Building the Robotic Workforce: Yondu AI's Michael Chen on Warehouse Robots, Teleoperation & Going from MIT to YC 18VC Podcast · Episode 3 · May 27, 2026 Guests: Michael Chen Source: https://www.18-vc.com/podcast/yondu-ai Lucas: Hello, friends. This is Lucas. Welcome to the third episode of 18VC podcast. In this episode, we spoke with Michael Chen, the co-founder and CEO of Yondu AI, a YC company that's creating the future of robotic workforce, starting with logistics automation. For those of you who are new in this channel, 18VC is a student-run podcast featuring student founders, operators, and investors betting on the youngest and smartest minds in the next generation. The host today will be me and Philip. Now, buckle up and get ready for the show. Lucas He: Hello, everyone. Good morning, good afternoon, and good evening. Welcome to another episode of 18VC Podcast. And a huge welcome to Michael Chen, our guest today, the co-founder and CEO of Yondu AI. Yondu is building the future of robotic workforce starting with logistic automation. So I actually got to met and found out about Michael and Yondu through Troy Lab's Demo Day. The startup demo is actually one of the largest startup demo events held annually at USC. And Michael actually showed me around his lab area and office space at Gardena, Los Angeles, where I got to see the amazing humanoid robots that are currently deployed in one of their early pilots. So Michael, you told me the other day that you always loved building, even from a young age, right? Michael Chen: Yeah. Lucas He: What's the first thing that you've ever built that you can still remember today? Michael Chen: Ever since I was a little kid, I've always loved building. That's one thing about me. I've built everything from a gasoline-powered snowboard to laser dog-fighting drones, to a giant train track-cleaning robot. I mean, the craziest things you can think of. But I feel like when I was really, really young, what I would love to do is build things to help people. Lucas He: Others. Michael Chen: So help others, exactly. So my parents actually had a Chinese takeout restaurant growing up, and I was always working there as a little kid, pretty much ever since my head could stick over the counter. And even when I was little, I wanted to make my parents be able to work less, and so I was always trying to build contraptions and machines to help them out around the restaurant. One of the earliest things I can remember building was- Lucas He: Mm-hmm Michael Chen: ... actually a machine to help fill up the sauces in the restaurant. Lucas He: Sauce machine. Michael Chen: And also, a little robot waitress that would kind of walk around the restaurant and take people to their tables. So that's some of the earlier things that I built. Lucas He: Well, how did it work out? And how did your parents react? Michael Chen: I think they were kind enough to let me try things, and they loved to see me build and explore my own passions, especially if they're related to helping out around the restaurant. Lucas He: That's so amazing. And just out of curiosity, what's your childhood dream? And are you living your childhood dream right now? Were you thinking about that you would end up becoming a founder or builder when you was a child? Michael Chen: Yeah. From a very early age, I always knew I wanted to be an engineer. I wanted to build things with my hands. I wanted to do something physical. I knew that from a really early age. I remember when I was in middle school, I said, "I'm going to build a jetpack to get to college every single day." Lucas He: A jetpack. Michael Chen: I never actually built the jetpack, but it kind of shows you how much I love to build and just create things. I felt like creating physical things was just so much fun, and I had this ability to imagine designs in my head- Lucas He: Wow Michael Chen: ... and be able to create and draw them on paper, and then bring them to life. I remember when I was in high school, I built this massive solar still to distill water actually from plants. And it was- Lucas He: Wow. Michael Chen: I started it from just an entire imagination of what it would look like, and I was able to source the materials and build this thing completely from scratch using just that initial vision that I had in my head. So I felt like I always had this ability to visualize things in 3D space, and then- Lucas He: That's incredible Michael Chen: ... and then build them. Lucas He: So that motivation to keep you building- Michael Chen: Yeah Lucas He: ... is it mostly an inner drive, or do you also draw inspiration and creativity from someone or something out there? Michael Chen: Yeah. I think it's innate. I love to just build and create and solve problems. I think everything I build usually starts from a problem I want to solve, and that's really core. But also, I'm inspired by those around me. I think these days, I hear a lot of hate for people that love entrepreneurship. Lucas He: Yeah. Michael Chen: And I think the reason-- It makes sense, right, what they're saying. People like, I think Peter Thiel have been quoted saying that you shouldn't back founders- Lucas He: Mm Michael Chen: ... that say they love entrepreneurship. Because starting a business is grueling, right? Lucas He: Yeah. Michael Chen: It's very, very difficult. Like Elon says, it's like staring at the abyss, right? Lucas He: Yes. Michael Chen: It's very, very difficult. But, I believe that you can be really passionate. That can be your dream. And it was my dream growing up. I remember when I was really young, I read actually Elon Musk's biography. Lucas He: Wow. Michael Chen: And I'm probably going to get a lot of hate for saying this, but he is still one of the people that inspire me the most today. The ability for one person to make such a enormous impact, that's what really inspired me- Lucas He: Yeah Michael Chen: ... about entrepreneurship. I always wanted to do things to solve major problems in the world, and entrepreneurship is a vessel. It's a mechanism to do that. And it's not just that, but you're providing people with jobs. Lucas He: Yes. Michael Chen: You're creating new value for the economy. Lucas He: Yes. Michael Chen: You're building things from scratch. These are all really appealing parts about business and entrepreneurship, and I can't lie, I was fascinated from a very, very young age. I was always one to want to start businesses of my own. Everything... Usually, it was small things, right? Lucas He: Yeah. Michael Chen: Small businesses- Lucas He: Yeah Michael Chen: ... even in high school. But when I was in college, there was a program called The Sandbox Accelerator program, and it's typical for a student to maybe do this program once throughout their entire college career. It gives you funding. You'll basically apply, and they could give you a few thousand dollars to support your startup idea. And you have to write an application. Lucas He: Mm-hmm. Michael Chen: So it's a lot of work to do it. Lucas He: Yeah. Michael Chen: And I think typically people might do it once or twice. Lucas He: Right. Michael Chen: They might get tired of it. They might continue with an idea. I did it every single semester- Lucas He: Wow Michael Chen: ... at MIT. I might be the only person to ever do that. And just because I love entrepreneurship so much, there's always a million ideas just flowing through my head that I want to implement, and I love channeling that creativity into a mechanism that you can actually use to make an impact in the world. Lucas He: That's incredible work. That's incredible work, man. Philip Zeng: Yeah. I think actually the answer is the question I'm going to ask, but I'll still put it up. So is there any specific people or experience that drive you in this more adventurous way of- ... starting your startup instead of going for a corporation job? Is there any... Yeah. What's your take here? Michael Chen: Yeah. I've always been inspired by the great entrepreneurs of our time. Reading biographies is- Philip Zeng: Mm-hmm Michael Chen: ... my way of learning from their journey and their footsteps. I'm actually obsessed with reading biographies. But probably more importantly, I'm inspired by my parents. They're both entrepreneurs. They came from China from a young age. They were in their 20s when they arrived, and they didn't speak English. They're in debt. They have no family, no connections, and they're able to go from pretty much scratch to get here, learn, work in a restaurant, kind of build up their finances from zero, pay off their debts, and open up a restaurant of their own, open up their own business, provide lots of jobs, and now they have two restaurants in Florida, and they're doing really well for themselves. Philip Zeng: Wow. Michael Chen: And that, to me, is entrepreneurship. That's the American dream. Lucas He: Yes. Michael Chen: That's the ability to g- go from zero to one- Lucas He: Yep Michael Chen: ... in an incredible way and provide value for you, your future generations. That's what's amazing to me about starting a new business, and so my parents definitely inspire me a lot. Their resilience, their hardworking-ness, their entrepreneurship, and their willingness to take huge risks when they're starting from nothing. Philip Zeng: So I think probably you learn from your personal experience when you're looking at how your parents are doing their own businesses- Michael Chen: Mm-hmm Philip Zeng: ... and that, I assume, encouraged you to pursue a startup way. Michael Chen: Yeah. I definitely think so. Working at a restaurant from a really young age impacts people in a certain way. When I was seven or eight years old, I was already working in the Chinese takeout restaurant. Taking orders, working the fry cook, cooking fried rice, noodles, and one of the most impactful parts of that experience was definitely learning just how to deal with immense stress and pressure. I think a lot of people just kind of take for granted that type of experience- Philip Zeng: Mm-hmm Michael Chen: ... but I know how much it impacted me throughout my entire life. Working in that chaotic environment where customers are angry and they might be yelling at you- ... because their order isn't ready, and it's a small mom-and-pop restaurant, right? There's only so much you can do. I think that is a transformational experience because for me, that made me resilient. Not everyone can handle that, and some people might give up, or they might go hide in the bathroom, or they might go home, right? Philip Zeng: Yeah. Michael Chen: But I stuck it out. Even as a kid, I learned to focus on the things I can control and tone out the noise and stay locked in, and get the job done, even when I was really young. And so that allowed me to be the person I am today, to be able to communicate with people- Philip Zeng: Yeah Michael Chen: ... go through difficult situations. Philip Zeng: Yeah. Michael Chen: The leadership of managing people in the restaurant kind of carries over to managing people today and just being a leader through difficult times. Lucas He: I think that's the kind of environment where you foster some of your greatest merits of tenacity and persistence that's able to carry you forward in this entrepreneurial journey. Michael Chen: Yes, 100%. I still peop- tell people doing a startup is not even as stressful or difficult as working in a restaurant. Lucas He: Wow. Michael Chen: I really think so. I stick to that because you're not standing in a hundred-degree environments where you have a huge list of orders, and you have to pack it and ship it- Lucas He: Wow Michael Chen: ... in such a short time window, and you have- Lucas He: Yeah Michael Chen: ... angry customers. Lucas He: Demanding food. Michael Chen: Exactly, yeah. And it's all family-run, so you've got to tell your grandma to hurry up and- Lucas He: Oh, God Michael Chen: ... cook this thing, and you've- Lucas He: Oh, wow Michael Chen: ... got to tell your uncle, and your uncle's getting mad at you, and then you got to tell your aunt to pack up the order. So it's the family environment I feel like probably doesn't help because everyone's mad at each other, and they're yelling- Lucas He: Yeah. Yeah Michael Chen: ... and they're... Lucas He: Yeah. Michael Chen: And it's all this chaotic- Lucas He: Yep Michael Chen: ... environment. Lucas He: Yep. Michael Chen: But if you can learn to deal with that, you can deal with anything. Lucas He: Yep. So prior to founding Yondu, you actually spent a summer in ARA developing autonomous drone system, right? Michael Chen: Mm-hmm. Lucas He: And then also one and a half year at MIT CSAIL, if I'm correct, doing open source robots research. How did these experiences and all the experiences prior to that combined sort of help pave the way for the foundation of what Yondu? Michael Chen: Right. My summer at ARA was a incredible experience. So this was when I was still studying at MIT. I went to go work for this company. I was in West Virginia, so the company is based out of DC, so West Virginia was the natural place to be to be able to test drones in big field, and the thing about West Virginia is there's nothing to do. Lucas He: Yeah. Michael Chen: Right? There's nothing to do in West Virginia. Lucas He: I'll agree. Michael Chen: I was living literally... around farms. Lucas He: Oh. Michael Chen: So I found an Airbnb that was kind of in the basement of this apartment that was surrounded by farmland. And I didn't even have a car, so I was sharing my roommate's car. And so I had to figure out something to do when I got off work, right? And I could cook and I could do whatever, but I started getting into figuring out how to connect ChatGPT to robotics. This was right around when ChatGPT came out and was getting popular, so early 2023. Lucas He: 2023. Okay. Michael Chen: Yeah. Right. So this is when it had taken off already, right? For maybe a few- Lucas He: Months Michael Chen: ... I think of it in semesters, right? Because I was in college at the time. But, aside from a few papers, no one had really connected it to robotics before. Now, I didn't know about these papers at the time, so when I was able to connect it to a drone and use my voice to control a drone navigating and tell it what to do, still today that sounds magical. It was an amazing, incredible experience. I got that working and I annoyed my landlord a lot and my roommates. But it was crazy. You could tell it to follow that person or fly in a particular shape or whatever that might be. And that kind of made a light bulb go off in my head. I thought, there are so many applications of this technology. By making robots, quote unquote, intelligent, there's so much you can do. There's so much potential that could be unlocked. And I proposed to the company I was working at, a really interesting project, kind of leveraging some of this technology on a fleet of drones. Lucas He: Mm-hmm. Michael Chen: And they loved it. They actually got funding from the- Lucas He: Wow Michael Chen: ... CEO of the company. We traveled to Washington, DC to pitch this to the CEO. Lucas He: For that project? Michael Chen: For that project. And all of the interns that were working with me at the company at the time- Lucas He: Wow Michael Chen: ... we were able to take all the interns and put them on that project. Lucas He: Wow. You became the elite intern in a sense. Michael Chen: Kind of, in a sense. Well, I took all the interns. We got to work on this project, we got funding for it- Lucas He: Wow Michael Chen: ... and we got to have the best summer ever. And you can imagine when I got back to MIT, I was just astounded and I was entranced by the- Lucas He: Mm-hmm Michael Chen: ... idea of making robots more intelligent and what it could mean for the future of robotics. I could imagine now a world where robots are truly all around people, interacting with people, working alongside people. And around that time, I was thinking, "Hey, I should start a company around this." Lucas He: Mm-hmm. Michael Chen: There's so many problems to solve with this new technology. Lucas He: Yeah. Michael Chen: I called up one of my best friends, Tamid, from MIT. Lucas He: Mm-hmm. Michael Chen: And it was a tough sell, right? Because Tamid, at the time, he already had offers from several of the top quant firms. He was getting offers for several hundred thousand dollar salaries, and it's a cushy life. So I thought it would be a difficult sell. And I call him on the phone, and I tell him about this idea, and I ask him, "Hey, would you want to start a company around me?" And at the time, we had no funding, no backing. It was just an idea. And he said yes on the spot. That was amazing. He and I ended up starting this company together, and we ended up going to Y Combinator in winter of 2024. Lucas He: Yeah. Michael Chen: So January 2024, we got into Y Combinator and started working on the company full time. Lucas He: That's remarkable. How do you and Tamid, your co-founder, sort of complement each other, in the execution, administration- Michael Chen: Mm Lucas He: ... or even just building? Michael Chen: Yeah. Tamid and I are really different people. So that it works out really well, because that's the best. I think when you have two founders or three founders, and their Venn diagram of skill sets- Lucas He: Right Michael Chen: ... intersect very little, I think that's great, because as a founding team, you need to be able to do a lot. Lucas He: Exactly. Michael Chen: You need to be able to solve problems in all different facets, from business to engineering to leadership, and people need to have different skill sets. Tamid and I complement each other really well in the sense that he is definitely a CTO, right? He's incredibly smart, incredibly research-driven, and has this amazing ability to explain ideas to people, provide technical leadership, while I have... And also complementary technical skill sets. Lucas He: Mm-hmm. Michael Chen: He has a lot of experience on the AI side, so he actually did one of the first compressions of a vision language model, at MIT. He did a lot of mathematical work at MIT, and I, on the other hand, had a lot of hands-on mechanical building experience. Lucas He: I see. Michael Chen: So I actually helped build one of the first open source bipedal robots at MIT CSAIL Lab, so that's the AI lab at MIT. Lucas He: Mm-hmm. Yeah. Michael Chen: And I had a lot of experience, like some of those projects that I mentioned, just building things from scratch- Lucas He: Yeah Michael Chen: ... as well as with entrepreneurship. Lucas He: Yeah. Michael Chen: So I was really familiar with the process of starting a company, of what it took to build a company, and we were able to leverage those complementary skills- Lucas He: Yeah Michael Chen: ... to be able to start and run the company with a very small team from the very beginning. Lucas He: That's called fit. That's what you should be looking for among co-founders- Michael Chen: Exactly Lucas He: ... at the early stage. Let's talk a little bit about Yondu. So humanoid robots could actually be used in a wide array of application and scenarios, and indeed, you are building a general-purpose embodied AI platform for the general-purpose hardware, right? But why did you choose a warehouse and logistics setting as the entry point, and more specifically, the task of bin picking? Michael Chen: Yeah. It was really organic. So we posted demos of some of the early technology we were using, so foundation models for robotics-We posted some videos using some of our early robots. So that was the HelloStretch robot at the time. We were picking items off a shelf, and when we posted it, and we had it on our website, warehouses actually reached out to us- Lucas He: Oh, wow Michael Chen: ... which was pretty crazy. It was totally organic. Lucas He: Wow. Michael Chen: Warehouses started reaching out to us and asking, "Hey, I have some ideas for tasks you could help me automate with this robot. What about mobile bin picking in our warehouse?" So that was the process of grabbing things off of shelves and putting them into carts. It's a really repetitive, tiring task. You got to walk miles every single day. And we thought, "Hey, this is a great application, but we're not sure if this is the right task for us to focus on. How about you introduce us to your friends, and we can see if other people are interested in this?" So he introduced us to his friends, his friends introduced us to their friends. Lucas He: Wow. Michael Chen: And that's when we knew that this was- Lucas He: It was onto something Michael Chen: ... something special. Lucas He: Yeah. Michael Chen: Exactly. That we were onto something, and there was a market for it. And that's when we started working on this particular application within the warehouse. Secondly, the warehouse space from a technicalist perspective is really the right place to start. It's a more controlled environment. You don't need to interact directly with people, and there are a lot of really simple, repetitive tasks that are shared across all warehouses. The picking, the packing, unboxing, stocking. This is present in every single e-commerce DTC facility. And finally, it's still very manual. Lucas He: Hmm. Michael Chen: I think people see lots of warehouse automation out there, but the reality is, over 80 to 90% of warehouses still have no automation. And that's- Lucas He: Why is that? Michael Chen: Yeah. And that's because a lot of the warehouses out there, they aren't satisfied with existing automation solutions. Lucas He: Hmm. Michael Chen: So you have really expensive automation solutions, like an ASRS system, for example, which can cost tens to hundreds of millions of dollars, and it might take 10 years to get a return on investment. And you also have, on the other hand, AMRs that are- Lucas He: Mm-hmm Michael Chen: ... really low cost, but they don't provide automation- Lucas He: Mm Michael Chen: ... of the task. It's really to help people walk less or make their lives a little bit easier, like a fractional improvement. And then in the middle, there's nothing. So for the vast majority of warehouses that don't have the throughput required to make these solutions make sense, there's no automation solutions for them. And that's where what we're building comes in. We're creating these general purpose robots that can be dropped into a warehouse, that are low cost. You can get a return on investment really quickly. And it makes so much sense to the warehouse owner because they've always been faced with solutions that either they can't get an ROI on or it takes 10 years to get a return on investment. Whereas this, you can do it within a few months to provide a full level of automation, and you can drop it in without changing any infrastructure, and slowly roll out these robots- Lucas He: Yeah Michael Chen: ... because of the fact that they're designed like people and can be dropped into existing environments. Lucas He: That's so interesting. Could you make some clarification with our audience who are outside of the 3PL warehouse space? What brownfield bin picking mean? Michael Chen: Yeah. So brownfield is what we focus on. It means the ability to implement automation without having to change the infrastructure. So traditionally, because automation solutions like purpose-built machines or robot arms that are hard-coded, these types of solutions require you to build infrastructure around them, whether it means changing your entire racking system, all the bins you use, safety measures, gates, whatever. There's a lot of infrastructure that needs to be built that can halt your operations while you're building them. Lucas He: Okay. Michael Chen: And they can also cost an enormous amount of capital expenditure. But I think the key is you don't want to have to shut down your facility- Lucas He: Yeah Michael Chen: ... for an hour, nonetheless, like a few months, right? Lucas He: Yeah. Michael Chen: That could be the end of a business. Lucas He: Yeah. Michael Chen: So we're able to drop in a solution that can start working from day one, so you don't need to wait for infrastructure to be built. Lucas He: So much better. Michael Chen: Yeah. Lucas He: And speaking of partnerships, I know you already have secured some early partnerships- Michael Chen: Mm-hmm Lucas He: ... and already have some early pilot customers. Michael Chen: Mm-hmm. Lucas He: How has those partnerships, for example, with ShipBots and some of your other early pilots, helped you actually validate the demand a- and further improve on your existing system? In other words, what are something that you only learn from actually deploying the robots in the actual warehouse setting that cannot be learned from simply lab results or inferred from market research? Michael Chen: Yeah. In the age of software being eaten by AI, I think deployment and hardware is the true moat right now. Deployment is incredibly difficult. You need to figure out how to wrangle this very, very new technology, which is humanoid robots or general purpose arms, and figure out how to deploy them into a customer site, integrate it with the constraints in the environment, handle the disorganized and messy environments created by people, and still be able to provide value. It's extremely difficult. There's things people never think about, like the networking involved. How do you get a mobile robot to stay connected to Wi-Fi throughout giant metal racks where you have no clue what's on those racks that could be blocking the signals? There's challenges with scanning barcodes. What if they're bent, or there's not enough light, or it's disorganized? How do you handle when there's a mismatch between what the database is telling you- Lucas He: Sure Michael Chen: ... is in the inventory and what's actually there? These are all problems we face on an everyday basis, and being able to handle that... everything from a printer being broken to a robot's wheel falling off, right? There's all of these things you have to be prepared to handle from a deployment perspective that make deploying robots incredibly challenging, but also sticky. Once you get a robot deployed in a customer environment- Lucas He: Becomes your moat. Michael Chen: Exactly. That becomes your moat, and you can expand within a warehouse. Lucas He: Mm. Michael Chen: And I will make the argument that humanoid robots are the greatest value-add service in history. Once you get a robot in a warehouse, you can have it do one task well first, but then you could have it do other tasks. You could have it do picking but also packing. In warehouses, things happen in cycles, so packing is usually early morning, picking is early morning. Lucas He: Yeah. Michael Chen: But then there's probably another cycle of orders that come in in the middle of the day. There's probably another picking and packing session, then there's a stocking session. And so things happen in these time blocks, and if you have a general purpose robot, you can actually have it do multiple tasks. Not only that, but because you have all these sensors built into a robot, you can have it do other tasks as well. One of our clients is using us for surveillance as well. Lucas He: Mm-hmm. Michael Chen: So you leverage those cameras to do surveillance. You can also leverage those cameras to do inventory tracking. Lucas He: Mm. Michael Chen: You can leverage the lidar to do a three-dimensional map that you can use to showcase your customers or keep better track of where things- Lucas He: Yeah Michael Chen: ... are in the aisle. Lucas He: Yeah. Michael Chen: So these are all things that could be independent companies of their own, but these are services that you can add and provide more value to a potential client using existing hardware. Lucas He: Makes total sense. Philip Zeng: Yeah. So you have talked about your robots can actually do different task in the warehouse. Michael Chen: Mm-hmm. Philip Zeng: How intelligent is it right now? Do you need to do a different fine-tuning or different setup for it to be able to do different task? Michael Chen: Yeah, I think one day we'll get to a point where you can just tell the robot, "Hey, do this packing or do this picking," and it'll do it. At the moment, we're not there yet, and that's mostly because for an industrial use case, you have to have such high accuracy and reliability that you need to fine-tune your model for a particular application. So that's where we're finding the most use right now. If you have it doing picking, you need to train it on the customer's environment. You have to post-train your robot foundation model on specific data that you collect in that environment, and just get it really good at doing that one task. And for a client, they care about just doing one task really well. Philip Zeng: Mm-hmm. Michael Chen: And that's the most important thing. That's your wedge. That's how you get into the customer warehouse, and if you can do that, then you can worry about how can we do other tasks, and that's a little bit further down the line. Philip Zeng: Yeah. That sounds cool. So actually from my understanding, you are using a hardware that is not produced by your own, but you buy from-the-shelf robots and then you probably collect the data, you train the model. Well, what's the rationale behind... Because there are other companies who are producing their own hardware. What's the rationale behind do not tap into that field, and would there be a selling of not doing their own robots- Michael Chen: Yeah Philip Zeng: ... in the long run? Yeah. Michael Chen: I'll go on record here saying that the biggest bottleneck in the next five to 10 years for humanoid deployment is going to be manufacturing capacity because the demand is there. It's an incredible demand. Every warehouse owner we talk to wants to have humanoid robots in their warehouse. We have a huge conversion rate, probably 50-plus percent conversion rate from initial conversation to a warehouse owner signing LOI or contract with us. Philip Zeng: Wow. Michael Chen: So there is going to be a tsunami of deployments over these next few years. Philip Zeng: Mm-hmm. Michael Chen: In our current day, there's a land grab. There are companies competing to get clients- Philip Zeng: Yeah Michael Chen: ... in the warehouse space because of the fact that there's a lot of companies doing a deploy-first teleoperation approach. If you can get a client and deploy robots in their warehouse, you can work with them and expand within their facility. Philip Zeng: Yeah. Michael Chen: And so I would call it a land grab right now, and truly manufacturing will be a bottleneck because the demand will be so great. By us not building out the manufacturing capabilities ourselves and being hardware-agnostic, we can leverage relationships with several manufacturers, and there are many, especially out of China- Philip Zeng: Mm-hmm Michael Chen: ... out of Korea. Even in the US now, there's so many humanoid and bimanual mobile manipulator companies because of the huge wave of funding that's come support these hardware companies. Philip Zeng: Right. Michael Chen: So if we can tap into this step change in robot hardware and we can leverage it mostly off the shelf, then we can satisfy the demands of clients around the world. That's number one. Number two is as a startup you have limited capacity. You got to focus on what you do the best, and for us that is training the robot foundation models and integrating the robots into the real world. That's a huge challenge in and of itself. To do that and focus on production, that's really challenging. Philip Zeng: Yeah. Michael Chen: So that's where we chose to focus and put our eggs, basically. So I would say those two are the main reasons. And then I think finally the last reason is not every warehouse needs a robot with legs and arms. Every environment has a different use case for a particular form factor or shape of a robot. So different robots have different use cases. You can have a robot that's more geared towards industrial use cases, for example, that has higher-end motors, longer usage, right, shorter cycle times. I think-That's more geared towards an automotive use case, but then you have robots that might be more compliant, that need to work more closely to people. And so you can have several different form factors that are able to satisfy the use cases that your customers have and for different types of clients. So that's what I believe. Instead of having one particular form factor in the future, there will be several different form factors that are all general purpose and can be sourced from different manufacturers that specialize in those particular form factors. So I see hardware-agnostic as a huge advantage for us. Philip Zeng: I see. So actually, talking about this hardware-agnostic- Michael Chen: Mm-hmm Philip Zeng: ... can you talk a little bit more about how is it actually like? Is it like cross-end effector? For example, if one robot use a, what's it called, a jaw or another one use a dexterous hand? Or is it in a different shape or form? How does this agnostic really means? Michael Chen: Yeah. We say hardware-agnostic because we use robots of different end effectors, different form factors, and we're able to use the same base model and just post-train on additional data that we collect with that particular form factor. And then once we do that, we can deploy it in the field and use it for the particular application that it's best suited for. So yes, we can use dexterous hands, we can use parallel jaw grippers, we can use different types of grippers, and robots that have different types of arms and configurations. Philip Zeng: Mm-hmm. And then also, I know there are robots with two legs. Some of them have- Michael Chen: Mm-hmm Philip Zeng: ... wheels to navigate around, right? You can also handle that. Michael Chen: Yeah. So we can handle it, and we choose not to work with robots with legs primarily right now, but we could. And that's primarily because in a warehouse environment, there's no use for them. Philip Zeng: Yeah. Michael Chen: So a lot of the folks you see that have legged robotics in a warehouse or factory environment, it doesn't really make much sense. You're losing a lot of efficiency by having all these additional components that use energy and can break. And for such a complex system, every single motor that you add is another failure point that you'll need your maintenance engineers to fix one day. And I believe the best approach is using the most simplistic form factor that can handle all the use cases that you have in mind. So as long as you can handle all the tasks presented to the robot, you don't need a robot that's as complex with legs and dexterous hands or whatever it might be. Philip Zeng: And then we can move on to talk about the data strategy of how you train your models. So from my knowledge, there are mainly three ways of training robotic models or how they collect the data. The first one is obviously, they kind of collect it from a human's movement, which I think it's called... Yeah, let me check. Yeah, I think it's called human motion capture. That is one of them. Like those guys who are shooting the films, I think. They look at different... A similar technology there, right? Michael Chen: Yeah. Philip Zeng: And there's another one which is AI-generated data that is kind of simulated data. Michael Chen: Mm-hmm. Philip Zeng: And the third one is, I think what you are doing there, that is teleoperation, right? And I also noticed there are several different startups that set themselves apart from other competitors by one of the three ways they choose. Michael Chen: Mm-hmm. Philip Zeng: For example, one that was called Galbot, Galaxy Robot, which is a startup in China, and they raised $1 billion recently. And they bet on the AI-generated data as the main argument. Michael Chen: Hmm. Philip Zeng: They say it's that the other ways are not efficient enough in generating the huge amount of data needed to train the model. So this is their argument. There's another one called Tars, which raised around $700 million recently. Also another Chinese startup on robots. They bet on the human motion capture data. And also we see Figure, which raised like $2.3 billion. They trains on a combination of these different methods. And we want to know a little bit more about your rationale of focusing more on the teleoperated data as the source. Yeah. Michael Chen: Yeah. I think the research papers have shown that a combination is the best. If you can have more data, it's almost always going to be better, and that's just because you feed these models a huge amount of data. The bigger the model, the better it'll end up performing as long as you have a suitable architecture. From a Yondu's perspective, we want to leverage what we're good at. We have real-world deployments. A lot of people don't have real-world deployments. We should be able to leverage those deployments to generate the highest quality data that you can't collect through simulation or through other means or... And I think that that's really crucial to our strategy. So real-world deployments allow you to access all of the edge cases that would never occur within a simulated or lab environment. For example, the flexible slippiness of a piece of clothing in a plastic bag. That's really challenging to simulate. That's also really challenging to get all the edge cases that could happen in a disorganized warehouse. So being able to capture those edge cases, that really valuable data, is what we prioritize and that we focus on. Not only that, but teleoperation data, because it's the highest quality... So teleoperation is using a person to be able to remotely operate a robotAnd it could be anywhere in the world. So Yondu has built out an amazing teleoperation stack that lets us control robots from anywhere in the world. And we believe that this is key because it's the highest quality data, because it matches exactly the robot form factor, and you can use the least amount of it to get the best result. You can always add in simulation data to get a better result, but if you haven't saturated the limit of what's possible with teleop data, it might not make sense to prioritize that. So for us, even though we have used teleop data strongly in the past and focused on collecting that, that's our main priority, we do leverage other types of data. So we have scraped over 10,000 hours of data, and we've annotated it using AI. Philip Zeng: Mm-hmm. Michael Chen: So I think we're probably one of the first companies to do this. We actually use AI to label all of our data sets. Philip Zeng: Wow. Michael Chen: Typically, it's a very manual process. Philip Zeng: Yeah. Michael Chen: You have a person sit there, especially for video data- Philip Zeng: Yeah Michael Chen: ... going through writing text labels for each subtask and each component of the data. We're able to use AI models to autonomously label thousands of hours of data very accurately and then feed that directly into the model. So we've been able to scrape different types of data that have been collected across different robotics companies and are open source on the internet and pair that with our very high-quality data we've collected in the field for our data mixture. Philip Zeng: Well, I think that itself of using AI to label the data could itself is a business, right? Michael Chen: Yeah. Philip Zeng: You could separate- Michael Chen: Yeah. Philip Zeng: You could spin out as a- Michael Chen: I know Philip Zeng: ... independent startup. Michael Chen: I tell people all the time that we have built several businesses- Philip Zeng: Wow Michael Chen: ... within Yondu just by creating incredible infrastructure around making our services possible, right? We've been around for longer than a lot of the other general purpose robotics companies in the space, so we've created a lot of infrastructure in-house from scratch that we can use as potential products in the future. Everything from data labeling to teleoperation- Philip Zeng: Wow Michael Chen: ... our exoskeleton, our VR teleoperation, these are all key parts to make our service work that could be companies of their own. Philip Zeng: Yeah. So probably for a better understanding of our audience, can you talk a little bit more about what is a teleoperation really looks like? Michael Chen: Yeah. So teleoperation means remotely controlling a robot, and that can come in many different forms. So people could use mouse to control a robot. They could use an exoskeleton. They could use VR. It's really a factor of mapping a mechanism for human input. It could be a joystick, a controller, keyboards, mouse, VR, and mapping that to robot actions. So the reason why there's so many different teleoperation methods is because their use cases are different. So, you can have an exoskeleton, which is very, very comfortable, and it's easy to use, but then the disadvantage is you have to build an exoskeleton- Philip Zeng: Oh, yeah Michael Chen: ... and you have to ship it to the user. Philip Zeng: Right. Michael Chen: So it's harder to scale that. You have VR teleoperation, which is kind of a happy medium, where you can very easily ship it out anywhere in the world, and your controllers can map to where the grippers are on the robot, and you can move it around remotely. But then there's also downsides. You get tired of staring at the VR world after a while. And then you have simpler methods, like a space mouse, for example. That can get three-dimensional space, but it's kind of complex to learn how to use that. And so there are trade-offs for each method. One of the big ones people are doing right now, which Yondu also does, is handheld grippers for data collection. So you can actually create kind of a replica of the robot's hand or a gripper and leave the arm out. Philip Zeng: Mm-hmm. Michael Chen: And because there's a camera mounted on there, you can replicate the view that the camera on the gripper would be seeing. And so you can hold that and start collecting data and use it as a way to simulate having the robot present. So that's from a research paper called UMI, and it's really, really cool in how a lot of robotics companies today are scaling up their data collection. They're sending these little grippers all around the world and collecting data like that. Philip Zeng: Wow. Following on, on the data part- Michael Chen: Mm-hmm Philip Zeng: ... are you using general video data, like videos from YouTube, featuring how humans are doing all kinds of activities? Are they useful? Michael Chen: Yeah. That has been shown to be useful through some recent research papers. And Yondu is working towards that as one of our key directions. So our robot foundation model is inspired by and utilizes world modeling so we can better understand physics. We can call it a world model VLA or vision language action model. And we also try and instill as much understanding about physics as possible. So segmentation, depth, pose, trying to get the model to be able to really understand physics and how the world works is super important to us. And a byproduct of that is our ability to leverage video data from the internet as well as eventually egocentric data as well. Philip Zeng: Yeah. Michael Chen: Egocentric being like a camera mounted on your head, for example. Philip Zeng: Oh. Michael Chen: So there's so many different types of data, and if you ask someone in the robotics industry, they'll probably tell you a really conflicted answer of- Philip Zeng: Which one is the best Michael Chen: ... which one is the best type of data to collect. But the reality is the more data you have, the better it's going to be. So, if you can figure out a scalable way to collect data-That's amazing. Different data types are going to have different levels of quality. So there's a defined tier, and teleoperation data is at the top. Philip Zeng: Yeah. I hear from another interview from some other robotics company founders, they said, the guy have done autonomous driving training before this one, and he says it takes about one million hours of visual data to do a fully autonomous vehicle, and it will take around maybe 10 times, which is 10 million hours of training data to do a general purpose robot. Michael Chen: Okay. Philip Zeng: Are you kind of agree with this kind of take? And how do you see you can scale to this huge amount of data? Because it's not like vehicles, right? You don't have any robots already doing the jobs and having the video ready out there. Michael Chen: Not yet. Philip Zeng: Yeah. Michael Chen: Not yet. I think that I definitely agree with what you just said. Robotics is a much more complicated problem than self-driving, and if you want to be able to handle the thousands of tasks that you do in a household environment, a truly general-purpose robot, you're going to need a lot of data. That being said, within a warehouse or an industrial use case, if you want to focus on doing one task really well, you need significantly less data. You can do it with a few hundred or a thousand hours of data, and that's totally sufficient in order to get really high accuracy and reliability, just because your environment doesn't vary that much and you focus on doing one task instead of thousands in all these different environments. Usually warehouses and factories, they look the same. And so Yondu's approach from the very beginning was we're going to leverage these simpler, quote-unquote, simpler. They're complex in their own right, but focus on doing one task or a few tasks really well within an industrial use case so that we can bring those to market while others are spending years collecting data and trying to figure out how to get the very, very complex task of getting a general-purpose robot working within the home. Philip Zeng: Yeah, of course. I'll move on to a little bit about the model part. Michael Chen: Yeah. Philip Zeng: Because as you already mentioned, you have training a fundamental model kind of thing, and you also mentioned word models and VLAs. I've noticed that some of the kind of industry leaders were Figure AI. They have their Helix VLA model, I think. And there's another physical intelligence, which I think not long ago released their Pi 0.7 models there. So from my understanding, their models come from different parts. For example, the Helix one. They mimic the structure of a human brain, right? There is a brain which works on how to plan a task. They take in visuals, they take in whatever language, others. Yeah. And also there's another one, it's called... I don't know. It's the part that controls your movement, right? There's also a part in our brain that controls our movement and help us keep balance. And I think from that Helix model, they're doing that way of design. Are you kind of sharing a similar philosophy of your models there, or are you doing it differently? Michael Chen: Mm-hmm. For us, it's a little bit different because we're not trying to pair reinforcement learning with language controls. For example, with Helix and Figure, they need to focus a lot on some predefined reinforcement learning tasks. For example, you need to tell the robot when it needs to activate the RL policy it learned to walk up the stairs, or walk or run or handle certain terrains. And then they care a lot about these sort of reactions, these lifelike reflexes. So that's a little bit different, right? For us, we want to do, as I mentioned, one task really well. And so what's the most important for us is creating a foundation model that does a really good job of being data efficient, leveraging the high-quality data we collect in the field, and creating a really good result for the industrial application. And that's why utilizing the robot foundation models that were out there, that wasn't sufficient for us, so we had to build our own in-house, and we had a lot of ideas for how we can improve it. One of the key ones was we wanted to create a world model VLA. We wanted to leverage the physics understanding of a world model, as well as take the best learnings from pose segmentation depth models and feed that into our model so it has the best understanding of the physical world. And because humans, we understand physics well innately, we're able to learn really quickly, and that's what our thesis was, going into the space, and that's been our main focus. So our foundation model that we've developed allows us to better understand physics and better understand how objects interact with the environment, and that gives us a data advantage. We can use less data to be able to get similar results as other companies. And then the last thing I'll say here is, in many ways, we're like a Waymo for humanoid robotics in the sense that we are focused on creating both the model, like Waymo does for self-driving, but we're also focused on the deployment, like how Waymo has cars in the real world. We're kind of unique in that sense because we actually focus very heavily on creating real-world deployments that we can turn into a data flywheel. Philip Zeng: Yes. Michael Chen: And the more robots we have in the field, the more data we can generate that's very high quality. And not only that, we can do it in an economical fashion. Philip Zeng: Yeah. Michael Chen: We can get paid for collecting data, essentially- Philip Zeng: Yeah Michael Chen: ... by doing the task within the facility.Makes total sense. Philip Zeng: Yeah. And about the word model part, from my understanding, I heard a little bit about what Feifei Li, Professor Feifei Li- ... is doing. They have a lab, right? Doing that. I think what they're doing there is generating a 3D asset kind of thing from probably your description, so you can navigate. It's like a game world, right? Michael Chen: Mm-hmm. Philip Zeng: You can navigate there. Is what you are doing different from their way, or is it similar? Michael Chen: Yeah, it's very different. Philip Zeng: Okay. Michael Chen: For us, we're still trying to output the actions. Philip Zeng: Mm-hmm. Michael Chen: And the vision language aspect of it, it's still doing the heavy lifting. Philip Zeng: Mm-hmm. Michael Chen: So what we're doing differently is we're taking learnings from the world model, from the segmentation model, the depth model- Philip Zeng: Mm Michael Chen: ... and we're using it to impact the loss that's used by model and better create outputs that are representative of the physical world. Whereas, for traditional world model companies, they're really focused on trying to predict or generate future scenes. Philip Zeng: I see. Actually, so logistics space is actually getting more and more crowded, right? Michael Chen: Yes. Philip Zeng: What I was just talking about with David and Philip, we're seeing the viral video Figure- Michael Chen: Yeah Philip Zeng: ... demoing their human interns versus their robots doing package sorting in livestream. And then we also know there are some other bigger companies like Amazon and Symbotic that are trying to move into the space. What is actually defensible about Yondu against this better-funded company that are trying to rush into the logistics space? Michael Chen: Number one is focus. We're able to focus on doing a few tasks really well and not spend our resources on the manufacturing, on the integration into the home environment, which is incredibly complex. We're able to focus on the warehouse and industrial use cases and do those really well. And we can build in changes into the hardware, into the model, into the classical robotics in order to get huge efficiency gains that you couldn't get if you're trying to just slap a general purpose robot into every particular use case. So, our thesis is different from companies like Figure. We believe there will be a few form factors that are the best for logistics and for manufacturing. There will be a slightly different configuration. That integration will be a huge part of the process. That's one of the things that Yondu has done really well compared to other companies. We've been able to focus on creating incredible integration into warehouse environments. So we have tie-ins to the warehouse management software, as well as the ability to integrate with other robots in the facility. So AMRs, like I mentioned, are a really common type of robot in warehouses. We're able to actually pick items onto the AMR in order to get greater efficiency gains. And nobody has done that yet. We're the first ones. That's a result of our focus on the logistics space, and a result of our understanding of the specific problems that these customers face, and our ability to actually deploy in the real world. How many companies can you name that are actually having several live humanoid deployments running every single day? You could count them on one hand. Philip Zeng: Yeah. Michael Chen: And we're one of those few. Even Figure, right? I think they did a deployment in a factory. They took it out, and they're doing a livestream- Philip Zeng: Yeah Michael Chen: ... in their facility, right? Philip Zeng: Yeah. Michael Chen: It's not a customer warehouse. Philip Zeng: Yeah. Michael Chen: So they haven't solved a lot of these challenges, especially with complex mobile tasks- Philip Zeng: Yeah Michael Chen: ... like what we're doing. Philip Zeng: Yeah. Michael Chen: Even the companies that are starting to do deployments today, if you look at them, they're trying to choose the simplest possible applications. Philip Zeng: Yeah. Michael Chen: I would call them toy or dummy applications because they're not providing that much economic value. For example, if you go to a warehouse, you can, of course, cherry-pick the specific use cases that are the easiest. Let's say you're trying to do packing, for example. You can always do packing- Philip Zeng: Mm Michael Chen: ... of only one type of item, right? That's very easy. Philip Zeng: Yeah. Michael Chen: You only need to know how to handle one item. Philip Zeng: Yeah. Michael Chen: But once you have a mixed bin of all sorts of items, that's when it becomes difficult. So it's very easy to cherry-pick deployment use cases- Philip Zeng: I see Michael Chen: ... just like you would cherry-pick a demo in a lab. Philip Zeng: Yeah. Michael Chen: And I would say that's what a lot of these other companies are doing. Whereas Yondu- Philip Zeng: Yeah Michael Chen: ... we are deploying in one of the most complex tasks, which is incredibly mobile, diverse, and dynamic. Having to deal with humans around them, having to deal with bins that are sort of disorganized and a wide variety of SKUs and different types of items. Being able to handle that showcases Yondu's ability to be truly brownfield and drop into existing warehouses, and adapt to differences in the infrastructure and the environment. And that's our most powerful asset, the ability to be hands-on, do the dirty work, integrate with the warehouse, and understand those key efficiency gains that we can change and integrate into our system to make it more efficient and better than truly a general-purpose robot- Philip Zeng: Yeah Michael Chen: ... that's trying to be slapped onto a particular application, or a foundation model that's just taken and just trying to be slapped onto- Philip Zeng: Yeah Michael Chen: ... a particular use case. Philip Zeng: That makes perfect sense. I'm sure that great founder like you also need great teammates to go out and build extraordinary things, right? And then as far as I know, you're also looking for fresh blood in your team right now, looking for robotic researchers and engineers, right? What are you not looking for among candidates? Michael Chen: Yeah. What we're not looking for is someone with 15, 20 years experience in the industry. We're not looking for someone that has been focused on the old ways, right? We want people that are new grads, that have maybe a few years, one, two, three years of experience in industry that understand this next wave of technology, that are passionate about humanoid robotics, that... are passionate about general purpose robotics- Philip Zeng: Yeah Michael Chen: ... that's what we're looking for. We're not looking for the old blood, we're looking for the new blood, and we're looking for young, hungry people that are passionate about what we're doing and are willing to work super hard to make it possible. Philip Zeng: Heads up our audience, your chance. Michael Chen: Yeah, we're hiring for full-time engineers right now, so if you're interested, let me know. Philip Zeng: Yep. Also having fun seems like an important part of Yondu's culture, right? Michael Chen: Yeah. Philip Zeng: Even to the point that- Michael Chen: Totally Philip Zeng: ... it's included as part of the cultural pillars. Its official website. Michael Chen: Yes. Philip Zeng: How does having that as a part of Yondu's DNA help the team members work better with one another? Michael Chen: Yeah. Having fun is super important. At Yondu, we try to have a basketball hoop set up outside, foosball tables set up. Philip Zeng: Wow. Michael Chen: We're always competing with each other, playing games, having fun, eating team lunch together. I think that's super important. Having a tight-knit team that's always in person, working in the office every single day is a very important part of our culture. We think that it fosters connectivity. Philip Zeng: Mm-hmm. Michael Chen: It makes us more productive and for what we're doing, how stressful it is- Philip Zeng: Yeah Michael Chen: ... dealing with real-world deployments- Philip Zeng: Yep Michael Chen: ... doing things that people have never done before, every single day, integrating humanoid robots into facilities. And the creativity involved in figuring out what's the next thing to do, what's our next project, what's our next task. This all is the creativity, the sustainability of working hard. All this is fostered by our culture, being able to have fun and work hard at the same time. Philip Zeng: Yeah. If you're going to describe yourself in one word besides a founder, a builder, or entrepreneur, what would it be? Michael Chen: Oh, my. That's a hard one. Philip Zeng: Yeah. We just excluded the ones that we just presented you a choice. Michael Chen: Yeah. I would say learner. I'm always trying to learn and improve. I think that as someone who's very young, I'm the youngest person on the team, and I have been since the very beginning. And I'm totally okay with that because every single day I'm learning from our peers at the company. I'm reading the latest papers. I'm learning from every facet of the company, across business to engineering, to AI, to deployments. Everything's a learning curve, to even management, right? Philip Zeng: Yeah. Michael Chen: Always reading books- Philip Zeng: Yeah Michael Chen: ... reading biographies, reading research papers, and learning across different facets of the business so I can be involved in the decision-making process, whether it be highly technical or very business and customer-focused. And I think that's just one of the most important things is never stop learning. Just constantly improving and never getting complacent. Philip Zeng: Yeah. And actually, we know that Yondu went through YC in winter of 2024. And how was that experience? What's the biggest change from that experience or what did you learn from that? And also, do you have any tips for some other friends? We got a bunch of friends who's also interested in applying for YCs, or do you have any tips for them? Yeah. Michael Chen: Yeah. As we talked about earlier, entrepreneurship is super hard, right? You are constantly alone. You're trying to solve problems. You have limited funding. There's a lot you have to do, and there's a very high risk you have to take. So of course, it's not for everybody, but if you're really, really interested in it and you're interested in applying for Y Combinator for the right reasons, what I would recommend is just make as much progress as you can before you apply. You don't need funding to get LOIs. You don't need funding to talk to customers and validate your market. I think that's the most important thing. YC wants to see that you've done everything in your power to validate before you need funding, and that, of course, they look at the team. So you have to have a great team, whether you're a solo entrepreneur or you have a co-founder or a few. You have to have the right team composition in order for them to back you. And a lot of times that means having some experience within the industry or being tied to the problem you're trying to solve. These are all really key aspects because at the end of the day, there will be competition. Philip Zeng: Yeah. Michael Chen: What's going to set you apart from everyone else? A big chunk of that in the early stages is going to be the team. So make sure you build the right team, you make as much progress as you can, and that doesn't mean building the product. It means exploring and validating the market. Philip Zeng: Yeah. Sure. And also, just a follow on. How that YC experience helped you most? Michael Chen: Right. Philip Zeng: Is it from knowing a bunch of people who are also startup founders, or is it a founding, or is there mentorship? What's your take there? Michael Chen: Yeah. I feel like when I was going through Y Combinator, there wasn't that many robotics companies- Philip Zeng: Yeah Michael Chen: ... in the batch. There was only a few. We actually had a small group with all of the YC robotics companies out there. Some of them don't exist today, so really fortunate to still be here and be kicking. But that experience was incredible regardless because we learned about how to speak with customers, how to do the go-to-market cycle. And I think probably one of the most important things is having that sense of community. Philip Zeng: Yeah. Michael Chen: People that push you, that are around you all the time, building and making a lot of progress. When I was going through YC, we were constantly compared to the SaaS companies. Which it's hard for a robotics company to compete with a SaaS company in terms of traction and customer growth. But we always felt like we had to push to try and be on the same level as them. And that type of momentum, to build that over the course of the three months that you're in Y Combinator, it carries on after you leave. That's probably the most important thing. And then the community. YC community is super supportive and helpful for each other. I, for example, this past weekend, I made probably 10 introductions- Philip Zeng: Wow Michael Chen: ... to clients for other robotics companies. Philip Zeng: Wow. Michael Chen: Being a part of that network- Lucas He: Yeah Michael Chen: ... is really amazing. Some people that I might have only met once or twice, but you're still able to help each other and provide suggestions and- Lucas He: Yeah Michael Chen: ... and advice. Lucas He: Yeah. Have you made any bad decisions as a CEO recently? Michael Chen: The worst decision that I ever made was trying to be an engineer as well as a CEO. Lucas He: Hmm. Michael Chen: I think that it's common, right? I think a lot of founders will say, "I'm an engineer and the CEO," or, "I'm an engineer and CTO." I think CTO it makes a lot more sense. Lucas He: Yeah. Michael Chen: But the CEO role, you have to make sacrifices, right? When I came out of MIT, I'm a mechanical engineer, right? Lucas He: Yeah. Michael Chen: I'm very technical. I wanted to build. Lucas He: Right. Michael Chen: And I thought that's how I'm going to contribute to the team, especially when it's small. I'm just going to build. Lucas He: Yeah. Michael Chen: I'm going to spend 80, 90% of my time creating designs and creating prototypes for our early grippers or whatever. Lucas He: Yeah. Michael Chen: We don't use any of that stuff today. Lucas He: Mm-hmm. Michael Chen: Right? The turning point for Yondu was when I decided to focus on the business side and really step into the CEO role. That's when our business took off. I think if you're a CEO and your founding team is focused on building for the majority of the time- Lucas He: Mm-hmm Michael Chen: ... they're not focused on the right things. As an early-stage company, the most important thing you can do is talk to customers, build out the market for the product, validate your hypotheses, and iterate as quickly as possible. If you're spending 80% of your time building the product instead of doing that, you're doing something wrong. Lucas He: Yeah. Michael Chen: It's okay if one of the founders is doing that, but not both. Lucas He: Wow. Michael Chen: Right? That was the biggest mistake I made, by far, was trying to be too involved on the technical side early on when I should've been involved more on the business, market, and validation side. Lucas He: Wow. I'm sure a lot of people will probably run into similar or same issues- Michael Chen: Mm-hmm Lucas He: ... as the one that you were going through. Before we wrap up, for all those student founders out there feeling uncertain and early, what would you want them to hear about? Michael Chen: Yeah, I would say timing is everything, and never give up, right? For us, there were dark times in the history of Yondu. There were times when we felt like there was a mountain of- Lucas He: Wow Michael Chen: ... of technology we had to build before we could even deploy our first deployment. And the problems that you're solving are often going to feel extremely complex. But the most important thing is to not give up. If you really believe in what you're doing, there's always a way forward. The most common way startups fail is they give up. Lucas He: Yeah. Michael Chen: That's the most common way. And if you have a good relationship with your co-founder, if you have a market that you're trying to tap into that you really, really believe in, you've got to just keep pushing forward no matter what. That's the most important thing. Because eventually, the stars will align, right? You will, over time, create your own luck. That's very important. For Yondu, for the longest period of time we were heads down building- ... making sure our first pilot run went really well. But once we got that working well and we started expanding to start targeting other tasks- Lucas He: Mm Michael Chen: ... now we do picking and packing and decanting for warehouse and logistics, we started getting tons of inbound. And now we're heavily oversubscribed on the customer side. And that's just a result of perseverance over time. Lucas He: Yeah. Michael Chen: Never giving up. Always focusing on what is the next thing to build? Lucas He: Mm. Michael Chen: What is key to the stack? Lucas He: Mm-hmm. Michael Chen: What's going to push us to the next level? Lucas He: I see. Now that concludes our entire session today. Thank you so much for your time, Michael. Really appreciate your word of wisdom and sharing today. Michael Chen: Thank you for having me. Lucas He: Yeah. Michael Chen: Bye. Lucas He: Bye. Michael Chen: Bye.