Underdog launches private on-device AI agent with payment-based monetization — no data centers, no subscriptions

Speaker 2: We have a lightning round because we're running late. We have Sigil Wen from Conway Research, the founder. Let's bring him in to the studio. Sigil, sorry for keeping you waiting. How are doing?

Speaker 10: Doing very well. It's been crazy.

Speaker 2: Yeah. Congrats on the launch. Break it down. What'd you launch? What are you building?

Speaker 10: Yeah. So I'm building Underdog, which is your private personal AI that runs on the device deep at DRM. And, oh my god, you guys have the sound bird. For the underdog.

Speaker 2: Yeah.

Speaker 10: Oh, god. Alright. What

Speaker 2: device are you most optimistic about? Today, in Strathecari, Ben Thompson was lamenting some of the frustrations that he'd had with Apple products. There's all this debate over privacy and pushback. Are we going to see Linux on the desktop? Are we going to all be using Android phones because, you know, we're we're we're in this AI world and we want the best? Or or are you building in a place where you'll be kind of hardware agnostic?

Speaker 10: You know, our goal has always been to provide free, capable, reliable, and private AI for the masses. Mhmm. And that means building for all devices. Mhmm. And so we're off the bat supporting all of Apple's products. Our iOS app is about to launch. Cool. You know, we're working with NVIDIA, so we're we're gonna ship to all NVIDIA and Microsoft Windows devices, Linux, Omari, Android, iOS. Cool. I think that you've already paid for the chip, and it's actually significantly more powerful. And, yeah, like, we're doing a lot of optimization to get more bang out of the buck Yeah. For the hardware that you already have. Trade offs of that optimization?

Speaker 2: I imagine that you could maybe take us back. You're probably not serving Astra or Fable level intelligence, but I imagine the model that you're serving is spiky in interesting ways. What's near for you... What what near frontier capabilities are possible on device? What do you wanna go to the cloud for or you need to at least in the short term? Obviously, next year might be a different discussion.

Speaker 10: Yeah. So in my manifesto, I write about underdog's law Mhmm. Where, basically, the frontier models of today will run on your devices in six months. And we can see this quantitatively Mhmm. Where underdog 27 b, which we post trained and r l on top of Quinn Mhmm. Quinn's 27 b model. If you look at artificial analysis, it scores better than Claude Opus 4.6 max, which is the frontier model from Anthropic from six months ago. Okay. And we're we're gonna see this trend continue where intelligence is getting a lot smarter, a lot faster Yeah. And we can make them run really fast on the devices you're runs six months later, but it's gotta be slower. How much slower is it? No. It's... So we do a lot of inference optimization and data efficient. We actually released Husky, which was an inference engine, and it is Oh, god. Four forty five x. That's good. We

Speaker 4: got our dogs barking. Yeah. So what... Like, where... You you said free. That stood out to me. Is that... That's because one, when it's possible, but what... Where where where and how are you gonna make money?

Speaker 10: Yeah. So right now, a lot of existing business models, they have to charge you their subscriptions or, like, sell them ads and, like, harvest all their datas to a data center. So we don't have any databases or data centers. It just runs on your device. Right? And what's what's fascinating, I don't know if you guys remember, but earlier this year, I was on here talking about Web four. And I was I was I was one of the first to talk about how AI can write and transact to the world. Sure. I built one of the first AI agents. I could earn some existence, acquire compute and stuff. Oh, yeah. And this enabled, you know, a lot of agentic commerce Mhmm. To happen afterwards. Like, Patrick Collison, the CEO of Stripe, is our biggest angel investor. And I think that the economy will be reformed, but with reliable and private AI agents that are acting on behalf of users. And the the business the business model side is completely gets reinvented. Advertising is such a dark pattern boomer business. Right? And the cool thing about... I

Speaker 2: am so skeptical about this. I mean, it's like it's gonna hurt you. It's like Android versus Linux. Like, yeah. I I I get that ads are annoying, but, like, one is way bigger. No. No. No. You can continue. We're... We'll defend ads until our dying day, but but continue to I spend his money, save both these ramps. There we go. Yeah. Yeah. He doesn't hate that. It didn't hurt. He doesn't hate it. It's fine. No. Anyways, continue. If an agent,

Speaker 10: the the amount of effort it takes for you to scan the entire Internet, the hundreds of vendors for, let's say, a flight Sure. Is... If you're a human doing it, it'll take you hours. If you're an agent, it takes seconds for you to do that. Yeah. And so Underdog can, you know, browse the entire web for different fights... Flights, look for the cheapest price, help you save hundreds of dollars Sure. And then it'll generate a unique card.

Speaker 2: Yeah. And then Oh, interesting. There we go. We've been talking about this. Okay. You monetize payments.

Speaker 4: I love this. As

Speaker 2: long as you're making a, you know, a rake on this, we're in. Yeah. So I was I was

Speaker 4: saying this about instinct because instinct and and Muse, they clearly are launching all these commerce partnerships but How will they launch? If you wanna... The the best way it seems like to to generate some some revenue Yeah. Is to just introduce a card so that you can get that interchange. Yeah. This is super cool. When when... This is cool because you also don't have the insane costs of Yeah. Yeah. Of Of course. Yeah. Sorry. Last question. Is

Speaker 2: it available? People demo it? How can people onboard? When is this when is this gonna roll out? Because you don't have the crazy inference costs maybe, so you can scale this pretty quickly.

Speaker 10: We... On Friday, we opened to an invite only beta. Okay. Over the weekend, more than around, like, 5,000 tweets went out of people asking for invites. Cool. And so I'm I'm pushing builds and improving it, and we'll be able to do everything that, you know, instincts and Woah. Shots fired. And, like, you know, it's... It already... Users have already been unsubscribing from, like, WhisperFlow, granola, superhuman because it's all on device. Why can't guys stream your data to data center? Okay. You know, when it can be just as good on device fully private. And Instinct and these these apps are are, like, major targets for, you know, advanced AI cyber attacks as we're seeing with, like, OpenAI, you know, and, like, we have met those class models that are all open source now. So I think this is the most important product to k. Build, and I wanna bring it. Pretty pretty savage to be the anti data center AI company. Yeah. Very interesting. Unique positioning because we haven't we haven't seen this yet. We bring the model to your data, not the data to your model. I like that. I under... I under... Good

Speaker 2: framing. Thank you so much for coming on the show. Have a great rest of your day. Very cool pivot. We'll talk to soon. And excited to try it out. Goodbye. Good to see you. Thank you. Let me tell you about Railway. Railway is the all in one intelligent cloud provider. Use your favorite agent to deploy web app servers, databases, and more while Railway automatically takes care of scaling, monitoring,