Beyond Automation: The Rise of On-device Superintelligence

AI is no longer just about smarter chatbots. Div Garg of AGI Inc. shares how autonomous AI agents are redefining digital work by acting on behalf of users across apps and workflows. From building real-world benchmarks to scaling adoption, we discuss why go-to-market strategy and ecosystem reach will determine who wins in the agentic AI era — and why simply having AI is no longer a competitive edge.

Div Garg

Founder & CEO, AGI Inc.

Div is an AI researcher and entrepreneur focused on advancing general-purpose intelligence. He is the founder of AGI Inc., where he builds autonomous AI agents designed to handle everyday digital tasks. His work spans deep learning, reinforcement learning, robotics, and computer vision, with notable contributions to efficient learning systems and real-world AI applications. He has also taught Stanford’s first course on transformers and collaborated with leading AI institutions, including OpenAI, DeepMind, and Apple.

• Div Garg's LinkedIn
• AGI's website

Thank you for listening! This podcast is a passion project of AScaleX. Know more about us through the links below:

• Angelique's LinkedIn 
• AScaleX 's Website
• AScaleX's social pages
  • LinkedIn
  • Facebook

Transcript

SPEAKER_00 0:00

Maybe like 100 years ago, people used hand weave like clothes and everything was like man human made.

SPEAKER_01 0:04

Yes.

SPEAKER_00 0:04

And now everything is just like machine made, right? You just you have like some sort of like sewing machine or whatever, like different industrial machines that must manufacture all the clothes for you. And the coding is going through a similar uh phenomenon. Like all of code was human-written where like a coder is like manually writing each single line of code for like 100,000 lines, which sounds crazy.

SPEAKER_02 0:27

Hi, everybody, welcome to Ascend Stories of Scale. This is Angelique Strauss, and with me today is Div Garg, the CEO of AGI. Hi, Div.

SPEAKER_00 0:35

Great to be here.

SPEAKER_02 0:35

Yeah, thanks for joining us. So let's talk about you and AGI and what you guys are building, coming from NVIDIA, Stanford, Cobalt, and all of these. Like maybe give us a background of yours.

SPEAKER_00 0:47

I started my journey working in a lot of uh big companies like Google, Apple, Nvidia Uber. At that time, a lot of top security AI projects and then spent a lot of time working on like cell-driving cars, robotics, always fascinated about like what's next in AI. How can this be used to benefit all of humanity? And that really inspired me. Like, how do I go and do things like this myself? My first company, Multown, we were like the first AI agent in the industry. We basically innovated like browser use, computer use. Started as a side project in my PhD just after ChatGPT came out. Uh we built the first prototype uh two weeks after ChatGPT was there. This was way, very early. Very early, yes. This was before like there was a GPT-4 model.

SPEAKER_01 1:23

Okay.

SPEAKER_00 1:23

Or like there was an actual open air API. That journey has been really fun. I was doing that for two years. We have been funded by some like top investors in the area, like General Catalyst, and then a lot of other strategic ones. When we started, the technology was very early. Like there was not much uh in terms of like uh agents or even like um like vision models. We realized very fast if it building a new product from like a technology that's very early, but that's very groundbreaking, it's it's it's kind of like you're building something brittle where like it's like hard to like scale because there's just so many edge cases, right? So when you look at like agents right now, like OpenClaw, yes, they're like very hypey, but it's very hard to make them reliable, trustworthy.

SPEAKER_01 2:02

Yeah.

SPEAKER_00 2:02

And uh and so from my journey at Meltown, we initially started with like, okay, can we build an agent that's a personal agent that uh eventually billions of people can use? And we realized very fast, like, okay, that's a very hard challenge with agents given their technology. Yeah. And we found a lot of product market fit fit with uh developers, enterprises, and that's where that company ended up going. So when I was starting AGI, I was like, okay, like I want to like start something that's basically built from ground up, where we're doing our own research, we're basically like collecting our own data, training our own models. And so we spent off AGI as a research lab, focusing on like okay, like how do you build agents that are very secure and very trustworthy?

SPEAKER_02 2:37

Yeah, before you actually go into that, I saw your blog about you know you saying like our app's dead. So let's let's let's unpack that a little bit. What are your thoughts on that and walk us through that blog that you just wrote?

SPEAKER_00 2:49

Pretty much everything that we do is starting to get automated. And so we already see that with like uh cloud code and cloud cowork. Yes. And so a lot of like engineering aspired, sort of like business uh functions. We think like the next wave is going to be like how can the same thing happen pretty much like for everyone in their personal lives. And we believe that it's going to be something where we totally redesign like the current interfaces. So can you have a like an AI device where I can just talk to the device, you can just do things for me.

SPEAKER_01 3:16

Yeah.

SPEAKER_00 3:17

And if uh AI can go and use any software, any app you have installed automatically, then you don't need apps in the form factor they are in right now. Because a lot of those things were made for humans. Yes. For me to click and type and interact normally. But I think that that way is going to be because like humans will not be like clicking and typing and using devices the way they have been in the past. It'll be more like a voice-driven interaction.

SPEAKER_01 3:39

Yeah.

SPEAKER_00 3:39

And now we need to start designing for this world. Where like if things can happen just by talking to it or by chatting to the device.

SPEAKER_01 3:46

Yeah.

SPEAKER_00 3:46

And then we start building something like I will call it like something like Jarvis.

SPEAKER_01 3:49

Okay.

SPEAKER_00 3:50

From the movie, like the Iron Man movies.

SPEAKER_01 3:51

Yeah.

SPEAKER_00 3:52

And it's like it's like a living, breathing interface and it adapts to the users, it personalizes to you.

SPEAKER_02 3:58

That's actually my next question. Can you walk me through what that experience feels like? Because a lot of AI solutions today are very early. It's actually changing how we work, it changes how we even like think and decide on things. On how you're envisioning this product, can you just explain in your plain words, like how does it feel to have basically this assistant work for you?

SPEAKER_00 4:21

For anyone who has used a human executive assistant, like having a human EA is almost a superpower because there's so many things you can delegate. Right now, like I would say only maybe 1% or less of the population actually have an EA. It's expensive. What we want to do is we want to enable like the remaining 99% of the population to have the same capabilities. Like in have an AI, like almost like uh it's living with me, it's it's working for me 24-7. Anything that I like, it's like all the small chores, like say like paying a bill, like even like something like ordering food, booking restaurants, like helping you travel, all of those chores can be fully delegated.

SPEAKER_02 4:55

What are the main problems do you think it solves? Because if you really think about it, like how it currently works today, it's working, right? What kind of value are you giving to your users?

SPEAKER_00 5:06

Big thing right now space in the space is like there has not been any uh killer products uh with agents that have actually taken off.

SPEAKER_01 5:13

Yeah.

SPEAKER_00 5:13

And so if you look at like a lot of the AI browser waves that happened last year, nothing was very sticky. Even if you look at OpenCloud, it's not very sticky. It's actually there's so many vulnerabilities where things can break and uh so many loopholes, right? We want to build something that like people can trust. Okay, like if I'm using this, I I trust the AI to do the right thing. It won't mess up, it won't cause problems for me, it won't send like a wrong message to my friend.

SPEAKER_02 5:34

Yeah. And uh hallucinate.

SPEAKER_00 5:37

Yes. And and that becomes like something like you have to like really, really think in the in and be thoughtful about like how you're architecting and designing the product. Like from day one, you have to have the principle, okay, like how does this product win human trust? How do we climb all the uh the la the last miles of like AI reliability? And that becomes a lot of stuff around like how do you build the right systems, benchmark them, evaluate them, make sure they're scalable to lens of people. And also that's a human challenge. I think we're still like maybe like uh two years away from having the students like really, really like be into something that I knew everyone trusts for everything.

SPEAKER_02 6:09

Yeah. I mean, can you also unpack to us like because I think a lot of people interchanging them, having an assistant and an agent.

SPEAKER_00 6:16

Yep.

SPEAKER_02 6:17

What's the difference?

SPEAKER_00 6:18

Yeah, I would say it's it's interesting.

SPEAKER_02 6:20

Um because sometimes it's also confusing. Like it's very interchangeable.

SPEAKER_00 6:25

Yes. Like I would definitely say it's uh it's a very ambiguous, like where the boundary between these two terms are.

SPEAKER_01 6:31

Yeah.

SPEAKER_00 6:31

And uh I would say agents is something that just became like the hypey word.

SPEAKER_01 6:35

Yeah.

SPEAKER_00 6:36

Like uh the buzzword, things are happening automatically for you. And developers really like the word because it's kind of like relates to like agency and being able to like do things. Uh, when you think about something like an assistant, that feels more consumer-y. Where like this is something that like a maybe a normal user is using and it's helping them as part of their daily life.

SPEAKER_02 6:52

Okay.

SPEAKER_00 6:53

Whereas like agent is something that's more, I will say, like, a developer word at this point.

SPEAKER_02 6:57

More enterprise.

SPEAKER_00 6:58

More enterprise or anyone like uh like doing a B2B kind of motions. Like prefers to use agents.

SPEAKER_02 7:04

Being very early on, right? Like I think your product is also ahead of the curve. Where do you see the competitive differentiation of this for what's available in the market today?

SPEAKER_00 7:14

One thing we're trying to do is uh we're building a lot of our technology from the ground up. And that includes like the models, the way we're architecting them, the way we're training them. And so what we've done is we've built like small models that can run directly on the hardware itself. So, can we make these models tiny enough that they can fit on your phone, PCs, or any other form factor? That allows to have like very, very cheap cost. And that's not possible if you're just using something like Gemini or like some other like um like cloud or other APIs. That gives us a big advantage in the market. We're like uh we're just not a wrapper company. It's not easy to like build the same thing, like high amounts of effort.

SPEAKER_02 7:48

And you did mention earlier privacy, I mean security, a very secure solution. How are you dealing with that right now in terms of security and privacy? If you remember when ChatGPT launched ChatGPT agents, that was actually one of the problems they had. For some reason, when you share your credit card information, it stays in the and then they they retract that they change a few things to fix that right away.

SPEAKER_00 8:10

Yeah, that's a great question. And so we actually think about this almost every day. And so this is like top of mind for us as a company. Uh this becomes a question of like one is like how much a user trusts the agent. Yes. Especially if it's new, very privacy first. Like if the models are running on device, you don't have to like worry about uh your data getting stolen, for example.

SPEAKER_01 8:29

Yes.

SPEAKER_00 8:29

And so you can even trust the agent with your passwords, credit card data, a lot of sensitive information because you can be guaranteed, okay, like this is running locally on the hardware itself, it's not gonna go into like a cloud. And so you don't have to worry about someone hijacking that connection or even like a third party that might like show advertisements or something. So just like having that level of trust, okay, that this is just fully private is great. And users love that, especially if like it has to do with uh visual data, your screenshots, things that are happening on the phone. The second thing is like also like around permissions. For example, you probably don't want the agent to like work on your banking app, for example. Because there's a lot of things that could go wrong. But like it's maybe totally fine if it's like maybe you know managing your emails.

SPEAKER_02 9:07

But I think we might get there too, right? In the future.

SPEAKER_00 9:10

Yes.

SPEAKER_02 9:10

If you really want to automate everything.

SPEAKER_00 9:12

You will get there. But I think we think about like trust as a ladder.

SPEAKER_02 9:16

Uh-huh.

SPEAKER_00 9:16

And it's like something that you climb up over time.

SPEAKER_01 9:18

Okay.

SPEAKER_00 9:18

So like even if you look at human relationships, we like if when you meet someone new, they're kind of like strangers, you become acquaintances, then you keep climbing the ladder until you have friends, closer friends, and so on. And so I think that the same thing will happen with AI.

SPEAKER_01 9:30

I like that. Yes.

SPEAKER_00 9:33

And so the first time someone uses an AI, they'll be like, okay, like I don't know if I trust this uh new thing. Like I'll give it some like simple tasks, it's able to do the simple tasks really well. Maybe like manage my calendar or emails. And it's like, okay, like yeah, like it's it's good. I trust it more. Now I'll give it something that's more that's more personal and sensitive. Maybe can it go and buy things for me? Can you can it order food? Can it maybe like book uh restaurants? Then I was like, okay, like now I trust it enough in my like sensitive zone.

SPEAKER_01 9:55

Yeah.

SPEAKER_00 9:56

That okay, like it can like manage my money and like from banking and like uh yeah, and then then it's in the level, like, okay, like really, really trust this. And then it's very hard to like replace that.

SPEAKER_02 10:06

Yeah, it's very sticky at that point.

SPEAKER_00 10:07

Yes.

SPEAKER_02 10:07

Yeah, yeah, yeah. Sounds good. I was looking at your website. I did sign up for the wait list because I was really curious about how everything's gonna unfold uh for your product. And I saw that you have 162,000 people on the wait list as of yesterday. Part of Ascend is talking about scaling. And I I want to ask you about how did you drive this demand this early?

SPEAKER_00 10:29

Yeah, well, just say we uh have seen uh like enormous like user demand of people who just really want something that can like just uh do things for them, especially on their phone. And uh so if you look at like products like Siri, they have been around for almost a decade now. But uh and they have a great vision, but just like technically they are it's just not good enough.

SPEAKER_02 10:48

Yeah, it's hard to talk to Siri. They don't she she's not listening most of the time or even interpreting what I'm trying to say to her all the time. So yeah.

SPEAKER_00 10:57

Definitely. So the thing is like there's a lot of suppressed demand from users who really want to use those things. Like everyone really wants a personal assistant on the devices, but uh it just has not been delivered properly. And we think this is the right era where this can finally happen.

SPEAKER_02 11:11

So, what are the strategies that you did behind there? I saw that you're doing some referral. Can you talk a little bit more about how you start that or what are the strategies behind you pulling 162,000 wait lists this early?

SPEAKER_00 11:26

Yeah, a lot of it just was like like we built like a really viral video, for example, and people love watching a launch video. I think that that helped us like just get above like punching through the noise and having a lot of momentum in the market. And we're also like also like the first product in this category. People have been toying with ideas, but we're the first thing that actually works, you can start using it. And everyone who has used it for the first time, they're like mind-blown, we're like, wow, like this is so good, like there's so many possible possibilities. And that's what drives this demand. We're like, okay, like uh people use it and they talk to their friends, wow, like you should try this out, and then they their friends talk about it, and that becomes like a viral loop where like we create a lot of word of mouth and that kept keeps spreading over time, and the more you use it.

SPEAKER_02 12:06

And then let's talk about your partnerships, because I think that also helped with at least the exposure of the brand this early. Recently you announced a partnership with Visa. How did you land? Don't have essentially the product yet, but you're already able to land this big partnership for AGI. How did you do that?

SPEAKER_00 12:23

We we just have been like uh like uh like early in the field and we've been very visible in uh like effortshift champion and like how we have kind of like uh built a community around us. That has resonated also with like a lot of our partners, like Visa, for example. We're working with them on agent payments. So if our agents are running on devices, they're doing things for you. How can we make sure these things are secure? So like can we make sure like the any like credit cards you use that's powered by Visa is uh securely like uh used by the agent for any transactions, nothing can go wrong. And uh we built a lot of like mechanisms there to ensure like uh proper encryption, maybe even like building having like virtual cards that an agent can go and use.

SPEAKER_02 13:04

And then in terms of when you're building, based on your experience, I I would assume like you also have a lot of a network uh supporting AGI and all of these things, like how does that help in building a company?

SPEAKER_00 13:16

I would say the best way to create a company is to have to think of it uh like a movement.

SPEAKER_02 13:21

Yeah.

SPEAKER_00 13:21

Like your goal should be like, okay, you want to create a movement and like uh enable something new, which is fundamentally useful and better for society. At that point, like a company is almost like the thing that arises automatically. So if you're able to succeed in creating a movement around an idea that really resonates, and then that just gives enough support because a company at the end of the day exists to basically like maybe like enable some sort of economic value to people around you. So like if that other business is consumers and they only want to use your services if they resonate with you, they like you, like like for your products. They believe in you, they believe in the mission. And the the more you can show that, uh, the more supporters you are able to build around your idea.

SPEAKER_02 14:01

Where are you guys right now in terms of like funding stage? And and I guess maybe we can talk a little bit about that, the challenges or things that you have to go through as a founder bringing a company this early on in the market.

SPEAKER_00 14:13

So I would say we raised our first round that was last year. Uh we called it that a pre-seed round. Um, and we are back to have some like really great investors, such as like Melo Ventures, a lot of phenomenal angels. Some of them are like VPs at like DeepMind, OpenAI, and then we have a very supporting network around us. I think we're as a startup like you, there's always like funding activities happening around you. And so we have like things we're doing there. And so we'll be like announcing something.

SPEAKER_02 14:37

Amazing. What excites you most in the next three to five years in terms of AI development?

SPEAKER_00 14:42

Fundamentally, the the way how we do things as humans uh is changing very fast. So even in like engineering has changed a lot. Like uh it used to take a lot of like uh an army of coders and engineers who are like handwriting everything. Now it's just like you know, you ask the AI to do things for you.

SPEAKER_01 14:57

Yeah.

SPEAKER_00 14:58

I'll say I will give one analogy here. It's like uh hand knitting, like a clothes, for example, versus just using a machine.

SPEAKER_01 15:03

Yeah.

SPEAKER_00 15:04

Maybe like 100 years ago, people used hand weave like clothes and everything was like man human-made.

SPEAKER_01 15:08

Yes.

SPEAKER_00 15:08

And now everything is just like machine made, right? You just you have like some sort of like sewing machine or whatever, like different industrial machines that must manufacture all the clothes for you. And the coding is going through a similar uh phenomenon. Like all of code was human-written where like a coder is like manually writing each single line of code for like 100,000 lines, which sounds crazy.

SPEAKER_02 15:26

If you could change one misconception about agentic AI, what will that be?

SPEAKER_00 15:31

I feel like agentic AI is it's a kind of like a buzzword. Like everything that's not a chatbot is is like people think it's like an agent.

SPEAKER_02 15:39

Yeah, yeah, yeah, that's right.

SPEAKER_00 15:40

And I think there's a misconception of like what that means. Like, what exactly is this supposed to do? Is it like some sort of business automation tool? Is it something like a personal assistant? What encompasses like an agent? That lack of definition has created a lot of ambiguity in the market. I think this is like one thing we are trying to change. Like it's like maybe having a proper definition, especially for the products you're building. Yeah. Like uh like what is expected out of it. And can we have it like maybe replicate a lot of human interaction? Like how humans interact with technology, how to do that do things. Can we do a lot more of like human-like interaction or build agents that are very human-like? And that's one thing that we are trying to create a category out of as as we grow the company.

SPEAKER_02 16:18

Sounds good. Because I think it's when also the SaaS boom happened, wherein there's every use case that you can think of, there's a software for it. Now we're going to the AI side. And so there's a lot of noise coming out of it. And so a lot of founders are also having a hard time putting their product out there because even for, for example, for B2B products, let's say I'm working with a marketer or the head of sales or whatever that is, they don't know which way to go because there's just so many products out there. How do you actually make sure that you're not part of the noise and actually changing something? Or at least like if you wear that founder hat and not as a founder of AGI, what will you look for in the AI company to know that there's really something going on about this company that you're worth looking out for?

SPEAKER_00 17:06

Yeah, but it's just like, is there something that's novel? Code breakthrough in the technology, for example, is there like some sort of like product innovation or like some sort of business innovation? So it's like it's is how does this company stand out of the pack? And I think that's very important because like right now a lot a lot of companies are rappers, the field is moving very fast. And so like all the foundation model companies, they're like basically like they're growing very fast, they're expanding their product lines. And a lot and that creates a lot of disruption because like if you look at the coding agent market, now everyone just is using cloud code. And there used to be like 10 plus different coding agents.

SPEAKER_01 17:39

Yeah.

SPEAKER_00 17:39

Is there some like technological or some sort of fundamental breakthrough that's defensible over time? And how can that keep growing into becoming something really big?

SPEAKER_02 17:48

And really sticks because there's there's really a lot of products that I've seen that more of a hype product, like everybody's using right now, but you can see that, oh, maybe two, three years, that's not gonna be a thing anymore because it's more of like a virality uh thing over like really value. So I I totally get what you're saying there. In terms of what to watch out for, you did mention that you have few releases coming out and also funding, closing out and stuff like that. What else should we watch out for from AGI in the next few months and where should they reach out if they're interested to knowing more about your product?

SPEAKER_00 18:21

Innovating a lot of uh cool new things, uh, especially around like as I mentioned, like on-device um intelligence. And so we'll be announcing some major partnerships there. We'll also have like a lot of like developer uh tooling that anyone who's a founder or building a business can use that technology. Yeah, we're very just very excited. We think the next big wave is going to be more centered on on-device intelligence, right? Like, how can we decentralize a lot of capabilities and make it like very, very cheap?

SPEAKER_02 18:47

Uh I was thinking of posting, but you touch up a topic that I'm really interested to talking about. Is AGI mostly like B2B or B2C or both?

SPEAKER_00 18:55

Um, we are almost like I would say like a prosumer product in a sense. Okay.

SPEAKER_02 18:58

So we're starting more Because from what I've seen on your website, it seems like you're B2C.

SPEAKER_00 19:02

Yes.

SPEAKER_02 19:03

But from that answer you just mentioned earlier, that founders can use it also, like that could be a B2B play.

SPEAKER_00 19:09

So I I was just curious, are you looking at both? Yeah, I would say we are going more like uh like I would call it product first.

SPEAKER_02 19:16

Okay.

SPEAKER_00 19:16

Like how do we build the best uh experience around an agent that actually works?

SPEAKER_01 19:21

Okay.

SPEAKER_00 19:21

And it's very reliable. And uh we're starting with like more like consumers who are like power users, need a lot of like uh help with productivity and like uh getting things done.

SPEAKER_01 19:30

Yeah.

SPEAKER_00 19:30

And down the road we're thinking of like maybe like exposing some of our capabilities to other developers. So if someone wants to build an app that has agent features, can they have a way to embed our agent as as part of their thing?

SPEAKER_02 19:42

Yeah, that they can integrate with you to actually leverage that to build their agents within their platform.

SPEAKER_01 19:47

Yes.

SPEAKER_02 19:48

Um, I have a few hot takes, just quick like seven questions, yes, no, or maybe or pass um that I'll just go through. Okay, let's do it. Will AA native companies replace entire SaaS categories?

SPEAKER_00 20:07

Yes.

SPEAKER_02 20:08

Is uh the biggest AA opportunity still invisible to most builders today?

SPEAKER_00 20:12

I think so.

SPEAKER_02 20:13

Is human in the loop just a temporary comfort blanket before full automation?

SPEAKER_00 20:17

Yes.

SPEAKER_02 20:18

Are we heading toward a world where execution matters more than ideas?

SPEAKER_00 20:22

I feel ideas will matter more.

SPEAKER_02 20:23

Will capital matter less than compute in the next decade?

SPEAKER_00 20:27

No.

SPEAKER_02 20:28

Will distribution matter more than product in an AI saturated market?

SPEAKER_00 20:32

I think so.

SPEAKER_02 20:32

Is AA first already becoming meaningless positioning?

SPEAKER_00 20:36

Um, pretty much. Like everyone's AF first at this point.

SPEAKER_02 20:39

Yeah, yeah. Everybody pivoted to that, right? Like uh all the SaaS companies became like, oh, we're AF first company, even though the company has been around for 20 years or something like that.

SPEAKER_01 20:47

Yep.

SPEAKER_02 20:53

Thank you so much for joining us. This has been really fun. Hopefully, you can do another one once you release all of the new things that you're doing this year. And good luck with everything. And we hope the best for AGI.

SPEAKER_00 21:05

Thank you.

SPEAKER_02 21:06

Thank you for listening at Essence Stars of Scale. This is Anjali. We'll see you next time.