Defense robotics startup Medias raises $30M seed from Kleiner Perkins to automate wire harness manufacturing for the defense industrial base

Speaker 1: play it.

Speaker 3: Three customers for you. That'd be fantastic. Sounds like a blast. Thank you guys so much for having me. Love your show. Yeah. Great. Good to meet you.

Speaker 2: Let me tell you about Codex. Codex is a powerful workspace for getting work done with AI agents. Whether you're writing code, analyzing data, creating content, or automating business workflows, Codecs helps you move projects forward from start to finish. And we have our next guest, Justin Seeley from Medias, co founder and CEO, taking powered robots to defense factories. Welcome to the show, Justin. How are you doing? I'm good. How are you guys doing? We're doing great. Sorry. We're running behind. Give us the latest news.

Speaker 4: What's up, guys? Just just announced today a $30,000,000 seed round with Kleiner Perkins and Kasinoa. So Good

Speaker 1: one. And

Speaker 2: coming out of stealth. And coming out of stealth. When

Speaker 1: did you actually start the company?

Speaker 7: Mid February.

Speaker 2: Okay. Pretty quick. How do you how do you actually lay a wire harness with a robot? Is that like a rote... Robot arm or, like Yeah. Great. Gantry or humanoid? What what what kind of robot are we talking about?

Speaker 4: Yeah. So a quick background. Basically, we're a bunch of defense tech guys. And so one of the biggest problems is getting technology and defense tech is like 80% IT and security requirements. And then you've got to solve the problem as well. And so basically the big problem is like, how do you use as much off the shelf stuff as possible? We just take off the shelf robots as is, get all that security, you get all that safety from it versus building all the hardware from scratch. Then you get to go deploy much faster. To the answer on the wire harness thing, we try to take... If a wire harness takes fifty hours to build Mhmm. You basically walk through and you find all the pieces that you can automate so that the artisan can build more faster is the idea. Okay. So you're building the wire harness, not necessarily installing. Because I think the... When when I think wire harness, the only, like, diagram that comes to mind is,

Speaker 2: the Cybertruck's 48 volt architecture, and you can kinda see it. It's, like, pretty neatly arranged, but it still looks like a bunch of... It looks like a nervous system or like a bunch of, arteries throughout the vehicle. And that feels like a lot of very fine tuned pieces, but you're laying out the wire harness kinda on the floor and then and then delivering it, and then they go and install it into the the the drone or the vehicle or whatever. Is that right? You got it. So basically, like, you

Speaker 4: build these things on large tables. They're called layup boards. Yeah. And there's just... Across The US and the world, there's hundreds of thousands of humans who sit there all day and manually build these things. It's incredibly manual. There's no automation, like, really works for this. Sure. And so, you know, as the first starting point, like, of the places we can kinda speed up Yeah. The defense and control base is is making these robots run, you know, when people don't wanna work on, like, nights and weekends. So then you get more capacity out of existing, you know, these existing artisans. Interesting. So how much, how how general

Speaker 2: do the AI robots need to be? Like how much of this can just be automation in the sense that you have maybe used AI to create a workflow that produces a wire harness at 99 reliability, and then you can just run it as a manufacturing line consistently. Like, you know, the original Ford assembly line is is AI in some sense. It's it's There are machines, but the the machine always puts the windshield in the right place. You don't just say, hey, make a car, and then it just figures it out. How general do you want to be in the short term, medium term, long term?

Speaker 4: Yeah. So the difference between deterministic and probabilistic robots, right? So if you're to build... I was a factory engineer ten plus years ago before I went into defense tech. And if you're going build more than 10,000 to 20,000 of something, you should just call ABB and build a factory line for it. Got it. Then you get deterministic every time. You get that well perfect every time. But the really nice thing is actually in the defense industrial base, we don't build hundreds of thousands of anything. We build like a 100 F-thirty five's last year and stuff like that. So actually VLAs and probabilistic is really good for the defense industrial base because that same robot needs to build ten and twenty different SKUs. And so that's where we're actually very bullish on this segment of the market because humans work at a workstation and they might build part x today and then part y tomorrow and VLAs and kind of like general purpose robot arms are really good at switching. There's no tool change between them. So

Speaker 2: I agree with that makes sense in terms of meeting the market where it is. But how do you respond to the the the proposal that maybe the future actually does look like just like 1,000,000 DJI style drones that are extremely commoditized, and we are moving into a regime of high volume manufacturing. And, yeah, maybe we'll keep making a couple f 30 fives every month. But realistically, wars will be fought won and lost with high volume manufacturing capability around extremely standardized pieces of the supply chain.

Speaker 4: I think it's kind of 20,050 something is like a good amount. It's still not hitting like mass production like you see in other countries. And so for me, it's kind of like, let's go hit that number first. Think we built like a 100 and then we kind of work up there. But also, think what you're seeing in Ukraine and others is like designs change pretty rapidly. And so you need re You can pivot between designs really fast. So that's also kind of changes like probably see is really nice. It's like, you had a tool change design or you retool the entire line to build that thing. Now you can kind of segment out the pockets that VLAs can just like an operator can just go like retrain the robot for maybe ten hours and then through a new training run, like, thinking of build the next thing quickly. Yeah. Got it. Last question. 30,000,000 raised.

Speaker 2: How much of that is, like, general operational expenses versus actually, acquiring robotic infrastructure, like effectively CapEx? I imagine that you're able to finance it if you do that. But what is the mix of use of funds for the next twelve to eighteen months look like?

Speaker 4: Yeah. Biggest thing is we we probably can pay for all of our CapEx through revenue. So we did with 7 figures in revenue within the first six months and basically scaled very fast. So the idea is like, let's go do 8 figures, you know, in the next six months and let's just keep crushing it. But the biggest thing is like, it's all acquiring AI talent and then scaling data. Like that's really good. And so most of that's for scaling data. And we need like hyper specific data where you're actually doing technical tasks. And so you have to train operators to do that, be very proficient. It's not just general data collection. You know, are kind of artisans in their own way. And so scaling that while also getting that, you know, all the software and AI and robotics talent you need in one place is what you need the cash for. That's awesome. Well, congrats on the fundraise. Congrats on coming out of Stealth. Thank

Speaker 1: you. You for doing. Great progress. Yeah. Gotta lead with the next time we'll be leading with the revenue. We'll hit the gongs for that. Great stuff, Justin.