Your phone is AI’s newest hardware

Ryan Donovan (0:08)

Hello, everyone, and welcome to The Stack Overflow Podcast, a place to talk all things software and technology. I am your host, Ryan Donovan. And today we are talking about what it takes to engineer agentic processes, tool use on the mobile devices. And my guest for that is Div Garg, who is the founder of AGI Inc.

So welcome to the show, Div.

Div Garg (0:30)

Hi. Great to be here.

Ryan Donovan (0:32)

Before we get into the topic of the day, tell us a little bit about how you got into software and technology.

Div Garg (0:37)

Yeah.

I've always been fascinated by software. I was like nine years old. We used to have a bunch of PCs in my house. I was always thinking around, see what you can build. Yeah. I think I was always fascinated, like what is this box, and how can it do all these things. And this was in India, and then I did a lot of stuff where there's a bunch of international Olympiads and so on, and they're coming to the US. And this was when I was 16 So this was when I came to the US.

I was doing a lot of stuff in physics at the time. So like a lot of international physics Olympiads, and my parents actually thought I would become a theoretical physicist, and I was actually thinking, I should try to get the Nobel medal in physics or something.

Ryan Donovan (1:17)

Yeah.

Div Garg (1:17)

And then it kind of hit me at that point, because one of the biggest things that's going to happen in my lifetime is going to be AI. And as soon as you have artificial intelligence and all these things are working, that is the next evolution of computers and where things are going. And it will be very fascinating to work on that. So that's actually what actually caught me into computer science in my undergrad. I was like, this is happening, and I got to actually go and study it.

That's what made me into a CS major, and eventually I ended up joining Sanford in the PhD program, focusing on AI and agents. So that was really fascinating. Nice.

Ryan Donovan (1:50)

Obviously, with the AI, it's branched out into agents. And I think ever since the open call, everybody's been thinking about how these agents can run their computers.

Div Garg (2:01)

Yes.

Ryan Donovan (2:02)

But today we're talking about running a mobile device, all on the mobile device. Is that right? Is that what you're doing?

Div Garg (2:08)

Yeah. Our goal is to imagine you could replace your whole computer with a phone. And theoretically, that's possible because all the computers are already there. Your phone's now having almost in front of memory, like this space.

And so the question is just like, how do you make your phone useful enough that all your work can be done from your phone? And can you have a smart enough AR to do this?

Ryan Donovan (2:29)

Well, and I think that's the thing. Anytime I talk about devices and program for devices, there's a lot of constraints for it, right? Especially when it comes to AEI., where even on desktop, you normally have a hop to the cloud.

How are you thinking about getting the AEI to live entirely on the device?

Div Garg (2:48)

Yeah. I think that's where we're doing a lot of innovative work as a company.

So fundamentally, we are a research lab, and a lot of our work is on on-device models. So we actually have the world's best models that can run on-device to take actions for you. And they can work across any third-party apps. They can ask you, like, can you call me an Uber? Can you order me something on Amazon? Get you food on DoorDash? And we'll just generalize to anything. So like there's millions of apps out there, and our AI just works across everything.

Ryan Donovan (3:16)

I mean, world's best device app is a big claim, big claim. So are you doing that by creating a specialized model that's just doing tool calling actions, that sort of thing?

Div Garg (3:30)

I know, it's more than that. So it's like it can actually mimic a human's interaction.

So it's like the industry term for this is like computer use or mobile use. So the AI has learned how to like interact with the screen. So like touch, type, scroll. And like whatever you can do on the phone, it can like basically like do the same thing as a human.

Ryan Donovan (3:49)

What sort of model is that? Is that sort of transformer model that's gotten all the press, or is it a sort of more standard deep learning model?