the first African-American astronaut to pilot a spacecraft. Her name was Sian Proctor, and she was in a spacecraft called Inspiration 4, the 4 meaning that's how many people went to space on her craft. Well, my husband asked the question to ChatGPT, how many people were on Inspiration 4? And it gave a list of 7 people and their names. And it was just funny because we're like, well, we know for a fact that there was only 4 and we know who those 4 were. But for some reason, chat threw out like, oh, it was 7 people and it was this random person that no one's ever heard about.

And this is so interesting. Yeah, it'll make stuff up. And like I said, it's getting better. And if you say only give me verified facts where you can provide a source and provide the source, right? You can, you can kind of limit it to say, all right, let's make sure we get good information. The reality is, you know, the tool is, is very powerful, but you, you have to, as the user, as the one in charge, you have to, you have to say, okay, I, I don't, I can't always trust this perfectly, right? I will ask, you know, I'll ask Replit to do something sometimes and I'll say, all right, does this work now? And it will say, yep, 100%. It will literally say 100%. This works. And I'll say, okay, prove that to me. I want you to tell me why it works. Show me in the code the pieces that guarantee 100% functionality here. And it will go through and it'll be like, oh, you know what? I actually found something, you know? And so learning how to question AI, even about its output, right? About like, and to challenge it to prove itself is a great way to avoid a lot of problems, Richard.

I mean, that's really what it comes down to.

Yeah.

And I get that. I think Josh probably alluded to this a bit earlier when he mentioned, yeah, if you have an established company, they have a lot of resources, they have a lot of documentation. A lot of things that they can build into that knowledge. But if you're starting from scratch and you start to start from scratch, or in Josh's book model as an example, is, is there a pirated site that has all those books? And I'm not saying that that's the case in, in Josh's friend's scenario, but are they ignoring the fact that this is someone else's intellectual property and how do you manage that accordingly so that you're not— Jon, you would have created a large company. And if I got all your documentation from that company and I created my own company from that while you're still holding on to that company, I'd be stealing your company from you, right? To some degree, that's intellectual property you built, that you managed, that you you went through and, and figured out.

So how—

that's kind of some of the concern that I have, uh, as, as we begin to build and, and, and ideate. And, and, you know, as you get into your images, you're going to create some images, uh, and are those images your images? Are they ChatGPT images? How do you copyright them? How do you protect them from other people. And those are some of the things that have been kind of running through my mind a little bit, especially when you start using other people's IP. I don't know if you're going on Temu and buying BYU gear, but I guarantee you that BYU— Yeah, yeah. BYU is going to be upset with that, right? If you go and do that and we we at the University of Utah, we spend thousands of dollars protecting our online sources, our, our protecting our trademarks online. You know, anyways, just, just a few thoughts and how we manage that properly and make sure that we're building something that's truly ours and not Ryan's.

Yeah. It's a, it's a great question, and it's really, it's probably closer related to the way that things have always worked than you think. Um, you know, if something is out there and publicly visible, it's free game for AI right now. It's in a copyrighted book. All right, now we're getting somewhere right now. We're starting to say, okay, what's going on here? So, um, Anthropic just admitted, and not that everybody knew that it happened, right? But everybody's trained their AIs on books, right? And so they just agreed to a $1.5 billion settlement with 500,000 authors, right? And they're paying all those authors $3,000 apiece, right? So it's not a lot, but hey, now they're legally trained on those books, right? That's the settlement that's been worked out. And the reality is most of this law has not been created yet. But the few— this is the other really, really interesting thing, Richard. The rulings that have come out of courts so far, for the most part, have actually treated AI more like a human than a computer. And what I mean by that is this. Let's say I ask my AI and I say, hey, will you create something that's kind of like the Death Star? Okay. What if I asked you, Richard Fairchild, to do that? Well, in your mind, you already know. Well, this is a big spaceship. It looks like a moon and it's got this. And, you know, there's places where spaceships can come and destroy the whole thing. You know, anyway, you kind of know what that means because you've seen the movies and it's available. It's out there, right? And you could create something that's kind of like it, but that's far enough away from the original that that's okay. Were you inspired and influenced by that? Absolutely you were. But you're not— if you're not copying it directly, then you're not going to be dinged for any sort of copyright or any sort of trademark or any sort of whatever, right? You're not going to be dinged on that IP. If you copy it exactly and you call it the Death Star, yeah, you're breaking a law, right? Or, you know, you owe someone some money, probably. Right, is really what it is. And so AI is being treated like that. And like, the fact that it's out there and it can see this and it can see public images and they're on news articles and these things, and like, it's there, we've seen it, we've been influenced by it.

And so because we've been influenced by it, we can do it. I personally think— trademark's a little bit different though. Yeah, it is. It's consumer confusion, right?

Somebody believes they're buying what you've created and that it's a It's the new Death Star.

Yep.

Yeah. You're, you're absolutely right. Yourself into trouble. You can. And so, so you are the user at this moment is the person responsible. It's not OpenAI or Midjourney or anything. We need to pull up a few of these images, ones that you guys can look at. But when you're creating that image, when you're creating that thing, you know, if you're, if you want to create an image of a stormtrooper chasing Simba, you can do that. But where you use it, how you make money with it, if you make money with it, all of those things, those are going to be up to you, right? And it's on your head for how you use it. OpenAI is not going to get in trouble for creating that image. Not right now, at least. I do believe, and I hope that the structure that's been talked about that I think makes the most sense is at some point, if I ask for an image of a stormtrooper chasing Simba, The AI will say, yeah, here you go. Here it is. And with that output will come some sort of license and the output itself, a small portion of what was charged for that output will go to Disney. Right? I mean, there, there are solutions to this. The reality is the infrastructure just hasn't been completed yet. And I think we're honestly 5 years away. From anything organized whatsoever. We're in the Wild West for sure. So because we're in the Wild West, you're, you're the boss, right? You're the boss of yourself. You're not going to be able to abide by the laws, right? It's the Wild West. There are no laws because it doesn't exist yet. Most of the laws don't exist yet. So you got to follow the same traditional laws that have been there and be careful, right? So it is what it is. It is what it is. And you just kind of have to deal with it. It's a very complex issue. It's really fun. Fun. But anyway, all right guys, let's spend a few minutes breaking some IP laws. Let's get into some image creation stuff and play with that. Quick question before we hop into that. With some of the things we were looking at today with the various APIs and having our own GPTs, how close are we to really having something that's agentic? Where it can take, you know, multiple steps. Is that something we can build in Replit, or is that something that's still kind of on the cusp of being usable? Now, you could build that in Replit. You could build that in Replit. It's a matter of building agents that do steps and then have waiting instructions, basically, right? You're just opening up multiple instances of an AI through an API. And having it complete something, and then it reports back, and then it has to wait until all of these steps complete, and then it analyzes all of the stuff it's done, and then it decides, do I need to redo any of this, or can I take this input and then move it to the next step? And so it's just a logic chain is really all those agents are doing. And so you absolutely can build it in Replit. It's not that difficult. Well, no, let's put it on the advanced side of things, right? But, but if you, if you understand how it works, Even if you don't understand how it works, let ChatGPT explain it to you. Let Replit explain it to you how they would build it. But you could just say, hey, I wanna build something like Manus that does this, this, this, and this for my company.

So.

Cool. Okay.

Nice.

Yeah. Good question. Okay, guys. Images. I'm glad we, we, we kinda had the front end of— I've, I've had this exact copyright discussion that Richard was talking about before, an IP discussion where you just, yeah, you gotta, you gotta know what you're doing. You know, I can, I can easily say, you know, give me, You know, give me a Star Wars intro inspired, you know, advertisement for GenAIPI, right? GenAIPI.org. Boom. Okay. And so that, that would then, you know, and sorry, I need to create the image for this. And we could actually probably ask it for a video ad script. But yeah, now it's gonna create an image on this. Now guys, I actually have done, I have done some image requests in ChatGPT where it said, sorry, due to IP laws, I can't do this, right? So there are some things in place, but most of those, those barriers have kind of been removed, right? One of the things that they really are careful about is protecting people. They really don't want you to be able to modify people as much as, you know, you can change like what somebody might be wearing or things like that to a certain extent, right? You can obviously, you know, take that a couple steps in the wrong direction. You don't want it to get pornographic, and ChatGPT certainly will not put out pornographic things, but there are platforms out there that will, that don't have those safety barriers in there, right? And so if you have kids, kids, you know, teenagers, whatever, make sure they know or you know that they can do some crazy stuff. Like they might have a crush on a girl. If they have a picture of that girl, they can undress her with AI. All right. That's an uncomfortable topic, but it's a real one, right? There's actually laws that have just been passed in the United States protecting against that kind of stuff. Like it's serious. And people have been cyberbullied with, you know, hey, this picture gets sent across school and it's a fake AI picture, but the face is the real person and it looks totally real. And, and, uh, and that causes, that causes problems, obviously.

Obviously you don't want that kind of stuff to happen. Okay, so, um, on a, on a less serious note, Jon, just like, uh, just to add to that, sorry, I don't want to interrupt you, but there's some, there are some, um, AI websites that you can use to generate models that look like yourself. They're very easy to use nowadays. They just, they use something called a model Some of them call them other things. Just, just a tip for the ladies or for any guy that has a lady that they might be making a thing of. Those things will pick up on certain things in the image that they think are important. So if you are, for example, a busty lady and 2 of your photos show a little bit of cleavage, you will start getting photos that always show cleavage or, or that put the girl in a bikini even if you don't ask for it. So the training data that you put in is super important. I just, I took a last year and it's, it's, it's kind of, or not last year, sorry, like last month I took a short webinar style thing about that. And they were showing the differences. This girl who was very curvy, she showed a couple of professional images of her in like a suit that just happened to show like the top 1 inch of cleavage. And all of the photos that it generated after that when she asked to make photos of her on a beach were her in like a string bikini, even though that wasn't at all her image or anything. So if you have that problem come up, take a look at the, the training photos that you're giving it and look for any clues like that.

Yeah, guys, it's really quite incredible what's possible with AI. Like, it can do amazing things. And you can take an image of yourself and, you know, especially for girls that love to do this, right, you can you can try on a new outfit, try on this new dress and try on this. And it looks real, looks great. And it's a very valuable thing. I spent 10 years, almost 10 years of my career in augmented reality and a huge application of augmented reality was can we do virtual try-on, right? Can we try on these shoes? Can we try on these clothes? Can we try on this whatever? And so I, I was building stuff that did this, right? Or I wasn't, but you know, the people I was paying to figure it out, we were working on this problem. So I'm, I'm, I'm pretty, pretty aware of a lot of what Abby's talking about here. And anyway, you know, Like all tools, guys, I mean, a huge amount of the internet itself is pornography. And so it's just we have— there's, there's people out there that are looking for that. And because people are looking for it, the internet provides it. People are willing to pay more money for it. And so AI very similarly can be applied in very, very dangerous ways, right? I mean, I worry for, you know, my daughter, you know, she's, she's almost 12 and, and going into teenage years where, you know, that stuff starts to really matter and could really impact her. And if someone does something bad and and takes a picture of her and, you know, she's a cheerleader and, you know, things, you know, bad things can happen and it can really, it can really hurt people. So watch out. That's just a warning, right? A warning to just be aware. Watch out for these things. Make sure that, that your kids know, right? And anyway, do what you want with that knowledge, but it's a very important thing. So here I was able to create a quick image, right? ChatGPT is actually pretty dang good. One of the things that ChatGPT is good at doing, guys, is, is putting text on images, right? My favorite image creator is Midjourney, without a doubt, right? It can create unbelievable images with amazing prompts. This was my prompt: Chief AI Officer. That's all it was, right? I was just looking for something cool. I was writing an article on it, and I ended up using— I think I used— I think I used this one, or maybe this one. I don't remember. It was one that kind of looked like this. But anyway, you know, I use it to make logos. Midjourney's outputs are incredible. Like this, I was trying to make an image I was going to use on the website and, and I discovered that I actually really liked— I really liked this image right here. I thought this was really cool. I just like the feel of it. Everything kind of looks good. I love this style. And by the way, creating these styles, right, this nice, amazing prompt. How do you think I created this nice, amazing prompt? I went into ChatGPT, right? And I asked it, I said, I need a good prompt for Midjourney that will accomplish this, right? And boom, that's what I got. And then inside of Midjourney here, you can actually turn this on and animate it and, you know, whatever.

There's—

it'll give you all these different styles and you can try the different variations and things that it does. And, you know, but now I could use this potentially as a, as an ad on Facebook, right? Like, hey, get confident with AI. Boom, I've got my ad. Right? It's awesome. Right? It shows a nice confident girl using her computer and these, all these things coming through and it's kind of magical. And anyway, it's awesome. Really, really useful. I love Midjourney. I think it's amazing. The one that I was talking about yesterday, yesterday is Gemini. I think it's just Gemini.google.

Come on.

I don't want.vom. Give me an autocomplete that is correct. All right, whatever. Um, did they— we're on 2.0 Flash. Okay, try now. I am going to do that. Okay, so I'm not gonna do this. Actually, I am gonna do this. I'm gonna say create an image. Um, okay, I'm not gonna do the video right now. We'll do that in a minute. Okay, but I can take this. Um, what— Nano Banana. I mentioned this yesterday. This is the newest image editing, you know, element, right? That's here. And it's just incredible. You can take existing images. I wonder if I have one here. You know what? I think I do actually have just this. What is this image? This is a phone. Sure. Whatever. We'll just do the phone. Change the background to a starry night. Okay. I don't know if it's going to give me a Van Gogh style or if it's going to actually just give me some stars because I said— if I said Starry Starry Night, it might get that a little better. Anyway, this is more capable than 99% of expert Photoshop users, like people that have been paid, you know, 6 figures plus to do image manipulation. Now Nanobanana can just change things for you and you can put real pictures, real people. Nice thing about Google is gonna have, it's not, you're not gonna worry about, you know, the naked pictures. You're not gonna have to worry about that kind of stuff, but you are gonna get unbelievable things. If you want to try a new outfit on, if you want to change this thing, if you want to change the background of a, of a video or something like that, there's a starry night. Perfect. So it took my product, right? My cool little iPhone here and put it on a starry night background, right? And so if I needed to do that before, I'd have to go in and I'd have to like outline this and cut it out. And you know, I could then composite two images together and do this and that. I could also, I could also say, let's just find an image of, you know, image of man. Okay. Images. We'll grab this guy, Indian man. Hold on. Is there a watermark on this or whatever? Yeah, there isn't. I'm not obviously using this for commercial purposes, so we're just gonna call this that. Okay. Then we're gonna upload this to here and we're gonna say, put this phone in the man's hand with it up to his ear. He's on the phone. Keep his stance the same otherwise, and keep the, keep the natural background. Okay. So think, think about this from a marketing perspective. You're, you're, you know, you have a physical product, you're selling mugs, you're selling, you know, clothes, you're selling cars, you're selling whatever, right? You can generate these images now, you Yeah, he didn't do— did not do a good job.

I feel like he's on the phone.

In the man's hand. It's ready for print. Yeah, it's ready for print. It's not in his hand. Make it be there. Anyway, there's plenty of other better examples I could certainly give of this. It's supposed to be able to do that really, really easily. ChatGPT can do this pretty well, but Nanobanana is definitely the best. Just put it on there. I love it. You can also have a lot of fun with this too. We're going to go over to Veo 3 really quick. Veo Flow. Come on, give me the— there it is. Create with Flow. Where is it, Gemini? Hey, there we go. That's a little better. Still not perfect. You know, the spacing, that hand, it's kind of weird. He's kind of holding it weird. The size of the phone is probably too big compared to his face, but you can keep going. But the point is you can edit these things incredibly, incredibly really fast. You can ask it to do a lot of really cool stuff. Um, okay, Flow is on this account. Flow guys can create unbelievable things that used to take, um, that used to take, you know, weeks and hundreds of thousands of dollars or millions of dollars to create. You know, if you wanted to create a little cutscene in a video game, right, that's what I created here, right? I was actually just messing around with like an Indiana Jones kind of style with music. So it's not perfect. I can work on some of these things. And by the way, I can create these prompts. I can create these prompts really well. I want to show you an IKEA prompt that came out of this. One of the things that Veo can do now, guys, that is mind-blowing and that really has flipped the industry on its head is the fact that it can do audio now. AI can make people talk. It can make— you can do just crazy things. And you've probably seen lots of posts like that, but I want to show one that's really cool. IKEA Veo 3 video. So look at this. Uh, here we go. Look how cool that is. Like, think about what that would take without AI to create that. That's a huge amount of work. And I mean, even the image in the, you know, the reflection here is correct in the mirror. Like, everything's right. Like, Guys, I've paid for stuff like this and it's expensive and it doesn't look this good. Okay, now here's one of the things that's really, really cool. Um, the, the shot for, or the, the prompt for this, um, IKEA prompts, okay, you don't have to know how to make that, but the prompt, we're gonna, we're gonna find it here. We're gonna get I don't know if it's this one. This is just VEO3 prompts. I think I want to actually find it. You can find it, it's not that hard. I mean, part of it was in that first one, but it actually didn't have the whole thing, man. Where is it? But basically the way that it was set up, um is— let's see if this is in like the description or something more. Yeah. So this is, this is a partial prompt, right? But basically it's like, all right, here's the description. We're going to do this. Here's the camera style, all these things. But the one thing I wanted to show that's not in this video is it actually took the— it took the— what do you call it? Just like the order of events, like 1 second at a time from, from 0 to 1 second, this is what I want to happen. From 1 second to 2 seconds, this is what I want to happen. From 2 seconds to 3 seconds, this happens. And then, you know, all the way through the full 8 seconds. Right now Veo can do 8-second clips. Okay? And so you can, you can break it all down into 8-second clips. But to create a big long description like this and a big thing, all you have to do is find a shot that you like or see something that you like and then say, okay, I wanna apply this to mine. Take this, this prompt, copy it, go into ChatGPT, and say, hey, will you modify this prompt so it works for my product? Right? Like, Jake, we could do a crazy cool ad for pools, right? Hey, let's make it do a thing like this. And hey, let's bring back together the rebar and the cement and everything comes together and then fill the pool up in the last second. And boom, we've got a beautiful, you know, amazing video of like, hey, this is what infinity pools can do now. Right? Now, whatever. So, but we can create that using VEO-3, using ChatGPT to help us write the prompt. Using ChatGPT to talk through, you know, what is going to be the right way to do this and getting the style right. And then you just try 5 or 6 options, and for less than $10, you've created, you know, a million-dollar ad, right? Or $100,000 ad, or whatever, you know, whatever it happens to be. Okay, but VEO3 is amazing. One other platform I want to show you guys, and we don't have time to like really do a lot of training with these. I didn't mean to do this. We already did Midjourney. I meant to go to Runway. Runway is one of my favorite apps by far. It is so cool. So they have a few new things. They can do a lot of different things. Let's go to all tools here. So you've got, you know, video creation, image creation, lip sync videos. You could create some audio in ElevenLabs with whatever voice you want, with your own voice, whatever you want. Okay. And then you could create a character, whether it's a real human or a 3D animated character like this. Bring it in here, give it the audio, and it will lip sync to it. It'll make it look perfect, right? Singing, speaking, whatever you want. Make that Suno, you know, that Suno song, take a picture of my sister and have her standing there singing it, and it'll just bring it to life, right? One of the coolest things it can do now is Aleph, where you can edit videos with natural language. Absolutely insane what it can do. I took this. Anyway, where's the— yeah, here's the video. I am dead inside. I am dead inside, right? So I just took this little clip from The Office. I am dead inside, right? And then I just said, hey, make him standing in front of, you know, the Raiders team, you know, whatever. And he didn't do it perfectly, obviously, but I am dead inside. Right. And so you can do all kinds of funny things and, you know, we're breaking all kinds of IP here, I'm sure. But again, I'm not going and selling this. I'm just playing around with it. I'm trying things out. You can change the backgrounds of what things are happening, right? Here's a volcanic landscape. It changed the background. Woo! Right? This— he was standing in— this guy was standing in front of a tornado before, right? In a big storm, right? And I had it change it to a volcano. And by the way, this video, this original video was created here inside of Flow, right? It was a fake video to begin with right here. That's what it used to be. And I had it switch it to a volcano thing. It's all fake. I share this because for a couple reasons, guys, I share these things for a couple reasons. One, you can get crazy fancy with your marketing on this stuff and save huge amounts of money and deliver amazing results. Two, you know, this, this class is also— we've talked a lot about the ethics and just the societal impacts and things like that. Probably about 1 year from now, probably 1 year from now, you will not be able to believe anything you see. Right now there's kind of a few telltale signs. You can kind of see through it. Sometimes you're like, wow, that looks real, right? I've seen, and, and there's so many people already on Facebook where there's some like huge, you know, dam breaks and a bunch of people, a bunch of houses get whatever and they're just like, oh my gosh, prayers, I'm sending prayers, this is horrible, right? And it's like, this is an AI video, guys. Like this never happened. Like, thanks for the prayers, but you know, these people don't need them. You know, they're not real, they're AI generated. And so, you know, understanding, guys, there's so many tools out there. We've looked at during this class, we've looked at, I don't know, 15 to 20 different platforms, and I could spend another 4 hours showing you another 15 platforms easily, or 20 platforms, and, and we would have only touched 20 or 30 of the 3,000 AI platforms that are out there. There's so many, they come out every day. It's impossible to know them all. What I hope you've gained out of this course is a framework for learning. And experimenting, right? And for being able to say, okay, I don't know how to do this, but I'm pretty sure AI knows how to do this, and I'm pretty sure I know how to ask AI how to do this, how to have it teach me how to do this, right? Because if you know how to do just those things, you now, you know, hopefully gain some confidence in jumping into Replit and building something. You've built— you build a game, you know, you've built some other things, you've watched me work on a bunch of different platforms. Today and we've talked through at least, you know, kind of, you know, in, in theory and structure, in framework, how we would pull some of these things together. Um, I hope you guys are like, okay, I think I get it now. And for the things where you're like, hey, you know, is there a platform that does this and this? That's why we've got, you know, you've got my cell phone number, you've got my email, you've got the Facebook group. Ask questions, keep learning, keep asking. Don't be, don't feel like, you know, if you're asking something that you're going to look dumb. Everyone's here to learn. Everyone's here to, like, do cool stuff. And every time someone asks, you know, a bunch of people jump in and I learn. I hear someone else's answer. I'm like, oh, actually, that's cool. I need to, I need to learn that, you know. I learned from a lot of the students that have taken this course and gone on to do great things. So all of you guys are gonna do great things. You've already done great things. I really appreciate really great participation from everybody in the course. It's way more fun when it's not just me talking, even though my voice is is kind of tired. I'm done talking for the day. I'm coming up to the U next Wednesday, 2 Wednesdays from today. I'm teaching.

There's 3.

There's an AI club and a business club and entrepreneurship club, and they're all coming together. And I'm doing a 2-hour session with all those guys. Be a couple hundred students up at the U. So, I'll be up there helping your kind. No, helping some great people up there that You know, I've taught some— I've taught up there the MBA students. I've taught the MIS students and lots of entrepreneurship stuff and things.

And I love working.

I love working with students. I love teaching. I graduated at Cougar. It's okay. Good, good stuff. Yeah. No, you know, again, I just— I say it all in jest. I say it all in jest. I cheer for the Utes. I want the Utes to win, except when they're playing us. That's it. That's it. Guys, as kind of logistical, we've got 3 minutes here. And I actually have a call I have to jump on right at the top of the hour. You're going to get an email today with both sessions. I'll have the YouTube links in there. I'll have the Google Drive link. I'll put the Facebook group link in there if you haven't joined it yet. My cell phone number will be in there. I'll also ask if you don't mind, I would love a review, right? Whether it's 1 star or 5 star and some text. And if you give it to me, I'll put it on, you know, I can put it on the on the website. Um, I have like 20 reviews that I haven't put on the website, so I, I, it's just one of my tasks I need to do. But, um, but yeah, and also if you, if you want to invite other people to take this course, um, I, you know, I have just kind of a standing $100, you know, referral fee. If you invite someone and they come, then, you know, I'll Venmo you $100, or, you know, Cash App, or whatever you want, or I can send you a gift card. I'll buy you something on Temu. So yeah, anyway, I would love that. And then of course, I would love to work, you know, we have at GenAIPI, we have a chief AI officer fractional service, right? If your company say, hey, you know what, I'd love to, I'd love for you to work with my company directly. And, you know, I, I and my team members will help you. I would love to, love to work with you guys. You know, if there's, if there's the right scenario, I'm already working with Jake and we're doing some really cool stuff and that I look forward to to continuing to do that. But, um, if, if you see other businesses around or whatever, you know, hopefully if you enjoyed what you did here, you can, you can point in my direction. Um, really appreciate it, guys. Um, I will send also the, the texts, the, uh, the conversations, the chats that happened in both of these courses yesterday and today. Um, just so you have that. I think it's helpful to be able to see all the prompts and what everybody, you know, all the questions that were answered, things like that. Um, somebody asked about the Manus prompt that I did for my inspiring biography. Before we hang up here, I am going to just copy that over so that we have it.

Let's go to the very top. I have a question about, I would love to hear all of your ideas. That was something I was hoping we would be able to get to, but we never were able to.

Is there a way that we can stay connected after this? So everybody's again on that. If you're in that Facebook group, everybody should be in there. The other thing is when I sent the pre-course email and when I send the post-course email, the cohort is together. You guys all have each other's email addresses. And so you can just, you know, you can reply all and spam us a little bit if everybody wants to share what they're doing. Happy to, you know, you guys can, you guys can do that as you desire. But okay. All right, guys. Thanks so much, Jon, Nick, Jake, Carrie, Abby, Ryan. Thank you. Appreciate it. Are gone. Cole. And thanks, Jon. And anyway, thanks guys. We'll see you. Awesome. Okay, bye.