How Mehul Nariyawala Built Matic, the Smartest Home Robot Vacuum

So let's pick a problem that is so complex and challenging that we don't get bored even after 25 years of working in it. So we knew from day one that this was going to take a long time. And you know, how do you find Nemo? You just keep swimming. How do you build a robot that people love? Well, you just keep building.

Welcome to Just Go Grind. I'm your host, Justin Gordon. On today's episode, we have Mehul Nariyawala, the co-founder of Matic Robots. Matic is focused on building autonomous products that combine technologies such as vision, ML, and robotics. They're solving existing real-world problems to build incredibly useful products. And Matic, if you haven't seen the video already, the Roadmap is phenomenal. It's incredible. It's the best on the market and you have to check it out. Today's episode is sponsored by Row. With Row, founders get a single pane of glass they can use to run and scale their business from day one. They're a longtime sponsor of this show. They are phenomenal. Founders get bank account, corporate cards to start running, expense management, bill pay automation, accounting to help them move as fast as they scale. Row is trusted by thousands of companies ranging from 2-person YC startups to brands like Perplexity, Cal.com, Splitwise, and more. The big thing with Row, every customer has a person at Row, 24/7 customer support, phone, text, email. They will be with you at every critical milestone in your startup's journey. To learn more about why startups choose Ro, visit ro.co today. That's rho.co today. And if you happen to be in SF, New York, where they do tons of events for startups, check out their events as well. Again, visit ro.co today to get started. Let's get to the episode. Miguel, welcome to the show.

Thank you so much. Thank you for having me.

Yeah, I saw Matic on Twitter or X, however you want to call it, and was instantly blown away after watching the video, seeing what the product could do, and you can't stop. stop looking at it and being like, wait, why did this not exist before? I can see why it's a very hard problem to solve, but it's so obvious in hindsight. How did this company get started in the first place?

Good question. So, so I think I sometimes tell our team that this is our— this is the 3rd startup that I'm doing. 2nd as a founder, 3rd as an early sort of person. And I don't think we could have done this one before having gone through a couple of 2s. First ones, and really it just one of the hardest lessons that entrepreneurs ignore and don't learn is to solve a problem. And that's because we're all nerds at the core, and we as nerds love sexy and cool and futuristic stuff, and we want to live in a future. So we our previous company was gesture detection over webcam, which was really really cool, really sexy. The idea was that hey instead of turning your finger or hand into a mouse, if you want to mute something, let's say this chat or a TV, why can't I just shush it? So we built this iTunes app that allowed the users to control a bunch of like gestures, just using gestures. It was great. It was hugely successful. But we were part of YC back then. And Paul Graham, this is 2012, Paul Graham was still there. And he would come to us and he would look at us and he says, oh, I know you guys. You guys are technology looking for a problem to solve. Have you found one yet? And he would say that again and again. And that was one of the hardest lessons we learned along the way. And In 2017, 2018, as we were leaving Nest, so we're part of this, we are part of Nest. So as we were leaving Nest and we had worked on security cameras and we had been thinking about homes, and it just sort of dawned onto us that, wait, there are 200+ self-driving car startups in the world. Back then everyone was doing that. And equal amount of industrial robotics companies in picking and sorting and robotic hands. But when it came to home space, all we had was basically Roombas and other disc robots, which are just copycats, but nothing genuinely innovative. And no one, there was literally zero home robots robotic startup. And that just didn't connect the dots. It just was like, wait, why? These are obvious problems. It's not like you've discovered some magic problem. Everyone hates doing chores in their home. Everyone hates vacuuming or cleaning their homes, yet no one is solving this problem. Why? So we started digging into robotics for the first time, and we realized that— and our conclusion was always that the entire field of indoor robotics specifically is built upside down. Which is we started with the sensors and we keep adding more and more sensors. But that's like having 20 eyes but no brain. Okay. And that sort of approach is fine to take. But if you think about outdoor world, which is self-driving cars, even for Waymo and Tesla, you at least have a Google Map that tells you where— tells the car where the road is going and GPS that tells robot where it's located. But in an indoor world, how does robot know whether it's on the right side of the couch or the left side of the couch? Whether it's in the one right corner of the kitchen or the left corner of the kitchen, right in front of the refrigerator, in front of a stove. And that sort of precise understanding is just missing. And we very early figured out that, look, indoor world is built by humans for humans with our— for our perception system to fit our perception system, which is vision-based. So given our computer vision background, we thought this was the right way to go do it. And I didn't mention my background, but Both Navneet and I have been in computer vision startups since 2005. He, my co-founder and CEO Navneet, has a PhD thesis in computer vision, and his PhD thesis is foundational piece of the classical computer vision. So single piece of paper has 55,000+ citations and won 10 Years of Time Award for CVPR. So we've seen computer vision evolve and we just felt like robotics needed computer vision algorithms first approach, and that's where we started out. And this was the problem that had to be solved. So we started with this idea that, you know, what if we can build robot today from scratch? What if Disc Robots didn't exist? What would we build? And that's how we did it. Now, to give credit to Roomba, sorry to go on.

Yeah.

To give credit to Roomba, I don't think what we are doing was possible until maybe around 2017, '18. I don't think the technology pieces of the puzzle existed. So what Roomba achieved in 2002 is actually mind-blowing for that time. sort of time slice of time and timeframe. It's just that in 2017, there was an opportunity to start from scratch and disrupt it and then come up with sort of a Robotics 2.0 product. And that's what we decided to do.

Well, it is always interesting to see the market though, as well, when you have these incumbents and they don't necessarily disrupt themselves, right? Even though the technology was available and existed, because then it's like, are you cutting your own market share? Do you kill off your own baby? Like, that's such an interesting, fascinating thing to witness where someone like you has experience comes in then.

It's a— and some of it is not by design. It's not like they didn't understand it. When we started out, I remember listening to interviews from Colin Angle, who was the CEO of iRobot, and they understand the problem as well as we do. I think the challenge there is that one is they are locked into this disk design language. And once you flip that design language, you're actually telling customers that, oh, initial product is not good. So there is a bit of Hey, do I disrupt my own self or not? Right. As you alluded, the second part is also that typically robots and roboticists are usually hardware background and you need obviously hardware to start the company. And that's an insanely good background. It's just that we felt like it was always 80% hardware team and maybe 10% with one or two algorithms or vision person. And I think there was a— it needs to be flipped where you probably need 50 to 60 or maybe 70% Perception team, because the intelligence and perception is the harder problem. Hardware is hard, but it's relatively figured out. Motors and actuators and things have been around. So it's really just flipping that thing and then vertically integrating to build a really, really good product.

One thing I wanna go back to though, with Flutter, that company going through YC, obviously Paul Graham, which is good timing to be able to have that insights obviously from him on that. But you had that experience. You were working at, you said on Nest and stuff. I heard before you also had to work with Tony Fadell on that. Tell me more about that experience working with him. He's like a legend.

Tony Fadell is the best product and marketing person I've ever worked with. That was like drinking through a fire hose. And he's insanely demanding, as much as Steve Jobs is. And he's extremely critical of your thinking. And we learned tons of lessons there. And it was very deliberate actually to go move to Nest. One, we, when we, when Google acquired Flutter and we got inside Flutter, we sort of took a step back and we're like, wait a minute, why didn't we just build our own hardware? Because even when we were doing gesture detection via this webcam, we don't have access to frame rates per second, we don't have access to auto exposure, Apple doesn't really expose any of those APIs. So we were sort of building algorithm for eye that was out of our control in some ways, right? So then we're like, look, if you really want to build a product, you can't arbitrarily say it's hardware only, software only, what problem you want to solve. So maybe we could have built a set-top device with camera and allowed people to control TVs with gestures and among the other things and really build a product. But we never even thought about it. And that's because we didn't have a background in hardware. So part of the thing was to move to Nest to understand how to build hardware. Also, that was also the first time, ironically, I worked on a product that people actually paid for because until then, all computer vision and software stuff was free.

Wow.

So that was a really great experience. And Tony clearly taught us, and there were 2, 3 lessons that were really important that I'll jump out. One is that if you have cutting-edge technology, you have to combine it with hardware, but you have to have an extreme focus on user experience. If you can do that and build a great product, which is just purely delightful, that out of the box customers just fall in love with it, then it's critical. And that's where the feeling part of it, how it makes you feel, is more important than the checkboxes of the feature. So every single thing you got to do, but can you do it delightfully? That's the tricky part. So this is how, not what features you're building, but how you're building it. That's critical. So that was one. The second thing he also taught us that, like, we would go and say, hey, we're building this camera, let's put gyroscope in it. And he would immediately turn around and say, great. Now sell me, why should I pay for this $4 for a gyroscope? Tell me what benefit I'm getting as a customer. Sell me right now. And they're like, wait, it's like, no, that's, that's technical reasons. No, no, tell me why as a customer I'm paying for this. Because each sensor you add does result in an increase in price, increase in BOM cost, increase in complexity. So he taught us that single sensor you add in hardware, assume 3 software engineers on the flip side as a permanent cost. More sensors, the more calibration, more sensors, the more failure points, more sensors, the higher the complexity of manufacturing, BOM, supply chain, so on and so forth. So each sensor actually increases complexity exponentially.

Hmm.

So you have to be very careful. And that was the reason why robotics stuck out and what I told you earlier stuck out because robots are Christmas tree of sensors. And you've got your proximity sensors and 4 LiDARs and 5 cameras and things and that. And that sort of complexity actually creates a much, much difficult software stack than people realize. So our thought was from day one that if we're gonna make it work, the only way we can make this work is vision only. system just with 5 RGB cameras. And that was a big bet and we did it and there was some evidence. And the evidence is that if you go and look at Kickstarter campaigns, there are Kickstarter campaigns after Kickstarter campaigns of robots getting funded, but almost always they never ship or by the time they're shipping, ship, they're prohibitively expensive. And that's because of that complexity that goes like prototyping on a multi-sample survey is easy, building a new product is much harder. And then scaling is another challenge.

There's one challenge after another. Of course. Also in the early days, so with Flutter, you have the experience. I always love talking to founders who are repeat founders, because then it's like, how did you decide to build the next thing? Because yes, there's like, you see, you're looking for a problem. But oftentimes the founder friends I have, they have so many different either ideas or things they could work on, or they have experience in whatever, like they're getting pitched stuff. There's plenty of ideas out there, things they could do. You have this experience that applies to, in theory, multiple things. Why this problem? Why not other problems? Why even start another company? Why not be like, I'm gonna go chill, work for a big company? Why that? I want to know where your head was at.

A great question. I think the answer there was Flutter felt like selling our baby a little bit, and we left gestures alone. And maybe we could have pulled— made gestures more ubiquitous if we had a right idea on how to do it. So yes, it was a good outcome, but it wasn't necessarily satisfying outcome. And satisfying outcome, meaning that for us, the best days of our career, if I could go back, are still the days where we launched something and people just loved it.

Yeah.

That was the case with Flutter app. That was the case with Nest Cam Outdoor Rural. So that's what we were chasing. And we took a step back and realized that, okay, we've been in computer vision for a while, way earlier than many people. 2015, '16 was really the tipping point for AI and computer vision, and it was going to skyrocket. So, and we've tried many iterations, but yet we have to combine anything and build what we thought was an iconic product, which is if you're in this space, you want to build something memorable, something iconic. An iconic product had 2 definitions for us having gone through things. One is that it has to be a product that comes to the market and changes the perception of what's possible. It's sort of, it has to open up imagination. So iPhones, iPods, Model S, these are all that, right? ChatGPT, similarly that. The second thing was that it has to be incredibly useful, not just cool. And the second part was the Flutter experience that we came from. And to me, one of the most iconic products, for example, is something we literally give zero time of day in a product called microwave in kitchen. And the reason it resonates with me is I moved to United States in '94. I had never heard of microwave until I moved to my uncle's place and saw one. And I remember thinking that, wait a minute, it's literally named out of physics technology, which is crazy. And then if you try to describe it, there is nothing cool or sexy way to say that it just heats fast. Okay, cool. So what, right? Yet it's a product that if it broke, you're going to be in a Target or Amazon within a few hours to try to replace it, because it's so insanely useful. And iPhone is obviously that way, and cars are that way, and iPod with the music was that way. So you have to build something that you can't live your life without, you can't fathom your life without. And that's the kind of product we wanted to build. So one was that, that also meant that we wanted to build something consumer. So, so I have been— my first job in Bay Area was at Salesforce.com. I've gone through the SaaS tree and just realized that I'm just not that excited about any SaaS problem whatsoever. And so that was the choice that let's solve a consumer problem. Let's make sure we're building a product. Let's make sure we're building a product that actually we would use ourselves and be the customers off, and then make sure there is a clear-cut market and clear-cut business model, which is we didn't want to be in a business of creating a market. That's obviously doable, but it's much, much harder. Typically, you're better off going in some market that already exists, and then you're just creating insanely great product for it. And so that's— that was the one. Then business model was we didn't want to kind of be in a— I mean, it's amazing that OpenAI figured out ChatGPT as a subscription model. It wasn't necessarily obvious that if you created AI, what would be the business model, right? 10 years ago, it wasn't obvious that, oh, the business model for self-driving cars is most likely going to be robotaxis, right? Tesla tried to sell to personally, but it's not taking off. Maybe it will take off eventually, but robotaxis seems to be the primary model. So, so in a similar way, we just wanted to solve technology, not take some of these other issues. So it's— to answer your question, we created these rules of engagement on what we would build. And then one of them was also that computer vision has to be the core competitive advantage. Because if we try and go to crypto, what do we know about crypto? Someone will kick our behinds. But if it's computer vision, maybe we can do some damage ourselves. So it was really just kind of creating this rule structure and saying within that what fits in, and this idea really just fit in.

You know, it's interesting because even a couple of the recent founders I've talked to, uh, like John Andrew from Wander, he has this amazing luxury travel company, and he has like a whole document of deciding what to work on. It's like all these questions in this whole thing because he's done it multiple times now. And every founder is different. I talked to another founder too, same thing. They have like a huge list of like, okay, before I start something, here are the things to think about, like questions to think about, because the strategy early on will dictate a lot of what you're trying to build and things you're— what's your next 10 years look like? What's your next 20 years look like? Yeah. With the early days of Matic then, okay, so you decide that this is the way you're going to go, you guys are going to work on this problem. Did you know how long it would take to get to market?

Yeah, actually, we knew the reason we picked this problem was other rules of engagement was whatever we do next, we're never selling. This is the company we want to build for the long term. So this is the last job we want to have. So let's pick a problem that is so complex and challenging that we don't get bored even after 25 years of working in it. And the category we picked is always the first category, was always the business category that we'd use. But we knew that this is a stepping stone to solving general indoor robotics, or at least home robotics.

Yeah.

And, and the reason I say that is because the challenge I described earlier, which is We realized that, look, if human worlds or homes are built for humans, then if you're trying to build level 5 robot, not just cleaning robot, any kind of level 5 robot, well, if level 5 robot for cars means that cars drive like humans, so then level 5 robot in homes must be, and indoor worlds must be that they behave like humans and clean like humans and do tasks like humans, right? So if that's the case, well, they need a perception system like humans. And that perception system is vision-based. So, and floor cleaning is a perfect beachhead for this as well, because by definition, if you're cleaning floors, you have to explore every inch of an indoor surface and clean it, which means you have to map every inch of an indoor surface. Floors also get dirty multiple times a day. So you have to explore that environment again and again and update your map on the fly dynamically on the device, right? And then if you just thought that, look, if we can teach robot To go from point A to point B precisely 10 out of 10 times, regardless of the changes in the environment, obstacles, or lighting, that itself is a huge value proposition because there is some sort of— I want to— I call it a misperception in the world that all mapping or planning or navigation for indoor world is a solved problem. And I kind of sometimes laugh at it because it sounds to me like a Fermi's paradox, which is if it's a solved problem, where are all the robots? Why aren't our homes and airports and hotels and grocery stores swarming with robotic applications and robots? And the answer is the precision is not there. Yes, you can do basic navigation, but precision was missing. So we just had this first intention that if you go into a new home or let's say new indoor space as a human, what do you do? Well, we explore, we kind of walk around and we do randomly. We don't know which direction we're going, but as we are walking around, we are building a map and actually localizing and knowing where we are. from that front door or wherever it is immediately. And initially, we don't necessarily have all the information, but within 3 days or 3 hours of spending time in it, we're actually going to get very, very precise mapping understanding of that space and take most precise route from point A to point B. So can we give robot that ability? Can it explore on its own? Can it build a map on its own? Can it self-correct on its own? Can it adjust to lighting on its own and always find its way around if something is blocked on its own? If all those things, yes, that's a great starting point and floor cleaning was the way we will go solve it. So that was step 1. And then the analogy is really like growing a human child, which is between 0 to 5 years, human child literally just learns how to navigate and perceive the world. 5 to 10 is where they get really, really good at semantics and maybe start getting better at dexterity and manipulation. And we start giving them basic tasks like move things around or organize your books or toys, a table or whatever. And then 10 to 15, and 15 is when we trust them with everything, whether it's cooking or driving or all kinds of stuff. So can we scale robots in that way? And can we along the way keep productizing it? And productizing is critical because when you build a product, it reveals the challenges that sometimes just doing the research does not reveal.

Mm-hmm.

So we always call it a product-focused research versus research That's what we do.

I have heard in other interviews from you that you're like, you kind of think eventually it's like the Jetsons almost, like you're saying with all these robots and all of that.

Yes.

So to that point, you view it as like Matic with this robot is like the first thing you need to build, get this right, and then you're able to branch out.

This was always the phase 1. We're still in phase 1 and probably beginning of the phase 2 where we build sort of this 5 to 10-year-old kind of robot that is helping out on various chores around the home and then kind of going into full-blown humanoid in homes, or full-blown Rosie. I don't even know if humanoid is the right form. I'm not gonna claim it is yet.

Yeah.

And the reason being is that it's not just about human body. It's also that you have kids and pets and you want something in a home that belongs in a home, that feels loving, cuddly, sort of that is going to take care of the privacy of your home, that's going to treat your home like a home, because it has to belong. And if it doesn't belong, it doesn't make sense to be inside at all.

Yeah. And if you've seen the movie I, Robot, everyone can kind of relate to that in terms of having these things.

Isn't it? And then I, which is, we talk about this quite a bit, like, it's funny that all the Hollywood robots that talk like humans are actually quite scary. Generally, if they look like humans, they're very scary as well. Anytime you have lovey or like people talk about, oh, I want R2-D2 or WALL-E, they never say I want HAL or I want, you know, I, Robot. So it's interesting. So we'll see how that plays out. There is an element of future, but we'll see how it plays out.

Take me through the early days of building this. Did you have to fund it? Did you self-fund it? Did you know you had to raise because there's going to be a lot of— tell me more of that because, you know, it's going to be such a long journey. I'm curious how you thought about it because it's a little bit different than other companies.

Good question. So, I'll kind of talk about the analogy of startup we have, and this is a bit of a crude analogy, but we've always thought of startup as a ticking time bomb. The day you run out of money, it goes poof, no matter how big or small you are. And having known that, we realized that, that from day one, that the best way to go do this is to take a look at the self-driving car space. And we saw 2 approaches. One was to do what Tesla did, which is build a product, sell a product, make money, have amazingly loyal customers. And along the way, you get to collect ancillary data and all these other things as a side benefit, which propels into second, third, fourth product. The other way to do this was to do what Cruise and Nuro and all these other companies did. And that approach was tricky to us because then you might be— because we didn't know when we were going to be able to ship or how long it will take. So then you are in a perpetual fundraising mode and depending on the market. And then to me, as long as you're raising another round to survive, all you're doing is adding more time to that clock. You're not necessarily diffusing the threat. So one of the principles was also that how do you pick a category and a revenue version where we can get to cash flow positive right away, that we can get to profitability in that Then you can build second, third, fourth product, and you can keep doing it for a long, long time. So that was one approach. But that also meant that initially we had to pick people who were going to be a little bit more patient and understand this company. And that's not to say anything wrong about VCs. VCs are actually amazing. The venture capitalist has driven so much innovation. It's— but that is rocket fuel. And best time to get the rocket fuel is when your rocket is ready. in Dogged. So that's where we initially bootstrapped ourselves. Part of the reason for bootstrapping also that as a second-time entrepreneur, it is easier, not easy, but easier than the first time to raise funding. That also means that sometimes you don't do critical thinking enough in the first time. So there is enough second-time repeat founder failures as much as the first-time failures. But first time your bar is really high to raise funding. So you have to put in a lot of thoughts And that forces you to go through thinking process. But second time, there is a chance you can go out with a deck and raise funding, and that might actually give you false premise that you're heading in the right direction. So that's where we were like, no, no, let's spend our own money. Let's establish the right genes of capital efficiency. Let's make sure that people know that there is a risk involved and we're taking the right thing, put money where our mouth is, so to speak. So that's where we bootstrapped. And then eventually we ended up raising funding, angel round. Which was led by Matt Rogers, who was— who's a co-founder of Nest. He led the round and we just raised from entrepreneurs and other angel investors, from Adam Nash to Jack Dorsey to Naval to, to John and Patrick Collison as well, very early on. And then even in 2021, when we raised our next round, I guess, which could be one of the largest seed rounds, I guess, that was there. But that one, we raised about $23.5 million. And again, we were very lucky to have known John and Patrick Collison for a long time. And then he introduced us to Nat Friedman and Daniel Gross, and 4 of them just decided to fund us.

I love it. That's awesome. Yeah. And it's funny. Yeah. How you go about this process, especially as a repeat founder, because there are different options. And like for a repeat founder who's had any level of success in their first company, you typically— yeah, you can like kind of raise on just your name and like the thing, if it's a good— which is like a small amount of capital to start with. You just have to prove it out as you get to like seed, Series A, whatever. We need metrics still eventually. So it's like there's more to it, but you'll at least have that kind of that swing from that. How did you think about that? And then from that, like in the early days, building the team, knowing that you needed investors that were patient, because this is gonna take some time. But you also need the right team to be like, okay, we're building this really ambitious vision of what we're going to do here. Are you on board or not? Like, tell me more of the team building, especially in the early days of Matic.

Yeah. And so honestly, it was hard. I wouldn't say it was easy. We were also in a time— I'll take a step back. In 2011-12, when we were doing computer vision and gestures, I think beyond Google and us, there was no one hiring computer vision people. It was just so many early days because even Facebook AI research didn't exist. Amazon research didn't exist. Okay. Between 2012 to 2017, AR/VR became really big. Self-driving cars became really big. So there was obviously a lot more demand from computer vision and computer vision skill set across the board. Okay. So one, it was hard, but two, the way we solved that problem was by saying that we're going to optimize everything for the long term. So instead of a 4-year vesting schedule. We had a 6-year vesting schedule for our equity. So, and we still do. And we talked about this idea from day one that this is the mission. It will be the 0-to-1 thing. It will take time and be very direct and very honest with people that if you're here, if you want to join something for a year or 2, we're probably not a good place and have that conversation. And then usually if you create enough of that things, people self-select very well. Yeah. And say I belong here or I don't. And we were lucky enough to find few who were just really mission-driven, who understood this is how the robotics would play out, that yes, this is the right approach. And we would walk them through every single thing, including our long-term vision on day one. And if they— if it resonated, then it becomes a no-brainer. And then, you know, I used to also have a— in this office we don't, but the previous office, my desk was right in front of the wall and the poster that I had on top of it was all— was finding Finding Nemo's just keep swimming. So we knew from day one that this was going to take a long time. And, you know, how do you find Nemo? You just keep swimming. So how do you build a robot that people love? Well, you just keep building.

So what were the complexities? Because obviously when you start it, you have an idea what this is going to take to build. You have this experience and everything, but what were the complexities as you got into actually trying to develop this product and figure out how do we make this work?

Great question. So There's complexities we knew and complexities we don't. So I'll cover both. The ones we immediately understood was no one has tried to build a robot with RGB cameras only and just doing computer vision using 2 cameras and depth detection. So we didn't know if we can do precision depth detection. We had ideas on how to do it, but we didn't know. So first proof of concept was to say that, hey, we can mimic how humans do depth on a camera as well, and let's prove that. So that was first. The second one was What would the ID look like? What would the design look like? This was a very big challenge because all we knew was that disc robots was not the ideal ID. And then the way— and this is another Tony Fadell lesson. He would always say that a product has to immediately give off its purpose. Okay. And the way example he would give, which is sort of counter, but we'll get a lot of customers sending us an email saying that, hey, you have this beautiful thermostat. It has a screen. Why don't you show me notifications on it or alerts on it or news reels on it or whatever? Because it's just right on the wall as a beautiful screen. And his answer would be, because it's an effing thermostat. It's not your TV or screen, right? So in the same exact way, we kind of did this thought experiment that imagine taking Roombas and sending it to the '60s through time machine and asking people what it is. And there's a good chance they won't be able to guess that it's a robot or a vacuum because it looks like a hockey puck or a speaker. So first thing we wanted to do was make robot look like a robot, make sure that it feels very friendly to pets and kids because it's inside homes. Make sure that it's a cleaning robot, that it's not going to get confused as Alexa on wheels. So these are all very much a thought process that we wanted to have. And so just creating that, and then we didn't know how many cameras we would need, how many sensors we'd need. So all those things were sort of rules of engagement we created. One of them was solve problems using software willingly, hardware grudgingly. So only add hardware sensor when we can't, when we've proven beyond doubt that it can't be solved using software. Okay, so these are all the challenges that we kind of created. So those were some of the known challenges. The unknowns and surprising ones was, and you still make mistakes and everyone makes their own mistakes, was one was real in any robotic system or any robot, you will have some element of SLAM, which is simultaneous location and mapping. There is a LiDAR-based version of it and there is a visual version of it called VSLAM. Now, mathematics and theory of SLAM system was pretty much worked out in mid-'80s. So we made an assumption that there has to be some open-source systems out there that we will be able to piggyback on and use it. As we started playing around with it in 2020, we realized that, wait a minute, none of these are even 70 to 80% accurate. That it just, they just get, robots get lost all the time. They fail all the time. The accuracy is just not there. And if you try to rewrite the open source, we'd have to rewrite everything anyway, and it's too big to run on the device. So we had to write everything from scratch and now— and use many, many techniques. But now, you know, this is one part I'll brag about on behalf of our team, but our SLAM system is probably order of magnitude more accurate than anything out there, period, in terms of state of the art. And that took a long, long while. That was just a grind. It just— it's just whack-a-mole. You just keep doing it, doing it, doing it until you kind of capture all the bugs and get there. The second, and this is a bit of a mistake, but second thing is we started out too early, or there are 2 more things. One is a mistake, one is just reality. We started in 2019. In 2019, it wasn't obvious which edge device chip was going to get win. We knew NVIDIA was great, but NVIDIA did not yet share back then GPU and CPU memory. So you had to have a lot, there is a lot of latency if you keep transferring data between back and forth. So instead we went with Ambarella as a platform which was the right choice for 2019. But as 2023 came around, it was obvious that we needed to move and we couldn't be able to, we wouldn't be able to do it because CUDA was amazing. They came up with the Jetson Orin Nano platform, which is just phenomenal. So we made that move that probably resulted in another 9 months of transition. And then third thing is right now you see amazing tools coming out of NVIDIA. They're building all this, you know, Groot and Omniverse and Newton. None of these tools existed for us. So we actually built every single thing from observability platform to end-to-end simulation to synthetic data collection, everything is built in-house, all the systems are built in-house, and we created all that infrastructure ourselves. So that was, that's the price you pay for starting a little bit too early. But we, we started with early also with eyes wide open, because in 2017, the bet wasn't that robotics is here, the bet was that robotics is coming. And we want to be ready when the wave hits, we don't want to sort of start run, get into the water and run with our surfboard when the wave is already coming, so to speak. So that's where having years of up ahead gives us a little bit of an advantage.

Even though, so you had to build the, essentially build the infrastructure yourself for some of this. Like, do you think it gave you any advantage in learnings or lessons because you did kind of have to do that versus like obviously more out of the box, you know, kind of already? Well, hell yeah.

Yeah, absolutely. Absolutely. I don't know yet, which is time will tell. But I do believe, which is some people do ask us this question that, hey, what advantage do you have from going to the second product from, let's say, Matic as a first product? And to me, that's not the right question. The right question to ask is, do we believe and do you believe that Tesla has a huge advantage in building Optimus now that they've built, solved the FSD And the answer is hell yeah. So if that's the case, well, we've just at least solved indoor robot for floor cleaning based on reviews and user feedback we're getting. So yes, based on that, we have tons of advantage and I can't really articulate what are those subtle learnings that we've done. Those are so, so many and so tiny, but it creates the genes, the muscles you build that absolutely helps in the long term.

Obviously there were like just delays and complexities and challenges. Like it takes longer always and everything takes Yeah, more money and more time, like, right? Especially in this hardware. But take me through what the team, your mentality, like when you're getting close to finally being able to launch the product. Tell me more about that. Because like, I imagine there's so much time and buildup and you have all these ideas and you've been working so hard to get this thing out. Tell me more of that when you're like, you're getting closer to launch.

So, so yeah, 2 things. I'll start with a joke first. We kind of joke internally that we are only off by one digit because we thought we would launch in 2020 and we ended up launching in 2025. So like, hey, you know, we were pretty close. pretty close. So, we just kind of laugh at it. But I think I heard Scott Belsky had this really good framework for Scott Belsky that when you start up an idea, obviously, you have a conviction about this thing and belief. As you're growing along that adventure, is your conviction getting stronger? Or are you actually getting less and less enthusiastic about about it. And if it's getting stronger, stick to it. If it's not, maybe it's time to pivot, right? For us, the more we talk to customers and the more we even started testing prototypes in customers' homes, it was just obvious that we were building— what we were building was right. There were signs. It wasn't perfect. Like, you know, robots had a bunch of bugs and things like that. But you see this very few here and there wow moments, sparks of brilliance, sparks of delight. And we constantly tested it ourselves. I constantly tested it ourselves. And I saw that ourselves. Like, there were times where I would look and say, oh my God, robot did that, it will blow my mind. And then you see, so you see that subtle behavior and then you have to keep work hard to make sure that your team sees it too, because a lot of people are young and, you know, and I don't blame them for being impatient. I think this is the longest job I've held. So I can't really blame anyone for saying that maybe this is taking too long. But you have to keep reinforcing this idea that, look, we're making progress. There are signs. Look at the signs. Stop worrying about what external world thinks as a sign or what external world thinks as a score. Score takes care of itself as long as you do the daily thing. So, so, you know, just keep swimming is another one. There was another quote I love from a UMBC basketball coach that says the work you will— you do in dark will reveal itself in light. So you have to believe that. And it really just boils down to do you believe and can you transition that belief and strong conviction to team? Along the way. And this is where I'm very also lucky to have a good co-founder whom I've worked with for 20 years now, because when I'm a little bit on shaky ground, he's strong and vice versa. And we sort of build each other up and that's necessary. I don't know how single founders pull it off. There are some amazing single founders and more kudos to them. They're stronger than I am. But yeah, but we're lucky. And then we found some really, really amazing early team members who've stuck around and who've kind of just believed that we can get there. And honestly, the Wired article was more exciting for me from a team perspective than my personal perspective. Like, we've seen that before, but when finally that came out and the review came out and we got like first 10 out of 10 in 17 years, apparently. Yeah, that was like, holy shit. Okay, now, now it's sort of worth it. That was a really, really good boost. Yes.

Yeah, because it is like you need the morale because like, you know, internally you're like, I believe in this thing. I'm starting this, I'm in it for the long run, but like the team is working their ass off too constantly. You're like, can we get some validation here for this work?

And some of it is starting that way from the beginning. So we do have this sort of, we don't necessarily have, I don't like the values. I think values are overdone now, but we sort of kind of put together a philosophy of Matic, so to speak, and philosophy of doing life works. And there are a lot of examples where we talk about this. And I think one of my favorite quotes from Jeff Bezos is that, you know, if you're doing something crazy, you have to be willing to be misunderstood for a long, long time. And that is the point of contrarian. That is the point of taking a different route than what's expected. Yeah. And then just believe that, okay, lessons we've learned about schlep blindness or working on unsexy things, those things are just there. And schlep blindness is one of my favorite essays from PG. So, so those lessons, you just have to believe that you're actually going in the right track. And then there is one other framework that YC taught us. I don't remember whoever told me this, but the framework, and this is how we evaluated a problem, that the problem you're trying to solve, is it intense? Is it frequent? And how many people in the world have it? So if I kind of apply to security cameras, let's say Nest Cams, which I worked on, if your home gets burglarized, problem is extremely intense. It's through the roof intensity. But how frequently? Well, turns out most people in the world would never experience it. And those who experience it may experience it once in their lifetime out of like, you know, whatever, 90 years that they live. So frequency is nonexistent. That means number of people who experience the problem is nonexistent. So then it's very, very hard problem. It's a bit like selling vitamin, very, very hard problem to convince users on why to do it. Versus cleaning floors was like, yeah, floors get dirty all the time. And when they're dirty, it's hard to kind of to, you know, sit in your home and be happy about it. And so it's intense problem, frequent problem, especially with kids and pets, because they're sort of like a— I call them entropy tornadoes. And they just create so much chaos. And so, and then number of people in the world where everyone in the world wants to live in a perpetually clean home with perpetually clean floors. So if we can somehow get a product that solves that problem, that would do it. So those are all the things that we had thought through this, and those beliefs kind of helps you along the way.

To that point as well, knowing this problem and understanding, you mentioned like the parents thing, if you have kids, for sure, if you have pets, for sure, like you very much so feel this. How have you thought about the go-to-market? How this is going to grow in terms of actually selling product? Because it's an amazing looking product. I've seen—

Thank you.

Seen the video, like it's incredible, truly. But then it's like, how do you think about going to market? Like, because you could do a lot of different things.

Yeah. So there are a couple of things we knew from day one, which is one is I knew for sure that we have to do this as a direct-to-consumer model.

Yeah.

The reason is because if I go back and look at any successful hardware company, especially in the '80s, '70s and '80s, from Apples and Ataris or Intels of the world, or even Dell, almost all of them were built-to-order model, which is they found the product-market fit, then they scaled, not before. On a flip side, with Kickstarter, a lot of people thought that successful Kickstarter campaign was a product-market fit, and they built 40,000 devices. And turns out after Kickstarter, It didn't work because you are you are getting earliest of the early adapter on a Kickstarter kind of thing, like top one percent of the one percent. So we knew that we had to keep testing this and rationalize that we have to go outside of the Bay Area bubble, keep testing it with people who are not necessarily early adapter. So those are some of the things we knew. We knew it will be built to order direct to consumers. At the same time, all the other all the tactics or marketing campaigns that worked at Nest never worked at Matic.

The reason is because world.

Surprisingly different in 8 years. It's like, it's mind-boggling to me how it has changed and learning there now. And I think we've found, figured out how to talk about the product, how to kind of get the content in the zeitgeist and understand consumer psychology and all that stuff. But it's an iteration. It was an iteration. I just realized that even marketing is the same thing. There is no perfect go-to-market strategy that day one you can prescribe and say, this is it. Just like as you're innovating on a hardware space, you're going to iterate to find a product-market fit. Even on the messaging and product side, you got to iterate to find that, that one thread that just works, and then you got to keep pulling it and double down, and it may change again in every 2 years.

With a product like this as well, then how do you think about pricing? Because it's obviously, it's hard because it's like a product that's clearly better than the other options out there. Like, how do you decide, okay, Based on what we want to be as a company, as our value, you can say values, but like as like what we see in terms of product market fit, like how do you think about pricing? How high you want to go with this? Like how is this going to fit into future products and pricing? Like how do you think about that piece of it?

Great question. I'll jump into it. I want to make one more comment on the previous question. I think Andrew Chan, who is a partner at Andres and Andres, wrote a really good blog post recently that all marketing channels are are crappy or shit, whatever he used the word. And it's because his point is that all them, there is nothing new that has changed over the last 6 months, maybe the LLMs and SEO of an LLM is a new one that you can potentially figure out and do it. But nothing has drastically changed. So none of the channels traditionally that worked very well are sort of great arbitrage. But then, you know, what's not ephemeral, or what's not reliant on a new product, a new platform is just a great So that one part of the puzzle, we always knew that the great product do have an amazing word of mouth and sell themselves. So first thing that we focused on was a great product. So part of the go-to-market was that super great product, generate great, amazing customer reviews, like first 1,000 has to love the review, love the product, and then amazing external reviews, and then you kind of piggyback on top of it. So that was the thought process. Now coming back to how do you do the pricing? So pricing is something we got massively wrong. Okay. Because you don't know the— you may not know the history. I was hell-bent. If you talk to our company, people will tell you that I was hell-bent on saying that we're going to make this work as software as a service or robotics as a service. It just doesn't make sense to buy the robot because ultimately it is about software and any software-enabled hardware does require changes as well. And as a user, you know, I don't want to keep purchasing new hardware every 2 years. Maybe we can just replace it if it's just subscription-oriented stuff. So there are a lot of these grand ideas. Then we launched in 2023, initial version, like very small launch and just on our own social media. And subscription model was like an organ rejection. It's just like people just basically said, why am I paying subscription? And it was, and what I realized is that people weren't saying why Matic. People are not asking why I should buy this product or why this product is great. They're just saying why subscription? And that's a wrong question that I want customers to ask. I don't want customers to worry about business model. I want customers to worry about why this product is a great one for them. Them. So that was immediately a wrong answer. And we knew, then we were like, okay, but we were— because we were thinking of subscription, economics was different. So we're like, okay, initially, maybe the product would be highly priced, similar to Tesla Model S, and we'll go down the market. So we priced it at $1,800 initially. And then that also realized that, okay, that's just really, really wrong price, because people who can expect or who can afford $1,800 product is very, very tiny. In fact, Maybe we learned this and I was recently able to articulate this, but there's literally zero ubiquitous consumer electronics device priced higher than $2,000. Only cars and mobility falls outside of that. Yeah, that's the only thing you're not going to buy anything, even TVs, you wait till they're maybe around $1,000. Then I also read iRobot, which they get credit, but iRobot apparently priced first Roombas at $199 because $199 or less than $200 is the price point at which you don't have to ask, discuss that with your wife or ask permission of your wife to purchase it.

Interesting.

So $1,800 is a very, very high price point. So that's also a feedback we immediately learned and we launched early version or we started talking about Matic in November of 2023, showed some demos to Varge and Vard and a few other places. And that's when it was priced at $1,800. And a lot of customers just said, this is just not going to work and that we don't want to do subscriptions. So we learned and then we realized that actually this is a wrong way of penetrating consumer market. Even iPhone, which is consumer electronic products, which is much more closer to our analogy than Model S, started at $400 price point. And then as it built trust and value and insane usefulness, it sort of increased the price point. And now there you can buy one for $500 or $1,500. So in the same exact way, we realized that we want to have— we have to nail the use case of somewhere around $1,000 just because any hybrid, which is both mopping and vacuuming robot, is somewhere around $800, $900 anyways. So that's the price point that we wanna start, deliver 10x value at a similar price point. And then as we scale, we'll be able to, we'll have to figure out how to bring the price down, but also create a premium version for, let's say, people who are interested in bigger homes.

Yeah.

Like we have a customer who literally took Matic to Australia and he's cleaning the basketball court. Which is not the intended use case at all. But yes, so those sort of things that people have indoor hangars and indoor warehouses that they're trying to clean. So there is a different use case along those lines.

Have you thought about that part of it? Like the enterprise version, like the commercial version of Matic?

No, no, no, we haven't.

You said consumer.

No, we've always thought about consumer. That's a great question. We always thought about consumer, but it's interesting how people use it and that kind of sometimes Sometimes it leads you into a certain direction and pulls and say, okay, where do you do it? But people, there are people who say, okay, hey, maybe there is a slightly bigger version where there is a bigger water tank or maybe bigger battery life or even more, even quieter motors because I just, even though you guys are the quietest, I still hear it and I'm just too sensitive. So there are elements with which we can push the needle a little bit or maybe fancier colorful ID that they want to have. that belongs in their home a little bit better than white and black. So there are elements that we'd have to do which kind of allows us to say serve different markets for different purposes or different segments for different purposes.

You know what I find comforting that for some odd reason you've gone through YC, you've had another company before, you've worked for like 24 Dial, all that stuff, and pricing was still hard. You still got it wrong initially. Like it's still challenging regardless.

So, I think one of my favorite books was The Everything Store. Okay. And the reason, and book is great and everything is great, but the part, first part is really, really good where the author, I think Bill Stone, he describes this point of view that I went to Jeff Bezos and asked him, you know, I wanna tell the story of Amazon. Can you, will you help me? And his question was, well, how will you account for narrative fallacy? Yeah. And, and, and I didn't know what narrative fallacy was, and many people don't, but it's this Nassim Nicholas Taleb idea that in hindsight, everything is obvious and chronological, and you can connect dots and you can tell stories. And Bezos's point and Jeff's point was that, hey, everything you will write about Amazon is really just narrative fallacy because in the moment, there is a lot of chaos and a lot of stuff is going on and how do you solve the problem? So, I tend to, and I have been to business school, but I tend to think that a lot of business school and theory on what happened in past has a narrative fallacy, unfortunately, to it. And you have to go back to the fundamentals and foundation and then context change, world changes. So things change along with it. And you can't really say that this is the model, this is how we're going to get it. So, so in fact, you're better off just realizing that you're not going to get it right and you're going to iterate than trying to get it right the first It's like no matter what you try, you're not going to innovate the first time. You will require iteration. So pricing, marketing, all that stuff, just realize that you're going out there, you're learning, you're getting the feedback and iterating.

Yeah, it's like what is the market going to say at that moment you launch? And it also can evolve over time too because competitors and stuff too.

And I haven't talked to, obviously I don't know Airbnb founders and I haven't talked to them, but when Brian Chesky says that Airbnb launched 13 times, I think what he's saying is that they traded. They traded to each launch, each messaging, each understanding, and they kept doing it again and again until They found the message that just resonated.

One of the last things I'm curious about, because we talked about this a little bit, but like the future and where this all gets to and everything with Matic, and this is like the first product that you have, and it's obviously made a splash so far. How are you even thinking about other products, product line? Like when do you launch second product, third product? Obviously, like in tandem working on things, but there's always limited resources. So you always have like this, like you can't do everything right away. So like, how are you at least like thinking about that?

Yeah, that's a great question. We absolutely do. I think the analogy is a war analogy that if you open too many battlefronts, you're not going to give enough resources to any of them and you're going to think so you want to win the war. But the way you win the war is making sure that whichever battlefront you won, you've secured it first. So there is an element of doing it. And then over the next 2 years, we have to make sure that we secure this battlefront scale and do right by the customers. And that is still going to be the priority. At the same time, what do we sort of have always believed that we start with the problem and work backwards. So what problem and use cases customers are constantly asking about, which one are coming up. So a couple of, I'll give you an example. And, you know, we don't know if this is what we're going to do, but we knew that customers tell us, and one of them is we've, I've been to about 250 or so homes, odd homes now, and talked to parents. And a lot of parents are just telling me consistently that, can you just give me toy cleaning robots? I don't really care about anything else. Just for love of God, like, you know, just give me toy cleaning robot because if I can have my kids' toys being placed properly, I'm already happy. So that's an interesting sort of pain point that is coming across. The other one, which is very counterintuitive, which I would have never thought about, is many people also came up, told us that, hey, and it makes sense because generation that I'm in and our generation below, kids tend to move much farther away from their parents than in past. And that means your parents are growing old on their own. So we had certain customers who were saying that, hey, my parents are in Florida. I don't really know how to keep an eye on them. I don't even trust the cleaners they come. They're obviously getting old and they're not tech savvy where I can give them a robot and they'll do it. So I actually want your robot and I want to control it remotely for them. But you do have a camera. So can you just take me— take a 10-second time-lapse every evening and send me and tell me that they're okay? okay.

Mm-hmm.

And you just keep an eye on them. And it's a use case that I had never thought about where initially, if you say elderly care, you'd think from the user perspective. So parents' perspective, not necessarily kids' perspective, but turns out kids are also worried about their parents and feeling sort of this guilt that I'm not there to take care of them. So is there a resolution to do it? And what is that problem and how does that look like? So those are all very interesting things that we sort of noodling in our head and then kind of going from there. And we all, one thing we do know is we're absolutely going to start with the problem and work backwards. We don't start with a use case. Internally, we talk about this quite a bit and I'll make one point and then let you ask it. But internally, we always say that no one actually wants robots. They want solutions to their problems. And I might be the only one who says that, says this, but I genuinely believe that R2-D2 is a stupid idea. Don't build R2-D2 robot because we already have R2-D2. That's our phone. We carry it with you. It answers all the questions, and we're going to get more and more of those devices, and you won't need someone following around to do that. Unfortunately, Luke Skywalker did not have an iPhone, so he needed R2-D2.

You know, because you've read— you mentioned Jeff Bezos a couple times here, and I actually, I wrote a deep dive on Jeff and read— basically, I read 3 of his books over the course of like 2 weeks and wrote a deep dive because I was very curious about all things Bezos. And one thing that obviously is the title of a book is Invent and Wander, and they are working on things constantly. They got the first product obviously right, and they— but they have branched off and done so many other things.

things.

How do you then look at like capturing the future while also simultaneously building Matic and focusing on Matic because you have so much to do with this now? But like, you know, you're kind of investing in some of the other things. Like, I know I kind of asked you before, but I'm just curious with like, stuff you've read about Jeff Bezos and other maybe influences. How do you kind of still do that?

So, the reason we read— so this was all in 2017, before starting Matic, that we actually went on a discovery process. And part of the reason was that, wait, because sometimes we want Like you know, did is Amazon Amazon sort of just because all the dots connected or was there a vision? And and the answer is that there was a vision. He wanted to build an everything store, but it just started as a bookstore. You know, Tesla was always going to be an energy company or or green energy place, and it just happened to build Roadster as a first product, right? So so there you know there is always this longer vision, like you know, and that applies to even. J.K. Rowling and Harry Potter, where she actually wrote a first chapter and the last chapter, and then kind of wrote that 7 books over 15 years, which is a fascinating journey to just write that and extremely complex, right? Like, it's mind-boggling to me. So we knew on day one, and the reason we did this at Flutter, we actually did not have a really clear-cut goal or vision on where we're gonna end up. We just felt like we can do gestures better than Kinect, and that was it. And we kind of started, but we didn't know where we're gonna be 5 years, 10 years, 15 years from now. Now, so this was a journey where we said, okay, as we read, we realized that no, there was a plan. And then along the way, you have to look at the world and you have to kind of figure out where you want to go and when is the right time and take some shots. And it may not work out, but some things will work out. So there is a, there is sort of this having a long-term vision and then there is a bit of opportunism that you have to display as well. So AWS to me is a very opportunistic expansion of what Amazon became. So there might be an opportunity for that as well. And you just have to keep your eyes open to customers and the world and not have a narrow, like a narrow, very narrow aperture. So it's tough because you gotta have a very wide aperture, yet you gotta focus and keep executing. So it's a judgment call. It's tough.

It's going to be fascinating to see where this goes over the next several years and beyond. And obviously you said you're in it for the long haul, so I'm rooting for you. Thank you so much for taking the time to come on the show today.

Thank you very much. Fun conversation. And if you're ever in the Asia, come over. We'll do a lot more, a lot more.