�鶹��

Div Garg, a Stanford University dropout, was thinking of building his own AI company when OpenAI came calling. He was faced with a choice: accept a near-million-dollar job offer from OpenAI to work on someone else’s projects or create his own AI company and tackle the pain points in AI that mattered most to him. 

He chose the startup path and has since founded , a startup focused on building AI agents that can run on mobile devices. The company raised its first funding round, an $8 million pre-seed/seed round, in June 2025 and is currently raising another round, details of which are still undisclosed. Garg wants to build something like an advanced Siri, he tells �鶹�� in a new interview. His app has received around 500,000 signups for the waiting list in three months. 

This interview has been lightly edited for clarity and concision. 

His beginnings

Can you start from the beginning of your career in AI?

I’ve been in the AI space for almost a decade. I worked at several big tech companies, including Google, Apple and Nvidia, on top-secret AI projects at the time, involving things like self‑driving cars, robotics, and related areas.

After that, I was doing a PhD in AI at Stanford, working on reinforcement learning and building agents. I’ve accumulated over 3,000 citations and probably more than 10 patents. I eventually dropped out of my Stanford PhD to start my first company.

What was the first company you started?

Initially, I started a company called MultiOn, which I ran for about two years. We raised over $30 million from top VCs in the Bay Area, including Joe Lonsdale’s 8VC, Catalyst, Foreign Ventures, and a number of VPs from OpenAI, DeepMind and others.

Recently, I spun a new lab out of that company, and that became AGI Inc. Now we’re focused on building a more research‑first and trustworthy AI product.

His startup’s focus

What is AGI Inc. focused on now?

We’re focused on building agents that can run on your own devices, on the edge, and bringing a personal assistant to every phone and every device.

At its core, it’s, ‘Can I talk to my phone, and can it do things for me automatically?’ We think the future is an “appless” phone, where instead of you manually using apps, everything just happens automatically through an AI assistant. We want to enable that future.

A huge portion of our lives is now digital. Most people spend something like 80% of their time on phones, computers, and other devices. There are countless small, repetitive, and boring tasks in that digital life. We wanted to build an AI that could automate those repetitive tasks so you can focus on what you actually care about.

Our core product is an AI that can operate your phone using natural language. Think of Siri, but it actually works. It doesn’t constantly make mistakes, and it works across any app.

You can say things like: “call me an Uber to my office,” “book me a dental appointment,” “order my favorite coffee,” and “reply to my emails.”

Basically, anything you do on your phone today, we aim to let you do hands‑free, using your voice, through the AI. That’s the product that originally had 160,000 people on the waitlist. We now have about 500,000.

Imagine everyone having a “super assistant” that handles all of those tasks automatically. We felt that it was possible and that we were at the right moment technologically to build it. That’s why we started this journey.

Why he turned down OpenAI

You turned down a near‑million‑dollar job offer from OpenAI to start AGI. Tell me more about that decision.

I’ve always been excited about creating something of my own that can have a lot of impact. Startups are uniquely suited for that.

In a big company, you’re often a cog in the wheel. It’s hard to have real decision‑making power; you’re usually solving problems that have already been defined and prioritized. I wanted to work in a new field where not many people were working yet, so I could approach it from first principles and do something genuinely novel.

Running my own startup lets us focus obsessively on making users happy, deciding which use cases we should solve, and building a truly great company around agents. That autonomy and potential for impact were key reasons I chose the startup over the job.

What deciding factor ultimately made you say no to a million dollars?

We were already getting amazing traction. Users loved what we were building. I even ran a poll on Twitter, where I have a large following of , asking whether I should join OpenAI or pursue the startup. About 500 people responded, and the overwhelming majority encouraged me to continue with the startup.

I saw that the work we were doing was meaningful and had the potential to impact a lot of people. You can always find a job later, but startup opportunities are time‑sensitive. If something looks like it could become really big and impactful, you want to seize that window. That’s why I decided to keep building the company.

Growth and virality

How did you build AGI to have 500,000 people on the waiting list? What was the draw?

We had very compelling use cases, and people loved the concept. We showed that we could automate many of the things you do in daily life. For busy people—whether business owners or employees overwhelmed by tasks—that idea of an AI “sidekick” is powerful.

We built use cases like automating lunch ordering, handling repetitive workflows in tools like Salesforce, replying to emails and automating LinkedIn. Those resonated deeply. We also set up viral loops through referral codes and sharing. People recommended us to friends, and we went globally viral in multiple communities, more than once, which helped build a strong brand.

What does it take to go viral?

It starts with understanding what the market really wants. Is your product truly useful and compelling? That’s the foundation.

On top of that, you need content—especially great videos—that clearly explain the product and make people say, “Wow, this is great; I want to use it.” If the product is compelling enough, going viral becomes much easier.

What are some challenges you’ve encountered while building AGI?

One big challenge has been working with large consumer hardware companies, especially phone manufacturers like Samsung and Lenovo. We spend a lot of time demonstrating our technology to their leadership—VPs, SVPs—and turning them into champions internally so they’ll consider embedding our AI into their devices.

Another major challenge is agent reliability. Today’s agents often fail; they have issues. We focus heavily on making sure our agent reliably does what you ask. For example, if you say, “Call me an Uber to my office,” it should do the right thing every time.

Is this something that could eventually show up on Apple and Samsung phones?

Yes. On the Samsung side, it’s easier because we’re already on Android and we have partnerships with Samsung. iOS is the next step. We’re working toward our first iOS product and plan to announce it soon—within the next month.

Competitive edge and revenue

How do you stand out from competitors?

We’re essentially the only product in the market delivering what we’re doing at this level: a voice‑driven AI that can operate your phone across apps and is live in a real beta with active users.

There are adjacent efforts, but we haven’t seen others build an agent this powerful and reliable on phones yet.

How much in sales did the company do last year, and what are you projecting this year?

We currently have at least $1 million in revenue and expect to be in the $20 million range this year, possibly more.

How long did it take you to see consistent monthly revenue?

It took at least the first six months. During that time, we ran many experiments to figure out what created real value and what got people excited enough to pay. Then we began locking in the product and features that people were actually willing to pay for.

What kinds of products and features are people paying for?

A lot of the value centers on hands‑free control. For example, you might be in the car and want to use your phone without getting distracted: automatically replying to meeting invites, sending messages, or looking up information.

We also focus on scenarios where Siri fails today. Many people try Siri for all kinds of tasks initially, but end up only using it as a timer or clock because it doesn’t perform well on other tasks. We’re building a more consistent experience that works across almost anything you want. It’s like Siri on steroids.

When things go wrong

As you’ve built this business, can you recall a specific instance when something went very wrong? How did you fix it?

Our system was good at tasks under about 50 steps. For workflows with 1,000 steps or more, it became easy for the agent to make mistakes. Some users tried it on risky tasks like online transactions or complex internal infrastructure. We got complaints — things like “It messed up my AWS account” or “It didn’t handle my software correctly.”

Our response was to create clear guidelines about what the product was ready for, and what was still future‑looking. We also doubled down on safety. AI is not 100% reliable out of the box; it improves over time. It’s similar to self‑driving cars: early Tesla Autopilot made mistakes, but over time, it improved to the point where it can navigate most roads. Our agents have followed a similar trajectory.

What are some things AI still cannot do well?

Anything deeply tied to banking and finance carries obvious risk. If an AI is interfacing with your financial apps, you have to ensure it doesn’t accidentally transact the wrong amount or send money to the wrong recipient. There are security risks: someone might try to hack your agent and trick it into sending Bitcoin or wiring $1,000 to a random account. Preventing that is a major focus for us.

Advice for founders and future steps

What hard, concrete advice do you have for founders?

Narrow your focus. Once you have a vision, you need to figure out how to bring that vision to life and who your target customers are. Don’t try to solve every problem for everyone.

Instead, identify the one core user who really wants your product. Get your product into the hands of the first 100 people who love it, then figure out how to retain them—how to make sure they keep coming back. Only once you’ve nailed that should you think about expanding your focus. Hyper‑focus on one thing you genuinely care about and do it extremely well, instead of chasing 100 things.

What are your strategies for retaining customers?

It comes down to experience. If customers love the product and the experience is excellent, they have no reason to leave — especially if it’s doing important jobs for them.

We keep improving capabilities and releasing new features. We also run a newsletter to highlight updates and upcoming improvements, keeping users engaged and informed about what’s next.

Looking ahead, what do you envision AGI tackling next?

Once we’ve fully nailed the experience on phones, we want to make the assistant better and more personalized. That includes remembering things about you and being proactive rather than reactive.

For example, the assistant might know you have a dinner tonight and say, “Can I help you find an Italian restaurant and book it? Here’s some info about the people you’re meeting.” Or it might see that you have a trip to Paris coming up and automatically organize a seven‑day itinerary with the best places to visit and book everything.

The goal is an assistant that doesn’t always need prompts — it knows you, understands what you like, and acts on your behalf.