← Back to search
Your AI App Isn't Finished Until You Do This
Leon van Zyl · 2026-10-31 · 28 min
Show full episode description
Learn more about your ad choices. Visit megaphone.fm/adchoices
✨ Episode Outline — click any point to jump to it in the episode
Problem solved
Coding agents like
Claude Code and
Codex build only the main functionality you ask for, skipping the landing page, legal pages, auth, data export, deployment and MCP access that make an app production-ready.
Benefits
- One-prompt builds include landing page, cookie banner, legal pages and docs
- Built-in remote MCP server so Claude, ChatGPT or Hermes agent can operate the app
- Auth, organizations, roles, account deletion and data export out of the box
- Deployment handled by the skill, running on a real domain
- Swappable tech stack via editable reference MD files
Use cases
- Feature-complete Trello clone for tracking YouTube pipeline built from a single prompt
- Claude Code connected via MCP listed assigned cards and moved a card to 'in recording'
- Connected-apps page shows and revokes Hermes agent, Codex, Claude Code, ChatGPT access with activity log
- Create Brand Kit skill generated a logo, color scheme, light/dark SVGs and design rationale HTML
- Transactional email (OTP, password reset), background jobs, payments and SEO covered by reference files
KPIs / results
- Entire Trello-clone SaaS built from one prompt
- Three skills packaged from ~20 years of full-stack experience
- Four cards read and one updated via MCP from Claude Code
Tools / build
- Start an App skill
- Review an App skill
- Deploy an App skill
- Create Brand Kit skill
- Remote MCP server for the Trello clone
- Vercel AI SDK + OpenRouter
Coding agents like CloudCode and Codex can build pretty much anything these days, but there is a catch. These agents will build exactly what you ask for. This means if you don't specify the tech stack, the agent will build the site on your behalf. This can lead to issues down the line when you try to deploy the app, scale it, or there could even be security vulnerabilities that could get you into big trouble. Your coding agents will focus on the main functionality mostly, which means all the small things get skipped. So although your app might look correct at surface level, it's actually missing a lot of things that make it an actual production app. I've been building software for nearly 20 years now, so I took my experience as a full stack developer and packaged it into three core skills. And with these three skills, you can build any web app, from a simple website all the way to a large SaaS application. But first, let me give you a quick demo of the type of results that you will get out of this app. And by the way, this entire app was built using a single prompt. Just to show you what that prompt looks like, it says something like use the start an app skill. And I want you to create a feature complete Trello clone. I am primarily going to use this to track the progress of all of our YouTube activities. But effectively, it's nothing more than a Trello clone. And that's it. So agent asked a bunch of terrifying questions, and it built this entire app and everything that you see here. Now, I am going to show you how to set up and use the skills during this video. But I just want to kind of show you what you get out of the box. First things first, we get a landing page. How many times have you asked an agent to build some kind of app? And all it gives you is sort of the dashboard and login page. We get this UI. But very important, we also get this cookies banner. Since this app uses cookies, in some countries, this is actually required. So this will track cookies for you. You can actually choose what you want to enable and disable in terms of using cookies. And you can accept these preferences. These will be stored in the browser, just like how you would expect them to in an actual app. This app is running on an actual domain. And by the way, deployment was set up to the skill as well. You don't have to do anything. So something I do want to show you, though. Let me scroll down to the footer. We get these help pages. And this is really important. Depending on the app that you're trying to build, the agent will actually set up the public-facing documentation for your app. If you've ever used SaaS applications yourself, you know that this is really a requirement. Now, for a simple app like this, there's really not much to cover, except for one thing. We have documentation showing how you can connect this app to AI assistants and AI agents like Claude Code or ChatGPT. This skill will create and set up a remote MCP server for your app as well. This means that AI assistants can interact with your app in a secure way. So basically, I don't have to sign into this app to use it. I can hook it up to Claude using a connector or ChatGPT using a plugin. And I can simply ask Claude or ChatGPT to interact with the app on my behalf. I firmly believe that exposing the capabilities of your app to AI assistants is a must in the current day and age. More and more people are starting to use agents to interact with apps on their behalf. And your app needs to be able to securely accept connections from these AI assistants. Scrolling down to the footer as well, we get all of these legal pages out of the box as well. Again, this does not apply to each and every app that you build. If it's a very simple one-pager website, you don't need all of this. But this skill will intelligently instruct the agent to add these pages if it's necessary for the type of app that you're trying to build. Again, this is not optional if you're actually trying to build and sell software. Things like cookies, the terms of service, the privacy policy, those are required. Of course, because we're building a SaaS app here, it does allow us to create accounts and to sign in with accounts. This does support social sign-in as well. Or you can just create a new account using your email and password. All right, after signing in, of course, I can create boards and manage boards. This demo is not really about the app itself. It's about what the skills give you out of the box when building apps. When we go to our user profile, we get access to the public-facing documentation. But what I really want to show you is what happens if we go to settings. Again, this is all coming from that very simple prompt that I gave the agent and nothing else. It instinctively knows that if we're building SaaS applications that have user authentication, we need to be able to give the user access to their profile and other information that might actually be a legal requirement. After signing in, I can change my profile picture. When we go to account, we can change our email and password. But look at this. We also get this danger zone. This means we can download our data and we can request to delete our account from the app as well. This means removing all of the data that this app has on our account. Again, this is not optional if you're building real software. This needs to be there. And agents will not add all of the stuff unless you explicitly ask it to. And this is also cool. If we go to connected apps, this will show us how we can connect AI assistants to our application using the MCP server. And this will also show us all the connected apps. In other words, all the AI assistant clients like Hermes agent, Codex, Claudeco, JGPT, whatever we have connected to our account. And if we want, we can go ahead and revoke access to those specific agents. We also get full transparency on what the agent activity was. And maybe let's go to the YouTube pipeline for a second. Here we can see all the cards on this little Kanban board. And there is an actual database sitting behind this app. There is proper user authentication sitting here. We even have organizations and workspaces, which means I can invite members to these workspaces and I can assign different roles and permissions to them. Guys, this is all just baked into the skill. You don't really have to think about it. All these capabilities depend on the app that you're trying to build. There's still a lot I can show you here, but I think just to wrap things up, let me show you what the MCP server allows you to do. From Claude Code, I can add my application using the MCP server. So I simply add that little URL. And this will give my agent instant access to a whole bunch of functionality within the app, like reading cards and boards, creating cards, updating the statuses, assigning users. It's all there. So this means in Claude, I can simply say something like, please list all the cards that are currently assigned to me. And this is really important. Claude will ask our permission before calling any of these tools. And we have full control over what the approval steps or plans are for these tools. Just if I open up these tools again, you can see that for each and every tool that we expose, we can tell the agent it needs to ask our approval. Maybe we want to block a tool outright, or maybe we want to always allow access to that specific tool. I should also mention that you might see this little disconnect button every year. In order to use this MCP, we need to authenticate ourselves in our application. This means that Claude will act in our application on our behalf. So Claude will only have the same access that you give it within the application. This is all very secure. All right, it was able to pull the card. So in the YouTube pipeline, we've got these four cards over here. And if we just jump back to the board, we can see those four cards. Please can you update the three skills video to in recording? And cool. So Claude just finished this. If I refresh the app, it was able to update this card on my behalf. But of course, this is a web app as well. So I could still just use this in the app, or I could hook it up to Hermes agent or whatever assistant I use. Now, there are just so many other things that I could point out. Like we've got shortcuts for power users. It's just a lot of stuff that's just covered out of the box. But of course, you're now wondering, how do I actually use these skills? And are they a good fit for my project? But first, a quick word from our sponsor, Oracle. If you've ever built an AI agent that works great in a demo, you know what happens next. The moment a real business needs to use it, everything gets hard. Who can access what? Who approves changes? Who keeps track of it all? And usually, that's where these projects get stuck. Oracle Agent Studio for Fusion applications solves this in a smart way. Your agents run inside Oracle Fusion itself. So all the security, rules, and tracking that companies already trust. Your agent gets that for free. And here's the part I like. You build with the tools you already use every day. That could be VS Code, Cursor, Determinal, Git, and coding agents like Codex and Cloud Code. There's also a public GitHub repo full of templates, started projects, and sample apps. So you're never starting from scratch. Simply go to github.com slash Oracle slash Fusion AI Studio. If you're already a Fusion customer, AI Agent Studio is included at no extra cost. And the GitHub repo is open to everyone. Go to oracle.com slash applications slash Fusion AI to check it out. The link is in the description. Thanks again to Oracle for sponsoring this video. Setting up these skills are really easy. I'll link to these skills in the description of the video. Simply go to this GitHub repository, then copy this URL. Then simply go to your agent. I'll just use Cloud Code for this demo. Paste in that URL and just say, please install these skills. And that's it. You can just send this message and Cloud will go ahead and install this. It should install it at global level, which means that you will always have access to those skills throughout all of your projects. In fact, I've got this bonus skill as well called create brand kit. You can use this to create your company logo and things like a brand kit. Let me just give you an example of what that actually looks like. I use that skill to create this logo for my cognitive website and it actually created this little color scheme and the brand colors and everything as well. So you can see that the logo and the rest of the content are very similar, right? So it created a brand kit and everything seems to match as well. So it just creates a way more professional looking vibe. But what this also gives you is different versions of your logo. So it's just like the logo with a circle or a light and dark mode version of it. You get these SVG files as well, which are vector files that you can scale to like billboard sizes or you can print them on shirts. And the skill also gives you this little HTML document just to kind of explain the reasoning behind the design that it came up with. So we can see the final logo up here. It gives you a little bit of sort of a reasoning behind the idea. It gives you, you know, all the sort of design that it went through to create that logo for you and a lot of different rules and like different sizes. So this could be a very useful skill if you're really starting a project from absolutely nothing and you do want the logo and everything to be available up front. But I digress. We're really here to learn about these three skills. Start an app, Review an app, and Deploy an app. And each one of these skills are highly customizable. These were built using the tech stack that I've used for many, many years and a lot of large companies and enterprises use the same tech stack as well. But it's completely customizable. So if you want to swap out one component for something that you like to use, it's very flexible. Let me just give you sort of a sample of what these look like. If I go into the start an app skill, we've got our basic skill file that just kind of explains to the agent how the skill works and how to apply these different rules. It will ask you clarifying questions and try to understand what you need from the app. But what you will notice is that the questions are in plain English. It's very simple to follow. It's not technical. All you have to do is have a simple discussion with the agent and the agent will figure out what the underlying tech stack should look like. This is all driven by these different reference files. So if you are building an app that requires any AI capabilities, whether that's a chatbot or an embedded agent with tool access, this is all described in the AI.md file. The auth system is detailed in the auth file, the databases, how to write public docs, the email sender. This is another thing. This app is able to send transactional emails, things like OTPs, password resets, verifying accounts. All of that is covered as well. It can handle background jobs. So if your app requires background processing, which could fail and it needs restartability, that's all covered as well. And I'm not going to go through all of this. Payment processing is included as well. SEO for search engine optimization. Everything is here. And if you ever wanted to swap something out, all you have to do is open up that MD file, like the AI.md file. And in here, I basically use two different pods. I either use the AI SDK from Vercel or I'll use Open Router or a combination of the two. So by default, the app will try and use the AI SDK with Open Router. But let's say you wanted to use like the Claude Agent SDK instead or the Py Agent SDK. It's very easy. Ask your coding agent to modify this file to use whatever AI inference SDK you prefer to use. The same thing with payments. By default, this supports both Polar and Stripe. But if you wanted to use something else, simply modify this file to use whatever app you want to use. And using this is really simple. After we've installed these skills, all we have to do really is in our project, we can just start a new session and we can run the command, start an app. And let's do something really simple. Build a simple to-do app. I know we're all tired of to-do apps, but I'm just trying to demonstrate how to use this skill. Firstly, the app will ask you a few clarifying questions. I'll just say, keep it as simple as possible. And then anyone should be able to sign in and use the app. This will prompt the agent to use a database for storing the user information as well as the to-do items. And of course, this also tells the agent that the app needs user authentication. So it will set up the auth system as well. And this is very important. The agent will ask us if we want to use a Postgres database or SQLite. These are all coming from a DB file in the start an app skill. SQLite is mostly meant for toy apps. So if you're never planning on deploying the app, you can just not use SQLite. If you are planning to deploy it, use Postgres. Should the app be able to send email like for address verification? If you say yes, it means that it will try to set up resend or some kind of email service. For now, I'll just say skip it. Something I like to do is in my agentic coding tool, I'll simply rename this conversation to start an app. Like so. In fact, with the Trello clone that I showed you earlier, it was called YouTube Manager on my machine. And there you go. There are the three sessions. Start an app for building the application. I then ran the review and app skill, which we will cover after this session. And I then ran deploy an app. And that took care of everything, creating the database in production. It even set up blob storage for storing files. Everything. It just completed all of the work for me. And for the design system, I'll just go to this website, styles.referro.design. And I think let's actually look for something like a clean SaaS maybe. And maybe to keep things interesting, let's pull up Slack. I'll go to extend it. Let's just copy this. And I'll just say, please use this design system. Otherwise, the agent is just going to come up with its own boring design. And by the way, if you want to learn how to build design systems, I've got a dedicated section on that in my Genticoding Masterclass. We've got several videos from the design system fundamentals through escaping AI slop and using tools like claw design and Google Stitch. I'll link to the Genticoding Masterclass in the description of the video. All right, let's have a look. So claw just finished everything. It tested the app using its integrated browser as well. And although this was a very silly demo, you know, we all used it to do apps. It really went all out. We've got a landing page. We also have this cookie banner down here. So if we click on more about cookies, it takes us to the documentation as well. So we've got our little Euro section. This is following that design system that we set up earlier. And of course, we've got our help page. This shows us all the public facing documentation on how to use the app. This also created that remote MCP server. So we do have documentation on connecting the app to AI agents. So yes, all of the legal documentation is here as well. We just have to tell the agent what the name of our company is, and then it will replace that little placeholder over there. Let's see if we're able to create an account or sign in. So I'll try to create an account that actually worked and it was blazingly fast. Now we should be able to add to-dos. Nothing fancy there. Now the thing I really want to check is what happens if we click on our user profile. Typically with agents build apps, these are really bare bones if they exist at all. So we've got our settings, connected apps. Let's actually go to settings. And look at that. How cool is that? So we've got our profile. We've got our account information. We can download our data, delete it. And as you just saw, this was a very simple prompt, but we get a very cool app out of the box. We can view the connected apps. We get instructions on setting up the MCP servers. We also get the systems page, which is really cool. It tells us basically, well, you know, if you want, you can hide this page, but it does tell you whether or not your system is up and running. So is your user authentication system set up correctly? Is the database up and running? So this actually is really useful. Now you are welcome to ask the agent to add anything else or make any fixes or any changes, but we can now move on to the second skill. This is the review an app skill. This will check for any drift. So is the design kind of going off the rails and did the agent add something that really didn't belong in the app? So there's like evidence and things like that. But the two important skills that I want to highlight is this will check for any security vulnerabilities based on the OWASP top 10. If you don't know what that is, it's a very common set of checks that we can do to look for any critical security vulnerabilities in our app, like broken access control, cryptographic failures, things like SQL injection, and a few more. We will also check for SEO. This is something I haven't really seen anyone cover when it comes to agenti coding. When you build web apps, you want your website or web app to be discoverable by things like search engines. And in modern times, we also want large language models to be able to find our apps as well and understand our apps. This skill is quite involved, but it will try to SEO optimize your app. It will also add things like the robots.txt file, the llms.txt file, and your sitemap. Now these are all really much search engine optimization concepts. But briefly, if I ask Google just to explain the differences, the robots.txt file will block or allow crawlers to access specific sites or directories. So if there are pages on your website that should not be available to search engines, the robots.txt will block scrapers from accessing those routes. The sitemap determines all the different URLs in your website that should be indexed by search engines. And llms.txt is new, but this basically creates a summary of what your app or website is about with all the important routes or products or whatever. But this is written in a way that's easily consumable by large language models. Running all of this is really easy. All we have to do in the same project is create a new session. Then let's call the skill review an app. And that's it. I'm just going to run this command and the agent will do its thing. I'll also rename this to review an app. And of course, the cool thing about this skill is it's again, highly customizable. If you want to add additional checks to your review process, feel free to add additional files. If you want to check additional things that I'm not checking under SEO, feel free to change these rules and adjust it accordingly. Maybe you want to check for additional security vulnerabilities in your apps over and above the OWASP.10. Again, simply modify this file. This is really meant for you to take it and make it your own. And I think while this is running and in the interest of time, let's actually have a look at the final skill. Deploy an app. This is really simple. Again, we'll just run the command deploy an app and that's really it. Now, if we go into the deploy an app skill, we can pull up references and there are plenty of different files in here that you can dive into. You can make changes, do what you need to do. Now, this is very much aimed at deploying projects to Vercel. But if you want to change this to deploy to Cloudflare workers or Hostinger or your own VPS, you can change these files accordingly. Simply pull these into cloud code or whatever agent you're using and ask your agent to adjust these files to deploy to whatever service provider you want to deploy to. Right, so I'll just rename this session to deploy an app. So it's really that simple. You know, you don't have to install the skills every time. You just do that once. Then from there on, if you want to start a new project, run the start an app skill. Once the project is built and you've tested it yourself, run the review an app that will fix up SEO and fix up any possible security vulnerabilities. And then you can run deploy an app. And this is going to set up everything for you in production. It's going to deploy the project files, but it's also going to set up your database blob storage. If you're using background processing, it will set up ingest. So it will really set up everything that you need for this app to work in production. And by the way, if you use SQLite to build a project and remember earlier, I mentioned that SQLite is really just meant for throwaway local projects. If you decide that you actually want to deploy this app to production, it will check if you've got SQLite and it will convert your database. So it will write a script to migrate your database from SQLite to Postgres and then deploy everything to a Postgres database in production. And by the way, if you've got no idea what Vercel is or what I'm talking about, it's a hosting company. It's a service that allows you to deploy your websites or projects to their infrastructure. In fact, the free plan is extremely generous and I'm not sponsored by them at all. I just genuinely use them for my own projects. But you will not pay a cent. So the database, the blob storage, the actual hosting, you will not pay a cent. But just looking at Claude Code, it's asking us a few important questions. So where should or what address should be used for our application? So you can actually buy a domain like a .com domain through Vercel or they can generate a temporary URL for you as well, which is free. So maybe while you're building this and figuring out what domain you want, simply deploy this to a free domain. And then once you know what domain you want, you can register it through Vercel as well or you can register it to another provider and simply link it to your Vercel account. Your agent can assist you with that as well. So for this demo, I'll just go with a free URL. This folder isn't a Git repository. How should deployments work? You can deploy your project straight to Vercel and for some of you that might be perfectly fine. I actually like to create a GitHub repository just so that I've got my code backed up somewhere else and then whenever I push changes to GitHub, that should trigger that update to Vercel. So I'll actually go with this option. It's Git init and in fact, let's actually go with other. I want you to create the GitHub repository as well. You can just create it as a private repository and then whenever we push changes to GitHub, that should trigger the updates to Vercel. All right, I'll just send that but I think selecting a third option would have done exactly the same thing. Right, check this out and this is really cool. So we're getting this deployment sheet. You will always get this when using the skill. This will just give you a basic summary of the things that we decided so it's going to tell you it needs a database and there's a couple of things it needs to do but this is the important part. What it will cost and it's telling us we won't pay anything. Everything we've set up in this project is part of the Vercel hobby free tier. Even the database is going to be free. So I'll just say sounds good. Right, so Claude actually completed the deployment. It is now busy just testing the sitemap and testing the app itself in the production domain but let me just show you what it did. It created this GitHub repository so how cool is that? Our code is backed up on GitHub and if we go to Vercel you do have to just sign into your Vercel account. You can go to the project and everything is there. The app was deployed and we have this temporary domain. As a reminder you can ask the agent just to purchase the domain for you or on the project you can just go to domains and then buy a domain or just transfer an existing domain over to your project. This also set up all of the environment variables so if I click on environment variables and if you've ever done deployments like this you would know what environment variables are. The agent set up all of these environment variables on our behalf. If we go to storage it also created this Neon database. If you needed file storage like blob storage you would actually see your blob store here as well. Everything is set up on your behalf but this is our app running in production. Anyone with this URL can access this app now. The cookies work so let's try to create this account and that's it. Let's try to add this to do if we refresh the page everything is still there and our account works. How awesome is that? One note is if you want to connect this app to something like Claude Code I do think you need an actual domain. I don't think this temporary domain will work. I've actually had issues with it in the past but as soon as you assign an actual valid domain it works. But I think for completeness sake let's try this. So I'll copy this domain in Claude I'll just go to add connectors manage connectors let's add a custom connector I'll just call this tick it's paste in our URL to the MCP server let's click on add now we just have to authenticate ourselves so let's try that. Alright this is going to tell us that we couldn't connect but this is purely because of the domain I've actually wasted a lot of time trying to debug this and the longer the short was that you can't use the temporary domains for this so as soon as you register an actual domain your MCP server will work. Again you can install these skills simply by giving the GitHub repo to your agent or you can just go to skills.sh if this is what you tend to use and copy these commands and simply pass these commands to your agents or run them in the terminal. I hope you found this video useful if you did hit the like button and subscribe to my channel for more agentic coding tutorials. Thank you again for watching and I'll see you in the next one. Bye-bye.