We Built THREE AI Apps Using ONLY AI in 58 Mins! 🤯 (Replit Agent Tutorial - Better Than Cursor)

Guys, what you're about to see is some of the most exciting AI tech breakthroughs that we've had in a very long time. Just like the title of this video suggests, you're going to watch me and my friend Mark build 3 fully functioning AI apps within the next hour. The first app that we're going to build is a simple pickup line app. The second one is going to be a blog generator app. And finally, the third one is a simple web scraper. Now, I don't want you to get caught up on what kind of app that we made. The reason that we made 3 is just to show you how versatile versatile this technology is, but you can apply the same process to any kind of software app that you want to build. And just a note here, neither of us are coders or computer scientists, and we were able to build these apps in under an hour using only AI. Now what you're about to see is the raw, unfiltered footage. I kept in all the mistakes that we made so that you can see the process that we took to debug these errors and move past them. Besides cutting out some of the waiting time so that we're not all sitting in dead silence, this video is as authentic as it gets. Okay, so you can access these apps plus all of the additional resources that I mentioned throughout this video using the links in the description below. So make sure you check that out. And without further ado, let's dive into it. What we're trying to do right now is just look through some popular AI apps with Mark here. And we are just going to recreate them using AI and Replit agent to show you how easy it is to actually just launch your own software. So Mark, let's just look through this list and see what catches our eye here. So basically AI for that is just a directory of like all the AI tools that's available. So let's just look through this and see what catches our eye here that we want to recreate. And we're going to choose 3 of them just to show you the whole process and show you a couple of different examples. I think like a writing one. could be really popular. What do you think?

Yeah, I think writing is good. Let's go maybe to more popular just to see what's real. There's— we have something like Rizz. We got Flirtify here. I mean, that looks pretty funny.

Oh, let's try that. Let's open that up and see what that comes up with.

What does that look like here? Okay, so update your flirting game with a little AI. So let's click into here. All right. So she or he? All right, let's try this out. So, um, let's try this maybe on you, Helena. 26, and you're really into what?

Um, AI.

Okay.

Obviously. Um, you should know me by then. What else am I interested in? Netflix?

And let's say YouTube. I think that's a fair bet. Um, hobbies on you.

Uh, cats.

Okay, I didn't know that's a hobby, but cool. Okay, unique trait.

Good at amazing at—

Good at amazing.

Good at amazing. I'm good at piano. I'm really good at playing the piano, which is not something that people know about me.

Amazing. There you go. And then how about, and she is good at deep convos, adventurous outings. Cozy movie nights. What you got?

I think deep convos is good.

Okay, let's do this. Let's see what happens. All right. Are you made—

Are you made of— Do you have a map? Because I just got lost in your eyes. Do you have a name? Or can you call— can I call you mine? Are you a magician? Because whenever I look at you, everyone else disappears.

Oh my gosh.

Okay, this is cheesy as hell, um, but—

I think this is doable though. I think we could totally replicate this, at least a low-resolution version of it. So let's do—

All right, let's do it.

Let's do that one. All right, let's go back here.

Um, I think the SEO writing one could also look interesting.

SEO, let's take a look at that. And let's go to visit website. All right, generating AI tool for one-click blog posts. Okay, so I think the only tricky part here is going to be maybe creating an image, but we could probably do the one-click part pretty easily. They're using GPT-3, so we could probably upgrade that. Tone of voice, language. Okay. I think we can get that on lock. No problemo.

Okay.

All right.

And for our last one, would you like to do?

Um, okay. Let's see here. That, but let's, uh, what about Interview Boss?

Let's take that. Um, sure. Yeah. Let's take a look at that one.

Ace your next job interview with Interview Boss. So let's take a look here. So it looks like there's some form of mock interview, there's some form of chat, and if you keep going down, there's like resume scoring. Feels like all we need to do is some form of upload function to be able to upload like some form of PDF for— um, I think it's fair game. This is also fair game.

Sounds good, let's get started. Which one do you want to do first?

I think it's obvious which one we want to do first. So let's take a crack at this one. All right. So we're going to fire up Replit Agent here. I'm just going to go back to this. And for anyone in the audience who's not familiar with how this works, right now, if you pay for the core plan of Replit, it's like a mini version of Amazon AWS. You will see something called Replit Agent on the very first page where You pretty much just give it your wishlist and it starts coding right away and trying to reach your goal. So—

Okay.

Now cheat code to actually starting is I will screenshot this page and I'll screenshot what happens when we click on the button. So I'm just going to paste my first screenshot here and the second screenshot, I'll take this. And this lets me be a bit lazy instead of prompting the whole thing. So I want to build a pickup lines app just like this with a blue color scheme, but with all the elements you see here. All right, so it's just going to say it's in early access. That means it's going to be buggy because it just came out, but we're going to click on continue. And now the really cool part about Replit Agent is that again, you don't have to have any coding experience whatsoever. It's going to come up with its own plan where it's going to recommend additional features that you might not have even thought about.

Wow.

So there you go. I've created a plan for a pickup lines app using Flask and vanilla JS featuring a blue color scheme. Yada, yada, yada. And then it offers all of these additional features. Add a generate new pickup line button to fetch new lines, which is cool because now it's giving— trying to upgrade this app where every single time you have to generate over and over again. So let's—

I like that.

Yeah, let's do that. Implement categories for different pickup lines. That sounds interesting.

Let's do that.

Um, add a rating system. I feel like this and this will be a bit more complicated.

Yeah, let's skip those 2.

Yeah, so let's click on Approve and Start, and this is where the magic happens.

If they go to replit.com, anyone can get access to this.

Yeah, the only caveat right now is it's available primarily to anyone that's on the Core plan. I think it's around $19 a month, um, but $19 a month for a personal software engineer is pretty good.

That's pretty good if you ask me. That can code in In like all languages this quick. Yeah, that's—

Yeah. So if you look here, you can see it's already created all these different files. And if you're even somewhat technically savvy, it's created CSS. So the design of the page, the HTML, which is the web page, it built the app in Flask, and it's going to give us its first draft very shortly here. So in this case, one thing that's funny is it looks like it copied The final page I gave it.

Yeah, I see that. And it didn't do the place where you can enter in the things about that person that you want to flirt with.

Yeah. Now one thing is it's super focused. It's going to ask you these questions here, like, can you see the upgrade flirting game page with a list of pickup lines? And are you able to copy paste them? Now I can, so I'm just gonna say yes. And it's going to check it off its list. Now its list has 3 things like you can see here. It's like a to-do list. So add a generate new picture line for new lines.

So it's not necessarily doing things in the order that you would expect the app to be working in.

Now, once it goes through the list, if we have to go back and say, hey, we actually want a page that does XYZ, we can do that.

Okay, it's crazy that this is how nerds flirt with people now.

Eventually it's just gonna be AIs flirting with AIs at this point. So I just want to talk about this, this part here. Um, this is super cool. So it's taking a screenshot of its own work while it's building it to make sure that it's doing a good job. So it's literally feeding screenshots of its work to a vision model to check, is it doing good progress against your original goal, which I find a really cool tool. Wow.

So that's really cool.

Can you confirm if the footer says build with heart icon? Yes. Okay, great. And yes, let's just say yes. So it checks it off the list. All right. So now it finished its first item on the checklist. So now it's going to try to put the generate new pickup button line. So let's see how that's going to work and where it's going to place it.

This is very impressive. And you can deploy this onto a domain?

Yeah.

After it's done?

Yeah, we'll get to that as soon as we're in a good place. It's literally 2 to 3 clicks away from being something you can share with anyone.

Wow.

And you can actually add a custom domain, so you can literally assign a custom domain if you want to create a SaaS app. You could theoretically do that.

Wow. So would you need hosting somewhere else though, like SiteGround or GoDaddy or something?

Zero. Uh, I just have to buy— I mean, hold on, if you want to buy the domain, yeah, yeah, you can just buy them on GoDaddy and then swap over the DNS.

DNS. Okay, gotcha. So it does the hosting and everything. You don't have to buy your hosting separately? Yeah, I mean, Replit can do all the hosting.

Yeah, yeah, there's a couple different options that we'll get to. It basically says, uh, like basically autoscale, which is like pay as you go, or you can have like a reserved machine that it's always on, but that becomes pretty expensive and you probably don't need it unless you're actually making a SaaS app. So—

Gotcha.

If we click here, it says, is there a generate pickup line visible on the page? Can you click and confirm that it adds a new pickup line to the list? So if we do that Let me see. Yeah, it is adding new pickup lines.

Wow.

There you go. So I'm going to say yes. And then hopefully there's only one more option there, which I think is going to generate categories. But the trick here is after we're done the checklist, can we make it start with that fill-in-the-blank design from before that we saw before we get to here? We'll see what's possible.

But the fact that it's working, that's crazy.

Yeah, literally, like, I'm not doing any prompting. Zero, zero prompting.

That's—

oh my gosh. Wow, it actually—

It's got a category. Nerdy, cheesy, or smooth.

That's amazing. That's actually amazing.

Yeah.

So if we click on— I know it's not done yet, but let me see if I click on—

yeah.

All right, can you see category selector on the page? Are you able to select different categories and generate pickup lines for for that category. So if we go here and go to nerdy, yes, we only have 3 nerdy. And if I click on generate a new pickup line, yes, yes. And okay, so I see a little duplicate here, but it is working. I'm okay. So that's a bug. That's a bug that we can tell it after the fact. So I'm just going to say yes.

The fact that we only been here for less than 10 minutes and it's gone this far. It's just mind-blowing.

Yeah, I mean, something like this is something you'd pay a couple hundred bucks on like Fiverr, Upwork, just to get an MVP up, right?

Right.

So, uh, back to your question originally, if you want to just deploy this, all you got to do is pick one of these. Um, the one I usually recommend is called Auto Scale because again, it's pay as you go. So if no one's using it, you're barely paying for it. And you click on set up your deployment. Okay. And then you can pick however many machines. Machines is a proxy for how much firepower you need. We could go down here and say like 2, and we can click on approve and build. And all you gotta do is just give it, um, a temporary domain. So this is a Replit domain that anyone can access. If you call it flirty-mvp. You'll see it's available. And then what's cool here—

Go with that.

It's created the databases behind the scenes where all this lives. It's set up all the secret keys on the passwords and set them up for you in the environment. So this is something that would usually take you at least 5 to 10 minutes to do manually after all the hard work's been done. So you just click deploy and then it will take around 2 to 3 minutes to actually end up deploying. And then you'll have a link that you can use in any window Share to anyone, is visible on mobile or even iPad. You can actually play around with it.

Wow, mind blown. So wait, so we're already done the app and we're already deploying it?

Yes, it's a flawed app and we can go back and we can switch a couple things, but theoretically, if you wanted to go from idea to an MVP and show someone something, we're done.

That's Crazy. That's crazy. Um, wow, I'm speechless. Doesn't happen very often.

Yeah. And then here's your link, and if you click it, okay, it's a web page. Functions. Go to nerdy, go to cheesy. You can generate on demand. So it's what it seems like is it's not using OpenAI to generate these. It built some form of database full of them that's kind of spitting out, but we can change that completely. Um, it's the idea that I can go from here in an incognito window where we're not signed in at all, and then it's all good and all the copies work. So if it says copy and I paste it in my little bar here, it works perfectly fine. Nothing impressive, nothing to write home about, but you did this in literally under 10 minutes.

Yeah. That's crazy. And then we can update the prompt so that it has the thing in the app that we just saw where you can put in some fields about that person.

So let's do that.

Okay, so—

Let's screenshot this again and then say, this is— you have to be very specific with it. That's what I tried multiple times. Create a landing page where you Emulate this experience of filling in blanks to generate initial pickup lines using OpenAI. So it's smart enough to ask me for my API key, which I will give it.

Okay.

All right, so I'm adding my OpenAI key here. I'm going to click on Add Secret. All right, and once we go through that, it has what it needs to actually build it. And what's cool is OpenAI has had different ways that you can call its API over time. It knows the latest and greatest because I think it has some access to Perplexity as well, which is super helpful.

Okay, it gives you a home button now.

Yeah, so now we have a home button, so I'm getting my, my hopes up. Okay, and it says AI-powered generator, so I'm just waiting for it to stop thinking because that means it's done with all its background coding.

This is so cool.

And there you have it. Let's go. So now we have the AI-powered generator. Now it doesn't look exactly like before. We could probably go back and forth, but we can say, let's do that again. 26 interests. Uh, if anything, it's actually trying to be more efficient instead of saying interest 1, interest 2, interest 3. So yeah, Netflix, YouTube. Piano, deep conversations. I got that right, Helena, right?

Sure. Okay. And what's my personality?

Very, very nice.

Okay. All right, let's see. Oh, it has an error.

It has an error. So this is going to happen all the time to you in whatever you use. So we can see here. Okay, I'll literally say, No, it didn't work. I got this error. Again, you don't have to have any coding knowledge to be able to look at error, give error to AI, and let AI try to fix it.

It even says, I apologize for the error.

Yeah, crazy. What's funny is sometimes it'll act like there's a team and be like, okay, I'll take this to my team right away, even though it's literally just one person. So, or not a person, one AI.

Okay, it restarted it already.

Almost there.

Wow.

Fingers crossed it works.

Imagine we got 3 working apps in less than an hour.

That would be so cool. Okay, let's give this another shot. Let's do again. It thankfully remembers, so let's do this.

Oh no, it gives you an error again.

Here we go. So let's go here and say Yeah, I'll give it a little cheat sheet. What I'll do in this case, what I would do typically, I just go to Perplexity and say, how do you call GPT-4o in OpenAI via Python or anything? Again, I don't have to understand what it's saying, but I can get Perplexity to do the cheating for me and ideally maybe copy paste it to remind it on how to do it.

That's a great hack because Perplexity is the best, like, search, you know, in terms of search, Perplexity is the best LLM model for that.

Yeah, absolutely. Okay, so I just used Perplexity to come up with what the right— what model we should be using because this is like a very old model. We want to use maybe GPT-4o, the newer model. So I just took this code that I found through Perplexity on how to actually make the right call. And I'll say, this didn't work. Use this code syntax below to call GPT-4o.

The fact that it even knew to ask for your GPT key and input that is just crazy because I built a Jasper lookalike in Bubble. Last year, and all the, like, for loops, like, I ran into so many, like, infinite loops when I was building it. And then it used up what Bubble called WU, or work units. And then all of a sudden, it was like, you blew over $20, because, like, I made, like, one coding mistake. And that was—

Yes. When we compare how annoying that is to this, this is nothing.

Yeah, for sure. And even with the no-code editors, it's like, if you don't know code, it's just hard to wrap your head around that logic. You know what I mean? It's kind of like, instead of writing a for loop in code, they just made a for loop with fields that you fill in instead.

Yeah, it's like a hybrid, hybrid language.

Yeah. So, if you don't know what for loops or while loops are to start with, I think even going to a no-code editor, you'd be like, what exactly am I looking at? Right? So, but like, this is exponentially easier. Easier and better. Yeah. Okay. So let's try and see if it's working now.

Fingers crossed. So this time we're going to add very, very nice and funny. Maybe that's the reason why it's not working.

Okay, okay, thank you for compliments. Let's see, fingers crossed.

Oh, it works! Oh my God, I clicked it like 6 times. That's why it's doing this. But it did generate the pickup line. So I'll just say, wow, good, good.

Are you a Netflix series? Because I could binge watch you all night, all day. Maybe we can share some YouTube laughs, tickle piano keys and dive into deep conversations. I promise I'm just as intriguing as episode 12. Cheesy as hell, but oh my God, it works.

It works finally. So if we say yes, that means it's finished its checklist. We could keep beating this dead horse, but I'd rather move on to a different app just to see what else is out there.

Yeah. And can we include a link to this so that other people can try it?

Sure. Yeah. Well, if we see too many requests, we'll cut it off. Otherwise we'll be bankrupt pretty quickly.

Yeah. Okay. Well, if you still see in the comments below, that means that we still haven't bankrupted ourselves with this app.

Not yet. Yeah. Okay, cool. So which one would you like to do second, Helena?

Let's do the SEO writing one.

Okay. So what we want to do here, and I think it's helpful to walk through how we both think about the approach. The approach is everything. Like we just saw, because I gave it 2 screenshots initially, it got confused and it copied the second screenshot. So let's try to be simple this time and straight to the point. So what's the best way? So we want to include things like language, tone of voice, ImageGen. I am skeptical that it'll work. We can give it a shot.

I think if we were using DALL-E 3, I don't see what the problem would be because that could just generate from the OpenAI API key still. Completely.

As long as it knows what DALL-E 3 is, then we should be good. So let's say, let's create an SEO enriched blog creation platform where you can create a blog from just writing a topic and clicking one button. Show— allow the ability to change tone of voice as well as language and add a Picture for the blog when generated. You think that's good?

Okay, yeah, let's try that out.

All right, let's give it a whirl. So again, it's going to go back to the drawing board, come up with a plan. I think it's complicated enough that I'm going to reject most things on the plan, but we'll see what happens.

I'm still mind-blown that You can just write 2, 3 sentences now, and then get a fully functioning app for like $20. Within 5 minutes, you actually get something.

Yeah.

Like, this is like, I'm just mind blown still. I'm still trying to wrap my head around this. It's crazy. That's— this is the most mind-blowing thing I've seen in a Yeah.

And we'll just get started. Yeah. So we can see here we have implement user accounts and save blog posts. We don't need that right now. Um, add social media sharing. Interesting. Uh, integrate with popular blogging platforms. Sounds complicated. Implement advanced SEO analysis tool with keyword suggestions.

I like the last one out of the 4.

I don't know what we're gonna get, but we'll give it a go. I know for a fact this is not gonna work now. You see here, from OpenAI chat completion? Yeah, that's not gonna work.

Okay, so it's basically still learned from the old way of calling OpenAI, um, and we just need to tell it to correct it to the new way.

Yeah, the irony is it's using the right thing for here, the completion, but I see something here that's a red flag, so Now it's marked it in red, which is interesting to me because I don't know if it knows it's wrong. And you see this time how it actually— it called it perfectly here, whereas before it called DaVinci 2. So it makes you wonder, like, why does it know and not know?

So it's redoing it now again where it's continuing— oh, it's continuing its work now that it has your API key, and then we can fix the We can fix the bug later with the OpenAI. Okay.

Yeah. Now, um, this library NLTK is for natural language processing. This is like what you'd use before OpenAI existed. So I'm curious what it's going to do with that. Imagine. Wow.

Oh my God.

Yes.

Wow. We already got a prototype.

Now, does it work? So if I say something like, Eating pizza is good for you. And I say humorous.

Okay.

And I say generate. Now when I click the button, it says an error occurred and I have no idea why. So what I want to do is laziness.

Probably the open.

Yeah, I'm going to screenshot it, but like the button, yeah, it didn't even work. So let me see here. I'm going to say, got this error? You probably imported OpenAI wrong. And the button doesn't do anything other than this error.

Another thing we could add to that is the SEO blog creator is putting in the keywords in.

Yes.

Field for keywords, maybe.

Now that's like the next thing. Implement advanced SEO analysis tool with keyword suggestions.

Oh, it's already got that on its agenda.

But I guess we could also add keywords input.

It's already one step ahead of me.

So let's try this again now with the new version. Fingers crossed.

Let's go.

That pause tells me there's gonna be another error.

Uh-oh. Let's see what it is.

No idea what this is.

This time's different. So we just copy it.

Yeah. Does it allow me to copy it? Okay, I'm just gonna screen— screenshot this and I'm gonna say— okay, I'm just gonna screenshot this. It's not letting me attach it.

Oh, there we go. Doesn't want to be attached.

We don't need NLTK anywhere, just OpenAI.

Because you did comment on it when I was writing the code, why, where it was going for it with the NLTK stuff.

Yeah. So one really good thing for anyone that doesn't code is if you see a random library that's literally not OpenAI and you're building an open— sorry, an AI-driven app, Then just ask either Perplexity or Google, what's this? And if it has nothing to do with it, I question why it should exist to begin with.

And it's good that we're running into these errors and people are watching how we would solve these bugs, you know, instead of just showing a polished version of the finished app. And they're actually seeing the process of how, how to get there, you know?

Yep. Yeah, the journey's full of potholes, but it's all about the end goal.

Yeah, it was still very impressive, like, that, you know, you basically have an almost free software developer on demand 24/7. And all you need to do is to tell it in natural language what you want it to do. I was at a conference once, and, and the person next to me, it was an AI conference, said, isn't that crazy that This is as bad as AI will ever get. And that just stuck in my head, that line. So yeah, this is as bad as AI will ever get.

Yeah.

Let that sink in.

Imagine a year from now, we probably won't have to do this error thing. It'll just know how to take the error, use Perplexity, use whatever, look for it, et cetera.

Yeah. So now it's asking you to test it again before it does the SEO optimization. So let's try it.

Whew. I feel like we're gonna get another failure.

Hopefully this time.

I don't see any action. I don't see any loading, nothing.

Hmm.

Yes!

Oh, it worked. It worked.

We got something. We got something. Okay, so it says, can you fill out the form? Okay, please check if the blog post has an image. Okay, so It generated content, but no image. Maybe use DALL-E 3.

I feel like I'm a kid in a candy shop right now, like watching this.

At least I love, I love the progress part. And like one thing we should do after this is done is make sure that when you click generate blog post, there's some form of like loading, cuz right now there's no feedback.

Right, right. So we don't know if it's working or not. That's great. Yeah, maybe like a spinning thingy. Yeah, a spinning thingy indeed. Yeah, very technical terms that we're using here. So when this was done, Mark, how did you get to the deployment part, or does it show up automatically?

Top right. And the thing is, once, once it's done the checklist, it automatically goes to the deployment tab.

Gotcha. Gotcha. That's how you got there.

All right, let's see if this got a bit better. Yeah, I definitely need a loading thingy ASAP.

Yeah, that would be our next prompt on this. And we'll include the link of all 3 of the apps we built today below. Just don't abuse it.

We got an actual—

The image worked!

It works. It works. Amazing. This is awesome. Okay, good. So we can check this off. We're going to say yes. And now it should go to the next thing, which I think was the advanced SEO analysis.

Yeah. So now it's— now we're trying to get it to write better because like what it wrote, I mean, it wrote something.

Wrote something.

Yeah. At least it's working. That's like this. We're doing this in real time. Like that was less than 20 minutes that I made that.

Yeah. And it's crazy. If you click on new tab here, like you can see it's It's— you can use it. It's a web page. And like, think of the hundreds of AI apps out there that make blogs that you just make in-house now. Wow.

And then you can also like upload samples of your writing and get it to like learn, learn that so that you have like a custom tool within your own organization that's going to write blog posts in your own tone and style. That you can, you know, don't even have to train your whole team on prompt engineering anymore. Just say, like, use this internal tool that we have. Yeah, exactly. You can also do it for proposals. And this is going to raise the valuations of your business when you have proprietary tools like this.

Yeah, completely.

Right. These are kind of things like, I mean, I'm not a Silicon Valley venture capitalist, but from what I on YouTube, this is what they're looking for.

Yeah, absolutely.

So, and I just want people who are watching this, like, to also not get caught up in the examples we're doing, but really use the process that we're doing and apply it to the businesses that you run. Think about the processes that you're using and thinking about how AI can be plugged into it. And then creating your own AI tools has Never been easier.

Yeah. And just, just this weekend, like, I built a lot of apps for myself, quality of life apps. So people send me tons of voice notes on WhatsApp. So I literally built an audio transcription thing that has a button where I download someone's WhatsApp voice note, I upload it, and it transcribes it for me. Very basic, but you can start building also quality of life apps for yourself to make yourself that much more efficient.

Yeah.

So one thing I want you to take a look at here is the fact that if you go through Everything I've been saying yes to. One of those things was I'm updating the JavaScript file to improve error handling. Now, this is something I didn't ask it to do. It's diligent enough to actually go through and try to be proactive for what might happen when we use this app, which is like one of those small things that an ideal employee of yours would be doing without you even asking. So I just wanted to add that in as a cool feature.

Wow.

So if we just say yes. Wow.

And if you want to brand this, I guess you could also like just tell it, add my logo in here, drop your logo. You know, you can also tell it to clean up the CSS, like I want the button to be pink or something, and it'll do that kind of thing, things as well. I'm just mentioning that so that people know that it's possible to do all of those things as well.

Yeah. And one thing that's super important, having worked with this a lot, is once you've given your first instruction, everything else, give it one instruction at a time once you're done all of these prototypes. So once we're done here, don't say I want this, this, this, and this. It'll look for the very first thing and start with that. So one major revision every single time.

Wow.

So now we are officially done with the build the initial prototype and we're going to the 2 of 2 step, which is implement an advanced SEO analysis tool. I have no idea what that's going to look like.

Okay, I'll be in for a surprise. Does it also do things like logins and create— allow you to create users as well? I think that would probably be a question that comes up.

Yeah, yeah, you can, you can go back and forth. I tried authentication and registration this weekend. It knows how to make databases to store user accounts to register. Sometimes it hangs up on that, you have to go in a bit of circles, but it's completely possible to do. It's just a matter of wording it properly. Sometimes if you word something vaguely, you don't get exactly what you're looking for.

Gotcha. And it can also allow the user to store all their information and create new accounts. It can do all that?

Wow. Yes, it can create a whole database.

Wow.

So if you go to the bottom here, PostgreSQL, it's built into Replit, so it can, it can set up a database within Replit itself that stores everything.

Bubble's going down.

I mean, Bubble and FlutterFlow, their days are numbered.

Wow, crazy. Okay, so can you generate a blog post now with a keyword suggestion? I don't see the keyword suggestions.

So we're going to scroll down and see if it— hey, there we go. SEO improvement suggestions.

Oh, that's super cool.

So for the purpose of time—

Cool.

Nice picture.

Let's go on to the next.

No.

Well, you know what?

We'll say yes just so we can kind of get to the end of this one. And the one thing I really want to do before we sign off on this one is we need that loading bar. I will not leave without that loading bar.

Okay, so let's do that.

We need to say loading bar when we click generate. blog post to know the progress of the creation of it. So now it says it's done. So if you do this and change it to, um, watching YouTube all day, and we make it formal and make it French.

Let's see. I don't see a loading bar.

That's no bueno.

Did you click the button?

I'll try it again. Unless it's a very small— maybe I should refresh, possibly? Try it again. Oh, there we go.

Oh, it has a loading bar.

Holy—

oh my God. That is insane.

Yeah, that means— so also, that's a— it's really good this is all happening in front of you guys. So like, if you make a change and for some reason nothing's changed, click this refresh button on the top left, and that might be the little thing that's off.

That's insane. Wow. Wow. Like, You literally just had to tell it what to do and it does it.

Yeah, the final test is making sure that when it finishes, it actually shows up. There we go. Wow, amazing.

Wow.

There we go.

Wow.

Okay, now I'm at peace with this one, so we can try the last one. And you said, Helena, you want to try something different, not the last one we picked. You had a different idea.

Yeah, let's deploy this. We can see, we can get the URL for this for other people to try if they wanted to.

Sure.

Again, guys, just don't, don't abuse Mark's API key or else we will have to shut this down. So you just have to go to the deployment, right?

Yep. And then again, choose autoscale. Set up deployment. Approve. And then we're going to go down here and then close— call this Blog Post Creator. And then we'll click deploy. And then again, it'll take a couple minutes and then we'll have a link and then you're good.

So if I wanted to use this on my own domain, where would I go to do that?

I will show you as soon as it's done because that's where it'll give you the option. When you click on that lock icon, it'll give you the option to start entering your own domain.

Okay, all right, so we just wait a second for that to show up. Yeah, so now even people who haven't studied comp sci in school can now create apps.

Yeah, pretty much.

So can you even tell it to do Stripe integrations to start charging money for your apps and stuff too?

I tried it on the weekend. It took a bit of back and forth and I had to give an example of how to make the call from Perplexity, but it was eventually able to get it.

Okay, so it just takes a bit of back and forth to do it, but that's still pretty impressive.

Yeah. So, all right, back to your question. If you want to go to settings, then you'll see this little link a domain. You can click on this, and this is where you can start the process of having your domain link be the primary app link.

And then you just change the DNS record with your domain provider.

Exactly.

Got it. And you can make this Replit app into like multiple pages too, right? Like you could do like first page is Is, um, you know, for something like this, you can even use it as a lead magnet. Like, the first page will be name and email, collect that, and then they get access to the tool, right?

Yeah, actually, it's a great lead. Yeah, yeah, real quick, um, if you go to Replit, literally the number one thing that they tell you that you can make is a waitlist generator. So this is built in, it comes with a database that stores everyone's names. So when it comes to lead magnet, It's done. It's doable out of the box. And if you want to know everything else that they're confident on making, click on View More and you'll see Local Landmarks app, waitlist website, link in bio, AI chatbot, like actual chatbot you can create on a screen, and then like a smart form and some form of visualizer here. Wow.

Kind of reminds me of what Neil Patel did. So he made an SEO tool where people can just put in their name and email and gave it for free. And that was his lead gen. Like software is one of the best lead gen magnets out there. And then what he did behind the scenes is look through who's using his tool, and then he would scrape off, you know, the top 1% and then convert them into clients for his SEO agency. So it kind of reminds me of that, like, but now you can do it so much faster because like, when he was doing that, software was still extremely, extremely expensive, right? But now you can do it such a small period of time. Everyone should be doing this for their lead gen.

Yeah.

So, Mark, instead of doing the third app as the Interview Boss, like we originally were thinking, I think it's just quite similar to the SEO one that we just did.

Yes.

So why don't we show them something a bit more complex? Let's try to connect the frontend of the Replit agent to a Make.com automation and get it to, like, do something. So here, what we have set up is a very simple scraper. So I have a webhook here. Which is a scraping webhook. And basically, the person can put in a URL and then have an HTTP request, which will take the website that was just scraped from the webhook. We still have to fill in the URL part. And then lastly, it's going to go into a ChatGPT module and summarize what that company does. So it's a very simple scraper. You can also, also like go to the text parser here, add in a regular expression to scrape out— scrape things like their phone number or their email address, you know. So you can really, um, get as fancy as you want here. You can, you can scrape that, that company's information, then write a cold email to them that's personalized to them. So the use cases here are really, uh, limitless. Um, so we're just going to keep it simple in this example and just summarize what the company does though. Okay.

Yep, let's take it. Let's do it. All right, so for this one, I'm going to do something a bit different. So if you've ever used ChatGPT and you use something called custom actions, you give what's called a schema that basically just explains what this app does, and it has the webhook that Helena just created. So we're going to include this in the description below. You can use this as your mini little cheat code here. And pretty much what it looks like, if I zoom in just a tad, is you have the name of what it is. So the Company Information API, you have the webhook that we just created. And then in this case, I'm telling it like the main method is going to be get company summary. And this is all generated by ChatGPT. I'm not that smart to actually make this on a short notice. So all I'm going to do is I'm going to close this, I'm going to upload it, I'm going to attach it here, and I'm going to say Create a simple web page that allows me to enter a URL and then send that URL to Make using the webhook in the schema attached, and then display the response from the Make automation on the page after Clicking a button called Scrape Me. And then we'll click Continue. And then I have no idea if this is going to work in one shot or not work at all, but we'll see how it works.

Yeah, yeah. It's good that we're like showing it live step by step and people see that this is all the back and forth that we had to do with the AI to actually get it working. And I think that's very beneficial to show that, and that it's not like one prompt and you're done. Not yet. I think we're going to get there soon, but right now it's still a few prompts to get there.

Yeah, yeah. Sometimes it could take 100 depending on what you're trying to build. But yeah, I mean, for the most part, on average, between 5 to 10 is decent to build a prototype of most mini apps.

But if you think about it, 5 to 10 prompts to build a web app. And I think for anyone watching who has ever attempted to build a web app knows that is an exponential improvement. Like, I've built a few apps, I don't know, like half a dozen on Bubble before. It took weeks to build one, and that was That was like an exponential breakthrough because before that it was like you needed a team of developers, like 10 people and 6 months to build any kind of apps to start with. Um, so it just shows how quickly technology is evolving.

Um, yeah, and one thing we haven't even talked about is we've been relying on this chat solely to actually build this app. End of the day, it's actually creating all these files, right? So if you know where, I don't know, the web page lives, HTML or the CSS, you can go to those files, put them into ChatGPT, and, you know, expedite the process by telling it, hey, make this blue, make this red. So all the assets are made here.

They're just—

this just helps you get those assets faster.

Right. And I want to add here, you can even take that code into Cursor. And ask Cursor to explain what, what that, those codes mean.

One thing when it was processing that I looked at was it says for some reason the webhooks EU. I don't know if this is a placeholder and it has some form of like secrets tab, which usually does. Um, I don't see anything here. So if it doesn't work, then that tells me that we need to correct that. So sometimes also when you're troubleshooting, it's helpful to go through the progress bar and see What it did, because maybe it missed something.

Right.

So if I get an error— okay, so see here, not found URL. Okay, so that's why it's good to look at it while it's actually generating. Uh, we need—

Right, because—

I'm just curious because it gives me the exact same error again. Okay, so it says accepted. We're making progress. So we're making progress. Did you get any history on your end? Because that tells me that you got a ping.

Okay, let's check that. Check it out.

Also, I just realized something.

Yes! Oh, I just realized I got a ping!

I realized something. You don't have a webhook at the end of your automation, so I get no response.

Oh, okay.

So now let's reset up Our, um, so do you wanna—

our automation?

Yeah, how do you wanna go through this?

Yeah.

I guess let's, let's make sure it works and then we'll go through everything we did to get there.

Okay, I'm gonna add in the webhook response in the meantime.

Okay, it did. Yeah. Oh, so did you make the change?

Yeah, yeah, I added in the response.

Interesting. So now I'm getting a 500 error. That's a Make error.

Let me see what the error is. Oh, I set the GPT tokens to 500 and there wasn't enough context.

Is johntheplumber.ca that crazy, bro? Damn. I guess there is quite a bit of text.

Oh, you know what? I'm going to do a text parser in the middle. HTML to text to scrape out all of the extra HTML code because all of that is taking up the context window.

Okay, I got you.

You see what I'm saying?

Yeah, yeah.

And then it's saying it's going beyond the context window.

Yeah.

Then I'm just going to do the text and increase the maximum tokens. Okay, now do it. Should be okay now.

I'll go back to the old one.

Okay.

Man, I hope this works. Yes. Well done.

Yes. Woo! We finished it though, bro.

I'm happy we made progress.

Okay, so we made a Make automation here. And what this Make automation does is that it's a webhook. It's going to listen for a URL and it's going to then scrape that URL, and then we're going to use ChatGPT to summarize what that business does and then return the result. And we can— we want to now connect this Make.com automation to a Replit agent on the front end. Now you can have a front end for all of your automations, and you can have a working app Within just a couple of hours. So again, just to walk you through what this is doing. So first we have this webhook, right? This webhook is going to listen from Replit. So whenever someone enters in a URL from Replit, it's going to call this webhook and this webhook is going to be triggered. The way that I like to explain what a webhook is to people looking at it for the first time, it's kind of like If let's say you order something on Amazon, you're not going to sit by your door for the next 2 days until that delivery driver comes and drops off your mail. You're just going to go about your day, and you're going to go to the door when your doorbell is rang by the delivery person, right? The same thing here. Basically, the webhook is kind of like the doorbell to your software application. It's only when that webhook is called, when this when something triggers this webhook, then the rest of this automation will run. So then when— so when this webhook is triggered, we're going to use this HTTP request to call the URL that the user has. And then we're going to use a text parser. So you can get the text parser here at the bottom. And I use the HTML to text parser. And this will scrape out all of the HTML code from the website and just return the text to us. And then I used the ChatGPT module and I just summarized what this company does. You can easily replace this with write a, you know, cold outreach email to this person. You could also add in another text parser here for match patterns to scrape out any email addresses or phone numbers on that web page. So really, like, the possibilities here are endless. You don't even have to do a scraper, you know, so you can connect it to any automation you want here. But we just want to show you the breadth of what this technology can do here. And then lastly, after we got the summary here, I just did a webhook response to send the results. So what we wrote in our summary here back into the webhook so that it can be displayed in our Replit agent web application. So that's a walkthrough of what this Make automation does.

Most of the errors were caused by the Make automation. Maybe we didn't parse out the URL properly. So that's all taken care of now. And my initial prompt was—

We ran into a few things getting the webhook to work and then determine the data structure for the webhook going back and forth. So when you start using multiple applications, then there's more variables at play. You have to figure out, is it— is it— is the error on the Make automation side or is it on the Replit side? So anyway, Mark, take it away.

Yeah, absolutely. Yeah. So again, this was our prompt and we'll include it down in the description below. And I gave it the webhook. After we went back and forth, we were getting some trouble that kind of Helena just went through how we managed to mitigate it. And every time I got an error, I would basically screenshot that error and give it back to it. When you get something like a 400 and it says client error, that usually means that it's on the Make end. It's not on the Replit end, it's not on the web app end. So even those clues, even if you're not a coder, if it says client, that means The other side. So we just kept sending those screenshots of those errors until we finally got this message that said accepted. And then we knew, okay, it's probably a good sign. So now we're at the final stage where if we start using an actual URL, like John the Plumber is getting free coverage here. If we take this and we throw it into the scraper and we click on scrape me, This should be kicking off the webhook in Make.

Yeah.

And it comes back with a summary of this entire site as to what they do. And you can see this is a pretty beefy site, and all we're using behind the scenes is GPT-4o Mini to actually come up with this.

Ain't that cool? Um, another thing I just want to mention is that when we were first trying to create this scraper, uh, it kept on Putting like hallucinating and putting in a random webhook for the Make automation. So just watch out for those hallucinations and when you see it happening, then correct it as well.

Yeah. Yeah. So when you go in progress, when you're first watching this, obviously it's magical, but as you're watching it, look to see what are the actual variables it's setting. So you can see here originally it made it like an eu.make.com link. With no ID.

And then we had to like enter in another prompt saying, hey, you hallucinated this webhook, put in this webhook. And we entered in the exact webhook that was from make.com too.

Yeah. And what's super cool about the progress tab is you can see the different changes that happened with all the back and forth. So you can see when I actually took a screenshot, you can see when we gave it the right Make webhook, what it did to override what it did. You can use this as your log to see what it's done. Now this works, and obviously what you could do to build on it is you could create this so now it takes 10 different URLs and scrapes all of them at the same time. Maybe it looks for phone numbers, maybe it looks for emails. So you can keep building on this, and the idea of showing you guys Replit Agent is that you can start really upskilling yourself and tinkering over hours and building something pretty robust.

Absolutely. Absolutely. Well, Mark, thank you so much. This has been so insightful, helpful, and such a game changer. One of the best, most mind-blowing tools I have seen in a very long time. And I hope everyone watching this will go and try it out. It's going to really help your business. Um, there's so many use cases as we discussed throughout this whole training. You can use it as a lead magnet. You can use it as internal tools that can help you speed your processes, the possibilities are just literally endless, right?

Yeah, absolutely.

Awesome.

I hope you guys enjoyed it. And, uh, yeah, if you guys want more applications, let us know in the comments below. We're happy to hop back on and crank these out and play detective if we need to.

Sounds good. All right, see you guys next time.