The New and Improved Way to Build ANYTHING with AI

The way applications are built these days is fundamentally changing. And in today's video we'll be looking at a novel approach to building things. Because on this channel you've seen me cover many text to application programs, but never before have we looked at something that is referred to as an ade, an agentic development environment. This is not a command line interface that builds stuff for you like cloud code. And it's not an IDE like VS code or something like Cursor that takes VS code and and adds AI on top. This is an application that is engineered from the ground up to build with both humans and agents working in unison. And in today's video, I'm exploring this workflow together with you and we'll be building a real world app that I personally actually need right now. Concretely, it's something relatively simple, but the look of it really matters to me because we'll be building a companion app for my upcoming book that is coming in early 2026. The book is all about introducing newcomers to AI technologies in ChatGPT. And this application we built today is going to be the digital companion to the book, so people can actually get some of the prompts from the book directly in their browser. We'll walk through the entire process together, from ideation to crafting the prompt, to building and refining the application. And before we get started, I want to point out that Warp, the application that we'll be using here today, is the sponsor of today's video. But as you know, we pick our sponsors very carefully and this is a new approach doing these things. And I myself especially enjoy working with companies that are really trying to carve out new niches within the AI space. And something that's labeled an agentic development environment definitely checks that box. So with that being said, let's do what I personally enjoy the most, and that's actually build something. To get started, you will need to install Warp on your machine. Luckily, this couldn't be any simpler. You download the installer, whether you're on a Windows or on a Mac, you run it and that's really it. All you need to do now is log in with GitHub. This would be my recommendation as it automatically links it then and we're ready to work. So what I'm going to be showing you in this video is not just the build of the application within Warp, but also the ideation stage. And there's multiple ways to do that. You can do that both within the LLM of your choice, or you can do it directly within Warp. If you do it within Warp, the entire workflow is self contained. It can plan, it can take screenshots, you can use various models and you never have to leave the application for me because I have my worked project set up. In this video I'll be showing you the process in ChatGPT, but I will also show the Warp variant on the side. And if you decide to do this in Warp, I really want to show you the settings that you should change in order for you to get the most out of this. So as we enter this idea stage, if you're working within Warp, you will really want to go to your profile and then hit settings here and then go down to AI. You can add different agent profiles. I'll just edit the default one that is set up right here by clicking Edit. And the crucial setting for planning that I want to bring your attention towards is the planning model up here where you can pick which model we want to use. So Maybe I don't want 03, maybe I want GPT5 with high reasoning. And then down here there's create plans which you might want to change to always ask, meaning that the agent will make plans before working and asking you about them. This is very similar to a workflow inside of any of the CLI tools if you're familiar with something like the codec, CLI or Claude code. And then as I mentioned I'll be doing this workflow within ChatGPT, but as we go through it sometimes we'll also split screen and show you what it looks like in Warp. If you want to do everything in one application and as per usual our build will start that with some ideation now to save us some time and also quite honestly to give me the space to think through this carefully. I went through this chat here with chatgpt5pro to flesh out what I really want from this application. I'll briefly show you the highlights of this so you can replicate this for yourself and also to exactly show you what my workflow in a situation like this would look like. I started set of my worked context project that includes a comprehensive overview of my company operations of everything we do at the AI Advantage, our goals, and a quarterly roadmap with the descriptions of the various products we have all in here. And then I always like to keep a Mission Vision values file inside of my work context in there I start with a simple prompt where I use voice dictation to formulate my first idea and I let that run through GPT5Pro. Now Pro is optional, but from all the models. It's my absolute favorite right now for the reason that it has the ability to consistently surprise me with suggestions that might go beyond the scope of what I ask for, but ones that I really like. Basically, I told it about the companion app for the book, I tell you about the structure and what I will provide, and I tell GPT5Pro that I want help in fleshing out this application. I end with some constraints like it not being overly complex and what the purpose of this digital application is. In the end, I just add a little bit of flavor by telling it to give me some creative alternatives and I let it go. As you can see, this is not overly specific and I don't even include the content that will be in the application or what exact format I expect. And this is how I often work because by not being too specific in my first prompt, I give it space to think. This is a brainstorming prompt, not a do this for me prompt. So it came back with a bunch of fantastic suggestions and some choices here. So for example, here in Point six it asks me, hey, these are the free delivery choices. You could do a static web app, do it, no code for notion, or you could do it inside of a specific tool like Obsidian. And then it gave me all of these creative ideas as I asked for. These are modular add ons for me to consider and so much more as GPT5 Pro loves to do. So briefly, I just want to show you what this would look like if we were to do the process inside of Warp. It's way more proactive in its planning. So it creates a project and the GitHub repository right away and then it gives me a plan. Now you can see that the plan includes free features and it proposes a technical architecture, Whereas here in ChatGPT it's just the nature of ChatGPT to be a bit more conversational. So it gives me way more options. And it's also the Pro model in here that just really gives me a lot of choices along the way. Whereas within Warp it pretty much tells me, hey, this is how I would do it. What do you think of that? And now I could accept or reject those edits, but as mentioned, I'll keep going here with the ChatGPT workflow. But the big advantage here is I get to choose from all models, right? So practically on different projects I would just try different approaches. Now let's get back into my ChatGPT workflow here and this is where I start zeroing in on my idea. I follow up with quite A lengthy prompt with some examples in here because we're getting specific at this point in time. And you can see I start up by saying create a prompt for V1 of the AI Builder app. Also there will be no login. Just to clarify that that was something that came up in here. Then I continued to describe this image that I found on the web and I said I attach an image of a prompt organizer layout that I like. No need for editing or adding new ones in this app of course. I just found this off a quick Google search and I like the simplistic layouts. The aesthetics on this are really important to me because, well, just think aesthetics matter. We're visual creatures and if I'm going to make something better look good and this is as simple as it gets, I can really see something like work with the different categories here on the left and then a clear interface on the right. If you're building apps, it's really helpful to have some mockups in the process. There's various libraries, but also a quick Google search does it oftentimes just make sure to add the snippet here to tell it what you actually like about this image and what you want to adopt. Then I'm very specific and say for 0.6 let's do a and if you remember, 0.6 was these three options down here. So we're going with the zero backend static web app. This will be perfect for the MVP just like ChatGPT said. And then I picked three features that I actually want in here. I want the dice roll where I can randomize the fields from the prompt formula to fill it in and to spark some inspiration. Also, the book includes some tone and style presets that I really enjoy. So the ability to choose that right within the website is ideal. And for people who want to use it in their automations or agents, markdown export is a must. Okay, quite simple. So basically I made some choices and gave it some concrete visual instructions. A picture says more than a thousand words and I use that to my advantage all the time. Okay, so underneath it's quite simple. I just gave it a part of a manuscript to show it some example prompts. Again, this is a beginner book that has really been refined to produce behavioral change in people. And if you're watching this video, you're most likely at the deep end of the AI pool, so to say. So don't worry about these prompts. This book is most likely targeted at your parents or grandparents if you're interested. By the way, I'LL put a link to the pre order in the description. But that's not the point. We're building this application in here and as soon as I hit enter, GPT5Pro fought for 9 and a half minutes and it came came back with something that I was actually really happy with. No further edits here. This is really everything I need. It starts with a gold building block, gives it the tech stack, it describes the layouts that I gave it via image in detail, maps out the logic, and then I can import all of the prompts and more via JSON file. It also has that functionality. So I'm just going to copy paste all of this. And now we're really ready to build because I have a very detailed prompt to start here. Now I can just copy paste the prompt and hit enter. And if you have experience with these apps, you will immediately see that this is a mix of many things that you might be familiar with. It looks like a terminal and has some similarities with VS code, but really it's a hybrid that is optimized for our human AI workflow. You will see it ask permission on various tasks like these bash commands just like a terminal would. And it does everything you would expect from agentic coding tool. And as I mentioned, it has a lot of similarities with other agentic tools like these task lists. And one tip is that if you log in with GitHub right away it has access and it will work in a brand new repo, meaning that every time you make updates you will have built in versioning. So even if you break something, you can just go back in time. And it's not just the interface that is different here. Warp performs exceptionally on various coding benchmarks. Of course the real world performance is what matters most. But some of these benchmarks score higher than cloud code, which is definitely not a bad thing. So at this point I want to point out that this tool really shines if you do know how to work with code, because it shows everything to you over here and you could also directly open the file to make changes right in here and there it is finished the task and it's writing the entire readme and that's it. It created all of this autonomously as you would expect. Now I'll just ask it to open it in my browser and it will start the development server and do all of this for me, just like I could do in a command line interface. But you might be noticing some of the differences that this required. 0 to null setup. I didn't have to install GitHub I didn't have to set up the repo, I didn't have to connect API keys. I didn't need an external application to look at what it's doing. It's all just in one Now. I am manually approving the edits, but you could just hit this and auto approve everything it's doing. And even some of these features are things that you just won't find in something like cloud code. Right now it's important to note that I'm not saying that it's clearly better. It will really depend on what you're doing. It's just different. And my hope is that for an actual project like this, you will get a feeling for what is different about it and then you can try it and make an informed decision for yourself. Okay, let's review the first version of the application. And here it is, the first version of our prompt companion. Okay, so we have the chapters on the left just like we wanted. Spacing looks good. I just say the color theme is a bit bland. I mean, especially if nothing is selected. If I dice roll. Yeah, I get the prompt preview here. Okay, so this is due for some changes. I want to hide these dynamic fields, I want to hide the examples and the return as should look differently. And then I also want to change the styling. So let's see how we would go about that in here. At this point in time, I would just want to commit this to GitHub so I have this version on there and I'll just auto approve all these edits and meanwhile I can show you around because if you open a side pane, you'd have various resources to make your life easier with startup prompts, workflows, and you could also add global or project based rules and MCP servers here. If you use some of the command line interfaces, you will know that these things are possible, they're just not intuitive. And all of these starter prompts you kind of have to keep in your head on an extra file. This really combines everything into one visual interface that is made for humans. And that sort of just makes sense when you're working, working with it as a human. Okay, so this is in git right now. Now let's make the change we talked about. No need to show the dynamic fields and example fields. And at this point I can switch up my model. So to edit existing code, I really like to use some of the more advanced models. Although this is a really simple task, usually when you're laying out a brand new project, even some of the more economic models Like Sonnet do that super well. But as soon as you start doing edits, that's when I personally like to switch up to some of the higher tier models. Okay, let me open this up, Project Explorer to really see the changes that are happening in here. So here it went into the app JS inside of my source folder, as I can see here, and it removed the buttons that I asked for. Now this is probably my favorite thing in here because you can toggle this open, you can see the exact changes that it made. It's not just going to disappear in this endless feed that a CLI interface often is, but you can really go in and see, okay, this is what it removed, this is what it added instead. And then it also removed some of the unused state variables that it set up for this. Now if I refresh the page, if something doesn't work as expected, I can just tell it restart the dev server. Okay. And there it is. It removed the two fields. This works well. So right now the main thing is the design. For me, this doesn't really work. And actually, as I was on a Warp website, I kind of like what I have going on here with their website. So how about I take inspiration from here just by taking a screenshot in the back and Warp, I'll paste that, no problem at all. And you can do something simple like adjust the styling to align with the attached screenshot. Let's see how this goes. Okay, let's have a look. Yes, this is more like it. Excellent, excellent, excellent. The prompt formula up here you can dice roll to get various examples. Okay, just this thing should work differently and this time we'll take a bit of a different approach. So I can just work directly with the code. So I can just look for this functionality that I want to change output format buttons. And I could for example, change MD to markdown, save and I don't even have to reload. And you immediately see the change in here. And if I want to review those changes, I have this View changes button where I can see exactly what just happened and it leads me to the code making my life as easy as possible, just keeping me in the loop with everything that's going on, rather than AI just running wild all over the place. Oh, and another thing is that I could open multiple windows here. So if I want to make this change in here and I want to start a brand new tab with the same app here. So in this one I might want to make the prompt format more prominent. And here I'm changing the first button you could work in parallel. And because it tracks the changes so well, I can make sure that they don't overlap. Okay, I did both of them at the same time. And here's the new version. I love how the formula, that's the key here is highlighted. This button works well. My final note here is moving this dice roll button. Oh, and here I actually caught it doing something I don't want. I don't want to review the preview card, just the title. So I can review the full changes here or refine this by saying only review the preview title, not the entire card. Now I can tell you that this is something that inside of a CLI code editor you would kind of just auto approve and then you would be reverting to a previous version after it does it all. That's a real advantage of this. More control, more overview. Because the interface is built for that. Okay, and here we see the change. Yeah, this is more like it. With the formula we can dice roll to get different examples and I think that completes the build here. Now, I might want privacy policies, more prompts, and maybe some social media links, but this design and the functionality aligns with what I wanted here. I'll just commit this version to GitHub and that's it for now. So hopefully this workflow introduced you to this new approach to a coding interface, because I think these apps are due for an evolution. If you like what you saw and you appreciate this mix of US powerful CLI and the editable code and the deep GitHub integration, a simple interface, then go check out Warp. Today it's available for Mac, Windows and Linux. And they actually prepared a deal that is almost too good to be true. Because for a limited time you can get their Pro plan for just $1 in your first month. All you have to do is go to the link in the description and use the code Igor to redeem that offer. And with that being said, my name is Igor and I'll see you very soon.