OpenAI Codex AI Agent Builds Custom Software in Minutes - Full Tutorial

Ever wish you could have your very own software developer who can write code in any language, refactor the code, write test cases, debug errors, fix those errors at any time, any day of the week? It is now possible thanks to OpenAI's new release called Codex. Codex is your all-in-one AI coding agent. I tried it out and it is incredible. Now, it is a game changer because it allows you to modify existing code bases and add on new features to it, fix any bugs you have, And this will make any software developer's job 10 times faster. Now, before we dive into the tutorial today, I just want to recap the current landscape. So if you don't currently know how to code, there are 2 main ways that you can launch your very own software. One is white labeling other people's software. So essentially, you take an existing piece of software and you slap your own logo onto it. Now, the most prominent example of this is GoHighLevel. The downside of this is that you don't get to make customizations, and it's pretty easy to figure that it is a white label at this point. And the second way that you can launch your very own software is by using vibe coding. So platforms like Bolt or Lovable can help you do this. Essentially, in just a few prompts, you can create your very own basic software. Now I want to emphasize on basic. These softwares that Bolt or Lovable can create are pretty rudimentary, and where they lack is the backend. So it's still pretty difficult using these platforms to have a robust database to have things like user authentication, being able to take payments, etc. Where Codex is different is that it allows you to take an existing robust codebase and be able to add to it and make your very own unique pieces of software from that. That is exactly what I'm going to show you how to do today. So unfortunately though, Codex is currently only available on the Pro or Team plan. I know the Pro plan can be pretty steep at $200 per month, but considering that you get an entire software developer anytime time of the day, never takes breaks. Like, that is still a very, very good value. So with that being said, let's now dive into ChatGPT and see how we can start using Codex. All right, so once you're logged into your ChatGPT account, you will see the Codex tab on the top left-hand corner. I'm going to click on Codex, and you can see that this interface will load. Basically, you just have to describe in everyday English of what you wanted to code. Now, the first thing we have to do before we start coding though is connect Codex to our GitHub account. I'm going to click on Connect to GitHub. Okay, so after we have connected Codex to our GitHub repository, we have to select the specific repository within our GitHub that we want Codex to work in. Now I want to show you a couple of things here. So once we get into this settings portion, we can see 3 tabs. The first tab is General. This is where you can give Codex custom instructions. Now if you go on to the Codex release page on OpenAI and scroll down to the bottom, you actually see the default prompt that is inside of Codex. If you want to make modifications to this prompt here, just make your modifications and paste it into the custom instructions here. Next, we have the environments. These are the current repositories that I have linked into my Codex account. And lastly, you have data control. This is where you can turn on or turn off allowing the training on your data. I suggest having this off so that your data isn't being used to train future data. This is just a privacy and security thing. All right, now let's go back to Codex. Now if you're like, Helena, I don't know how to code, I don't have any GitHub repositories, I'm gonna now show you how you can go into GitHub and download entire softwares that are open source. Then you can use Codex to make minor modifications to it so that you have entirely unique pieces of software within a matter of minutes. Like, this is just blowing my mind, guys. The cool thing about GitHub is that it's also a marketplace. It's a place where different developers can collaborate and share code with each other, and a lot of this code is open source. I'm talking about entire softwares that are currently available. So if you go to github.com/trending, you can actually get some inspirations here. Like, there— this is all code bases that you can download. You can also go up at the top if you know the software that you want to search for. Now, there are like so much open source code. Like, if you want your own version of Bolt, did you know like Bolt is actually open source. If I wanted to copy Bolt, I just click on the fork button. So fork basically means clone. I already forked Bolt, that's why it's not letting me do it again. I could also fork DeepSeek. So if I just search for DeepSeek here at the top, I can see that like all of the code that I need to run DeepSeek locally on my computer, I can do so by just pressing fork. Again, I already forked this, so it's not gonna let me fork it again. But the cool thing about like things like this is that now Now you can download all of these code and make modifications to it using Codex. And then you got a powerful piece of software that you can create now with no code. So for example, for something like DeepSeek, what I can do is maybe add in my own fine-tuning instructions, my own custom prompts, or my own training data. And then I have my very own customized LLM that I can now slap my logo on because I can just build on top of the existing open source code. Okay, another example here I can show you is called Twente. And basically Twente is an open-source CRM software. Like currently I'm paying $497 a month for HighLevel. If I wanted to create my own CRM with my own customized features, what I can do is build on top of these open-source code that already exists. So Twente, like if we clicked on their website, this is what it looks like. It gives you a dashboard where you can store all the companies and people and tasks that you have going on in your company Like having a custom CRM like this for your own company is great because you can have all of your data centralized into one place, right? It also has these tasks that you can move into a Kanban board. You can sync all your emails together. Like there's a lot of things that you can do with this exact existing platform. Now to create a copy of all of this code, just go up here and click on Fork. You can also check the ratings here. So I know like this code is a good codebase because it has almost 28,000 stars and And it has 3,000 people who have already forked it. So once you have forked this particular repository, what you're going to do is go back to Codex and then go up here to environment and then click on create a new environment and then search for the repository that you have just forked. Let's say it was Bolt, for example, right? Then we're going to go down here and add in any environment variables or secret keys that we want to be included in this code. We don't need any in this case. All right. Once I've done that, I'm going to click on Create Environment and wait a couple of seconds until this is all loaded. All right, so you can see here now that I have my new environment created, and you can see I have a new repository here. Okay, so now that we have added our new environment, we have to make sure that we're working with the right repository. To make sure of that, just click on this dropdown here and make sure that the right repository is selected. So we just added in the bolt.new repository, which a fork or a clone of the bolt.new code. And now we can make it our own. So the first thing that I usually do when I'm working with a new repository is just to figure out the lay of the land, where things currently are. So the first thing I'm going to ask Codex to do is just to explain this codebase to me and the structure of the code. And then you can see here, once we have put in exactly what we want the AI to do, we have 2 options. We can either ask it a question or ask it to code. So because this is a question, I'm going to click on the Ask button here, and you can see that now we have this task running. Another task we can do is find any bugs in the code and refactor the code to make it as efficient as possible. So refactoring means taking a look at the code to see if, if there's any duplicate codes or if there's any code that we can condense from this To this, whenever you can shorten your code, your code is going to run faster. So usually, if you're a project manager who is in charge of creating software, you will leave some time on your project schedule for refactoring and testing. And now you can have like AI do this is just quite incredible. So I'm going to click on code because this is where I wanted to code. And then another task that I can put here is this is a fork of bolt.new. Vibe coding platform. I want to make this my own, so can you suggest 5 to 10 different features that I can add into my own version? All right, so this is a question, so I'm going to click on Ask. Now while the AI is writing all of this code, you can actually click into any of these. So I'm going to click in the first one just to show you, and you can see the process it's going through to work on the task that you wanted to do. And then I can click here where I can click on View Log, and I can see all the steps that it's going through and taking at the moment. At the top here, I can also see the 3 tasks that I have it working on, and I can toggle to another task here. And you can see that it's now working on the refactoring code one. And I can toggle on the progress of the third one here as well. And you're not limited to 3, by the way, you can have like 100 different tasks being executed at the same time. Like if you currently have a software, just take your entire roadmap, like every single task that you have on there. Now you can put all of those tasks in here. And the AI is going to like do all of those synchronously for you, even if you have a human developer, they can't work on 10 different things at once. At most, they might be able to do 2 at once if they're good at multitasking, but usually they can only do one at a time. So this usually will take about 2-3 minutes for this to complete, and I'll be back once everything is finished. All right, so it's been about 2 minutes or so, and we can see that all 3 of our tasks are now complete. So let's take a look at how well it did. All right, let's take a look at the first task that we told it to do, which is to explain the code structure. And we can see that it has done that. We— it explains to us what bolt.new is, and then it's highlighting how this tool is different. Then it's going into the project layout. It's going to the app folder first, then the functions, etc. Then it's going into the web container integrations, etc. All right, so it did a pretty good job of explaining this code to us. Now let's go on to the second task that I did, which is to find bugs and refactor the code. And let's see what it came out with here. And you can see the notes and summary of what it did on the left-hand side panel here. It told us what it was able to do and what testing it was able to run or not be able to run. Like if you hover over the 2 curly brackets, you can even see the code that it's highlighting or the code that's being affected. And then it's telling you what files it has changed or not changed. So here we can see that it has changed 4 different files. It's taken out one line of code and added in one line. The first one, it's added in 11 lines of code in the second one and taken out 2. And then you can look at exactly what it's done here. So you can see that it wants to take out this line of code, and then add in this line. So green means it's wanting to add in that line of code. And the red means it's taking out that line of code. And you can see exactly what it wants to change. You can almost think of this as like a Word document with track changes. And once you're happy with all of the changes that it's made here, then just go here and click on push. And you have 4 different options. You can create a new pull request, you can create a pull request as a draft, you can copy the GET apply, or you can copy the PATCH. So when you're happy with all of these changes, what you do is you go up here and you click on push, and then you create a new pull request. And then what this will do is that it will then push all of these changes that was in Codex into GitHub. And then once it's in GitHub, you will have to merge the pull request for that code to be then inside of your very own repository. So to recap, what we did was we forked or cloned the codebase from bolt.new so that we can create our own version. And then now we're creating our own customizations to it that is only going to be found in our very own version. All right. So I'm in GitHub now and you can see at the top I have one pull request. Let's click on that and we can see that it has come from Codex. Ain't that cool? Like if you have multiple team members working on this, you can see who's making these changes as well. And it clearly labels that like this one's from your AI coding agent. Like it still blows my mind. And then it's briefly summarized what it did. Like Codex wrote all of this, so it fixed memory leaks and updated dependencies. And then we can click into it. It summarizes what it has done. It also summarizes the testing that it has done before it's committed the code. And now if you want this new code to be merged into your existing codebase, you just press merge pull request, put in a brief description, then confirm the merge, and boom, you got the new codebase now integrated into the existing codebase. How cool is that? And then now let's go back to Codex. Text here. And then let's go on to the 3rd task we got it to do, which is to suggest some suggestions on features that we can add into our own Bolt version. Okay, so let's take a look at the 9 different ideas that it has come up with. Um, let's see which one we like here. Okay, so after reading through these 9 different feature ideas, I really like the 3rd and the 5th one here. So let's get it to work on both of these. So literally what I'm gonna do is copy the 3rd idea, new feature to add, and then paste in what I wanted to do. Then I'm gonna go back and copy the 5th idea here and new feature to add, paste in what I wanted to do. So those The previous feature was being able to have some preset project templates, and this new feature is being able to configure the AI model and choose the different models, adjust the system prompt, letting the user tailor the assistant's behavior for their need. And I'm going to press on code again. I mean, like I said earlier, this is such a game changer because now you can create an entirely unique piece of software. With complete back and front end using Codex and open source code. By being able to modify these open source code and seeing what it can do, like what changes can be made, you can create something custom, something unique to yours, being able to put your logo on it and not just have it be a white label where it's the same for everyone else except the logo. All right. So it's been a few minutes and you can see the 2 tasks that we got it working on has already been completed. I've also added in a third task here, and I just told it to change the name of the platform from Bolt to Helena Bot as an example, right? And it's going to work, and it's going through all of the code and changing any instance of Bolt to Helena Bot. Like, that's how easy it is to start renaming these open-source software to your own name. All right, let's take a look at what it did inside of the 2 features that we wanted to add in. So again, with every single feature that you wanted to add in, it will have a summary as well as notes as to what it has done, and then it will also show the documents that it has changed as well as the new code that it has added in. Once we're happy with all of this, the process is the same. We will push the code as a new pull request into our GitHub, then we can merge all of the code together, and we have a unique piece of code that we can now own. It's a robust front and backend system that is not just like a logo that you slap on, like a white label software. Now this is a unique piece of software with features that can only be found in your own platform. I hope you guys are starting to see the possibilities. Right now you're only limited by your own imagination and creativity. I've already given you 3 ideas here. You can create your own version of DeepSeek, fine-tune it, add in your own training data, and then boom, you got your own specific LLM. You can also modify OpenAI's source platforms like Twynt, and then you got your own CRM tool that you can sell or use internally. Or you can do what I did here, which is modify the Bolt.new code and boom, you got your own Vibe coding platform. Like, how cool is that? So I hope you got a lot out of this tutorial. I hope you'll give this a try, whether you have an existing software and you just want to speed up your progress, like throw all your tasks that's currently on your roadmap onto Codex and get them done 10 times faster. Or if you don't have a software, clone or fork an open source software, make some modifications to it, add in some new features, and you got your own software. Like, it's mind-blowing. Okay, um, thank you so much for joining me for this tutorial. If you're new to my channel, please make sure you like and subscribe. And if you haven't done so already, join my free AI course where I will teach you everything about AI software development as well as automations and agents. So I can't wait to see you inside of my free course there. It's going to be a great course teaching you how to monetize all of the new AI skill set that you have and not just learn about it in a theoretical fashion. All right, that's it from me. Thank you so much for joining. I'll see you guys next time. Bye for now.