Episode 143: Build Your Own Mobile Apps with AI

Co-Host

Aytekin Tank

Founder & CEO, Jotform

Co-Host

Demetri Panici

Founder, Rise Productive

About the Episode

What would you build if an app only needed to work for you? Aytekin Tank and Demetri Panici share how they use AI to build personal mobile apps and rethink the software they use every day. The conversation includes native iOS development with Codex, a podcast app built with Jotform, and a personal teleprompter demonstration. They also discuss where custom apps fit into content workflows and internal tools.

Because development as a product is such a difficult thing, right? You'd have a team working for six months to create an app that's full of bugs. Today, you just prompt it, maybe at night, and have an app ready in the morning.

Hi, my name is Dmitri Peniji, and I'm a content creator, agency owner, and AI enthusiast. You're listening to the AI Agents podcast, brought to you by Jotform and featuring our very own CEO and founder, Aytekin Tank. This is the show where artificial intelligence meets innovation, productivity, and the tools shaping the future of work. Enjoy the show.

Hello, and welcome back to another episode of the AI Agents podcast. In this episode, we're going to talk about making mobile apps with AI. We have a lot of different tools to show you today. Obviously, we know that we can build stuff with Jotform, but other tools out there, like Claude and Codex, do a really good job as well. So we're going to dive into a couple of examples of how you can take your idea and turn it into a real mobile app right now. How are you doing, Aytekin?

Hey, man.

I'm doing well. These days, I'm constantly vibe-coding websites, apps, and mobile apps. It's this idea of building products for yourself, for one person. That would have been so weird in the past, right? Development as a product is such a difficult thing. You'd have a team working for six months to create an app that's full of bugs. Today, you just prompt it, maybe at night, and have an app ready in the morning. This is something very unique to this year. I'm constantly building apps for myself, just for one person. Building software for yourself is very cool.

Yeah, it's really fun. I'm loving every minute of it. My main inspiration started with an issue I had with some of the work I do. The free versions of apps weren't that great, and the pro versions weren't crazy expensive, but I still thought they weren't that good. They did what I wanted, but I know from working in software that this isn't complicated to make, especially with AI now. My first inspiration came from recording. I record all the time and use a teleprompter with an iOS app for scripted videos. I thought, 'Why the heck do I have to pay eight bucks a month for this teleprompter software?' I picked the best one I liked and thought I could make something pretty close for almost nothing.

I also had personal inspiration for other apps. I wondered how good Todoist was and how hard it would be to make a to-do list. I also made a personal finance app. I want to talk a little about how these tools work. I use iOS. The nice thing about iOS is that it's a great experience to code for natively on a Mac. Developers use Macs because the simulator lets you test bugs and analyze code in a real-world environment, clicking around instead of just coding and testing. What you see in the code and what you experience in the app aren't always the same. I primarily use Codex because it has had simulator connectors, views, and computer-use features longer than Claude Code. I really enjoy Codex, but recently I've started to like Claude Opus 5.5. I like the Codex experience, but I think the Opus 5.5 model is really good, so I'm working with both to find the best workflow for planning, implementing, and testing my mobile apps.

Yeah, I used to be all Claude Code, but lately I'm all Codex. It feels more flexible and fast. I'm still all in Codex these days.

What I'm finding is that whenever I'm working on a project, I'm just going to build an app, like building a website. It's very similar, and you can easily build progressive web apps and install them on your phone, too. For example, I was curious about my personal investments. I have investments in different places, so I gave Codex my statements and asked it to build me an app. After a lot of back and forth and feedback, it created an amazing app. I can see how much tax I might owe and the gains on different investments. I have so much more visibility, and I can prompt it to create any chart I want and compare it to the S&P 500.

The great thing about building apps with Codex or Claude Code is that they can also do research. You can join your data, build widgets, dashboards, and charts that do calculations, and ask it to research tickers and compare them to the S&P 500 or whatever you want. It can do everything, which means you can have something productive. When you combine your own data with research or publicly available data, you can build something really useful to you. It's very easy to write the code. I'll ask it to do something, get started quickly, and deploy it as a page in a browser. Codex lets you log in with your ChatGPT account, so it's protected. I even bought my own domain and created subdomains for each project.

Then I'll log in to these dashboards and keep track of my projects and to-do lists. I might use Notion or other project-tracking tools, but building your own app is so much more productive and flexible. You just say what you want, and it happens. It's magical, like having a genie.

So, it's amazing.

Yeah, it is amazing. To get things started, I'll open my simulators and showcase what I made. Let me share my screen. The first app I'm going to show you is right here. I'm going to use the simulator itself. This is my phone simulator. This is a native iOS app, which is cool, but there are ways to make apps that aren't native. I'll go to the other app first to prove your point. This is my tool called Qstream. It's essentially a teleprompter app, but this is my Todoist clone, which I made as a test of my own skills. How does this work from a back-end standpoint? It's a Next.js app deployed on Vercel and hosted by Supabase. I used ChatGPT and Codex, along with my Mac connection, to deploy it to my account. Google's developer accounts are pretty simple. I think you get three or four personal ones before you have to start paying. You pay about $100 a year to post to the App Store and get unlimited deployments of different apps on your phone. I paid for it because I planned to make an app public at some point. This is an entirely vibe-coded app built with ChatGPT and a little Claude Code. It was done a couple of months ago, and it's a really solid baseline framework. It has its own MCP server, so I can do chats and things like that. You can see the check marks, and there are undo buttons. The database connects to my Supabase back end. It even has Google Calendar integration and different views you'd expect from a great task manager like Todoist. It's completely functional.

Why did you build a Todoist clone?

Because I was annoyed that I was paying for the subscription.

Okay.

Yeah, that was pretty much it. I thought I could use the $200-a-month subscription to make a Todoist clone.

Did you just start that way?

Well, I did, but the process of building it obviously wasn't short. It took about a month and a half to get something pretty close. Once I had the baseline, all I had to do was compare the simulator to the actual phone preview. With a Mac, you can use phone mirroring. I connected it to my phone, and then it could analyze and take screenshots of Todoist and what I'd deployed, and make the relevant code changes. I had a lot of different chats going at the same time, asking for changes to the bottom bar's style and glow, among other things. If you're using a product and make a request, they might build it two years from now. But if you're vibe-coding a product and don't like something, you just say so and make it the way you want. That's what I meant by building a product for yourself, for one person.

Building a product for one, for your own needs. There's something magical about that.

Yeah, it was my task management system. I wanted to test myself and see if I could make a clone, and I pretty much did. Granted, it's not as stable or perfect, but for one person it's close enough. If it saves you five or ten bucks a month and you have a bunch of subscriptions, that's a lot of money. It's nice to have the option not to pay for these things. I enjoyed doing it and wanted to test whether I could make an MCP server, connect Google Calendar natively in apps, and build all these different things. I'll show a couple of other apps later. My starting point was making an app to organize my day. Another reason was necessity: Notion's mobile app is so bad. The load time has been a consistent issue. It's not a native app yet, and apparently they're working on that, but the load times are bad because it's like a wrapper. The bundle sizes make it take a long time to open and write down a task. If this is my mobile and desktop to-do list app, that's a problem. I considered using this for my whole to-do list system, but I didn't because I like how Notion interacts with other apps. Long story short, it was out of necessity and curiosity about whether I could do it.

All right. Nice.

Mhm.

Yeah. I'm going to showcase our own product because we also have an app builder. I find myself building apps with it a lot, in addition to Codex, because it makes progressive web apps. They're basically web pages, but they also work like apps on mobile phones. What I like about it is that it's very useful, especially when you have forms. The app I decided to build was an app for our own podcast. Why not build an app for our own podcast? This is my prompt. I copied it. Let's go.

We can see the prompt here: 'Create an app for AI Agents podcast.' This is all free, part of a suite of Jotform products. The prompt asks it to find the podcast on YouTube, include the most recent 20 episodes, make it possible to browse and watch videos in the app, and create an about page. That's my prompt. What I like is that it's much faster than Codex or Claude Code. It builds the app very quickly. If you need forms in your app, you can easily do that. It already got our banner from YouTube and has login and everything. It says app generation is complete, though it's still creating episodes and pages. Yeah, it's already done.

Let's see. Let me close this. We can preview it here and see what it looks like on mobile. We have the banner and navigation. We can see the episodes and watch them right here. I can click on one and watch it. If I go to the home page—okay, it got stuck here. Let's leave it. We can see the host page.

All right. Yeah, we have the host page as well.

It went to YouTube and the web, did the research, and created a database right within the app. It pulled 20 episodes. I can ask it to bring more if I want. It has descriptions, publish dates, durations, and so on. Because I asked about the host page, it found the podcast hosts and placed them here. I can get an app link and customize it to be private, require registration, and so on. Lots of things are already prebuilt. If I had app users, I could see information about them here, including things like push notifications. It supports many things that native apps support, and we can see how it looks on tablets and desktop, too.

That's so cool.

It's pretty good already. If I ask it to add other things, like a contact page with a form, it can do that too. I can talk to Podo. Podo is our mascot and has been for a very long time, but once we had AI, it kind of became our AI.

Yeah, Podo came before things like Muse, Rockbot, and Dot.

Yeah.

Now I feel like they're kind of copying from us, right?

Mhm.

It was very quick. It added a contact page as well, hopefully with a Jotform form. Let's see.

Yeah, it's already got a contact form and is ready to use.

This is it. With the data, you can build all kinds of things. You can see contact form submissions here, too. It's a nice little app builder that works really well because when you're building software from scratch—

You mentioned it took you about a month to do the Todoist clone.

Yeah, yeah.

Mhm.

In this case, we already have all these pieces. We have the database, feature listings, and you can customize and edit the listings and add columns. We have forms, navigation, login, and everything, so it's already ready. We had this app builder many years before AI. We've built all the features someone might need for an app with or without forms. It's fast, stable, and works well. If you're not building a Todoist clone, but just a simple app with forms and data, I think Jotform App Builder will help you with those use cases. But if you're building a Todoist clone, definitely use a code editor.

Yeah, I agree. Everyone's going to have their own unique set of requirements for a tool. I want to call out the benefits of what you just showed. Todoist has a back end of forms and a back end that's beneficial. Here's an example for someone who may want to do this. We're not big fans of Google Forms, but the positives of what we have at Jotform are that you can take something like what you just did. I just had my high school reunion. Imagine the form and post I received for the reunion invite being built on something like this. They send out a blast text saying, 'Remember all the times we had?' and then you fill out the form to say whether you're coming. That's the kind of cool stuff you can quickly do with Jotform apps, and you can see who RSVPs and who's coming.

You can build features like that. It's really cool to see, 'Oh, my friend is coming, so I should go.'

Exactly. I had friends I was wondering about whether I was going to go to this thing. Then I thought, 'I guess I will,' because I didn't want to miss being with my friends who had been with me through so much of growing up. Right now, I'm going to show you another app. I'm just making sure I get a proper script in here. I'll share my simulator. This is another one I mentioned earlier, called Qstream. This is one I might put on the App Store once it's really refined. I could charge a dollar for it and have a bunch of people enjoy a quick app. This one didn't take long to build; it took maybe a week to get to where I was happy with it.

This is a teleprompter app. It syncs with either your Google Drive or iCloud. It's a native iOS, iPadOS, and Mac app. It works on your phone, iPad, and Mac. It's not available on Windows, but it gives someone like me or you a native experience across different devices. A native tool looks and feels better. If I make a new script, it starts like this. I'll paste in an example script to show you how it works. I'll press save and sync. It's a simulator, so it won't actually sync to my phone. Let's switch to horizontal. You can see the horizontal change. When I press play, this is the teleprompter. Imagine it sitting here on my setup. If I press enter, three, two, one. It's got this line and settings for which direction you want it to flow, down to up or up to down. It has settings for margins and sizing. For me, having it as close to the center as possible is better, so I don't look left to right.

What other settings are there? I made a bunch. There's a clicker, which won't work on the simulator, but it lets me press it to stop and go without touching the screen.

Yep. It's really nice that I was able to build something that connected to my Bluetooth clicker for another application.

Yeah, I love it. I also gave it access to record me on camera. I have a lot of cool little features here: text size, line spacing, margins, speed, how long the countdown is at the beginning, whether you want the guideline, manual scrolling, inverted manual scrolling, mirroring, and customizable keyboard shortcuts. There's also sync. It's pretty much my go-to teleprompter app now. I've played with this.

Just record.

It's Swift.

Swift? Oh, yeah. It's Swift. It's a native app, so it's using Swift.

Yep. That's the nice thing about this. On iOS—

X, or sorry, Mac Xcode is what's used for Swift. All you have to do is install Xcode, and Codex will take care of most of it. You just say, 'I want to make a native iOS Swift app.' Usually, my best route is to do a bunch of research at the beginning, get a baseline that looks fine, and then you're off to the races. This also works on iPad, as I mentioned, and that's a really cool experience. I enjoy being able to make it for multiple devices. I find it interesting to tweak and make different applications for different formats.

Here's an idea for you.

Yeah.

Set it up so the AI actually designs a video, right?

Mhm.

It could prepare a video and a script. It plans what to do for the video, the whole idea and everything, then creates a video and a script for you. Then it asks you to read the script. You're just the human helper following the order to read it. Then it publishes it on YouTube.

Right?

I mean, that'd be pretty awesome.

That'd be pretty epic. I think it's possible. Fully agentifying everything I'm doing would be incredible. Imagine that. That's an idea. I'll think about it. I might not do it, but I'll consider it.

I like that. Maybe it's going to skyrocket your content workflow. Maybe it'll go through the roof.

But yeah, that's the really exciting thing. I really enjoy everything I'm doing with AI and content. There are frustrating times when I'm like, 'Ah, it's not quite right.' Maybe we can talk about this in another episode. We've reached the point where you really can sound like yourself with AI. Luckily, I have a lot of data to pull from. I have, I'm trying to think of the number—that's why I'm stuttering—like 5,000 pieces of ongoing content of myself. It's pretty easy to understand how a person talks when you have all of their talking. You can really do that. I'm thinking of other episodes we can do, because everything coming out now at Adweek is fascinating.

Yeah, I'm working on making a skill.

Correct. I'm working on making it right, like you speak.

Correct. The only issue right now isn't that it doesn't talk like me. The issue is connecting it to other skills. It's fine if it's trying to send a message or something, but I've noticed it gets my language right, and I don't know if I would format the text like that. I wouldn't want a social post to start how I naturally speak. I wouldn't want to start a social post with, 'So, last week I was working on this.' I'd want a scroll-stopping first sentence, then talk in a slightly different format than I would speak, but use the words I would use. That's the hard thing about AI right now. A lot of people say, 'Oh my God, it sounds just like me.' That's cool, but it also looks like I'm just reading you gab with your friends. That doesn't help me make a good social post or script. Everything we're doing at Adweek is amazing—not just us personally, but every piece of news that drops. I just go, 'Holy crap.' There was a tool yesterday on Twitter, an MCP and a skill similar to what I did with Upward, trained to copy stuff. That means more clones of things can come out. For internal tools, I'm sure you at Jotform have your own project management software and maybe an enterprise account. Do we need that? Is this something we need anymore? So many things are happening.

We use a tool called FullStory. I don't know if you know them. It's really useful when we release a feature because we can watch people use the product. The problem with FullStory is that at Jotform, you can speak to Jotform, like Podo. Podo can speak, but FullStory doesn't have voice support, so you can't record the voice and watch it together with website usage. So we cloned FullStory and made JotStory. It's a FullStory clone plus voice features, and we're using it more and more instead of FullStory because we asked for the feature, but they didn't build it. We need to watch the quality issues with our product and the mistakes our AI makes. I don't actually like the idea of cloning products because it's not cost-free. You have to maintain the software instead of someone else maintaining it, keeping it stable, and keeping your data secure.

Yeah.

Instead of having someone who specializes in that do it, you're doing it, which isn't great. It's better to pay for it, but sometimes vendors drop the ball. Now we have the option to clone them and make a better version if we have to.

Yeah.

Yeah, I agree. In closed situations, with little apps here and there, there's an opportunity for cloning when it's just a UI/UX thing housed somewhere. But I think this is going to lead to a lot of competition. Tools will realize they're charging too much, or that their feature set isn't outstanding compared with the rest of the market. They'll have to use AI vibe-coding tools to become competitive. That's exciting. A lot of these tools are going to be stuck. People like me have other ideas and things I'll probably show in other episodes that a solo person can make now, which they couldn't before. I'm sure you know people with ideas, whether in business, where they never felt they could get investment, or something niche like software. Someone might have an idea for an app, and now we can spin it up. A year and a half ago, a year ago, even six months ago, the average person couldn't have done that with the quality we'd expect.

Yeah, it's freedom. It's a freedom of ideas. Anything you want—the possibilities are endless. You can build any kind of app, anything. It's just so free, with so many possibilities, and I'm really excited about that. We're going to keep covering it in the coming episodes. Some exciting stuff is coming up.

Absolutely. With that being said, thank you so much for listening to this episode of the AI Agents podcast, and we'll see you in the next one. Bye.

See you next time.

Stay Ahead with the AI Agents Podcast

Get the latest insights on AI agents, their future, and developments in the AI form industry.