Roundtable Series w/ Charlie Andersen, Joe Biancaniello, Samuel Reeves, and Chris McDemus Ep 4 | Founding Philly

Welcome to an exciting edition of Founding Philly, where we bring together local startup founders and ecosystem partners at the Fitler Club to discuss their entrepreneurial journeys. Gain deep insights and inspiration from Charlie Anderson of Burrow, Joe Bianchanello of JP Morgan, Samuel Reeves of Fort Robotics, and Chris McDemis of Cosano Connor. All right guys, well thank you so much for being part of the Founding Philly Roundtable series. I'm excited to have you guys. So the Roundtable series is a little bit different than what we've done in the past, where we're bringing together founders, investors, ecosystem builders from the area to have a candid open conversation. So I appreciate you guys joining. With JP Morgan's guidance, your startup can go from growing pains to growing without limits. With tech that powers your growth and bankers across the globe to help you disrupt every market, you can go from seed to IPO and beyond. JP Morgan, let's build your future together. Cosano Conners nationally recognized emerging business and venture capital team partners with entrepreneurs, investors, and growth companies at every stage. With more than 925 attorneys across 33 offices, Cosan has the experience and resources needed to help businesses scale, raise capital, and excel in today's dynamic markets. As one of the nation's top law firms, Cosano Conner is uniquely positioned to support your team success. To get started, we'd love to just do introductions. Charlie, you mind starting us off? Yeah, of course. Yeah, so I guess I'm Charlie Anderson. I grew up on a working farm about 45 minutes west of here, now in New York, New York, New York. I grew up loving sitting at a tractor cap, pushing a bulldozer cap. I didn't want to get out of the cap and do work by hand. So fast forward a bit. In 2017, I was working for a company called C&H, or Fiat Industrial, Steer's Largest competitor. I thought that that would be a really fun techie company. It was kind of like working for code, I guess, digital cameras came out where we didn't want to do autonomy. Therefore, in 2017, I put my job, found a couple of co-founders, including Terran Vibor in Philly. Since then, we've literally gone from three guys in a dog working out of an unheated barn to they were on a team of 60 and shipped at about 25 rubles a month here. And the company I run is called Burrow. You said three guys and a dog? Three guys in a dog named Meg. So literally unheated dog? A dog? Yeah, so it was my dog growing up, literally with what I say, literally an unheated barn. I moved into Philly because I did not have heat. I need a place where people can work with heat. That's what kind of led me to be here. And my team is now from all over the place, but we've had a great time building company, Philly. Great. Joe? Zach. Thanks for having me again. I think it's the fifth time I've been on the podcast. Joe Bianca-Nello. I'm an executive director with JPMorgan's commercial investment bank, specifically within our practice called Innovation Economy Banking. Focused on technology startups from Philly to Pittsburgh and everything in between. So I've been at the firm 12 years and I've lived in Philly my whole life and so happy to be here. Samuel Reeves, founder and CEO of Fort Robotics. We're a safety platform for physical AI. About 650 customers across every kind of work site that you can imagine. Essentially every machine is becoming a robot in some regard. And the kind of one thing that unites them all is the need for safety. And that's what we provide. Nice. So Zach, thanks for having me. First time caller. So Chris McDee-Mess, I co-chair the Emerging Business and Venture Capital Group of Cozen. I've been working with tech companies for 30 years. So I cut my teeth on the commercialization of the internet. Now that makes me feel like an old dude. But yeah, so I've been doing it for 30 years. I've been on the law firm side, big firm side. I've been on the private side on a couple of management teams help build and sell a couple of companies from the chief legal officer role. And yeah, I mostly play around on the, I spend most of my time doing venture financing work and M&A work in the tech space and then all the kind of day-to-day blocking and tackling stuff in between. So it's great. Well, thank you guys for that. So Troy, you were going into a little bit about your background and how you ultimately landed starting Bureau. Samuel, I know you have an interesting story of how you ended up with Fort. So you mind similar to what Charlie did to share a bit more how you ultimately ended up with Fort. So I came to Philly to go to college. I thought I was going to be like Joe, the financial world. And I did a couple internships in financial companies and then I got kind of taken away by this idea of technology entrepreneurship for big world problems. And I did this independent study at Wharton on companies that created products that solved big world problems that aligned their business models with solving the problem and trying to align through the kind of profit objective the solving of a problem at the same time is making money. And I fell in love with that idea. I was like 22 at the time and it was like a romantic notion of using my business degree for good. And this was like not long after the like international campaign to ban land mines got the Nobel Peace Prize. And so there was a huge effort to ban the use of land mines. But then once they were kind of slowing down in their usage there was this big question of like how do we get rid of the world's land mine problem. And the fact was like around 2005 there were like 65 countries in the world that had a giant problem of land mines in the ground. And the way the world is getting rid of it is super manual. And so in robotics now we have this term called the 3D's, "Dulderty Dangerous." And landmine clearance is all 3D's like in space. You have people crawling around on their hands and knees in these minefields which are inherently super dangerous. And they're digging the things up by hand. And so the magnitude of the problem just didn't fit the way that the world was going about it. So I come about with my you know fresh eyes and like energy around solving a problem through a for-profit company. And the obvious way of addressing that problem through a company was a robotics company. So I got to robotics via a problem I wanted to solve which I think is similar to the way that Charlie found it. But is not necessarily similar to way a lot of the robotics founders find it. I think starting with the tech and then finding a problem is unfortunately a kind of reverse process that a lot of people go through. But I found robotics because I wanted to solve a problem. And this was probably 10 years before there was an adventure capital in robotics. And so we became a DOD contractor. US Army understood the problem we tried to solve like landmines or something they dealt with since the beginning of landmines. And they had money to spend on robotics development. So we became a US DOD consulting company that was developing a landmine clearance robot. And we got this giant construction vehicle kind of like your C&H experience and added like 2012 era autonomy to it. Which was like almost actually the first was like a bulldozer. It was like a a a a a skid steer loader from the terror. Yeah, yeah, yeah, that's exotic. Yeah, exotic. No, there's like two percent share. Yeah, I mean, they were hungry to sell us a machine because they had to. It was like, so like they sold us a cheap machine. We got this autonomy that was 2012 era machine autonomy, which was like really bad. And the arc of AI 2012 level autonomy is like really bad autonomy. And so we had this thing that was 10,000 pounds. It was like running around our shop with these people around it. Nobody in the driver's seat. Now when people do robots, they like start with somebody in the driver's seat. We didn't have a safety driver. And so it just like the the danger of this thing like really hit you. It's like this thing is big enough to kill me and the people are around it. And it's just like kind of going crazy. And so we dove into this question of like how do you take a machine that's big enough to kill you running AI connected to the internet working around people with no fences between the people and the machines and how do we make that safe. And so we looked around for solutions you could buy off the shelf. There wasn't anything. So we created a safety stack ourselves. And that was a lot of the development we did was creating a safety stack. And we we became a UN contractor. We worked all over the world. Africa in the Middle East to clear thousands of miles of roads in Africa with the UN did a lot of awesome work in UN PCPing missions and like eventually got from lab to field and and with hand heads down into this very hairy problem. But while we're doing that like tons of the people popped up like VC started funding robotics companies and OEM started like their kind of development efforts for autonomous machines. And everybody started having the same issue which was like holy shit I bought a robot and I don't remember you can curse I bought a robot and it's really dangerous and we started hearing from our friends in the community like hey you guys we're having the same issue you guys solved it already can we buy things that you built and we said sure a few times and we didn't think anything of it and then they happen a bunch more times and they're like oh actually the huge opportunity for the stack that we built is in this horizontal platform. that does safety for everybody else so that they can spend their time on hauling grapes around or digging holes or moving boxes or whatever because safety is hard and tedious and you don't have to do it. If it's not your business, then you shouldn't have to do it. - I would realize though, once the robots figure out that you're controlling them, that you're gonna be built in from day one and have all the right kind of wrappers around the AI and they still die if you unplug them. (laughing) So there's other ways. There's a way reinvent electricity. - It's such an incredible story 'cause I think you hear a lot about these pivot stories with founders and their journeys but I think the fact that you didn't set out to build a safety layer within robotics and that's ultimately how you got there in such an awesome story. When was it or what was that like a ha moment when you were like, you know what, we may not be building for minds anymore but there is this massive opportunity now and that holy shit, we may be on to something. - There was a not a ha moment that was kind of slow before that but it wasn't millions of dollars in but it was a few customers in. It was kind of an annoyance at first. Yeah, extra money, how could we turn that down? - But then DARPA comes in and DARPA was doing the robotics challenge where they had humanoids, we were just talking about humanoids, that were, I forget what the task was, like climbing a ladder and turning the valve and opening a door, those were like three tasks they had to do. And it's just right after the Fukushima power plant melt down and they realized that they had a humanoids robot humana robots that could climb a ladder, open a door and turn off a valve, they would have like saved the meltdown. So, this is circa 2016. - 16, 16, 16. - And so DARPA comes in there like, we want to buy your product for all the teams that are doing the DARPA robotics challenge. We were like, oh, okay. So we were the only and standard, standard and only safety technology required for the DARPA robotics challenge. And that's kind of the gift that keeps it on giving, giving because like all of those teams are now starting companies, they're all over the place, you know, they're working for companies and they've taken our products with them. And that's part of how we have 650 customers. But that was kind of like the first non-1Z2Z sale where they're like, oh, that's amazing. - Maybe a thing. - What's interesting is that we're talking a lot about robotics companies and like what you guys are doing. I think when people think about robotics, I mean, in today's world on social media, think about like the human noise that we were just talking about. But I think there's their actual hardware side of robotics and then the software side of robotics. Could you go more into where your company kind of falls into that? - I can jump on them with first. I think traditional robotics, maybe a decade ago and prior is hardware and software. Robotics today really is hardware, software, and AI or training date or physical AI effectively. And there are some huge implications of that. The first implication I think it's most notable here is it means a robot can go not just in warehouses and factories, but also in the great outdoors. The second implication is I think touches on both of our companies is if you, I think once you go outdoors, you're in a way less constrained space. Like there's a lot more things that can happen. And things people tend to learn, I think this is definitely true of robots. They tend to learn by making mistakes. You don't tend to learn by being told exactly what to do and just doing it. You tend to learn by going into the real world and making mistakes and improving. And I think in my company's case, we build a robot that drives around safely in your people, indoors and outdoors in completely unconstrained areas. And I think we're, as it relates to robotics specifically, what does really weird point of time but aren't any winners yet? They really aren't. We talk about human noise. And these are sub-hundred systems companies in that space. There's no big winners yet. And so the whole space has this feeling of Mount Everest in 1952 or PCs in 1978 where there may be a couple companies that really figured out. And you have 650 customers. I have about, probably 110 customers today, about 600 plus systems running, about 2000 to 4000 users a day on big days. And so we're starting to figure out, where in this kind of early inning of robots being everywhere. And I think that 10 years from now, there are gonna be robots, you know, I love it. Everywhere you go, with the probably gonna be pretty boring. They're gonna be like, they're gonna, again, dull dirty dangers being really boring stopped. You're gonna blend it in the back. And I'm gonna be like, "I'm gonna notice them, "it will be everywhere in my opinion." I mean, in which early days though, for sure. - And do you feel like, 'cause you are, like you have full stack from software to hardware, right? - Yep. So I, like, I gotta think robots are hardware, software, and then training data or AI. So it's like, you're answering, I guess a system driving through the wheel has gotta answer, where am I, what's around me, and then what do I do, or how do I behave on that basis? And there's AI in all of those layers. So, you know, where am I, I love your, if you're indoors, you can't use GPS, you gotta use lighting in very feature sets. If you're asking the question, what's around me, how do I behave? Do you drive, like, in a warehouse, if it's flat, you drive, if it's tall, you stop. Once you're outdoors, I don't know, it's a ditch free space. What do you do with reflective water? Do you drive through spider or sweat? So like, there are all these weird scenarios you can kind of encounter, and it used to be that you couldn't handle those, 'cause you didn't have AI. With AI, you get out there, you see it, you make mistakes, and you learn, and the goal is to design a system that can run a lot to get the most data, to therefore be the best. - And then, in terms of where your company sits within the hardware, software, AI stack, where do you guys sit? - Yeah, so we're the safety layer. We started with safety-rated control over wireless. Control systems have been out there forever. Like, every airplane fly on, cardio drive, excavator that you maybe operate, industrial robots in factories, you name it. Anything that moves that's industrial grade has some kind of control system on it. But when you start moving these things around on the work site, you don't have the ability to wire them into a control network anymore. So all of a sudden you have this need for this, that same value of safety-critical control, but over a wireless network. And nobody ever thought of safety-critical wireless control. There's wireless, which is dominated by the Ericsons and Cisco's of the world, and then there's control systems, which are dominated by the, you know, continental's and Bosch's and Rockwell Collins and Rockwell Automation's Siemens. And like those two haven't really cross-pollinated. So our first bite at this space has been to create these virtual control systems that give you the value of safety-critical control, but over any kind of wireless network. So we can do safety-critical wireless on Bluetooth, Wi-Fi, LTE, our own radios, the public internet, you name it. And still offer that same level of reliability. And that was kind of like the first step in creating this larger safety platform, because like if you're gonna do governance of machines and you're gonna tell them to stop if they're about to do something bad, then you have to be able to remotely control them. You can't be sitting on them. So wireless control was our first step. Now, next step is to take that baseline of control and add some kind of perception on it. But we have to do safety perception. We have to make sure that our perception is able to be certified to a very high level of safety, because otherwise people like Charlie do perception every day. And there would be no need for us. But if we can offer a level of safety-rated perception, that means that people like Charlie don't have to spend as much time worrying about whether their perception is safe. Then that again reduces the amount of time that our developers have to spend thinking about safety things. So overall goal is to have somebody approach a robotics or physically eye project and say, I'm just gonna not worry about safety so much because I have fought on the machine and I'm gonna focus on my application. And to do that, you have to be able to control it anywhere you are in the world and you have to be able to sense through perception and reasoning that bad things are about to happen and then use the control to prevent those things from happening. So we kinda, we said, depending on your stack diagram, we said like kind of sandwiched in the middle of the robot stack. So you have the drive system in the wheels and then some of the compute and then you have for its layer and then you have the sensing and the AI and the other applications above it. So we're kinda in the middle. - That's one's to be questioned. What if you hit a red button on a Ford control with the human eye? What does it do? Does it just follow? - We actually have a few different types of e-stops that we offer depending on how they configure it. So one is kinda like something bad has happened and if you don't do something about it, then we're gonna cut power and then the other one is just cut power. So you don't want them just cutting power and lay like all of it, lay like all of it. - Right, right, right. - All in good. - So we're actually a part of the humanoid safety standard and the other mobile robot standard but we're voting members on both and that's been an important thing on humanoid safety is like you can't just cut power and have them fall over and then lay about. They're doing enough flailing as it is. - Yeah. - Like this motion really so. - Yeah, that could be dangerous. - Right. - So we were talking a little bit about physical AI before you mentioned it a couple of times now. Could you define physical AI? - Autonomy applied to something that moves. It's really motion that. that differentiates AI in the digital world versus AI in the fish school world. And I think people obviously are familiar with like Chatchy BT and all of those more consumer facing AI products. How would that differ from the AI or is there like AI for robotics versus like open AI or a throbbacry, anything like that or how would you differentiate those for someone who listening to this? That's a great question. There are efforts now to create AI applications that people use to make robots more intelligent. But there really is no go to AI for physical AI. There's kind of a malgamation of different kind of models that people use to create robots and their own physical AI. But there's no kind of open AI for robots. And there are a few very, very well-funded efforts to try to do that. I think you would probably have a perspective on this as well. I think most robots, most physical AI in the future are still going to be an amalgamation of a bunch of models. There could be a large open AI style, a large model, not open AI style, but large and then model as a component of that. But the physical world is so complex. That's too hard. You need a kind of symphony of smaller models doing other things that are kind of stitched together into one objective. There's this thing called Moramex paradox. It's a common thing that people have. It's the idea that it's easier to replicate your higher level thinking than the reptilian portion of your brain. So it's easier to train something to write poetry than set something up to drive to the physical world. Applications of that, a lot of the large language models are trained on all of humanity's knowledge that exists on the internet. So I don't need a poetry reading video. If you ever ask a Gemini or something to create a video, something, a frequently created video with a car driving sideways or something that's really stupid in physical space in the real world, that's because it doesn't have a good data set for that element of things. I think we're in this weird point in time with robotics where right now this is fleet building days. You need to get the biggest fleet out there as fast as you possibly can because the more robots you have, the more data you have to train stuff to get it really good at moving through the physical space. And so I think that the winners today are the ones that have the biggest fleets because the more robots you have, the more training data you have coming back, the better your model is becoming the more capable those robots are. And I feel it's for my team. Everywhere we have processes, five terabytes of data per hour of runtime per system. That's five terabytes per hour per robot. And each one has something called a novelty engine. So it's basically throwing out stuff that's already seen and just sending novelty back. So things that hasn't seen before. And I think we're at this point in time where I think we're like digital cameras in 1990 right now. It's like 320 by 240 pixels. You can kind of make out, hey, it's a Coke can versus a Pepsi, but not really. And you start imagining 20 years or no, 30 years or no, it's going to be nuts. You're going to have things in the real world that can see every single thing better than you as a human, that are cloud connected that can go and do virtually anything. And we're just in the very early days of that. >> You know, I think you're right about the dearth of data, of course. I mean, chat, DVD had the beneficiary of all of human language since Gutenberg. And it was a large language model. So there was a huge data set to trade on. Trade on. And then the dearth of robotics, for sure, this is a well acknowledged problem. I think there's one way to get it, which is getting a bunch of robots and generating it yourself. There are other ways to get it later. >> Yeah, man, when there is the whole simulation or digital twin or there's a whole realm of things, I still believe it's a fleet building. I should have looked at it like Tesla. Tesla's like, go build the fleet first. Build the fleet, get it, and then get it. I think that there are two schools of thought on this. To me, it's the pragmatist versus the perfectionist approach. Pragmatist approach is build a big fleet that is gross margin attractive. You can actually make money off of and get data back. And the perfectionist approach is build a really cool simulation tool and perfect there, then go out. And I'm more of the pragmatic Google than the perfectionist one. >> So nobody's been collected. There's nothing out there where they've been collecting that data and not even cars, like a rudimentary data set of some companies collecting valuable data from cars. But correct me, if you got a room full of data from mobile, it might not be. It's super helpful to you on a farm. >> Some of them might be, but not all of it. >> They're pretty good data sets. It's also like, think about on-road at the time. You've got lane markers. You've got stop signs. It's actually a pretty structured environment. Once you're off-road, I don't like anything. Anything, I've literally anything. And so I think a lot of these, a lot of roads we're describing are going into, yes, sometimes semi-structured, sometimes very unstructured things where it's like going to a mine and going into, I mean, I'm in vineyards and nurseries and deep bow yards. A lot of these funky areas where you also don't, you have no reason. No one's making a video game of them either. Nobody's making a video game of how to drive through a deep bow yard. What would you do in that? Because there's a good video game of them, because there's not lots of systems running it, because there's not lots of, the data sets just don't exist. >> So just out of curiosity is that why, I mean, I trolled through both your sites before this, and all the borrow examples that I can find tended to be on a property that had straight lines in some way, like a vineyard, a warehouse. Is that some way that you're trying to leapfrog a little bit on collecting that data? Like it's not a structured environment, like a road, but it's structured naturally in some way. Is that help you at all, or is that just a, is that more a function of like the type of customer that sold value in your product? >> Probably more than that. Like nothing specific or deliberate there. >> Yeah, it's just the way vineyards are. >> Yeah, just the way a lot of. I mean, like rows and roads, and there's, like the more you can have a little bit of structure, the more you can follow it, that is a benefit. But I think things weirdly get, I guess what we had going for is we're going slowly, we're off road, we're near people, because you're near people, you can have someone that can like intervene if something doesn't do the right thing sometime. So there are a bunch of benefits there. But lots of structure and clean environments has not been a benefit of the scene. And it also, like, I think we were, one of the biggest problems we're into is like, do you stop for it or not? Like I like tall grass. Like that's a pretty seemingly simple problem, but like, was it, is it actually like a crop or is it something you should drive through? And those are surprisingly hard problems to answer, especially when you've got, you know, 1800 pump vehicle and you've got a toilet of mow or need a much stuff like that. >> Majority of our applications are off road for kind of the reason you're saying is, because it's easier to structure things. If you can build a fence and say, no humans can come in here, then you can kind of limit your liability, you can limit your complexity if you're on public roads with, you know, Joe Pedestrian, then like, it's a whole other thing. >> That's Tesla's problem now, right? >> Right. And which is why you've had, you know, so many companies kind of go by the wayside trying to chase autonomous vehicles. Whereas even some of those founders that got to the end of the road, dad, yeah. >> Into the road. In autonomous vehicles have kind of pivoted their careers to being off road because it's something you can limit. You can limit the operational design domain. But, you know, in the AV example, just kind of going back to something that we were talking about before, I think sometimes you need a giant moonshot vision in order to catalyze the amount of investment. Like, to your point about the whatever, 2015 to 2020 RNAVs and how much investment was poured in, I don't think that would have happened without this kind of idea that, you know, you could reduce the millions of deaths and you could give us back hours per day and you could reduce the kind of environmental problem and you could reduce number parking lots and, you know, the kind of number of hours that cars are just empty and all of the things that the AV people talk about. I don't think you could have raised that amount of money without an expansive vision. And, yeah, so it took longer and then a bunch of the technologies had to be pivoted over into other markets where they found value, but, you know, kind of differently than they originally thought. But I think you had to have that big vision to motivate the investment in it. So, humanoid are probably going to take longer than any of those companies are telling their investors that it's going to take before they have, you know, a good, good, good, Jillian dog lose a revenue, but will it get there eventually and will we, will society be transformed when it does? Yes. You know, is it going to be exactly in the return on investment models that anyone's expecting? No. So, hopefully, hopefully everyone realizes that, but it's still net positive over at the long term. If you're scaling a business, building a team, or just looking to surround yourself with people who get it, you should know about Fiddler Club. It's where Philly's founders, investors, creatives, and leaders come together in a space as ambitious as they are. With workspace, wellness, dining, and curated events all under one roof, Fiddler Club is your invitation to connect, grow, and belong. Step into your ultimate third space, where connection spark and ideas thrive. Fiddler Club dot com to learn more. So, you mentioned the capital needed. I think it's an interesting topic. We talked about physical AI. How has kind of fundraising for a capital intensive kind of market been, and then how has the introduction of physical AI maybe have changed that since it was previously? So fundraising in the physical world has always been a challenge. People have always preferred software to hardware. I think maybe in the past two years, maybe in the past year, hardware has become more acceptable because you had Tesla in space. Right. Thanks to Elon. And thanks to Andrew, you have hardware companies that are making return on investments. So now it's plausible for early stage venture capital to get into hardware. They used to say like 20 years ago, hardware is hard. Hardware is hard and it would prevent investment from happening. So it's always been hard. It's easier now. But we've only raised $60 million, which is like compared to an AV company or a humanoid company is like a drop in the bucket. So we haven't needed as much investment. But every single dollar has been hard. Every single dollar. And I think I would say like, I think that a lot of the space is kind of like SaaS companies when SaaS was first starting out. Like you got to sell the tower first before you can sell something to run on it. That's a lot of these companies, you've got a physical thing, you got to sell, and then the autonomy that runs on it. And if you don't, you have to build the fleet to sell the software that runs on the fleet. It's really impossible to just sell software if you don't have something to physically run it on. Oh my god. And then, and a related point to that, there have been a lot of people that have tried like retrofit style things. Those tend to be signing up, at least in my domain, for a ton of like variability and customization a huge scope. And so it has seemed a lot easier to build the whole thing from top to bottom and sell the hardware and child recurring forward as opposed to just build the thinking layer but then retrofit it onto trucks and buses and tractors and all these other things because all of those things are pretty variable. That's interesting because I feel like I thought about as it can be a hardware company, robotics or software or both, but you're saying if you're just a standalone software, there's even more challenges. That's certainly my opinion. I used to work for the second largest tractor and construction equipment company in the world. And typical permutations of build on like a bulldozer tractor is like two billion. So you can build it two billion different ways. And so how do you, I don't know, like the tire size is different and so they got to calibrate your odometry on your autonomy layer to go with the right speed and it's a different height. It's too big of a problem. Whereas if you control the whole thing top to bottom, you can, you can integrate something that actually works and maybe in time it gets so generalized but going on to anything. But to me right now the question is how do you, how do you, with less than a hundred million dollars get to 30 or 40 million dollars in revenue as quickly as possible? That's kind of, that's ultimately what we're largely trying to solve for and I felt controlling everything top to bottom likely is the faster way. You probably are going through that logic because you've been asked many times why it's just built software that just goes on on everything's already out there. But it's like as if it's just been in, I mean as a robotics company aren't you really trying to solve like a utility problem in the end? You're not trying to solve like a software problem, right? It's like a utility problem. You're trying to move through it. Yeah, exactly. Well, originally now you're moving out of it. I think for me like there's, I think there's, you think of like like sizes of machines. There's a lot of machines that already exist or very mechanically official. There's a person that sits in the cabin does a lot of work really quickly. There's like this, this form factor of smaller things that work in your people and it's a size that like, it is no reason to exist without a ton. Like why have a, why have something the size of Disney's Wally or R2D too? That's not a ton. There's no reason to. But once it becomes autonomous then there's like this new form factor that should exist in my opinion. It's like it drives in your people. It does a bunch of you know, perception and caring and manipulation type things and what there's a really cool point in time where you can build that. And to me then and also like you think about companies in the PC space early days it was a lot of kind of like hobbyist stuff that was put together didn't work very well. Then you have like Apple emerges and Windows emerges. You have some like standardization of layers. I think we're kind of at that like, you know, we're at the point where you've got like a seven inch laptop all the way up to like a 25 inch laptop and all the form factors all over the place. They're going to start to coalesce around a couple things in my opinion. So I think it's a lot easier to to win in that stage if you build everything, not just build one layer. Yeah. Just doing just a hardware or something else. Yeah. I mean, I think there's there was the assumption that you could build software and easily deploy it because that's the way it works in the IT world. But the IT world is not moving. Is the big difference. Well Apple build an operating system right and a computer. Yeah. Right. Then at some point Microsoft said, well forget the computer. I'm just going to build the operating system. Yeah. Dell said forget the software. I'm going to use their operating system, but I'm going to build the computer. Yeah. But I guess in I think in the robotics case, I can kind of see the logic in like you really kind of have to be like an apple in the sense that you got to build the operating system. I think this this this project. At this stage. Yeah. I think it's time that we'll change that. This particular point of time that you want to build you want to you want to have control over both sides, not just the software. I mean, at some point Microsoft realized they weren't they weren't going to win a hardware game. They just you know, well, none of those could kill you. There's the thing that's true to like these all could kill you, which is why you know we were safety. Keep saying that. Safety. I just want to remind you safety. Safety. Well, we believe that safety is the thing that that unifies all of this and provides you know, it doesn't provide it now because I think operating systems are going to look different in physical AI than they do in digital AI or anything that's digital. But I think safety can become the operating system for physical AI because it's the only thing that is universally important across every single one of these applications. So that's why that why we're in safety. You can't just yet kind of like deploy software and and have it work on the device. But what if you could what if every one of these devices could be like an iPhone and have a bunch of different applications from a bunch of different parts? It doesn't not like the internet of things kind of concept like that somehow you're going to be able to like take all these sort of immobile things and like hook them, connect them up in some moment. Yeah, whatever. These are things. I mean, what about sort of things? You know, they are things. Yeah. No. And Zach, there's a there's a really simple example for the person listening at home that just kind of like was so simple. It kind of blew my mind when I first got into this robotics responsibility within thanking versus software, right? So you remember being in school, if you took an IT class information system class or whatever or even if you're just playing around your home computer, you copy paste the instance of Microsoft Word and another folder. It's another instance, right? Software. The gross margin on a really good software company is north of 75 and north of 80%. So just think about that. You're copy pasting. You're getting another instance of a software. You're getting another instance of software. The cost of goods sold, again, gross margin, your revenue, my shirt, your cost of goods sold your gross margin. That's why it's so high. You repeat that multiple times to multiple customers and then you embed that software into their operating model. That's why it's so profitable, so lucrative. But the child who's point earlier, software had 30 years of sort of evolving into what eventually became almost everything in BC in the early 2000s, late 90s, was software, right? Robotics, you have to build, you have the R&D to work prototype and then you have to have your early product and that takes machinery, hardware, right? And so your cost of goods sold to your first customers is high. And so your gross margin, the idea is to get it up over time and then you have the software layer where you could charge your current revenue and build a whole stack that allows you to sustain your business model. But for the people thinking at home, we don't have a lot of experience in this. I remember, again, just being elainment at the time and being like, wow, I can just copy and paste an instance of software. That's what a software company is doing to their customers essentially. I mean, there's R&D and development costs and it's not, I'm not saying it's easy to build a great piece of software that works. I'm just saying, once you have something for your application, very capital-efficient. So you've been working with robotics companies for a while now. How have you seen the market change over the last five, six, seven years? Well, I haven't been doing it for seven years. I've been doing it for almost five years. And I've also worked with software companies and internet companies that we call them. But I have a specialty now in what we call applied technology. So that's everything that, it's really everything that encompasses the intersection of hardware and software. It could be your networking or your semiconductor, your chip company. It can be robotics. It could also be sort of an internet of things. There's a whole variety, but where we're hardware-meat software. I've seen a huge shift to the stainless-point-n-er in the narrative around excitement in robotics. Post the chat GPT moment. Because we talk about R2D2, for example, science fiction or Wally, which is happening now. Right, it's right. We all grew up with this idea of a future where there are robots among us. And so science fiction, prognosticators and writers and novelists and directors for movies, they've been playing with the idea for a long time. The first I believe assembly line robot was invented by GE in the 50s. So think about that. But fast forward to the last three years since chat GPT created, walked to me since opening I created chat GPT, right? Not getting too technical. The key breakthrough there was the creation of transformer architecture. Go back to that time. Now you have in late 2022 this idea of artificial intelligence as an overlaying so much of our daily life that's now in what we call the leg cice or society, right? People are thinking about it constantly. Physically AI is a category term to define AI in the physical world, robotics. So for me, five years ago when I was first getting the robotics, it was still this abstract thing that's like, oh, it's so hard. I don't even want to think about it. It's so difficult to do. It's so expensive. Is it practical? Is it going to be able to scale? Now we have a world where AI is in the hearts and minds of everyone. And then the good thing about humanoid think too is that people are thinking about it. It's like being put in our on our radar that robots are coming amongst us. And then you do a little bit deeper dive and you learn about the things that aren't that sexy, but incredibly practical and impactful, like an autonomous, but great telling robot. I mean, AI being on the mind of the public is awesome. We had hundreds of customers before chat GPT. And so like a lot of people in this space have been working for a really long time to make their creations work. And like the new machine learning techniques that came along with like the emergence of chat GPT have certainly made it easier to build robotics. People have been kind of toiling away, like quietly for a long time, which is why things look like they happen overnight, but they never really happen overnight. Yeah. Like lots of people toiling away. But we are kind of to your point earlier about the operating systems and the PC revolution. I think we are kind of where the internet was in like the mid 90s, where the creations, the robots or the websites, and you know, 30 years ago, like they worked now, which is a new thing. Like they didn't always work before and like Charlie's worked really well. And like many hundreds more than many others. But like broadly across like our portfolio, finally the robots work. They do the job, whether it's like picking apples, digging holes, moving boxes, you know. But they're not broadly in general like scalable. So they're kind of at this adolescent stage. I call them adolescents. Like the robot adolescents is where we are right now. And to grow up to become scalable, they need to become like broadly safe, secure, reliable, and economical. And the industry is not yet those four things. And so I think the journey that's going to happen over the next 10 years is this kind of platforming like happened. You know, in software and the internet really happened over the past 30 years. But I think the next 10 years of physical AI are going to be really pivotal. For how physical AI platforms and it's through that platforming, that kind of standardization of different layers of the stack. That are adopted across thousands or hundreds of thousands of robots that's going to drive that kind of growth and safety security reliability economics. So I mean it's it's it's actually kind of a watershed moment that they work now. Because everything else is it's kind of it's just work from here. It's not it's not a big use. It's not like oh my god it might never work. It's like it's going to work. It's going to take it's going to take time. Do you guys feel like you're leveraging skill sets that have had been had disappeared for a while? Like I felt like for like maybe two decades we were so focused on like building in a digital sense. Whereas like I feel like you guys are probably re leveraging things that like Henry Ford had a deal with, right? I mean like like what is physical distribution? Like how does it work? Like the car works. But now how do I build hundreds or thousands of cars and how do I get them out into the market place and how do I convince somebody that cars better than a faster horse? You know like physical service. Like we're just we're just opening up a premium service, managed services, professional services, premium support services because we have so many units in the field. And as they encounter things like things happen. How do you economically get things up and running? There's a big question. The things happen. I think it's like in software you get like a unit test. It's like you build the test and then you test your software against the test. And it can be it's pretty, it's almost like a light switch. Like does it turn on or off? You talk about robots in the physical world. As you start and so my team has a little over 600, I'm about 20 half a month right now. The more robots you have, the more variability of the encounter. And so even the simple question of like how do you, so scaling in this world is about as hard as it possibly gets. Because the more you have the harder it gets for them to do the same thing every single time when encountering a much much greater permutation of variability with things. I think that's what's so hard about these companies. A lot of these companies can't get over like the first about five or ten systems is one thing. Then you go to like 50 or 100, then you go to like 500, 600 plus that's where things get we the really really really hard. And AI is the single biggest tool that enables you to do that. And then the second thing around physical stuff like then you've got hardware margins. And in fact that's really hard. Like just build it just even like how do you, how do you, I got a lot of robots that I'm going to see they'll have like numbers on the robots that are pretty big and prominent. I know if you've got a big sticker on the robot with like 86 or something on it that that thing's got personality versus unit 87 and you're not scaling like there's just no way. Because you, the things have got to be you know universally the same always behaving the same way in the variable world inside final thing that I'll shut up. The making mistakes. I think a lot of these companies can't make mistakes. And so I look a lot like a lot of the a lot of things I think within like self-driving car companies or within within my room or humanoid just if you can make a mistake how do you get the data to know how you make a mistake to improve from that. And so I think it's really I think that the and then in a world where a lot of people are describing kind of like perfectionist. I think if you if you can't admit you make a mistake. How do you how do you describe how good your stack is. So we do we talk about autonomous miles per falter for per user and prevention. We can drive around 27 to 30 miles per mistake that we make and we've done that over 850,000 hours at this point. I think we're kind of in like this slightly like snake oilish space where people actually talk about how reliable they are. And so you're kind of like well you say you got a bunch and like how reliable isn't really hard to know. So it has all the hallmarks of like early technology days or sometimes people are maybe way over selling what they have relative to what it actually does. And there's not even like standards for how good or how safe or how reliable is your thing in a funny way. Yeah, but all that will come later. I mean like we we we always lag on we society. And lags on regulatory and how to judge success and metrics of reliability and things like that. So. Do you mean that's on speech speech and robots? Yeah, like I think voice is coming to robots. Absolutely. It's an exciting thing like I think that everyone loves humanoid because humanoid is like it has the human figure. But you're going to soon have something you can like talk to and tell it to do something and it's going to be able to talk back. Absolutely. I think that's like super exciting. That's actually of all the elements of that's probably this single part of the space that most excited about because it's really done that in a big way. And can you use I know you were saying you guys can't use like an open AI model on like a robot but like there are a lot of voice model. You can run some of the large language models of robots. Oh, you can. Yeah, you totally can. So you have to paint a server to do it. You can definitely run that so you can you can put together. It's basically putting it like the higher port the higher level cognition of your brain with something actually works in the real world. So you can say, hey, like keep this far mode. Hey, like don't run out of batteries. Make sure you charge. Go to bed. You can you can give those style commands and have a robot figure out how to do it. Which starts to get pretty exciting. And then you might be the question like what should robots voice be? Is it like I love the terminator? Is it like David Attenborough? Is it like what what what should it's garbage? Or is it more like art to detail? And can you use those voice soundation models on robots? Yeah, totally. Yeah. Oh, yeah. So you can there's like some technical nuance, but you can run portions of them and then ping a server and get it back. That's going to be fun. I was just able to click the technical nuances very complicated and I won't claim to know really understanding it very well. But the large language model gave way to the vision vision language model. Right. So you know, show me a picture of a cat's moon. Yeah, that came out. If you remember that came out. I don't know how many months after the first GPT chat GPT instance and what robotics companies like physical intelligence or school day. The foundational model robotics companies, what they're trying to build are vision language models that work across robotic systems across different form factors effectively. So it's kind of just for the listener, the simple LLM to VLM to VLA large language model to vision language model to vision language model to action. It's a crazy world we're going to have voices going to be important. I mean, because if human needs are important or if human is make sense because the world around us is constructed for humans. And so the robot that goes in that world should it's easier if they if they are. like I human why wouldn't we think that we're going to talk to them. So it absolutely makes sense. From our perspective, you know, we started with this control layer and some of the endpoints on our control layer are human interface devices. So we picked a video game controller as a form factor for one of them because, you know, kids in the military, when there was the first robot, was kids in the military played video games and the training time was like zero for a video game controller that is built on our safety networks. And so we are working on voice application for robots. For us, it's more of a question of how do you take third party AI models and wrap them in enough safety that they can be trusted because like no one company is going to make the, you know, the language model that works for hospitals in Portugal as well as, you know, great fields in California. And so we need to have an open approach. Our approach is an open approach. And so being able to interface with third parties and do it safely is kind of the meta question that applies to voice as well as other kind of third party AI that needs to get into the safety. So when we think about the Fili star scene, how do you feel about building in Fili as a founder? What has your experience has been, Chris Joe also from you guys just being in Fili in this market? What are your perspectives on Fili as an ecosystem or innovation economy start up scene? And what's your thoughts? Creator, Philadelphia. What's your thoughts? I'll go first and you can, yeah, if that would so, so I guess I started working on relax here in 2017. At the time I was a geolaging polycyme major in college and the MBA, no clue how to build anything in the software space. So I like a meal in the way I could weld, I do all that good stuff. I found that Fili from a talent perspective is actually fantastic for software development. So a lot of my team comes up to the other grass. That pen and Lehigh and Drexel, there's a lot of very, very good people here. It hasn't been a very easy place to raise money specifically. So I've had to go afar to bring capital. I guess I raised $46 million to date of that $46 million box about 75K. It's come from Fili of which 25K came from my mom. So like we're talking about money. Thanks mom. Yes, thanks mom. But like it's been a lot easier to go to London in New York and Chicago and Sanvers. That's going to raise capital. But I think what the Fili space has been great for is certainly for talent. And then the other element I would say when I think about the Fili, I mean there's in Fili, there's us, there's Fort, there's Ghost, there's Exyn, there's some, there's a sign of their bunch of like, actually very, very, very high quality companies. And I think that the, what I think what Fili forces companies used to be pretty gritty. And like build something that actually works. Build something that people actually use. Work with local suppliers of which there are many if you're doing a physical thing to do it at lower cost. And in a more nimble software, heavy as opposed to expensive hardware heavy way. And so a lot of the weaknesses of the region are in my view actually sources of strength. And then I think what Fili has really struggled with is actually just marketing, explaining the actual merits to be here. And so to me it's been a great place to build a company, not obviously a great place to build a company. And you know, I don't actually don't, I think I maybe have one, I don't have a single customer in Pennsylvania at all. I've shipped 600 systems globally. I've raised like an almost no capital here. But I have tons and tons of talent here. You know, tons of talent here. And it's been like a fantastic place to build something. Yeah, I, I grew up in Texas. So I didn't, when I came to college I did not expect to be here this long. I didn't expect to be anywhere else. But I didn't expect to be here either this long. So I think it's because I started one company straight out of college. And then the other one first company led to the other company. And so you know 20 years later. Still here. And so I think echoing capital like it is super challenging to raise capital here. I mean, we also have next to zero Philadelphia investors. So you do have to get on a plane or train to raise money. Which I do think relates to the gritty piece, which is like the atmosphere of constraint kind of breeds ingenuity and breeds a kind of like we cannot fail mentality in turn. Internally and pragmatism that comes in that. So when you look at like we don't have a lot of robotics failures really. We have a few companies that started and people will like we're all still going. Which I think is pretty incredible given the larger robotics ecosystem. So I think there's a grittiness. I would echo the talent. You know, there's a ton of talent here. Really high quality talent. So there's the capital is definitely a problem. But that can be solved by getting on a plane. That atmosphere of constraint is definitely a good thing. Talent is definitely a good thing with some exceptions. I mean there are roles here that just don't really exist. Like product management. It doesn't. It's not like a thing that this region does very much. But generalized in engineering hardware. You know building things like we have this multi hundred year legacy of having done things like that here. Which is pretty special. I think culturally like we don't have a culture of startups like like when you look at the largest employers in the in the town. They're not companies that started as young tech companies and then grew up in the big tech companies. So there's not like that journey that many people share. So there's like a little bit of reeducation that has to happen when somebody goes from a more traditional business into a startup. And so we've gotten kind of used to onboarding for that. So that would definitely make it easier if there were capital and then there were kind of a culture of startups. That's super true. Like the tradition of like big company startup and also from school to startup. Both of those are like unique journey that you got trained people. Which is doing more. He's going to do it. The tax structure just doesn't work. Like period full stop. Like we can argue about it. But like revenue tax and gross top line revenue tax makes no sense. The net profits tax and the wage tax unbelievable. And like I would love for this region to win in physical AI. But it just won't happen with the current tax structure. Just one. So that's something we could work on. But otherwise, I mean, it's a delightful place to live. Like I don't know. I don't know about you. But like sometimes I randomly see him on the street. And I ran and see you a lot. And you I just met you. I just so much. We'll see each other in the future. But it feels like kind of it. It feels like a large city that's also kind of a neighborhood in the same way. Which is really special. It's actually I actually feel even though I see a bunch of people that I don't actually know for some reason that feel kind of it. Oh, walking around. Maybe it's I mean, living the same neighborhood for 20 years. But I think that that that kind of mix between big and small is is special. I think the livability is special. Like of all the East Coast cities were probably the most livable from a kind of just physical size and cost of living perspective. Great restaurants. Great culture. Great museums and concerts and whatnot. So there's a lot to there's a lot to love about living here. I think if we I think we've heard we reformed the tax structure. Got some kind of financing infrastructure that involved you know angel investing in real early stage investing like on a really big scale. We could just kill it in physical AI. What do you think? What do you think? I've taken more more. I have my experience you guys will have more umbrella experiences. Well, I would I would just say that I completely agree with the ethos of grittiness that Charlie was referencing. I've lived here my entire life. Went to college here. You know that that is a very positive aspect of the way you define a Philadelphia in or Philadelphia. For decades and decades and decades and a lot of that has to do with previous of the workshop of the world you know the country was founded here etc. I also there is a charm to living here that same it was also looting to in terms of like the it's provincial it's communal like it people know each other it feel it's a very big city overall. You actually just pay attention to the density and the size of the city but you know it's all that is like you everyone ask you where you went to high school if you're from here right it's like and there is there it's it's it's like the kind for me it makes me think of your greatest strength is also a curse because we have that provinciality that is charming it also comes within a humility which is a virtue but we don't call us around a new way to market ourselves in a really positive way or in a really I just say scale or impactful way like we still talk about we were talking about earlier like we love our sports teams but like that permeates everything conversation about Philadelphia but the if you actually think about the talent we have here the university treasures that we have here and and the actual capital coupled with the humility and the grit that we've always had I think all of us could work to come up with a better way to position this area. The vision of it it's like what do we want to be everyone wants to be a top start-up scene or ecosystem I don't think everyone's really focused on well what does actually mean. and how do we get there and what is that actually, what's that vision? Or I mean, there's so many pockets and it kind of, it's too kind of just out there and knowing actually executes on it. - I mean, I agree with everything all these guys have said. I mean, I think there's a lot of reasons that Philly's a good place to start a company. And my experience is I think true to both of yours, which is like, you know, the really good clients that we've represented in Philly that have grown, you know, they tend to raise the majority of their money outside of Philadelphia, you know. We probably have more experience raising money on the West Coast than we do sometimes on the East Coast for some clients. - But I mean, that's, I think the bigger money problem, I think when you get to a level that you guys are at, you can capably raise money outside of Philly 'cause you've got something interesting. I Philly probably suffers a little bit from like an early stage money problem. Like there's not enough early stage money on the West Coast. People write checks for, you know, something that just is super esoteric at that stage, where you will not get that on the East Coast or frankly, like in Philly. And I think to get to that point, what you really need is a couple of companies that have large exits, which you've had in Philly. But I think they have to be companies that start it as startups and remained as startups through the exits so that you've got like 20 or 30, you know, sort of like original employees that walk away with like seven figures and are young and want to start more companies and they want to angel invest and they know the ecosystem. They know how to build something. Like I think you need a couple of those to happen. And then I think you need people, you know, to these guys' credits, they've stayed here. I know clients that happen, right? And I think sometimes you run into that problem where, like, you know, if somebody grows up and they want to play pro football, it's kind of like, well, if I want to play pro football, here's the college. I have to go, I can only go to these colleges, you know? And a lot of people have that mentality where it's like, well, if I want to build a big company and exit for a lot of money and raise money, I got to be in Silicon Valley or I got to be in Santa Monica or Boston or whatever. And they feel like that's a reason to leave Philly. You know, we need more people like these guys who say, no, I don't have to go to whatever college to play pro football, like, you know, I can get there on my own, where I am, so. - Totally agree on the ecosystem of exits, kind of spotting new starts. I mean, when you, you people start companies in San Francisco, they, you know, they have 10 other friends lined up to put in 10K each and then introduce them to, you know, 50 friends that put in 50K each and introduce them to somebody that, you know, writes $2 million, pre-seed check and that happens in a week. - Right. - Right. - Right. - Yeah. - And, but to catalyze that dynamic, we have to have a structure that means people stay when they get big. Right now they don't stay and they get big. As evidenced by the top employers in the city being vast majority non-profit. - Yeah, so. - All right guys, well, I know we're way past the time. So thank you guys for being part of the Roundtable series. I really appreciate you being part of this. It's great here and your stories and your journeys and really looking forward to what's next for the fully start-up scene. - Thank you. - Appreciate guys. - Thank you. - Thanks, everyone. - Thanks, Zach. - Cool. (upbeat music)