#26: Draw in mid air and an AI makes an app

This is going to be a fun one. We're about to get a demo in the Vision Pro and you know I'm hot about the Vision Apple Vision Pro and about spatial computing, but this, this lets you do some weird shit, lets you draw in the air and make an app out of it. What the hell? Hi, Jacob.

Hello. How are you?

You're calling me from your Vision Pro?

Oh, yes. Can you see me? Can you?

I can see your Vision Pro. I can see what you're seeing and.

Wonderful.

You're the founder of Glowy. Tell me a little bit. Well, let's get a demo because I'm going to stick this demo into a longer interview where we go for it.

Yeah.

What are we seeing here?

So this is Globi. So basically you can draw whatever you want to build. So Globi lets you develop pretty much a different type of software, but software that you can export to multiple programming languages and you do it just by drawing on the screen. And it's pretty much apps from Doodles. So, well, we could say like, hey, can you create a website with, you know, something about San Francisco?

So you're talking to it and drawing to it. This is clearly a bunch of AI watching what you're doing on the screen as well as what you're talking to it about.

Yes.

This is awesome. This is magic, isn't it? Do you wake up in the morning and go, man, I'm developing magic.

Yeah, it's unbelievable. And you see we got some bridges over here and that's pretty much you could build anything. You could make it functional. You could say like, hey, but that's pretty simple right now. Yeah. And what's cool about this, right? So you can also translate it to multiple programming languages. So for example, you could do SwiftUI. And SwiftUI is Apple framework, right? Yeah. And then you get code that you could use to build this kind of like, to build this creation into like an iOS app or vision Pro app even. And I would say this is kind of helpful for people who, let's say experienced in web development or if you're a UX designer and you have experience in some front end frameworks, maybe. So you could build exactly the same thing but for iOS or Android. So we support SwiftUI for iOS. Flutter is a cross platform framework. IOS, Android also web and Kotlin is Android. And next JS. This is like a professional level framework for web development.

This is awesome. What kinds of things can you build with AI? You can go into some complex forms, couldn't you? If you're really good at drawing, I guess.

Yes. So yeah, basically first step, figure out how to draw. Right. Second step. Now you could build software this way, but.

Here'S a question. Could, could I prototype somewhere else like using one of the line drawing apps and, and bring that in and, and start with that.

So yeah, over here we have some designs that we built in Figma. Right. And so we gonna touch it here and we're gonna ask Globi, can you please build this game menu, make it also functional. So when you press start button something happens. So let's see if it works or not. But I would say the state of development right now that it's very good for prototyping or quick prototyping. Okay, so we get something. So when we click on it doesn't really work so well. But we have the first version of our app and now we could translate it to Swift and then you, let's say you copy the code and put it on Xcode. And for Android it's like Android Studio, you download the project, right. And then you copy the code into the project. There's like certain areas for globally allowed to write code and that's how you build your app or a game.

This reminds me of a lot of Visual Basic. Visual Basic gave you a 2D form and you could drag and put a button on the form or a dropdown menu on the form or a menu bar on the form or a chart or a bunch of different things. Right. Back in the early days of building Windows apps, this, you use your voice and you draw in the air and then you can still stick code behind it. So it's almost like a Visual Basic app builder, but it's for the modern world where you're going to be wearing a pair of glasses soon.

Yes, yes, I remember. I was also, you know, when I was a kid I was building something using Visual Basic. Yeah, yeah, it's.

Could you show us some other.

It's kind of similar. Yes. So, well, shall we then stop the demo? Vision Pro.

You can, if you have them in Vision Pro, you, you can show me what you got in Vision Pro. Otherwise we can do an interview. I mean this is really cool how if you're better at drawing that, it obviously gets better, doesn't it? Or does it try to build, build.

Something, maybe something else here. So the, the, the tricky part, right? So I also, if I were able to connect my computer here, you know, I would be able to demo the whole workflow but because I'm streaming this to my computer, so it's impossible to connect a monitor to vision, sadly. Yeah. Right. So. Well, but I have this prerecorded too, so maybe we could do it another way. Okay, but maybe do you want to build something? Like maybe you could prompt it and I could draw it or we can attach something.

Oh, what kind of app would I like to build? Well, let's say we're in operation and we need to build an app for the robo taxi. So could you build an app that says find me a ride, something like that. Find me a Tesla that'll come to my house.

You know, like app for Teslas. Right. Or how to do it.

Something like an Uber app, you know, where you want to start out saying, well, can, you know, I don't know what kind. Here's a question for you. Since you, you've been working on this. What kinds of things does it really do well? What kinds of apps do you find.

It really does well with Concurrently? It's very simple, kind of like one screen applications, but it also could be functional. So for example, you could build something like unit calculator or any type of calculators. You could build some menus where you just.

Let's build a calculator that probably is easier for you to demo than a full blown Uber app, which requires, you know, maps and all sorts of fun things.

I don't know. Maybe. So we, we could try. So we, we cannot predict, right. We cannot predict what the AI is going to generate for us, but if.

You ask it, you know, click on a button and then pull up a map. Could it pull up a map like from Google Maps?

Not, not yet. Okay. Not yet, sadly. Okay. Yeah, so, but, so well, we could, we could say like. So, yeah, definitely you need to learn how to draw. Yeah. Right. So before you start coding and this is the major technology shift, you know, how about Tesla calculator? I don't know. Oh, okay. Please build a functional Tesla calculator that calculates something interesting like mileage. Right. Is it good? Yeah. So yeah. And we have this on Vision Pro and obviously we have it on the web. So we also working on like Mac app and iOS app and Android app that. Okay, so there's something. Oh, okay. So AI thought that this is like not buttons, but pictures, which is interesting. Can you see it?

Can you refactor, like if it gives you something wrong like, like here. It might not have been what you were dreaming of. Can you say, I wasn't thinking about features. I was thinking about buttons with numbers on them that would let me click on The.

Yeah, you could, you know, you could press edit here, right? And you could press edit and you could say like, hey. Oh, okay. These rectangles represent buttons, not pictures. And then. Okay, something.

Yeah, very different.

But yeah. And again, like as pretty much with interacting with any AI. So we kind of have to prompt it well. Oh, there we go. Tesla. Right? So, well, there's Tesla picture and some buttons, but there's no calculate button. Just up. How do I dismiss it? Oh, there we go. So, well, and they have a simple page, kind of like calculator about Tesla. And then we could switch it to other programming languages like Swift, and then we copy paste it to Xcode and run exactly the same for like iOS app. Right. Or Vision Pro app, which is very interesting concept generally. So yeah, that's pretty much maybe the demo for Vision Pro, right?

Yeah, that's fine.

Okay, just a second. Oh, okay. So, well, and over here I have like this open demos. Okay. Shall we do that as well?

Sure, yeah.

Okay. Just like, let's make it full screen. Maybe just. Okay. So this is Globi. Globi is available on the web and on Vision Pro right now. So basically what it is, you just tell it whatever you want to build and you draw and then it gives you your first version of your app. It could be functional. In this example, you build a contact app with a functional search and then you translate it to multiple programming languages. So specifically you could build iOS app or like a Vision Pro app, for example, over here. Then you can translate it to Android using Kotlin. Or if you prefer, you could do it using Flutter. So Flutter cross platform framework for good, for rapid prototyping. So over here it's like a news website. So what's cool about it? You could also solve leetcode problems. Are you familiar with leetcode?

Oh, yeah.

So, well, you could just attach a picture of a leetcode problem or just describe it and it will solve it for you. So over here, Globi can also create another version of Globi. So you could use our clone tools. You can download Globi from the website and basically you could deploy Globi to any hosting you like. And this is like a Vision Pro demo that we also did live. So over here it's like a simple recipe app that you could build. So well. And the idea is super simple. You start with one button and you draw and then you press build and you get your prototype. So it's pretty much V1 of this. Right? So we are working on improving it, increasing the complexity. But now it's very good for prototyping, so very good to. If you want to figure out and you want to see, you know how to make those simple apps for like different platforms. Yeah. And then you can also ask follow up questions over here. As you can see, there's like, like you generate Swift and then you press on this refresh button and then you could say like, hey, there's an error, fix it. Or like, I don't like this button. That's also possible. So yeah, that's pretty much it will be.

That's awesome. Who are you?

I'm Jacob. I'm a founder of Globoam. So we are the creators of Globi. Globi is the first draw code AI assistant for easy app creation. So if you want to develop an app, you talk to Globi and basically the way it works, you figure out how to draw. You draw whatever you want to build, you attach your designs and then it exports to different programming languages.

That's really cool.

I was living in San Francisco for more than 10 years. I was a. Yeah.

It'S really cool because you draw.

Yeah, I was living in San Francisco for more than 10 years, so I was.

Yes, it's really cool because. And sorry, we're having a little latency because my thing's crunching the video. It's really cool because you're like HyperCard or Visual Basic or Visual Studio for this new AI age where you can just draw on a form, talk to it, prompt it and it builds it. It's freaking insane. Do you wake up in the morning and think I'm building magic?

Oh yeah, absolutely. I'm very, very excited where all this like foundation models go. Right. But also we're building globally on the kind of like application layer and it kind of becomes more and more powerful every day. So now we have this web version that you go on the website, you draw and it kind of builds it for you and we have Vision Pro version. You know, I, when I was just like building apps for myself, you know, 15 years ago, I wish, I wish I could have these tools, you know. Yeah, back then, but the old way.

I mean if you go back to the 1990s when I worked at Visual Basic Programmers journal, Visual Basic gave you a form with a toolbar and then you would drag tools like a button or a drop down list or a data entry field, something like that onto the form here. You're just drawing that and telling the computer to generate it, right?

Yes, yes, exactly. So. Well, and this is kind of like simple by design, right? So we working on democratizing software creation and we've been around for quite a few years and this year we've decided to build something like super simple, right? It's just like one button tool max and then it turns out this idea with drawing whatever you want to build, it's fun, right? And it's possible to make something functional using these foundational models and some RAG and other stuff it could make over time these generations will be more and more complex. Well, it's very interesting. I wake up every morning to improve this technology and I'm very excited about it every day.

Yeah, you're really the best example I've seen of how AI is about to change. Well, everything really. Right. I mean it's certainly changing how you can design applications and interact with them. Right. Where do you think this world is going? It's a great demo. You could do something very simple. But tell us like what is it going to look like in five years for people?

So I would say the foundational model is going to be going to get better and better. Right. And every time there's an update on foundational models, it's also going to lift up a bunch of applications, including these software development applications. And again currently we support OpenAI and GPT4O model. And on Vision Pro we have Mistral like model that works locally which is kind of cool. But also our next update is an next month or so. So we're going to support a lot of different models so we can give people choice. Right. And there's also there's some like techniques like RAG that kind of like adds multiple, you know, additional context to what you're building. And for right now our tool is just like good for quick prototyping. So just like within like 30 seconds you could build something interesting, right. And just like play with it or see how it works, maybe ask globif like follow up questions. But I would say by the end of the year, right? And especially maybe five now I would say you just will be able to wake up, think about your day and globe will be able to create a functional app, for example, for the day. So for example you have all these things in your calendar, bunch of interviews. Imagine people currently they post a social media because it's very easy and accessible. But what if people, pretty much everyone could build a useful tool for the day, for the month, for a year and just share it with their friends? Well, I think that's where it's going if it makes sense. So you just kind of think about your tool and then kind of Globi, or like any other AI tool, gives you different options where you could swipe right, left and right and like, pick whatever you like. And then you get like fully functional kind of app or like fully functional software. Yeah, kind of like in your context.

In the demo, we were building all sorts of different kinds of apps. Like, we were able to build a calculator, right. And stuff like that. Really cool. What's going on underneath the hood? What. What's doing all the work, you know, because. Because clearly it's watching the screen, it's watching what you're saying, and then writing a lot of code and letting you build an app. Tell me a little bit about what's going on underneath the code.

There's like two. Two versions of Globi right now. So it's the current demo that we did. It's on Vision Pro and they also have it on the web on Vision Pro. It's built using SwiftUI, the framework for building iOS and Vision Pro apps. And also it uses GPT4O. And also if GPT4 is not available, it uses Mistral like models. So it kind of was the model from Hugging face, right. And then kind of like it works without Internet, so we just kind of interested. Super interesting. It fascinates me. But it turns out the Vision Pro, right, so it does have M2 chip, right? So it's like pretty powerful chip that could run LLM locally. It's just so cool. And I got this LLM that like 4 gigabytes, and it only can do like one instance. If it does the second instance, it kind of like crashes the whole thing. But you see where we're going. So it's getting more and more powerful. So I would say Vision Pro, right? It's perfect. V1 of spatial computing. Right. I'm super interested to see what's going to be in like five years. Yeah.

Like I would say on Monday, Apple's announcing a new AI strategy. Everybody's expecting to see a new Siri. We're. We're expecting to see a new version of the software that runs on these Vision Pros, right? A version two of the software. You're a developer, so. And you're really pushing this stuff harder than I've seen other people push it. What are you hoping for on Monday?

So, well, if there's like a new foundation model or new API, I'm definitely going to be one of the first developers who supports this. And that's how we build our Software. Right. So we are on application layer and we would like the most comfortable position for our companies to be on the front wave. Right. So, and I hope CD is going to become a little smarter. But you know, maybe there's like a new API, maybe you know, there's something new for like Vision Pro. I don't know. How about you? What do you think?

I'd like to see you be able to build 3D applications, spatial computing applications. Like, you know, make the coffee table in front of me a football field and then, you know, do some stuff with that, put some buttons on it or something like that to let me control it. It'll be interesting to see where, where things go. You know, I'm not, I'm not too expecting really crazy stuff like that. Like, like, like you said, I'm hoping they, they lay down a strategy for how we get to that world. Right. Because that's, that's what we really want to do is draw something. Hey, put a, oh, put a hydrant over here on the floor and put a tree here, Right. You know, and have it build it around you. That's, that's what I call the holodeck. Right. And we're clearly heading toward a 3D environment that we could talk to and do things like that. Like, and that, that's where I would be real excited to watch. Yeah, you work in that holodeck and build something really, really even more mind blowing than drawing on 2D screens. You know, a little user interface for a 2D application. I want to move the world into a 3D world. Right. And so I'm hoping that people, that people like you can build new 3D experiences very, very quickly using this kind of technology. But yeah, yeah, you know, you're right.

Well, before moving to San Francisco Like 10 years ago, I was like indie game developer actually. I built like 3D games before and I built like over 50 titles and most of them failed. But I got two successful applications. So it was Glow Puzzle, second game in America and another successful game. It was called Office story. It's like 3D business simulation. And what I'm excited about, I just, I use these models of this Minecraft style characters. So I combine this with draw to code. So it's not released yet, but this is like one version that's kind of like I'm building under the hood and I have maybe this demo on Twitter. But basically you press the button and you have this like your AI agent as a 3D character kind of like shows up, right. And you could say like, hey, now you're an Android developer and I put you in the kitchen or in the bathroom and I'll come back to you in 30 minutes or six hours and you better build this thing autonomously. And this is kind of like a next version of the Vision Pro adaptation for Jawtecho that I see. So it could be like the simulation environment or maybe it could be like a startup simulation where you could build a real software. So you build real software, but then you have this simulated world in kind of 3D, right? So that you kind of learn how to market your software, maybe how to raise funding for your startup, something like that. So it would be super cool. And you know, we have already a lot of things to use to build this world, but overall it's getting better and better each major release from major companies.

It's an awesome that's coming. What does spatial computing mean to you when you look at developing for the Vision Pro, how do you see that world and how do you see it as different than computing that people are doing on a phone or on a laptop or something like that, 2D computing?

So, well, I think it will be more convenient to build a lot of different things, right? You can have like, if you're a developer, right? So you can have a lot of different tools around you, kind of like you could use to build your apps or I don't know, for like everyday people. It's very interesting already, right, so it's like Vision Pro is like V1, right? So probably it's not perfect yet, right? But also, if you remember 2008, iPhone, right? So, well, it was maybe like four years until kind of like big businesses started to be built on top of iPhone applications. It was just like a toy like version of it for like four years. And I would say Vision Pro could be in a similar category. So maybe like in five years we'll have more lighter version, more powerful version of special computing device that a lot of things could be possible. So right now on this version, right? So I could put 4 gigabyte version of foundational model in there, but maybe next version is going to run like 30 gigabyte or a bigger model basically and become faster. And this screen will be way better and it's going to be lighter, but I think you got to be patient. But maybe five years from now, the way is going to be a little more different than today from that perspective.

You do your demos drawing and you're better at drawing than I am. My drawing is even worse than yours is and yours is Pretty bad. Can you, can you bring in an, a screenshot from something like an Uber app or as some app and say, hey, copy that. And then instead of calling it Uber, let's call it Tesla. You know, something like that. Could you, could you take an existing app and start working with it or do you have to start with a sketch?

So yeah, you can attach your designs. So you could like bring a screenshot of the app or let's say you would design something in Figma or other design tools and you can maybe screenshot it and give it to Globi and Globi will work with that and generate code for you. So. Yes.

That's awesome. Tell me a little bit about the company and the business plan. You know, because it's cool to build a demo, but you, I, I assume you're, you want developers to actually buy this and use this or license it or whatever. Tell me a little bit about the business and how that's how that's going to go.

So. Well, yeah, currently we have a website, so you sign up there and that's how you start your software development journey. So we target regular people and that's kind of the onboarding part. So the demos. But you could build already pretty complex applications. So if you dig deeper, like read the documentation and kind of go explains the code that it generates. So because it's full discorder, it's like a V1. Right. So. Well, and we do have, you know, there's some lack in the guidance and documentation maybe, but we are working on the next versions where you'll be able to generate fully functional projects just starting in super basic input with drawing or attaching your designs and the company. So we've been around for quite a few years, so we have four great investors. So we raised a little bit of money. So we are very capital efficient. We have three people and a bunch of AI tools. So we are kind of small, small, small team and yeah, so well, we are in San Francisco, so our headquarters in Knob Hill. Oh yeah. That's pretty much global. And so I wake up every day and work on Globi.

That's cool. What are you learning about AI and how it works differently than the Visual Studio days or whatever tools you were using to build 2D apps or build apps that didn't use AI. What are you learning about building apps this way? The computer science behind it.

So. Well, I think, you know what's different, let's say in 2024 versus like 2023 for our company, is that I was actually so we had like a slightly different product last year was kind of like it wasn't drag and drop, but it was like a couple of blocks that you kind of mix together. And then we just realized that we over engineering things and not demoing the product enough. And in 2024 we just simplified this using large language models. Right. And I started doing almost daily demos. So it's like five demos a week. And I would say that's how we got an influx of new users who never ever thought that they will be able to build software. Right? Yeah. And I think that's kind of like the impact of AI tools or foundational models for our company. So I've decided to start demos on almost daily basis and I started simplifying things as kind of like lowering the entry barrier for almost everyone who are interested in this topic. Because I'm a builder myself. So I was a founding team engineer in two startups in Silicon Valley and indie game developer before. So it kind of changed my life like working on software. And it's kind of like a superpower, right. To be able to code something and we bring in this superpower to everyone. So that's our mission.

I love the mission and that's why I'm talking to you. I want more superpowers for human beings and AI certainly lets us bring that. Have you learned ways to talk to the AI that work better than other ways? You know, for instance, my friend Dylan talks about emotionally blackmailing the LLM and sort of saying, hey, if you don't get this app correct, I'm gonna get fired tomorrow. Something like that. All of a sudden it gives you a better answer than if you just ask it for. Build me an app. Right?

Yeah. You know, like we've done a lot of experimentations, but behind the scenes. Well, and now we have this new version coming up that going to support different foundational models. Gemini, it's almost done, but there's no date to announce, but it's coming soon. And Entropic Mistral models, what's recently interesting thing that I built. It's kind of like a debate app that you set a debate topic and GPT4O kind of debates. Gemini. And I'm going to Brazil this summer and I was just like, oh, shall I go to Brazil or to Europe? And GPT4O was advocating for or like globally powered by GPT4 or was advocating for Brazil and Gemini was advocating for Germany. And it's kind of like, it's very interesting what you could do. Right. But also from a Technical standpoint, right. So there is also what we're playing with is just like rag where you get context of your project. So let's say we are not building the foundational models, right. But we are using pretty much everything and we playing with everything in our kind of, I would think of all these foundational models, like maybe databases, right? And we're building something on top of databases and all this code that we built over the years. So we kind of put it in kind of argument data models. And with the mission to make not just quick prototypes, but actually fully functional software that you could use in, I don't know, production maybe, but it all kind of takes time, right? And all these foundational models kind of being built, they compete with each other and we using everything so well, we're trying to catch up. And I would say that's why I really like being frontier of on the front wave of technologies. And I was doing this in 2008 for iPhone. I'm doing this. I'm very excited about Vision Pro and this 3D world that I could build. And with AI, just like our goal is to use whatever is coming up like new update from foundation models and play with new techniques and try to make sense of it and use it in our context. So it's been awesome. It's really interesting world that we live in.

Getting a look. I love your your demos. I've been seeing them on X for a long time and it's just fun to be able to draw and talk and have something built. We should check in after WWDC again and see. Okay, you know, did something there catch your eye or were you disappointed or were you happy? You know, because you're the front line of the Vision Pro developers. You're one of those people who is pushing it hard and I appreciate that a lot. Where do we learn more about you and your company.

So well? I started being active on Twitter in 2024, so it's Jacob Ellen. My last name reads like LLM, but it's I L I N so well. And the company is Globum, So it's globe om.com and the AI engineer that we're building is Globi. So if you want to develop an app, you talk to Globi and find us on Twitter. I do almost daily demos.

Love you and keep it up and I can't wait to see what you build over the next few years because you're really a creative dude and I appreciate talking to you. I could talk to you about it so much. Been waiting for hours about what you're learning and what you're building.

Yeah, I'm down, so. Well, feel free to invite me next time so we could discuss.

Absolutely. I'll definitely be watching you and keeping you up as part of this show. Thank you so much.

Thank you.

I appreciate it.