My Conversation with Mehul Nariyawala, co-founder of Matic
In this interview Mehul Nariyawala, co‑founder of Matic, discusses the philosophy and practical challenges behind building a vision‑based robot vacuum. He emphasizes a product‑first, problem‑backwards approach, iteratively stripping away features to achieve simplicity, and the importance of deeply understanding customer needs. Mehul describes the painful transition from prototype to mass production, detailing supply‑chain hiccups such as camera delays and a motor‑noise issue caused by a supplier’s glue change, and how the team diagnosed and resolved them. He outlines Matic’s end‑to‑end customer experience strategy, treating support as part of the product and avoiding scripted responses. Reliability testing, calibration, and longevity checks are presented as essential to protect early evangelists. The conversation moves to the constraints of hardware scaling, the forced deliberate growth that acts as a moat, and contrasts hardware’s long compounding timeline with software’s rapid cycles. Mehul reflects on market dynamics, noting that unsexy, regulation‑heavy categories (e.g., thermostats, smoke alarms) deter competition, while easy‑entry spaces (security cameras) flood with rivals. He explains why robot vacuums present a rare, high‑revenue, low‑competition opportunity, and shares personal challenges, especially cultivating patience over a multi‑year venture.
Chapters
like, why do people love their Model Y or Model 3 or Model S in an absolutely fanatical way? What I've learned over my career is that there is no such thing as a perfect product, but there is a simple product and a complex product. If you look at Facebook, why do I download Facebook today? Is it to connect with my friends? Is it for newsfeed? Is it for Reels? Is it for stories? Is it for group chats? It's for— what is it for? Purpose of the product has to be clear. When you look at your t-shirt, you know it's t-shirt. When you look at watch, you know it's watch. When you look at camera, you know it's camera. So we wanted to make a robot that looked like a robot, and that's how we Today I have the pleasure of sitting down with Mehul Nariyawala.
Correct. Yes. That's great. And he is a co-founder of Matic. Matic is a company that makes these little autonomous robot vacuum cleaners. And the way that this interview happened is I just kept on seeing people over the past probably 6 months now just posting about loving this product. And I know that any time that I see that, I know that there's a huge amount of thought and care and effort put behind the scenes and intention put behind the scenes to make something like that a reality. So let's just start off with like, why are you creating this and how do you think about it?
Oh, I think years of trying. There is a quote by Elon Musk, which I really like. He says, every innovation is number of iterations and you just gotta reduce the time it takes between iterations. We just, I, from longest time I remember, I was fascinated with technology company because of just beautiful products. Many people forget how great Apple was in between 2000 to 2010. And Apple is still very, very good, but it's not Steve Jobs great. Steve Jobs great was something else. And I remember trying iPod for the first time and feeling the magic. And then I remember buying my first Macintosh computer in 2005 and immediately falling in love with it. love with it. And it was just this detail and this touch and this craftsmanship that got there. And at some point we're like, you know, why do— which is one of the things that I kind of asked ourselves is why do artists get to have a pride in their products? Right? You hear these stories that Leonardo da Vinci was never happy with his paintings or this artist, or I think I'm forgetting his name, but some of these painters, popular painters were never—
Picasso.
Picasso, right? They were never happy with their work.
Perpetually dissatisfied.
Exactly. So then why are we not creating our own artwork through this product? Ultimately, there is a, like what Steve Jobs said, is make something wonderful. And that wonderful itself is storytelling. That wonderful itself is making their impact. So some people make an impact through their art or their song, their books, in your case, amazing interviews. We wanted to make an impact through building something that people would absolutely love? And how do you make that? And it's about that feeling. And feeling is very different because there is probably a t-shirt or shoes or a shirt that you absolutely love, and you can't explain why. And it's just a shirt. It's just a t-shirt.
You just have an emotional bond to it.
Exactly. So how can you do that with even a physical product? Because it is possible. We humans sort of yearn for that bonding and connection. So can we do that? And that was the goal, that how do you build a product that is iconic? And that was the day one day we started. That was the first thing we wrote down. on a piece of paper. And there were a few of these things, but one of the things was there are products that touch users, and then there are products that users touch. We wanna build the former, not the latter.
Yeah, and normally you'd, like the Paul Graham school of thought, is you want to basically start and ship fast, and like create the first version, and then immediately get it in the hands of users.
Yeah.
You took the exact opposite approach. You basically spent 6 years iterating in darkness, in effect, before shipping your first.
Correct.
So how'd you decide to do that?
There are 2 types of products you can build, I feel like, in the world. One product is the kind of product that comes out and you say, wait, why do I need this? What is this? The no-prick, it doesn't exist yet in anyone's mind. So if you kind of think about first cars or first cell phones or even first phones, all kinds of iterations are fine and you just wanted to see if the technology is working. And that are minimum viable products. In the case where you go in an existing market where there is customer has this preconceived notion, so what is MVP for a new electric car. It's not just four wheels and a, and a seat. It needs to work. It needs to work. It needs to have your ABS. It needs to have windshield. It needs to have a starter. It cannot you know come out with a crank that we used to have in the early cars. So in the same way, we realized that we are going in in a product category where people have expectation of certain things for robot vacuums or robots. So we need to not only meet that but exceed that. So in our case, we need what we refer to as minimum lovable product. So what is that and how do you build that? And that one was just iterate. trading. And the reason we trade is because we felt like the entire space of indoor robotics, not just robot vacuums, were built upside down where all these disc robots were essentially automation without intelligence.
Hmm.
They just sort of bumped around. So the way we thought about it is that first Roombas in 2002, they came out and they were phenomenal for 2002 because there was no computer vision back then. There was no technology available. But really what they were, were, what they were is you put a blindfold around your eyes and you kind of, you know, Start walking. Start walking, and you keep bumping into each other, kind of like a pong game. And if you bump a wall enough time and keep crossing, you'll cover the entire room. And it was a great idea for that time. Then came out sort of like this single-pixel LiDAR robot vacuums that are there popular today. But single-pixel LiDAR is literally, you still have your blindfold on, but you have this one hand that is extended out. So if your hand hits a wall, you know there is an obstacle. But if the wall is sort of higher or lower, you're still gonna bump into it if your hand doesn't catch it. So we just felt like everything was just adding sensors and this blindfold, but no one was really removing that blindfold to build true understanding of the 3D world or context. So there was no— robots were basically— robots were— the way we thought about it is all these robots have 20 eyes but no brain. And can we actually give it a visual cortex of the robot? So that's how we started thinking about it.
Before we started, you were talking about Tony Fadell. And when you were working at Nest, there was people telling him, why don't you put the notifications and stuff in the Nest thermostat if people are going to have this device in their home. Why don't you just try to present information and all these things? And he said, it's just a fucking thermostat. And it's just supposed to do this one thing and do it really well and then kind of disappear. How did you kind of take that same philosophy to this?
Internally, we talk about again and again that no one actually wants robots. This idea that you want to buy a robot is, in my mind, a misnomer. It's false. People want solutions to their problems. And if you happen to solve that problem using robots because robots are the right way of doing it, But no one really wants AI. They want solution to their problem, which might be writing a great email or creating a great video or creating a great image. And AI is just a means to an end. In the same way, robots is a means to an end. To me, one of the worst ideas in robotics is to try to build R2-D2. R2-D2, no one really wants R2-D2. You actually already have R2-D2. It's called your iPhone. Luke Skywalker did not have one, so they had to have a dumb robot following him around. But now you do, so you don't actually need that robot to follow you around. So what problem are you solving? And that's the way it starts. And actually, I remember listening to this podcast, I don't know, maybe in 2016, 2017, I can't remember, but it was Kevin Systrom's interview from Instagram. And someone, whoever the interviewer was, asked him this question that what is the one piece of advice that you give that no one ever listens to or most entrepreneurs reject? And his answer was solve a problem. And this is because, not in a wrong way, not that entrepreneurs and engineers are not thinking through solving the problem. It's just that ultimately at the core, we are all in this field because we are nerds. We get excited about technology and we want to solve cool problems.
Might get excited about an elegant solution.
Exactly. It's like, you know, it's really, really cool to work on Vision Pro, but reality is that no one knows why they need a Vision Pro at the moment. It doesn't solve any problem yet. It may. And there is a world to be figuring out. And we did So we weren't— you asked us how'd you get here? Well, first company we built was this company called Flutter. And you mentioned Paul Graham. Flutter was gesture detection via webcam. So think of it as a Microsoft Kinect, but just using RGB cameras. And we had this insight that instead of turning your hand or finger into a mouse, if you're sitting in front of a TV and you wanna mute it, instead of finding that mouse button and air-clicking it, why can't I just shush it? 'Cause we already have a body language. So let's actually teach computers to use this body language. So we applied to YC, we got in, And the entire time we were part of Y Combinator, Paul Graham would always come to us and say, hey, I know you guys. You guys are technology looking for a problem to solve. Have you found one yet? And he would repeat that again and again and again. And we would just say, wait, wait, but we have this happening. He's like, yeah, that's not our product. And he kept teaching us that. And it took us a while, but it took us 2 and a half years to kind of sort of grudgingly agree that what we were building was really, really cool technology and really cool UX, but it's not a product because no one wakes up in the morning and says, today I'm gonna buy gestures, today I need gestures, right? Versus people do say that today I wanna buy a robot vacuum cleaner that keeps my floor clean, or today I need a robot vacuum cleaner. So, but it's because it solves a real problem.
You already had one successful exit prior to this, and that gave you some cash, and you're trying to build a like enduring business that you're never gonna sell.
Correct.
And that's, I think, a different way of thinking than, than most founders think today. I think most of the best think about the company that they're just going to die building. But that's— it's a different— especially if you don't have any cash, it's kind of difficult to think that long term.
Absolutely.
What did having that like initial starting amount of money enable you to do from a like just start and iterate perspective?
So the analogy I always talk about from startup perspective when founders come to me is that you can listen to all the interviews, you can read all the Paul Graham essays, you can watch all the YouTube videos, But starting a company is akin to trying to learn how to swim or how to ski. It doesn't matter how many videos you watch. It doesn't matter how much you learn theoretically. You got to jump in that pool and try to swim, and that's going to be hard, and you got to struggle through it, and you got to iterate. In the same exact way, you can't get on the ski slope after watching a bunch of videos and just know how to ski. You're going to fall. You have to have that muscle memory. So in the same exact way, most— like, there are obviously amazing founders who have gotten it right the first time. But I think the way Elon Musk did it is the right way, which is he also started multiple companies become— before doing it. So it is an iterative game. And for us, more than money, the learnings. So prior— so both Navneet and I, my co-founder and me, met at this startup called Light.com, which was the first computer vision startup in Silicon Valley back in 2005. And we— every single computer vision idea in world that is commercialized today, we tried it between 2005 and 2010, and we were just way ahead of the curve. But there was so much learning that enabled Flutter, and There was so much learning that allowed us to ultimately decide that we should go to Nest and understand how to learn how to build products and how to do all of that.
Did you basically stop your own learnings and decide to stop that company because you wanted to go learn from someone that was great at doing it?
It wasn't exactly that, hey, we're going to stop this because we want to learn. It was more along this realization that, hey, Flutter is definitely technology, not a product. It needs to be part of a platform. Luckily for us, both Google and Apple were interested in acquiring us. And we're like, look, if we don't sell it to Google, Apple, or Samsung, what platform is left? So we ended up getting acquired by Google. And once we got there, we took a step back and said, okay, what are the things we didn't know? So one was that, hey, we have yet to work in our career on a product that people actually pay for. So I made this decision from day one that whatever product I work on next, I want people to pay for it. And if they're not paying for it, I'm not working on it. The second thing we wanted to also understand is that we were dealing with gesture detection via webcam and we built algorithms, but we didn't have control over the web cameras. So in your PC or laptop, you don't have control over frame rates per second. You don't have control over autofocus. So it's like building an algorithm or brain for eye that is wavered. So we only knew half solution and we felt like there was a limitation. So when we kind of got to the other side, we're like, look, next time we want to solve a problem and we don't want to arbitrarily limit ourselves to say we're just going to solve problem using software or hardware. We wanna be able to learn it and we've never done hardware. So let's actually transition over to Nest because one, Nest has paying products and two, we can actually learn how to build that product and hardware itself.
What was it like at YC back then?
Very different. Oh, well, I haven't attended it today, so I wouldn't know, but it was very much sort of a very scrappy, authentic environment with PG. And PG's amazing. He's a really, really good dude.
What was the most memorable like experience or story that you have from him?
So we did not actually get into YC with Flutter. We got rejected at the application level. And I have this rule of thumb that I don't take answer no as an answer for the first time. I'll at least try again. So I guessed a bunch of PG's email address possibilities and sent him an email and I got one of them right. So he responded with the same thing that I told you earlier, which is, hey, you guys are technology looking for a problem to solve. And I was like, yeah, but you know, Oh, that is a technology that even John Collison loves. And John Collison and Stripe back then were still 5 or 10 people company on Ramona Street. But the only reason I knew them is because few days before our application or week before the application deadline, they had a Meet a YC Startup Day and one of them was Stripe. And we ran into John and we showed him a demo and he loved it. And I just connected these dots that if it was part of Meetup YC Startup Day, then maybe Paul knows him. So I just threw a name and turns out he knew John very well and he called John and we got in eventually. So we got an interview and then we got in and we meet him 2 or 3 weeks after we got in for our first office hours. And we told him this story on how we got in and he's like, really? I gave you guys an interview and you guys got in?
Yeah.
Oh, please don't tell this to anyone else because I don't want to be inundated by emails. And then he said, the second thing he said is that, hmm, I wonder how many babies we throw out with the bathwater. So it was just this idea. And I remember thinking like, wait a minute, he has no recollection. And it just surprised me. But I get it now.
I remember with this company I did, I know that YC like tracks all of the companies that they've rejected and then like went on to succeed. And so I applied with Relentless just so that I would get rejected so it'd be in their dataset.
Nice. Nice.
That's awesome. Anyway, didn't get the top 10% even. Anyway, yeah. So when you were working with Tony Fadell, what was the environment like at the Nest? You know, at Nest?
I think it was really energetic. There was an excitement about everything that we were doing. It was a very interesting time because it was also transitioning from Google to Alphabet. Yeah.
So I remember, I think I read Build, which I think is his book.
Yeah.
He was talking like there was entire like chapters dedicated to this whole very strange cultural spin this out or are they going to reorganize it?
Yeah. So there is this stat I think I read in— like, I have a lot of people in Silicon Valley don't respect this, but I do have an MBA and I went to Booth, University of Chicago Booth, and it does help in its own way. Not needed, but it definitely helps. And I remember reading in one of the case studies, or maybe it was a Harvard Business Review article, that 97% to 98% of the acquisitions fail. And they don't fail because the products weren't aligned or the strategic thinking wasn't correct. The company failed. They fail because of cultural mismatches, that people have their own ways of doing things and it's like oil and water. And Nest, it was mini Apple. It was built entirely in a very Apple-ish way where you have a strong product sense, strong secretive culture, strong ownership, and you drive and you build these things relentlessly. On a flip side, Google is very much Democratic culture instead of sort of like this dictatorship top-down culture, and that was oil and water. And you could see that. And what I remember telling Navneet the most is based on how people behaved, I could tell whether they joined Nest pre-acquisition or post-acquisition. Because if they were pre-acquisition, they were just absolutely, absolutely devoted to Nest. There was just the sense of pride and ownership and thinking that it was just phenomenal. terminal. And if they were post-acquisition, they were sort of questioning whether they want to buy into this Tony Fadell philosophy.
They were really coming to Google.
They were coming to Google. They had a Google— they had bought into the Google culture. So for them, it was like, okay, do I let that go and buy into Nest culture, or do I not? And it's not for everyone. It's an acquired taste.
Is that part of the reason why you kind of, for this company, you set out with a goal of not ever having it acquired, just building a standalone thing that is kind of like your So it's the best thing that happened to Nest obviously is a crazy acquisition and all the team that they won, like $3.2 billion is nothing to scoff at.
But I also feel like the worst thing that happened to Nest is also that it got acquired by Google. And for a lot of Nesters, including me, it sort of is a little bit painful to see what has happened to that ecosystem and the products and how they're not getting updated. So it's unfortunate. And I understand Google's reasoning behind on why they do what they do.
Yeah.
But at the same time, we've poured our heart into building this product. So being able to see them in the world, it's amazing. And it's kind of crazy. I think one of the products that Google is no longer going to continue is Nest Protect, the smoke alarm. And there are literally threads on Reddit where people are saying, oh, this Best Buy has 10 left. Let me go buy them before they run out. So when people are scrambling that for a product that we built 10 years ago, even now, then kind of tells you that how much they love it or how much thought was put into it. And it's kind of sad that it's not going to continue.
It kind of reminds me of Elon, I think before Tesla started working on the, well, actually, I think during the work on the Roadster, at one point there was this whole EV mandate a long time ago about like, you need to make more sustainable cars. And I believe that whenever that got repealed or something in the like early 2000s, um, I forgot which company it was, but they basically recalled all the cars and then smashed them. And there was like a candlelit vigil held with all the, all the people that bought these cars.
Huh. That, that I'm not familiar with. That's interesting. That was a big reason.
That was a big reason why he, like, he used that story a bunch of times when he was talking about the electric vehicles.
Okay, I'll have to go into that. That must be prior to Tesla, but I'm not familiar with them.
Yeah, it was almost 6 years before you guys actually shipped your first product. Can you just walk through like what it was like in the early days and especially kind of figuring out what are the right areas to attack and things like priorities to focus So we started very early on robotics and we started because of, as I said, 2017, as we were talking about earlier, in 2017 there were 200+ self-driving car startups in the world.
And we looked around and we had just been thinking about homes because we had been at Nest. So I had been looking at all the problems inside home. And one of my problem was that I had a golden retriever dog and our joke was that he sheds twice a year, 6 months each. So robot vacuum was a big need and I tried all of them and they were all underwhelming. And I knew that they were underwhelming because algorithm and computer vision was not great in them. And that's where our background was. So we sort of just looked around and we're like, wait a minute, there are 200+ teams that think they can solve self-driving problem, but then not a single home robotics company is out there that's trying to solve, build a robot that doesn't need to bump.
And this is like a real, you know, people are buying this product, but it just kind of sucks.
It just kind of sucks. And at that time, time market was already growing. There were total amount of robot— I think robot vacuum market was already $4.5 billion category. It was growing super fast. And when we looked into it, the entire category's net promoter score was literally -1, which means people were disrecommending this product. Yet every year new robot vacuum would come and every year people would try it again because it was just an intense need and we just couldn't understand why you couldn't build a better one. So that's how we started thinking about it. And ultimately we realized that, wait a if we can have level 5 cars for driving, why can't we have level 5 robots for homes? Well, if we have level 5 robots for homes, what does it even mean? So if level 5 cars means that cars drive like humans, then level 5 robots must be that they behave like humans, they navigate like humans, they clean like humans, they manipulate objects like humans, they understand contextual context inside home and changes inside homes like humans. Well, if that's the case, shouldn't it have a personality? like humans, and that perception system is vision-based. So can we give it eyes and brain instead of just all these random sensors? And each robot is a Christmas tree of sensors. So how do we actually simplify it instead? So that's how we started thinking about it. And we just felt like floor cleaning category was great because people were buying this product. They weren't all that great. And we just felt like if we can build a great product, that's the way to earn trust, and that's the way to build license to go to second, third product to keep solving Apple of home robotics. But at the same time, we didn't think it would take 6 years. I'm sure you've heard this point of view that if I knew how long it was going to take, I would've never done it. So the joke inside Matic is always that we were only off by one digit because we thought we would ship in fall of 2020 and we shipped in fall of 2024. So we're like, oh, we're only off by one digit. But reality is that there were so many unknown unknowns that we didn't No, but at the same time, it's kind of very interesting to think about it now because in 2017, it was clear that AI is coming. It was also clear that AI computes on the devices is going to skyrocket because my co-founder and CEO Navneet helped spec out Google Coral TPU from Nest perspective. So we knew that AI chips are coming as well. And if you combine this technology, robotics will get enabled. So we made this bet that, okay, we're gonna start on it now so that when the wave of robotics comes, we are ready to serve versus running into water at that point in time. So that's how we started thinking about it, but we were way ahead of the curve. So I think I walked you through our simulation rack and end-to-end testing rack, and all those tools that we have to build ourselves are actually getting built by NVIDIA and open source now. There are companies getting started just to build these tools. So if you're starting robotics companies today, you actually have a bunch of those tools available to accelerate fast. We were so ahead. We didn't have any of this stuff. So we had to build everything internally and do it. And there are advantages of it, but that also meant that we had to do all the work ourselves and grind it out. And that took us a time.
When we were just walking around the office, you said when you had a Roomba, I think it was a Roomba, it like, there was one day where you were out and about and it just like goes to some part of your carpet and it starts, you know, just sucking up the carpet and keeps on running. It just like burns this or, you know, sucks up all this carpet. What was that Kind of like when you first started working on the product itself, what was the like MVP? Because you actually, even though it took 6 years, you got this entire stack of like hundreds of these prototypes.
Yes. Yes. So we, this time we did differently because we had done with Flutter, we were just building this technology and we started and we didn't think through this. So when it came to Matic, we actually sat down and we built a problem deck before we built a solution deck.
The problem deck.
And what I mean by that was that what problems do customers have? at a very high level as a father and a family man and a homeowner. I want to live in a perpetually clean home with perpetually clean homes, clean floors, sorry. So perpetually clean home and perpetually clean floors, that's requirement number 1. I don't want to do it.
With the least cognitive load. You want a clean environment with the least cognitive load.
Yeah, so that's the 3rd one, which is I don't even want to think about it. So I want to live in a clean I don't want to do it. I don't want to think about it. Those are the 3 things. Is there a product that delivers it? The answer was no. But then at a very tactical level, why do robots constantly get stuck on wires or choke on wires? Why do they fall down the stairs? Why do they bump into my amazing furnitures? All these were tacticals. And one of the problems was that actually the robot was Dyson's 360i robot vacuum. So one of the first Dyson robot vacuums ever. Fantastic. And it got onto one of our nice rugs and it didn't know that this was a nice rug and kept on, and suction was really high for the torque. torque built into the wheel, so it got stuck and entire patch of the rug, it just sucked it up because rug shed. So then realized that all these robots are just, for a lack of a better way of saying it, are just dumb. And how do we actually build an intelligent robot? The one that just works, that you don't have to pre-clean for, that you can just trust and you don't have to babysit. And that didn't exist. And that's where the point of view came that, okay, that's how we thought about it. But then what does What does it even look like? So there were a lot of requirements. So we knew what we didn't want to have. So we knew that—
Do you want to just go through like what all those things were?
Yeah. So from day one, we're like, okay, it cannot be a flat or circle robot, partly because if circle or disc shape was the right shape, all the manual vacuums would have shipped that way. But disc shape by definition is very bad in terms of going to sides and corners. And if you flip the robot, actual vacuum is 2 inches away from the wheels. So the cleaning area is just very bad and very minor. So that's why the efficacy isn't great. But ultimately you buy a robot because you want to clean. So let's make sure that we're solving a problem. So we kind of talk about, as I mentioned, no one wants robots, they want solutions to the problem. So that was one. Then disc robots are something that my dog and my daughter were really, really afraid of.
Hmm.
And we're like, how do we make a robot that people are— kids and pets actually love it, that belong in a home? Then how do we make sure that it doesn't look like this ugly appliance because many homeowners go out of their way to hide appliances behind closet veneer. So you really don't want this giant dog or ugly duckling sitting around in a living room or family room as well. So how do we make sure that it's friendly, it's good looking, that it feels like it belongs to a home? How do we make sure the robot actually looks like a robot? How do we make sure that if kids run into it, they're never going to get hurt, that it can go over various terrains? And one of the thought experiments we had done, which I told you about, was Because we thought that let's imagine a world where we take Disc Robots and we have invented teleporting machine and we can send it back to '60s and ask people, what is it? There's a good chance they won't be able to guess that it's a robot or a vacuum. So we wanted to make sure that robot looked like a robot. And that was actually one of the lessons that Tony Fadell and Matt Rogers taught us at Nest, that purpose of the product has to be clear when you use it. So when you look at your t-shirt, you know it's t-shirt. When you look at watch, you know it's watch. When you look at camera, you know it's camera. But when you look at Disc Robot, you're like, wait a minute, is this a vacuum? Is this a speaker? What is it? So we wanted to make a robot that looked like a robot, that it was friendly and people got used to it. And that's how we started thinking about it. So a lot of iterations were initially figuring out what problems are that we wanna solve, can we solve them? And just iteration on, is the big wheels better? Are the 4 wheels better? 3 wheels better? And just continuously going through and solving each problem. And with each iteration we learned and we got better. And better.
You said that a lot of vacuums, like people associate this the the loudness of the machine with the amount of suction, and so you you know vacuum companies would just artificially inflate the like decibels to make them extremely loud, so they sound like they're suctioning. I remember when I was a kid, I you know vacuumed all the time, and or I was at least asked to, and it was extremely loud, but it didn't a lot of the time it didn't even vacuum.
It didn't even exactly. So so and and not only that they have switches right? It's like you go higher suction and go zoom, and you're like wow. Wow, it must be so great. But turns out actually noise, noise has nothing to do with suction power, has nothing to do with it. It actually is about airflow. But even going back to the history of vacuums, vacuums were invented as carpet sweepers.
Hmm.
Vacuums did not exist for the world that was hard surface. In '60s and '70s, we had this manual swivers with just brush roll that you would kind of drive and it will pick up things up. So in the same exact way, the analogy I use is that if you have If you have a car and it's dusty and it has this fine dust, let's say you went to some sort of a national park, on your way back, no matter how fast you drive, that dust doesn't go away. But if you go come home and just gently nudge it with your finger, it comes off. So it turns out on hard surfaces, if you really want to agitate dirt, you have to have a brush roll that spins it and scrapes it. It just needs agitation. And once it has an agitation, all you need suction for is to take it back to the bag or or the bin. And that does not require crazy amount of suction power. The amount of suction power that Dyson or some of these other vacuum companies have is great if you have sandy carpets, but most of us don't live near beach and we don't have sandy carpets. So just simple amount of brush roll is great. And what was amazing is that all these settings are preset.
Hmm.
But the world has changed. We have thick pile carpets, low pile carpets, shag rugs, you know, Iranian rugs, and some of them require different kinds of suction power. So we were like, look, just the way our brain thinks and our fingers sort of adjust how we should do the task, why can't we build a robot that can dynamically adjust based on a floor type what kind of suction power it should use, what kind of brush roll speed it should use? And because it has an actuating cleaning head, it can go up and down just to the right amount of a thing instead of squishing it.
Ideally, you've got your list of all the problems. But hopefully, not everyone has done things completely horribly wrong over the course of the history of vacuums.
Sure.
things were you able to learn? Like with Dyson, they designed, you know, handheld vacuums, a lot of them that are like delightful to use. Like they're really delightful to use, but it's not even necessarily, you know, that solves like the enjoyment of the actual taking the action and cleaning your home. But maybe you don't actually want to clean your home. Maybe you just want the home clean. And so what were you able to take that was like good learnings and adopt into your product?
I think there are some ones, but one that we really like, and I didn't know about it until I started doing research, is Miele. It's a European brand, but they're actually very well liked and very well loved vacuum, and those who use it swear by it, and it's because they're quiet and the airflow is great and they're very gentle on your floors and carpets. So that one is something we kind of looked at and said, okay, that's a good inspiration. But ultimately, and Dyson was great from the perspective of branding, perspective of design, there is a distinct identity to it, which is really, really cool. But at the same time, it wasn't something that people swore by the way iPhone did. So there was an element of saying that, okay, what are the products that people absolutely love? Like, why do people love their Model Y or Model 3 or Model S in an absolutely fanatical way? And how do we do that? And how do we create that bond with the product? So those were some of the things we had been thinking about along the way. But obviously Dyson suction is amazing. If you really want to clean a sandy rug, that's the best vacuum you're going to get for your money if the suction power is the priority. But if you're trying to build a robot A robot is mostly going to clean and it's going to take slightly longer when you're not in a mode of cleaning. And no one likes noisy guests in their home. So if you build a quieter robot, then people will use it all the time. And then you can get to this perpetual clean point of view, not once a day or once a night, which is can you continuously clean? And actually the inspiration of continuously clean came from this movie Passenger with Jennifer Lawrence and Chris Pratt. And they're on this spaceship.
The ship. Yeah, exactly.
He wakes up and then after 3 months he's kind of bored and then he's eating in a cafeteria and he drops cereal and this vacuum comes out and just immediately cleans it. And then he's like, hey, hey, stop. I want to talk to you. And it goes away. And what it does is it picks up a spoon off a— And then drops it on the floor. And then drops it again. It comes out. And I remember like, continuous clean. Like it should just remain pristine. So that's where the idea came. But once the idea came, actually we realized this was also true for India. So what I mean by that is, you know, the way to think about that is in India or any third world country, you can afford domestic help and we could afford it even as a middle-class family. And I remember that my clothes not only got washed every day, they got cleaned, ironed every day, okay? Homes get broomed and mopped twice a day. Dishes got cleaned after every single meal because there was a physical human being matching the Entropy as the homes were getting dirty, and clearly we can't do that everywhere in the world. So robots are perfect to do that. They don't get tired; they don't get bored. So they can just do the same thing over and over again. That and that allowed us to live there. But then we also realized on a flip side that every single appliance built in United States and first world are built as a batch processor. And what I mean by that is you have to collect your dirty dishes until dishwasher can be full, and then clean them. You do collect laundry till laundry. The washer and dryer.
And it's completely full, yeah.
Yeah, yeah. So, it's like, because you're trying to make sure that you're using it efficiently, you're collecting entropy. And that doesn't make sense. Like, if you wanna just clean one dish, you're shit out of luck. If you wanna just clean one piece of clothes, you're shit out of luck, and you gotta just do it manually. But because—
Same thing with the floor. You just collect entropy, a bunch of shit, like, you know.
Exactly.
Strung about.
Exactly, 'cause you don't have time. You only have time over the weekend, especially busy parents. They may not have time to continuously clean. And kids, you know, kids and pets are tornado of entropy generators. The rate of entropy just skyrockets the moment you throw kids and pets into the home. So that's where we realized that how do we build robots that are continuously cleaning on our behalf? Because when our homes are clean, we feel this peace, this zen, but that only lasts for a few minutes because homes are constantly getting dirty. Entropy is always there.
I want to talk about even though you didn't necessarily ship product, you did iterate a huge amount. And I assume that you had this operating in your own home or somewhere else. And from what I can tell, the very first version right next to you is made out of wood.
Yeah, that was just like a student project. I don't think any mechanical engineer would lay claim to it. That was just me and Navneet very early on tinkering. And then we had a couple of interns who were tinkering with us. And we just said that, okay, let's just build a robot. And what does it even take to build a robot? Because we had no prior experience in robotics. We were computer vision guys. We just sort of built it that way and see if we can make it even work. And then we were like, okay, we know how to do it. So the next 3 prototypes, the white ones and 2 black ones, those were some of the proof of concept prototype.
Mm-hmm.
And proof of concept was that, can we make it quieter? Can we build, at that time we're like, maybe we need cyclone vacuums. So we also built cyclone vacuums. Can we make sure that just with 2 RGB cameras we can map and navigate and teach robot how to behave intelligently? Can we build an interaction method on it? So even in that prototype, we built voice and gestures interactions there, We haven't shipped yet, but we kind of pointed out and proved to ourselves that yes, this all can be done. Now we just got to do it at a level where it scales and just works again and again. And then we went out and did a lot of research from users' perspective, and we realized that most people actually don't even want— so these robot vacuums are built with this metaphor that I want to clean my entire home. Actually, most people don't want to do that. Most people on a day-to-day basis don't clean their entire home. They just clean their living Living room, dining room, family room.
Whatever they see.
Exactly.
The attic is always completely messy. Kitchen.
Yeah. So, it's just things that they're constantly getting dirty versus your bedroom, which will get dirty, but once a month or maybe once, 2 months is enough, or once a week is enough depending on how often you want to do it. So, we realized that the living area, which is kitchen, dining room, family room, they were always getting dirty much faster. So, can we clean it from that perspective? So, that was another one. But then, a lot of homes So Roombas and vacuums were obviously invented in a world that was wall-to-wall carpet in late '90s. But over the last 20 years, everything has shifted to hard surfaces and thick pile rugs and wires and thresholds. So wheels needed to be bigger. People also wanted mopping. So we just sort of kind of understood what users wanted and then kept iterating. And initially we did 2 separate iterations. We said, okay, we're gonna actually build mechanical systems and that will be separate than actual vision and intelligence systems. So first robots, We just wanted to use a remote control and say, if you're gonna have an Xbox remote control paired with it, can we use that to control? Does it clean well? And it was this idea that when was the last time truly great mechanical engineer looked at cleaning systems from scratch? Well, that was probably Dyson 30 years ago, but the world has changed. So how do we build it? Same thing, which is when was the last time someone really redesigned the mobility system of the robot? 'Cause every single robot vacuum is exact same set of wheels and exact same set of bearings. So can we switch that so it go over thresholds and rugs very well. So there was an entire design sort of— we used to have this diagram which felt like 5 different streams. One stream was ID, another stream was perception, another stream was app, another stream was hardware, and within hardware, ME and EE. And the idea was that we each work individually and eventually the streams connect.
How did you initially interact with users?
Did you just go and ask people how they So first was just reading every single review and feedback I could find on robot vacuums and history of robot vacuums and history of vacuums, just understanding how they evolved and why they evolved. So that was one. The second thing was at some point doing research that what is your need? What are you looking for? And realized that the target customer for us was families like us. And families like us were especially busy parents with kids. For them, the biggest pain point was deep cleaning. day-to-day cleaning. It wasn't even a deep cleaning because they almost always could afford or would hire someone to come and deep clean their home once a month or once every 2 weeks. So it was just this day-to-day cleaning where instead of spending time with kids, they're cleaning floors because kids sit all the time, or they're wiping tables. So that daily chores is what we really wanted to go after. And that's how we started thinking about it. And then along the way, it was just using the product in own homes. Then once we got to a point where we could showcase it, I believe I went out and did at least 50 different home demos in different homes. But it was both, it was dual alternative. One was to just show the product and have them to react. But 2, I wanted to see if it's working in different homes as well. So in the context of this demo, we got to understand that it is working in a bunch of different homes. And then we at some point did about 20, first initially 20 customers. We gave them the robot and they used it for a year and then we built another 17, Give it to an early set of customers. So just kept increasing the amount of customers who would have product and then seeing their reactions. And some people, like first 20 products we ever shipped were really, really bad.
And was this just like for free?
Just asking?
Just for free.
Yeah. And we actually divided. So first initially was free, then we're like, okay, let's get 5 customers who were actually paid and let's try them and see if they're keeping it. So just kind of slowly, slowly increasing the stakes, if you will, to test. But first products wouldn't work in low lights. First product wouldn't work in nightlight.
It would get stuck randomly.
They were just really bad. And then some people obviously stopped using it, but then there was a set of customers who just kept using it in spite of it being so bad. They just figured out when it works extremely well and they kept using it and they kept using it for a year. And in many cases, even refused to return the robot to us even when we had a newer version. They're like, no, no, no, I want to keep this janky version.
And I was just like, check it out. They figured out what it would do.
What it would do and when it works. And they're like, oh, I'll keep it in my different floor, I'll put it in my basement, but I don't want to give it back to you. And that was really the interesting part. We're like, whoa, like here's the product that is not good in our imagination and people are still loving it. So that was your early—
Kind of the brick.
Brick idea that it just, it just problem is so intense that people don't want to just clean it all. And even if it works 50% of times, they're pretty happy.
You know, you said the first 250 iterations were basically like completely 3D printed.
Correct.
Why did you kind of make that decision and how did that enable you to iterate faster?
And that was really, that's a great question, but it is that quote that, hey, we want to iterate faster and everything is in iteration. The reason we did it for 2 reasons, which is if you try to injection mold any of the parts or even soft mold any of the parts, it's expensive. It also, because we don't have really a good manufacturing industry left in the United States, a lot of times you have to go outside of the country or get those parts from China or different parts of Asia or different even states in United States. Maybe Detroit has something there, but all of that takes time. And that time that you have to place the order, not only it's expensive, but it takes longer to come here. So 3D, the idea initially was that can we just 3D print and iterate faster? And luckily for us, just around the time when we started, 3D printers started really getting really, really good. First Prusa and then now Bambus. So it was very easy to just build it from that perspective and say, okay, let's make sure that our design is functional. And once we got to a point where it was functional, that's when we said, okay, now is a good time to start thinking about injection molding. So it was both doing it cheaply and doing it fast, and how do you do both?
How much did you spend making those first 250 iterations prior to doing the first injection molded model?
So overall, we probably spent order of magnitude less money than companies, or maybe 1/6 of the money that companies typically spend in getting to this stage, especially hardware company. We're still about 70 people team entirely vertically integrated. Typically companies get to about 300 people by the time they're at this stage. So we know that we are 1/6 of the size. I wanna say maybe around like $15 million. Yeah, I don't have exact number, but that was the total burn.
Yeah.
As we iterated through those first years. But it's still relatively cheap considering considering how much robotic companies are raising these days.
Did you end up ever going to China? Because I know that you are like, most of your parts or like assemblies are coming from China now. Did you end up going to China and talking with a bunch of suppliers and factories and stuff?
Absolutely. So we have, you know, we actually have team in Hong Kong and Taiwan now. We did go initially. I think the first time we may have visited is probably 2023.
So 2023— So when you were actually spinning up the supply chain?
Yeah, once when we got to injection molding, sense because even if you try to do it from US, sometimes you just get vendors who themselves get parts from China or different parts of Asia. And initially, a lot of supply chain was China, but actually over the last year we've switched it out and now we go to Vietnam, Taiwan, Malaysia is another country from which we have CMs in. We are also trying to spin up Mexico.
So there are different parts that we go and try How much of the product did you actually have to make from scratch versus, you know, parts off the shelf just happened to be there?
Great question. I think as much as possible we try to use off-the-shelf parts, but the type of product we build, as you can see, hasn't been built before. There is an element of building custom parts, but things like motors, things like batteries, things like cameras, these are all off the shelf. We didn't want to use any of the things or even, you know, PCBAs or I guess NVIDIA's GPU. So anything that we can buy easily from outside, we would do it. We didn't want to create custom sensors or custom parts in that scenario. The plastics and the shape of the robot, those are custom. So that's where the design comes in.
I imagine if you're going through hundreds and hundreds of iterations over time, you kind of have these probably serendipitous unlock moments where you notice something's fundamentally shifted.
Yeah.
What were those?
The moments I remember the most is first time it cleaned. So just using even a—
How many times did it not clean?
It's like being able to just move this up and down, which is you have this vision and you put it together and it cleans, but it cleans in a lab environment. If I can take it home and just even with the remote control, if I can drive it and cleans and room felt clean. And we always kind of joke that the first customers are obviously of our better half, better house and our wives. So if you can't make them happy and if they're not happy with the cleaning performance, we have a terrible product because they're biased and they're willing to love our product. So that was the first thing when we would use it. Wifetest. Yes. Your mom test, which is, is she going to love it? Is she going to be happy with it? So that was the first one. Then the second time I remember, like, the other unlock moments were when all of a sudden it would navigate under very sort of complex environment. In their case, just a dining table with chairs inside it and just navigating around it, weaving around it without bumping into it, which is really cool. Like anytime robot goes underneath the table and finds its way, it's really, really cool to me. So that was amazing when it made some of those very, very human-like turns or car-like turns into it. So that was really cool. The third, and then the one that I remember the most was, I remember first time I got a bag full. It was really, really exciting thing that not only I've used this robot, but it cleaned enough that our bag is full. And I brought it and showed it to entire team and it's like, okay, this is the first bag. And I took the picture and it's this proof that robot works.
Well, how long did it take to get the first bag full?
That was probably 2023, January.
So 6 years.
6 years, yeah. 6 years to get one bag full.
Do you guys track how many bags, like I imagine, you know, every single bag you guys are personally shipping. So you track like over time, how many used vacuums are there?
Over time. So now on average, people use about 2 bags a month, but we've been like, I think we're going to end up shipping somewhere around 17,000 or 18,000 bags over the next few weeks. So yeah, a lot of dirt getting collected and thrown away.
We clean about half a million square Do you have any context on how big a house is typically? Like how many square feet an average house is? Is it like 1,000 square feet?
Around 2,000.
2,000?
Around 2,000.
So it's like roughly 250 houses a day getting cleaned?
250 houses kind of space to getting cleaned. But once people put furniture and stuff, it's a very narrow space and people don't use it. But on average, people use the product 7 times a week. So almost once a day. They clean that entire space once a day. So that's great. And as I said, we do about half a million to a million square feet a day. And how do you get the product into the house? We've crossed $100 million. And altogether, I think robot has driven 75,000 miles inside homes.
I love this line from Brian Chesky where he's trying to design the 11-star experience. You know, like the 10-star experience is everything goes right and it's perfect. But then the 11-star experience is like, assume not only, you know, what is almost like impossible today, but hypothetically it would like delight customers and stuff. What is the 11-star experience that you're trying to drive towards?
11-star experience that we're trying to drive towards is as the robot comes into your home and says, hey, Ty, this floor cleaning thing, I got it. You never have to worry about it from this moment onwards. If it turns out that there is a stain that I can't clean, I'll tell you about it. If it turns out that there is a home, a room to, one of the room door is locked, I'll tell you when to open it. I'll tell you when we haven't cleaned underneath the table, which is big, and moved it out of the way for me to go clean in 2 weeks or 3 weeks. And ideally, when you're sitting in a home relaxing, it says, Hey, Ty, I really need your help. Can we do deep cleaning today? I think it's required. So this idea earlier you alluded, the cognitive load of keeping floor cleans is robots. It's not something that you ever have to think about. And that's really interesting because very early on we realized that we as humans are creative species, not a repetitive species. And if we can give people their time and energy back, then hopefully that would result in much higher productivity. And that's precisely what has happened in the history of civilization, that you see productivity to be pretty flat until 1900 arrives and it skyrockets with the Industrial Revolution. Not only that, but one of the stats that blew my mind a long time ago, as we were doing research, is that if you just dial back 100 years or maybe 105 years now and look in 1920, on average, people had traveled 30 miles from their homes. Now, you're from Alaska, I'm from India, we're in California doing this. Can't even fathom how that world is, but that changed because traveling through plane or trains or cars became much, much easier. So, technology shrunk the distance and gave our time back because otherwise it would've taken years for us to go from, even on land, from Alaska to California, correct?
You no longer had to get on a ship and experience the treacherous waters of the Atlantic.
Exactly. So that just blew my mind. So it's like, okay, if you can shrink time or basically give people their time back, then they will use it for more productive means. And that's the hope that how do we enable customers to be person of their dreams, whether that's better parents, better caretakers, better artists, better singers, better piano players, whatever they wish.
Over the course of the 6 years where you're just tinkering, of in darkness before things have shipped. Uh, what were the biggest moments of just like extreme pain?
There were a few. So, so first one was, uh, when the pandemic happened. Um, I remember in March of 2020, we had no idea what was going to happen. We shut down because we didn't know COVID was there, and we hadn't raised a lot of funding till then. We were still relatively bootstrapping and had some angel investors. Uh, so I remember thinking that, wait a minute, if we shut down for a year, what happens? And then we thought maybe we can make it work. So I know a couple of engineers who took 3D printers in their home and tried iterating. But then in June of 2020, we got back to office. Within 3 months, it was fully—
That's actually really fast.
That was really fast. So we really— by May, there was enough literature out there that said that you could safely work with the mask on. So in June of 2020, we came back. And part of the reason we came back is because we just weren't making enough progress. So the question was either you go back to the company and wear masks and work that way, and we did that for about a year and make progress, or you die. And so that was one really scary moment. The second one was— The second one was a different one, which is there are times where you don't anticipate certain challenges. So I talk about this a little bit, but one of the fundamental technology that allows robot to know where exactly it is inside home is this algorithm called Simultaneous Location and Mapping, SLAM, right? Now SLAM algorithm theoretically were solved in mid-'80s.
Mm-hmm.
And there have been tons of SLAM implementation since then all the way till 2000. So one of the problem, one of the assumptions we made when we started this company was that SLAM is a solved problem. There has to be some open source library that is doing really, really well. We'll use it. I think it took us entire 2020 and working on early part of 2021 to realize that none of the SLAM open-source libraries were even remotely good enough, that they were sort of 80% accurate at best. And then we decided to just write the whole thing from scratch ourselves and used bunch of neural network techniques as well as classical ideas. But we got it to a point where I believe our SLAM system at the moment is order of magnitude better than anything out there. And that was just pure grind of 3 years. And the way I describe that is a sort of iPhone with like touch interfaces, pre-iPhone and post-iPhone, which is touch interfaces did exist before iPhone, but you had to jab your finger with all the power you can muster, or you can smack those styluses in order for input to come. And then iPhone came and it's just beautiful and it's silky and it's just touchy.
It's almost like you forget.
You forget.
Like beforehand, if you're using a stylus and you're slamming your phone. I remember this when I had my first Kindle Fire.
Yes.
It wasn't super great.
Yes.
And it would miss a lot of the actions that I was trying to take, which like completely sucks you out of whatever you're trying to do.
Exactly, and it's very frustrating. And that's how robot vacuums are. There is other part of it too, which is something that not just most robot vacuums, but any indoor robot, if you go and look at any indoor robot, they can map and navigate, but those maps are like blueprint plans. They don't have contextual awareness. But the part that is actually worst is that if you'll see their demo, they almost always start from their dock and build a map. That's because dock is a reference point. So these are relative maps. To us, that was always a wrong way of doing it because that's like saying that I can only navigate my home if I enter through front door.
Like I came to Alaska and I can't navigate if I don't understand where my house is in Alaska.
Yeah, exactly. Just like I can go through front door, but if I came through side door or back door, I don't know. Or if I didn't go through Canada, I don't know how to navigate. There is no around way, right? And that doesn't Does that make sense? So we wanted to build an absolute map, which is just like human being. You're in this office, if you're walking around, if I blindfold you and take you to the manufacturing area where you've been only once, you'll immediately know where you are. So can we teach a robot to do that? And there is entire classic kidnapping a robot problem in computer vision. So we wanted to solve this in a very, very meaningful way. So that's where we pushed the needle. So that's the first thing. So that was really tough. That took us much longer. So that was one of the reasons why we had to keep iterating for so long because software turned out to be much harder than we anticipate. Precise precision.
Was the hardware easier than the software for this?
Hardware, I don't wanna say easier, but it was ahead, which is it was building hardware or building vacuuming system. It is hard, but it is deterministic. It has been done before versus what we were trying to do with this robot, just vision-only system for indoor world hadn't been done before. So that one is slightly— we didn't know when we would get there. It was indeterministic.
I think this is kind of like the Tesla situation where I remember Elon came out and said, we're not going to— we're just going to delete all the LiDAR from the cars. I mean, they had for years, they were building this hardware with the understanding that they would have LiDAR. They were even trying to create like invisible LiDAR. They would create like little spots on the car that the LiDAR could see through that you wouldn't even notice. It was, you know, there was no interruption in the paint. But for a long time, you know, he was kind of like shamed or people like to poke fun at him. And then it kind of worked. Like eventually people are driving around self-driving cars. What was that like for you? Did you have people that were kind of detracting and/or giving you negative feedback on that?
There was this, I don't think people were detracting, but there was this thought process that, hey, if you just add sensors, you can faster and you can get there faster. And if you just do that, it'll work. But it's not just about making product work, it's also about making business work.
Mm-hmm.
It's also about getting, making sure that customers would buy it. Correct? Now, first Roomba. So the misconception is that Roombas were the first robot vacuums. They weren't. There was a product from Electrolux. They actually created this disc robot and built a robot. And I think it came out in 2001. I'm forgetting the name of it. of the product, but it was priced at $1,400 and it failed because back then $1,400 was obviously a lot of money. So then Roomba and iRobot built it below $200 and they priced it at $199. Now, any guesses why they decided to price it below $200?
I'm not sure.
So it turns out that, and this is psychological barrier kind of thing, but turns out that below $200, you don't have to ask permission from your wife or significant other Has there been studies done on this? Yes. So they did, they actually, that was the thing that they actually, I remember talking to Dr. Rodney Brooks and he's, they did that study and they actually wanted to price it at $149, but then it got a little expensive though. So they ended up doing it at $199, but that was a threshold. Then I've also heard from someone at GoPro before that if you build a gadget, any kind of gadget, which is priced somewhere around $100 and $200, That's toy money. You'll have 50,000 people on the internet absolutely buy it. But as you go above that price point, things change a little bit. So we knew that you couldn't have a robot that is really expensive.
Mm-hmm.
There was a second element of it. As I said, we did a lot of research and understanding of it. And if you take a step back and think about it, there is literally zero ubiquitous consumer electronics device that's priced higher than $2,000. Beyond $2,000, you are in a consumer space where professional gamers might do it or professional graphic designer might purchase it.
But your market shrinks by over 10x.
Exactly. Only thing that consumers pay that is higher than $2,000 is cars. Cars after 100 years of utility, which is you don't have to— no one questions why I need a car or whether it's going to be useful or not. So after 100 years of proven utility, proven usefulness, coolness, it's still not an impulsive purchase. If you're trying to buy a car that is costing $10,000 or $15,000, it's still a considered purchase. You think twice, thrice, 4 times about it. So if you're building a robot and if it's not going to be priced cheaply, it's game over. And we observed that there were Kickstarter campaigns after Kickstarter campaigns where robots would get funded, but by the time they were supposed to ship, either they never shipped or they were prohibitively expensive, and that just killed the market. market completely. So we knew that we had to make this robot affordable and accessible on day one because there are a lot of these robot vacuums at $800 price point. And if we came out with one at $8,000, it wasn't going to work. And the thing with the sensors is, and this is something we learned from Tony and Matt at Nest, but we had this rule of thumb that a single sensor you add in a hardware, assume 3 software engineers on the flip side as permanent cost. So more sensors, bigger the team, more sensors, more calibration, more sensors, more complex the supply chain, higher the BOM cost, more complex the manufacturing, more failure points. So with each sensor, complexity actually rises exponentially. And so we came to conclusion very early on that either we're gonna make this work just using RGB camera and absorb complexity in software, or we won't be able to build a product that is commercially viable, economically viable. So we made this bet along with Tesla that vision-only robotics is the only way to make products viable. And so that's what we really went after.
I know for you, even if you are gonna be kind of in darkness for a very long time, you do really want this like fast iterative loop. How did you kind of keep pressing on the gas even if there's not a really obvious metric to look at and point at and say, this thing is going up into the right and we need to make make it go faster?
I think we did come up with certain metrics to go do it. So first one was like, let's just make a robot work mechanically and let it fit cleans, right? Or let's just have robot map my home. But I remember for the longest time, I just kept saying that, hey, let's just clean a rug. Can we even clean a rug? Let's say there is no obstacles in it. Is it going to work autonomously? Then can I just get a robot that cleans my room with all the obstacles in it without getting stuck. So it's just like, how do you take this, which is constrain the problem in a very minimal way and then keep solving it? So initially we're like, we're going to build map manually. And if I have manual map built, will it clean? If the answer is yes, now actually build map autonomously. Then we were like, okay, it's going to clean carpets very well. Can we clean carpets and hard surface oriented room very well? So it was just again and again. So even which is constraining the problem. So when we started shipping in November of 2024, our robot actually would not clean edges of the room. It would only clean interior. It wouldn't clean underneath the kitchen cabinets, which is referred to as toe kicks. All those things we shipped after we shipped the robot. So that was just software updates, but it's like arbitrarily constrained, not arbitrarily, but functionally constraining the product and saying, can we just do this? Let's get there. And then we add another layer, another layer. So it's this idea that you're trying to climb the quote unquote Mount Everest, if that's the way you think about zero-to-one product, what is base camp 1? What is base camp 2? What is base camp 3? Can you define those milestones and can you hit those along the way so you know that you're making progress?
Yeah, it's a little bit like understanding that this is the first version is not necessarily the thing that's going to be mass adopted, but that gives you enough data to get the second and the third and the fourth.
Correct.
Do you, when you have that kind of philosophy, what was the first version that you were saying, we're going to ship this, it's not going to be perfect. We're also— I think a lot of especially like AI products today, like people will come out the gate and they'll promise basically like this perfect solution to, you know, a plethora of different products, problems. How did you think about basically distilling that down and saying, we're not going to do a million things, we're not going to do under the kitchen cabinets, we're not going to do underneath it, you know, underneath the couch or whatever, but we are going to do a few things really well. What was that first set of things that you decided? We're going to promise this.
That's a great question, actually. I'll take a step back. I think one of my favorite quotes ever was by Jack Dorsey, and he always talks about this idea that you have to make every detail perfect but minimize number of details. As much as I knew this quote is actually extremely hard in practice because you just sit there and say, wait, how do I— like, literally, you're going to ship a robot that doesn't clean edges? What if most of the dirt is edges? Then I still have to manually do it. Why would people buy it. So it's really hard in practice. And then I remember reading this essay, I think Paul Graham has this essay, like 16 mistakes you make as an entrepreneur, 18 or something like that. And number 3 is shipping too early. And number 4 is shipping too late. Okay. So how do you get it right? How do you get that thing in the middle? And the answer is we didn't get it right. We had to ship partly because after 6 years there was this gigantic pressure of we just got to ship. We had also announced our product in November of 2023 and we thought we would be able to ship it. So along the way we switched from Ambarella as an SoC to NVIDIA. That's a separate story. But because of that, we were 6 months more delayed than we had anticipated. So there was also this intense pressure from customers who had paid for the robot and reserved it and saying, when is it coming?
Yeah.
You guys are never going to ship. So there was this intense pressure of shipping and then we're like, okay, we clearly don't have edge cleaning. We We absolutely don't have this. Maybe we reach out to our customer base who have already placed orders and ask who is willing to take the robot as it is. And turns out there are a lot of people. Some people said, hey, I'm going to wait for another 3 months. And some people said, nope, I'm ready, just ship it.
Did you preemptively just say, this is what our robot can do, this is what it can't do, this is what it should be able to do very soon, but it just can't do it right now?
Correct. We did that. We went into customers and we were very direct and very honest. that this works and this doesn't work. And if you really want— care about what doesn't work, please don't— But if you want to be an early adopter— But if you want to be early adopter and you just want to try it out and you want to take it as it is, we'd be there to support you. We'll iterate, you know, you'll see us making progress right away and you can start taking it. And many customers who are just early adopters, and especially those who have an understanding on how hard this challenge is, they took it and they've been extremely helpful. And they're like, yeah, I see it. I'm happy to wait for next software update. And then they give us a lot of feedback. So there are many customers who want to help you innovate. There are many customers who want to be part of community. And if you can find them and if you just set the right expectation, they will actually help you. One of my favorite product of all time and favorite company of all time is this tiny burger place called In-N-Out that no one ever talks about. And the reason I find it absolutely amazing as a product is because it's a freaking burger. It's not rocket science. It's not taking Starship to space, but for 80 years, they've never changed a menu. They've never changed anything about their process or stores for 80 years. You've never seen any advertisement for In-N-Out anywhere. Yet each In-N-Out store does 10x the revenue of each McDonald's stores. And they have gone through 3 generations of owners. And it's gone from literally 1950s, which is completely different world. And it literally started 20 miles away from where McDonald's started in in Los Angeles. And how do you build a product that 80 years you don't change anything and yet people still love it and they just rave about it? And sometimes it is just making a promise and meeting it. 'Cause most products don't, most products overpromise and underdeliver. So if you can just make a promise and deliver it in a meaningful way consistently, that's good enough. That's where I was, we were talking about Starbucks earlier, Chipotle. You know, no one will call Chipotle the best burrito in the world, or no one would call Starbucks the best coffee in the world, but they make a promise and they deliver it consistently in terms of the taste. taste and price point in the same exact way. If you can just do that, it's really, really amazing. And that's actually much harder to maintain over time.
Yeah, I actually think that Chipotle is an interesting example of kind of a company that probably fucked up in some way. And the reason why is because a couple of years ago, I remember this idea of like a Chipotle burrito is you got your burrito and it's going to be like this big burrito and you get a lot of food and all this And then over time, like the portions kind of started getting skimped. And then, you know, online people would, you know, they did this study where if you recorded the person making your burrito, on average it was 50% bigger than if you just asked them for the exact same ingredients in the exact same way.
Wow.
And so there was that big of a difference. And of course, like over the course of, you know, the past probably like 10 years or let's say 7 years over, like people, millions and millions of people are seeing that in their mind and like the erosion of the customer experience, like no one goes to an In-N-Out and out today and says it's a drastically different experience than it was 10 years ago.
That's correct. Which is at some point, if you do not— customers are not stupid. Customers are really, really smart. Every single one of us expects what's going on. This is not '70s where you can just create some advertisement and people will trust. This is the world where people have expectation and they observe it. And you're correct that if you erode trust in a very, very small way, it will backfire.
Death by a thousand cuts.
Yeah, death by a thousand cuts. and cuts, right? And then there is also a bit of a— I also dislike this idea. So, okay, I'll take a step back. What I've learned over my career is that there is no such thing as a perfect product. It doesn't exist. Perfection is a mirage. But there is a simple product and a complex product. And simplicity is much, much harder to maintain. So In-N-Out has kept that simplicity. A lot of companies fail to do that over time. One of the points that I always make is, I've I have a— my dad just passed away, but my parents were in 70s. My mom is 75, my dad was 80, and they would visit from India. And every year when they came back, Uber app is different. And every year they come and iOS is different. And a lot of it is just basically designed for design's sake versus genuinely making it simpler or better. And that always bothered me that why are we changing things that are not needing to change. So there is a bit of a sort of this arrogance of design, and I don't mean to criticize anyone, but there is this element that we need to learn that simplicity is the goal, not perfection. And you shouldn't keep designing things or reinventing the wheel to get to that sort of newness, just kind of teasing that.
Counterintuitively, if you have a complex product, like let's say the Apple phone, it's not super, it's not, extremely simple. Um, but people over the course of like a decade plus learned to use it.
Correct.
I remember they came out with the, uh, the Apple Glass update where it was just like, they, they also like did a whole bunch of redesigns on like the photo app and all these other things. And I'm like, we've already trained our grandmas and grandpas, you know, they've already gone through this process of helping them understand their, their, their device. And suddenly we've just created this like, Matt, you know, just taking a grenade and just dropped it and all of that experience. And it's I think that they just pulled a Chipotle and just like destroyed, eroded a lot of customer value.
Correct.
'Cause part of it is just that consistency.
Consistency. It's hard, but don't change things. So at least in 2017 when we were starting out, I created this sort of like a number line where you have complexity on one side and simplicity on the other side. And what I did was I said, let's take Facebook apps and add them. And in 2018, at least I felt like Facebook was the most complex app and then probably Facebook Messenger and then Instagram and then WhatsApp. And WhatsApp today is still probably the simplest one. And the way I kind of think about it is if you tell people why do you want WhatsApp, or if you sort of living under the rock and you downloaded WhatsApp today for the first time, you'd know what it's for. It's to communicate with your friends and family, your contact book.
Your purpose is clear. Immediately see your contacts, you click on one and you're able to message.
Message. It's clear, it's not necessarily the prettiest app anymore, but it is simple and the purpose is still clear. If you look at Facebook, why do I download Facebook today? Is it to connect with my friends? Is it for newsfeed? Is it for Reels? Is it for stories? Is it for group chats? It's for, what is it for? Same exact thing, Instagram, when it came out, it was really clear that you got the product to—
Like share photos.
Exactly, fix your photos, make them nice, share with the world. Now it's what? Is it about the stories? Reels? Is it about posts? Is it about communication? I don't know what is it for. So for a new user, it becomes overwhelming when you actually add a lot of that stuff. So simplicity is harder to maintain than the other way around.
I was briefly mentioning like an interview that I really want to do with Pavel Durov. And he has managed to create a company. I think it's worth like $40 billion, has a billion users, generated a billion dollars plus of revenue last year. And it's just like, and he's got a team of 30 people that work on it total, including himself.
That's a great constraint. And they're all like remote. Yes.
They're all in different And the only way that you get hired at Telegram is, is there's this website called contest.com.
Okay.
Um, and, and basically you have to just solve these like extremely difficult like coding challenges and all this stuff. And then like the very best of the best of the best that are killers at this, they get, maybe get a job at Telegram. Um, but with that sort of product, it's, it's like deceptively simple where you just kind of intuitively understand what it does. And then with Facebook, they may have 120,000 plus, you know, people working at Facebook. And what does that do? That says like there's all these different teams trying to create value inside of that company. What does that mean? That means new features, new, new this, new that. There's only one designer at Telegram, and that's Pavel Durov.
That's— and that's exactly right. And that was the point for WhatsApp as well, right? WhatsApp was just 48 people when it got acquired for $19 billion. Instagram was just 13 people when it was first billion-dollar acquisition ever. So they were simple products and they remain small team. And I believe WhatsApp is still probably the smallest team inside Meta or Facebook. So constraints are great and constraints forces you to think there. So when you only have, let's say, 30 people like Telegram, you would say that, is this feature that's going to help 90% of my users? If the answer is no, not build it. But when you have, I don't know, 3,000 people, you say, oh, 300 million people will actually use it, but that's still just 30% of your billion users. So you can always frame the number in a way where it feels like it's an important feature, but it's And then that discipline is really, really, really hard.
I wonder, have you ever seen the fuck around, find out chart?
Yes.
I wonder if you can kind of do the same thing with like customer experience and the amount changed. And it's basically the fuck around, you know, how much do you want to change and how much do you want to see impact how people, you know, customers interact with your experience?
It's a, I think a lot of people, the best way to say it is Chipotle, or if it is getting away and its stock price at least feels like it is getting away with this, right? Or some other company, if it gets away away. It gets away because they almost create a monopolistic environment and you don't have an alternative. And when you don't have an alternative, people keep using it. So that's unfortunate part of it, which is sometimes that sort of behavior where you know you've cornered the market loses that mindset as well. But discipline is really hard. And I feel like the best companies or the best entrepreneurs over time go after simplicity.
How do you kind of distill that in your culture? Because you can kind of keep it in your mind. Yeah, right. And I know that the best companies, like the torch of the company is kind of carried by the founder, right? Yeah, the soul of the company. And that's why when the founder leaves, typically the company is drastically different. Or if they get acquired and they're no longer really the dictator of the experience.
Yeah.
They're no longer able to direct it. How do you think about instilling that in your, in your culture of just radical simplicity and trying to make sure that we don't add things They don't need to be added.
Unlike Flutter, where we didn't think about it, at least here we sat down and we said, how do we build a company that when it grows up, let's say with, I don't know, a gigantic amount of revenue or whatever, we'd still enjoy working in? And what that meant was how do we make sure that the goal there is also still going to be, still going to be shipping iconic products, even if it's a 5th, 6th, or 7th product. And for that, you almost have to think of company as a product as well and keep crafting it. And that iteration and crafting doesn't go away. And to install in your culture, you just have to keep repeating and prove it and keep preaching those things even when it's hard, especially when it's hard. And we do that, but it's still hard. I think the company that is a model company, at least in my mind, that has done this very well is actually Netflix.
What about them?
Netflix is still— so Netflix started at the same amount of time. So we did this experiment. I'll take a step back. We did the experiment. We said, okay, inside Facebook, WhatsApp is simple. Facebook app is complex. But can you do that at a company level? If you do it at a company level, what's the measurement? What's the metrics you'd use? And one of the metrics might be that, hey, it's profitability per employee, but that's really hard because of GAAP and all that stuff. So what about revenue revenue per employee. So at that time, at that point in time in 2018, we were like, okay, let's take company that has been around for at least 15 years, has been public for a while, is pretty big, and let's actually figure out which one has the highest amount of revenue per employee. And I started with this idea that it would obviously be Apple because Apple is just so amazing in many different ways and the products are not necessarily cheap. They're premiumly priced. Turns But it was Netflix when we did that math. And it was Netflix because the amount of people inside Netflix is, I think, I believe they're still around like 11,000 or 12,000 people versus Salesforce.com is 100,000 people. Meta is probably somewhere close to that. Apple is 200 plus, Google is 200 plus, Microsoft is 200 plus. So there are these gigantic companies. And I bring up these companies is because Netflix started just around the same time. And it remains small and it remains small very, very deliberately. So one of my business school colleagues is a chief product officer at Netflix now. And she told me, Eunice Kim, she told me that there are only about 50 product managers in the entire Netflix. So that discipline is really, really clean. Then another example is I read somewhere that for first 20 years of Netflix life, they used to have a free trial. You went to Netflix website and you signed up for new trial. When they decided to get rid of it, it wasn't that you just commented that section out of the code. They actually put together an entire team to delete everything from their system that was around this particular feature that they had. And that's important because what you're doing is by deleting, you're simplifying.
Yeah.
When you just archive or put it in a code, you're not simplifying it, right? I think there's some stat that Windows is what, 2 billion lines of codes or something like that, and you can't really duplicate it, or Word is that way. So being able to kind of control the beast is hard. So how do you actively simplify it? And by deleting things. And those were some of the things you have to do, but it's hard. It's not easy. There are not that many companies that got there. I think that requires patience, that requires deliberation.
In-N-Out is probably another company that has done that very, very well, What have been the moments in your life where you came across something where you felt like extreme intentionality?
Ooh, I think it started very early on. I used to be very much a movie buff growing up, and I think movies was my initial passion. And I just felt like great stories, even though they were the exact stories that had been told some multiple times, there was just a craft about it. And there was this thing where you couldn't glue like a I remember like maybe sometimes in college, I can't remember when, but I would literally sit there and look at or watch movies and say, which scene would I cut? Not because— and I had watched that movie 5 times already and I would sit there and thinking like, which one do I cut? And I wouldn't cut anyone. On the flip side, sometimes I get people mad because I kind of say that I, as much as I like Star Wars, I would say that Star Wars are the worst edited movies because every single scene kind of stops and then And then you see three rockets flying around, and big giant ship of Darth Vader. And you hear see Darth Vader or big giant Resistance ship, and then Resistance happens. But every single scene changes that way. So initially I didn't notice it, but once I started noticing it, it got really I got really mad because I just felt like I left. It always destroyed the link of the story that it was episodic. It just moved from one episode to another episode. It was just tiny shorts that changed with each rocket. flying back and forth. But then it was— but even then, like, Star Wars for '70s was amazing, uh, what they did. But that's when I think I started thinking about the intentionality. It was first with movies and then kind of came to the product after I got hands-on Apple products.
Did you take any lessons from Steve Jobs? Like, did you decide to do anything differently than you would have otherwise because you had him as An example.
I— so initially I came with this idea to Silicon Valley that product trumps everything, that a great product, great experiences are there. I think it was both Apple as well as Pixar. So doing something consistently is really, really hard. And then the first 10, 15, whatever number of movies from Pixar were just so mind-bogglingly amazing. And it's not the same Pixar that it used to be that 10 years, but how do you do that? So that was Pixar story was really fascinating. fascinating to me as well. And there was so much intentionality in every single movie. And how do you come up with hits after hits after hits? And it's really, really hard to do that. So that was really— and then the other story that I really gotta admire is I absolutely love Harry Potter. I've read it 3 times. And the reason I like that is because it took J.K. Rowling 15 years to write those 7 books. And each book reveals more about the previous books than you thought it would. So to weave in All these clues where when you read a 6th book, all of a sudden, second one makes more sense to you. That's mind-boggling. And to be able to stay with that for 7 years and tell that story in a very, very concise fashion, relatively speaking, it's so amazing.
It's like just adding depth.
It's just adding depth. And doing that, 15 years is a long time. So I remember thinking like, absolutely, that if I think of it as a product, Like she built this amazing 7-book product that is just, I don't think, you know, many people will try but won't come close to it. So that was just fascinating to me. So that's where you see the intentionality. So if you kind of look for it, it's everywhere in anything people do across the board that is fascinating. And to me, I just kind of said, let me just think of it as a product.
Yeah.
So I translate everything as a product and then stories emerge, then the intentionality emerge. you kind of think of it. But I absolutely remember reading, I didn't start reading Harry Potter till the 6th book was already arrived. And then I remember being obsessed about it for a long, long time.
I love that that's kind of like your favorite story, because that translates almost exactly into what you're actually doing at the company level, is you build the first product and that product builds on itself. It's just basically like a stepped-up version of the first one. And then there's a 3rd version. How are are you thinking about kind of planning out this multi-decade journey of building a products company where every single product that you release just adds depth to the story that you've already been building?
Yeah. And so I think Charlie Munger said it, right? Or someone else said it's that the— or no, I think Einstein has this quote that the 8th wonder of the world is compounding. Anything that you do for long-term compounds and it takes time. There is no shortcuts. And we realized this because I think I told you, that we were part of Y Combinator batch with Flutter in 2012. The most successful company out of that batch is Gusto, the payroll company, which is, I mean, Tomer and those guys have done phenomenally well. But I remember in 2017 thinking that even then it was like 5 years into it and they were just compounding versus we sold Flutter and it was gone. Same thing, I remember running into Stripe's office, as I mentioned, when there were 5 people on Ramona Street And now it's a $100 billion company. So you see the value of compounding. And this is the lesson that I learned later in my life. Even Amazon, Everything Store, like Elon's Tesla and SpaceX, like until 2015, probably no one paid attention to those companies. So anything worth building over the time, we learned that it requires patience, it requires building compounding, and you want to do it that way. So that's where we, in 2017, we knew that We were never gonna sell this company. And the goal wasn't to quote unquote build robotic vacuums. That was the problem we wanted to solve as a step one. But the goal was to, as I mentioned, build products that give people their time and energy back. And we did this research very early on where we realized that families in United States and Western world on average spend about 45 to 60 hours a week doing home chores, okay?
Really?
That's 3x bigger than the time a family of fair—
That's like a full-time job.
Driving, yes, exactly. And you don't realize this because you spent 15 minutes here cooking and 15 minutes here doing two dishes or 15 minutes here just doing vacuuming. But it's a consistent time sink, and that time is gigantic. So at some point, we realized that hey, if we can build a product that gives people their time back, that's amazing. And that was really the mission: that how do we build products that give people their time and energy back? And and that mission was. never about just this technology or one robot. It's about building, solving all the problems. And then we kind of laid down how do we solve these problems in a sequential way. And very early on we realized that there are 2 approaches to do it. One is to do what Waymo, Nuro, Cruise, a lot of this company did, which is build dummy robots and start collecting data and build these self-driving cars. And it takes 15, 20 years to do it. For us, the goal was to build great products. And that's where we like the Tesla approach because the way to think about Tesla approach is that they're building cars, they're building, they built cars, they sold cars, they have amazingly loyal customers, they are generating revenue and data collection and building FSD or full self-driving is an ancillary side benefit of selling these cars. So they are making an impact that they change the world in terms of gasoline engine or ICE engine to EVs even before getting to FSD. So how can we do it where we are making sure that we are product advertising it every single way. And ultimately, whether your technology genuinely works or not, the test of that is not writing a paper. Test of that is can you ship it to customers and it works? So if we can ship it and make it work, then we weren't trusted to make the second, third, fourth products.
For you, just internally, when you're deciding whether or not to ship something that you know is not fully fleshed out and you're literally emailing your customers, those first few signups, And you're saying, here's what we can't do.
Yes.
And here's the small set of things that we can.
Yes.
And if you want to sign up for those small set of things, we'll ship you your product.
Yeah.
You know, what is going through your mind in making that decision and deciding to just ship?
The great part about building companies is that no matter how you get it right, it's kind of mentioned— I mentioned, right, like you have to swim, be in a swimming pool. And sometimes when you're in that swimming pool and you decide to have have a courage to go to the deep end, you have no choice but to get to the other side. And that teaches you certain things in the same exact way it was constraints. So we, the way we think about it is in November of 2024, we had no choice but to ship. And in all honesty, but we had to do that because I had come to conclusion by that time that if we don't ship this year, we don't have a future, that we may not survive as a company. So either you ship or you die. And this is the beauty of startups. And this is why I love it, which is it's very binary. You ship or you die. You do things with less resources than the other companies or you die. You innovate or you die. It's a very binary thing.
It's kind of like fighting back against entropy is like the default state is if you do nothing, the brand kind of disappears and you kind of disappear and the meaning kind of disappears. So you have to keep on watering it.
Yeah. And we do a lot of things because we don't want to die as a human being, as a species. A lot of our innovation is just being based on survival. So even for companies, when I sent that email, it wasn't that, hey, which is either customers will accept or we're dead anyway. So it's like, are we ready where customers will accept? Pretty easy decision to make. Which is we have no choice. So you get to a point where you have no choice. And I think I mentioned early on to you that if you had come to our office in April of this year or April of last year, right before Wired us 10 out of 10 perfect rating, I think most of our team would've told you that our product is actually shit. And people genuinely believed that it wasn't good because day-to-day you only look at the problems. You're constantly trying to improve it. So you're solving the problem and whatever is done, product does really, really well, gets forgotten. So then you don't think about it. But there are little elements that we always had in mind. So first one was this idea that if it's a robot, And if we are building genuinely intelligent robot, then day one we knew that if you take anything out of the box, it's object, it's physical thing, it's not smart. If it's a robot and it's intelligent, it has to roll itself out of the box. So that's where some of the idea was that how do you build a box where robot just rolls out? And then say, how do you say, how do you get people to smile and be friends with it in the first place? 30 seconds. So can we say on a display that, hello XYZ family, or hello Nariyawala family? And if you did that, maybe it puts a smile on their face. So it was the idea that how do you make it friendly within the first 30 seconds? And the way we could become friends or someone seems friendly is if they come and say hello to you. So why can't robot say hello? And then we're like, okay, a lot of kids still may be apprehensive about it because it is a new object coming into your home. So at some point, I realized that, you know what, kids love stickers. So you ship stickers and the moment kids put stickers on the robot, it's a friend instant. Before even it moves around. You built a bond. You just built a bond right away. And you personalize it. So you actually feel good about it. Like that's how we love our pets. The first thing we do is we name them and we personalize them. We give them colors, we give them things that we want. And that's how it works. So if you just kind of think in that direction, it gives you certain ideas. It was just, yeah, pushing in that direction.
When you first started getting like customer feedback from those first few orders, how did you kind of take that feedback and get to the next group of customers where you can ship that future version?
I think we've always talked about this internally, that single negative feedback is worth 100 positive. positive feedback.
Mm-hmm.
So we had to balance positives, and you can't just play on positives and not worry about other things. But we almost kind of got to a point where we said, no, no, no, give us every single feedback, especially things you don't like. Partly because even when I was doing, as I said, I was going to friends' homes, 50 different demos in real homes. In person, people are super nice. They don't want to tell you what they don't like or like about it. So reality was to say, no, no, tell me what is critical and pay attention to that.
Give them like a comment box where they can say, fuck you.
Yeah, exactly. Really. And then the second one was also get to, which is when we wrote this email to customers saying that these are the things that works and these are doesn't work. We made sure that we were actually sending that email to customers who had paid for it. Because if you have paid for a product, you're your expectations are far different than something that's free.
Mm-hmm.
And you're far more critical about whether you value that or not. So that was the criteria as well, that if they return it, we know the answer. No matter what they say, if they return it, we know where we stand. So return rate or them just not using the product, those are the signs.
So we would kind of combine it with what we expect users to do if they genuinely liked it versus if they Like I said at the start of this, the reason that I came and did this interview is because of the just positive, spontaneous customer feedback that I saw on X and elsewhere. What were those kinds of moments like? You know, Toby Lütke famously said, Matic is a really special company, or like, where do I buy one? What were those days like?
It was really, really, really, really satisfactory in some ways. Which is, so when we started this company, both Navneet and I also sat down and said, okay, what are the best days in our career? And the reason we said iconic products is because best days in our career, the most memorable ones weren't the day we started a company or the way we sold company. In fact, what I do remember about Flutter getting acquired or selling Flutter to Google was relief, not necessarily joy. Jubilation, right? That was the feeling I had. The jubilation always came when you taught something meaningful for the futuristic, or when you built something futuristic and got into the hands of users and they loved it. And that part was really exciting that you craft something, you build something, you pour your heart into it, and then other users get hands on it, and it's just amazing. So from that perspective, when it started finally getting to a. Notoriety on X or people started loving it, it was really fulfilling. But it was both surprising as well as not surprising because we still haven't shipped 50% of the ML features that we wrote down back in 2019. So it was surprising that people were loving it so much. And I think it was surprising because as I said, things that it does well, we just take it for granted because we're so focused on improving the negative things. So it was really amazing from that perspective. that all these decisions we took and we were just taking for granted, people are really recognizing it. So that was great. That was absolutely amazing. And ultimately the reason I mentioned 100 million square feet clean, it's still, so I think, what is that? I think it's like 12,000 hours cleaned if I'm not mistaken, or 12,000 hours saved, sorry.
Wow.
12,000 hours of labor saved, which translates to about 100 million square feet as we kind of do the thing. So that's really fulfilling that you're really attaining your mission. So it was really great. It was really great. And especially because we tried to launch in 2023 and no one gave a shit. And we tried to launch in— we launched again in November of 2023 and no one cared. And I think Brian Chesky is infamous for saying that Airbnb launched some 13 times or something.
And if no one pays attention, then you can And just keep on launching.
Keep on launching. Exactly. So it was really good to finally see that, no, no, no, we are doing something truly innovative. It's not just another robotic vacuum that we built this from very grounds up, from first principles, that is the vision-only robot. And all of a sudden everything that we were talking about that, hey, it's sort of like a Tesla FSD for home robots, people started saying that. And that was really fulfilling and fulfilling in a way that, okay, now we know absolutely foolproof way that we're going in the right direction and we absolutely have to double down and keep going.
I think probably right now you are at the stage where it's Harry Potter and the Chamber of Secrets, but not quite even past that.
Not even quite even past it, yeah.
What does this kind of look like going, this journey look like for the next 15, 20 years of kind of adding more lore, deepening the story? How is this going to unfold?
I would go back to our sort of axioms and what we talked about it, which is can we keep solving more problems?
Mm-hmm.
So more than can we build a great product, it's more along the line that, okay, if we solved floor cleaning and let's say we nailed it completely, now maybe can we take it to the next level and maybe we'll do, I don't know, will it be a toy cleaning robot? Will it put the shoes back in its place? I don't know answer to those questions, but what is the next intense problem that customers have that they absolutely wanna solve? And can we do it to a point where they again have a delightful experience and no longer have to think about? So in that lore, I would love to have a set of pain points that customers have and then say, we take this one and then we take this one and then we take this one and we keep going down. And what we build as a product is just means to an end. So to me, that's how we think about it, that we start with problem first and work backwards versus starting with robots and working backwards.
Yeah.
Which is, let's say at the moment, and I absolutely love everything Tesla does and everything Elon does, but even if Optimus was available, great, why would I buy it? Am I buying it for laundry in my home? Am I buying it for dishwashing in my home? Am I buying it for everything in my home? Am I buying it for babysitting or maybe my elderly care? What is the purpose? And it's not necessarily clear yet. I'm sure they have some guidelines and intuition that they're going after. So it's really just productization is the key piece of the puzzle to me. And there is actually a really good— so one part we learned over the years, and this is, I forgot to mention when we talked about SLAM, but initially we got to our first prototype in 2021, working prototype, and it took us another 3 years to ship. And initially we thought that, hey, we would be able to ship this product faster Because mistakes that we make inside home are trivial. But turns out if the task is trivial, people's expectations of precision are higher. So with today's AI, for example, we collaborate, which is if it gives us 80% of the video right, we are mesmerized. Or if it gives us 80% of the app coded already, we are mesmerized and we're happy to take it to the 20%. Or 80% of the email is already drafted and you'll do final touches and put it there, right? So we're happy to collaborate because it takes us years to learn how to code or learn how to be an amazing director or be amazing interviewer, so on and so forth. But learning how to navigate your home without bumping, you don't go to school for that. Learning how to vacuum, you don't go to school for that. Maybe 7-year-old can do it in a far— in a very precise manner. So it turns out that if the tasks are simpler, people just People just want to delegate. They don't want to collaborate. And if you're delegating, the bar for accuracy is much higher. So we get email if single popcorn is left behind and says, hey, your robot didn't clean that, it didn't pick it up. And people lose trust right away. So precision bar was much, much higher and it took us a while to get there. And that's one of the things we are learning again and again that in this scenario, as we build the lore, it has It has to just work. People want to be rid of this task completely, and there is a lot more work to be done to get there.
I think it's a little bit different with Optimus where they're going to deploy it in their factories and stuff first, and there's going to be real value driven or not, and they'll know because either the cars will go off the line or they won't. But do you think that the model of you basically try to jump to book 7 immediately, And then you say, we're just gonna ship this thing and it's gonna be perfect. Does that even work?
I mean, Waymo got there, so I don't think it's not, it doesn't not work. It just takes longer time. Like you can't—
It's like even Apple started with iPhone 1.
Even Apple started with iPhone 1, and I'm sure there will be 5 Optimus version, but, and will it get there? Sure, it will get there. It's just a matter of when, not if. So there is a future in which humanoids are around. There is a future in which Rosie the Robot is available. The question is, when do you get there? And if you're a startup, can you survive until then, until technologies to get better? So anytime a startup, one of the questions that entrepreneurs or at least investors love to ask is, why now? And to be entirely honest, the product that we built, we couldn't have built that in 2012. Technology didn't exist. Compute didn't have weren't available. 3D printers weren't there. And one of the actually things that we got really lucky on, and one of the things we've talked about quite a bit is that we are entirely Rust language shop.
Mm-hmm.
And Rust wasn't great back in 2012 as well. It only became really, really good in 2019, 2020. So you do get lucky along the way where certain things open up and get available. So for example, you know, Netflix and Reed Hastings never wanted to build a DVD company. They wanted to build a streaming company. That was the goal, deliver, deliver movies over the internet.
Create like a seamless experience.
Seamless experience. But reality is that they couldn't have done it in 1997 internet or even 2002 internet. It's only 2005 or 2006 where broadband became ubiquitous and you could do it. So sometimes infrastructure has to be around for you to be able to deliver that product experience that we want. Like Instacart was tried as Webvan back in early 2000s, right? Or DoorDash for that matter. And didn't work because the technology and infrastructure and the communication devices weren't good enough yet.
Yeah.
So sometimes you have to have a wait. So I think Andreessen Horowitz says this best, Marc Andreessen, that as an entrepreneur, the challenging part is that whether they will get timing right or not is very hard to predict. But investors can keep investing in the same idea again and again. And sure enough, Sequoia invested in both Webvan as well as Instacart. So it is possible to go do that. So in that scenario, if you start early, then you have to survive. Now Optimus and Elon can make that bet because they don't have to worry about survival because they're already generating tons of profits and there is a cash cow.
But they also have, like you were talking about earlier, the muscle memory. They have, like in the same thing that you were describing is you guys are like a products company and you don't want to basically work in dark You can't start a business for 20 years and not ship a single thing and then say, here's this perfect thing, and then start shipping.
Correct.
Over time, you're building this muscle memory of you've built one product and then you've shipped it to 500 customers or 3 customers, and then 1,000 and 5,000 and 10,000 and 1 million.
And for them, it's great because they are the customers. They have these amazing factories. They are absolutely aware of what are the efficient systems in terms of their manufacturing or what are inefficient systems. And where human labor is absolutely necessary and where the labor shortage is hurting them. So for them to build Optimus robot and kind of go after it actually makes 100% of the sense because they can build for their needs. And then it goes back to this idea that if you build something that works for you, hopefully more people will have a same exact need and then they will do it. So I think it's brilliant for them.
Some of the wonderfulness about the world is that humans are mostly the same.
Yes.
We like to be special and we want to be unique, but mostly Obviously we all want DoorDash roughly the same way.
Yes, yes, exactly.
Or Uber.
Exactly. But startup, to kind of continue, we did those. So in 2017, it was obvious to us that if you're trying to build the way Nuro, Cruise, or some of these companies did— okay, I'll take a step back. The day you run out of money, you die as a startup. No matter how many customers, products, how much revenue you have, it doesn't matter. If you're If you don't have any money left in the bank, you're dead.
There's that wonderful line that all startups and all companies die for the same reason. It's just lack of cash.
It's just lack of cash, exactly. So my very, very crude analogy for startups is that every startup is a ticking time bomb, and the day you run out of money, it goes poof. Now, as long as you're raising funding from outside, you're just adding more time to that clock, right? You're not necessarily defusing the threat. The only way to defuse the threat is when you get to cash flow positive on your own. Now you're generating your own cash and you don't need any external resources. So if you're a startup trying to build for 10 years, when you don't— if you're a startup that is trying to build indeterministic product, you don't know how long it's going to survive. And it's kind of fascinating to me when we look at self-driving car space that the only 2 companies that are shipping and thriving today are Tesla, which has a car as a cash cow, and Waymo, which has AdWords as a cash cow. So both of those companies had these profits that they were making that they could generate. So survival wasn't based on shipping potentially, and you could keep building for a long time.
I feel it's kind of interesting that you say that because I feel like the DNA that both of those companies have is that DNA of shipping. Like, you have to ship, you have to create something that's valuable in order to defuse the clock. And you maybe have, but these other organizations that do raise like a couple billion dollars without having to fuse the clock, they didn't actually ever build that muscle memory.
That's, that's a great point. I honestly, I didn't, I didn't think about it, but you're correct. Absolutely correct that they actually have a genes to ship as well.
That's kind of what I'm saying is like, are you doing an almost, if you just raise like a couple billion dollars and you never ship anything, is it basically doing like an organ transplant of, you know, and does it actually like accept? Do you go from suddenly having a culture of never shipping, raising huge amounts of money and—
Yeah, you don't get there. No, no. That's a founder has to get there, founder has to do it. Usually, usually it's if you don't have that gene and if you've been in that research lab oriented environment, you don't end up shipping. So with Claude, I mean with Anthropic and OpenAI, it's exception to the rule than the reality. But yeah, you're absolutely correct.
What thing do you do that almost every other founder that you know doesn't do? but you think is right?
I'll go back to what Kevin Systrom said, which is think of a product first. I am not— which is we preach again and again and again inside this company that we're solving customer problems. Start with the problem, work backwards. Don't build cool, build useful first. These are very, very unintuitive for whatever reason to many, many early engineers in their career because you are just so mesmerized by technology and possibilities of technologies. So for us, one of the things that I do again and again is that, is this really solving the problem? Is this really going to be useful to customers? The other thing that is somehow very unintuitive is putting yourself in the shoes of customers and just thinking, is this complex or simple?
Hmm.
Which is asking why. Like, why are we doing this again? You know, why is this simple? Why do we need to have this tap? Or why do you need to have these features? Are you sure we need it? That's another very, very counterintuitive thing, surprisingly, because most of the time, which is a— there was an article from Steve Jobs that I read a long time ago. I can never find it again. But in that article, he talks about this idea that most people think of a solution and they just go build it. And it's like, at Apple, we don't do that. We build a solution, we put it there, then we go away for 3 days and come back and say, is every single single feature or piece of button we added to this hardware product absolutely needed? And it's like we keep peeling the layers of the onion until the essence is left. In the same way, we think of a solution, we put it there, and then we keep removing things to get to the simplicity. And it's like, why do we need 3 buttons? Can we get away with 1? And that requires iteration and that requires time and you have to walk away from the product. And that's actually still very very counterintuitive, which is it's easy to say, oh, let's just add an option for it, or let's just add a button for it. But instead it's very hard to say, no, no, how do we solve a problem where it just works for customers? Or how do we do the hard work on the other side?
Well, let's come back to Matic and where you guys are at right now. We're in your office. There's a whole bunch of parts and assemblies and sub-assemblies and all these different things that you guys have stacked around and you're just starting to ramp manufacturing now.
Yes.
What's the journey kind of been like going from having not shipped a single unit to shipping thousands of units?
It's so, so it's very much fun, but it's, it's, you know, again, quoting Elon Musk, he's right about a lot of different things, especially products. And he has this quote that says factory is the product. And he always talks about how scaling Model 3 was much, much harder than actually building Model 3. And we are in that scaling hell at the moment.
Production hell.
Production hell, exactly. And building something that consistently just works again and again is really, really hard and we're figuring it out. And I think Navneet has a really good framework for it. So Navneet, my co-founder and CEO, again, he always said that it's very easy to imagine if you build one product, it's easy to imagine how we're going to build 10. If you build 10, it's easy to imagine how we will build 100, which is it's easier to imagine order of magnitude growth. But going from 1 to 10,000 is hard. That's very hard to imagine. And that's still true in software as well, because even though people build apps, people build website, if you all of a sudden went from 100 users on your website to a million users, it's going to crash and burn. It requires different set of tools and systems. comes to deal with 1,000 users concurrently on your website. So in the same exact way, the challenges for us has like we shipped 300 units in Q4 of 2024. We've shipped 3,000 over the last 2 months of Q4 this year. Like 10x growth.
So 10x.
So we did add a lot of magnitude. Now the goal is to go to 60,000 this year. And it's both extremely fulfilling and as well as extremely challenging.
We were walking around and I think you said that 20% of the assemblies that you're getting from China are basically just malfunctioning or like they don't work in some way. With that, you have to get really good at basically unfucking shit.
Yes.
Right? So how have you started getting good at that?
Great question. So there are 2 parts of it actually, you know, to be entirely honest, there are set of challenge or quality issues in your product that at the small scale don't surface. So when you try to build 1,000, they show up a lot more than when you try to build 100. So there have been challenges like that about reliability and quality that as we got to scale, we realized. So we have to solve that as well. But there is also communication things. So things that are really, really surprising that has happened. So we, endured 60 days of camera delay.
Hmm.
Which our cameras come from STMicro, which is a French company, really big, really popular. They're amazing. But even they face their own delays. So we endured 60 days of camera delays. Then at some point we get our motors from a really popular Japanese company. I'm not gonna name them. And they've been phenomenal for 2 and a half years. And then all of a sudden we started getting these robots in 80% of the robots were just failing our noise tests.
80%?
80%. This is right around the time in May as Toby Lutke and all these guys are tweeting about as we're getting lots of orders.
So the fires are just like everyone on Twitter is like, go buy here. And then it's just like fires are cropping up everywhere.
Fires are cropping everywhere. Like 80% of robots are failing. We're like, what the hell is going on? So initially we're like, we must have screwed up something into our manufacturing process. So we kept digging into it and ultimately realized, no, no, no, that motors that have been amazingly reliable for 2.5 years of us iterating all of a sudden are just noisy.
Hmm.
So we reach out to the CM and we said, hey, your motors are noisy. Did you change anything? And they're like, no. And we're like, no, no, no, let us show you. So we had this old motor and new motor and we proved side by side that these were noisy. So they started digging into it and turns out unbeknownst to them, their own supplier had changed the glue that goes on the impeller on the motor, which results in higher friction, which results in higher noise. So figuring that out and then dealing with it. So then we had to replace all those motors and we had to fly them in and getting there. So those are all the challenges that you don't think about on a day-to-day basis as we kind of build it. But those are the things that we really have to think about every day that things, it's like there is a quote somewhere, right? That man plans and God humbles. There is nothing more humbling than trying to build hardware assembly line.
With that in mind, especially for the first hundreds or thousands of units, you can kind of have a much higher touch experience with customers. You can maybe even be personal customer support. That doesn't obviously scale to millions, but you can do it at the early stage. And so when you have a customer and something goes wrong, what is the process that you kind of create to solve that issue? As fast as possible.
So I think this goes back to our thinking that we never thought of product as just Matic robot itself.
Yeah.
This goes back to Tony Fadell, Tony Fadell's teaching. And he always said that the product entire experience, you have to get the entire experience right, which is from the moment they hear about Matic or your company all the way till they stop being your customer, which is completely stop using your product, is entirely a journey. And every step you have to get it right. So what happens when they come to your website? What happens when they purchase the product? What is the experience when they ship the product? How do you— how did they feel? Are they feeling like you're being very straightforward to them? What happens when they actually get the robot and unbox them? What is the first 24 hours, first 30 minutes, first 24 hours, first week, first month? And what happens when robot actually doesn't work as it's supposed to work and they've paid for it and now they're in this crisis situation? How do you handle them? And you're not gonna get all of these things right. Some robots will ship in spite of your trial with some quality issue, but how do you handle it? It becomes critical. So to me, customer service has never been a separate thing. It's part of the product experience, which means it has to be done in a very, very meaningful way. And every single customer has to be looked at in detail. So we try to avoid, for example, any sort of templates. We try to avoid any sort of cookie-cutter answers on how you are Here, how do you do it? Every single customer, we try to explain exactly what the problem is in detail, because worst is when you call a customer service, then they walk you through a script. So we know what bad customer experience looks like, so we try it. And then at the moment, anytime a customer has any sort of issues that is there, obviously I can't look at every single customer tickets, but I do get every single one one as an email. So I get email alerts on everyone. So as I'm browsing or even—
You want to keep a pulse on what's happening.
Exactly. So I do do that. And then there is a process we placed in that anytime you see issues like X, Y, or Z, escalate it up to me. If customers endured some sort of experience that was really, really out of their point of view, can we go and— can I go and reach out to them? So every single customer who wants to return a product, I'll reach out them and ask them why, what happened. And that's a learning opportunity for me. Any single customers, there are routine issues, there are maintenance issues, and then there are issues that were unexpected. So I try to go and dig into why and what happened and where we didn't meet expectations and make sense of it.
What I noticed is you just have a huge amount of reliability and testing before any one of these robots goes out.
Yep.
How did you come up with that?
I think it was also iteration. Even assembly line could be a product. And we just kind of said, okay, what are the point of view? So for example, we looked at calibration, camera calibration. That's critical because if the cameras are not looking at in a right way, entire system fails, right? And a lot of it is because we just keep testing robots ourselves as well. And we realize that here are the failure points. So as we do reliability testing, as we do longevity testing, we know what the failure points are. points are, typical things are. So then you can come up with that, okay, what are the things we can test that would change it? And really the answer to the question is that in the first 10,000 or 1,000 customers are going to be your evangelist for life. And if they receive a bad product, they're not going to be happy with it.
And if you didn't care.
If you didn't care, it's even worse. So we're much rather off taking pain and double and triple check everything and hand them and understand what's needed, then we nod. And then as we understand that there is a reliability, we can relax certain constraints. But as I said, you just have to assume, go after with this assumption that things are not going to work as well as you want. And at least at the critical point, there are tons of reliability checks, but critical points, where are the 5, 6, 7 tests we can do that says robot is working exactly the way it's supposed to work?
I love this line from Brian Chesky. I don't know when he said it, but he basically said every founder, when they start their company, they like almost unanimously just want it to take off immediately.
Correct.
But there's something like special and useful about companies just not taking off and not working for a long time. It allows you to kind of build different muscle and different understanding of the problem and things. And so I imagine that with one of these types of products, it's almost if most of the people that will eventually adopt don't want to buy it day one, because then you can actually spend that much more time and that much more thought and care on those first few orders to make sure that the thing is nailed before you scale up.
Yeah, that's a great— it's actually built into hardware. It doesn't matter how, like ChatGPT can go from 0 to 800 million users in 2 years. In hardware, it's impossible. There is nothing we can do that will happen. Not only that, but some of the numbers that we want to shoot for, we have to buy certain parts 8 months, 9 months, 10 months in advance. So even if you get to a point where all of a sudden, like, you know, in May of last year, all of a sudden we saw demand skyrocket, we couldn't do it. Every entire projected sales for the month of December prior to Christmas that we could have shipped for was sold between Black Friday and Cyber Monday, just 4 days. So everything we planned for 7 months and it's like, we're going to ship this unit. Turns out demand will be severely under estimated demand and everything got sold in that 4 days. And then come December 2nd, we couldn't have— if the order came on December 2nd, we knew, we anticipated that we won't be able to ship it prior to Christmas. So we couldn't really do any push anymore and we changed our timeline right then and there.
Telling people.
Yeah. And there was nothing we can do about it. Could we have sold another 2x, 3x more robots? Sure.
But you couldn't have sold them in the right— by Christmas.
Yeah, we couldn't have shipped them by Christmas. So the advantage of hardware is that your supply, if you find a product market fit, you will be supply constrained, which means you are forced to grow deliberately. And that deliberate growth is necessary because every single level of growth uncovers bugs, uncovers issues, and it gives you time to push it. So it's sort of like a forced discipline of compounding, whether you want it or not.
It's interesting that you say that Because it's once you hit that inflection point where there's just way more demand than there is supply, suddenly the product goes from being the actual product to being the factory.
Being the factory, precisely.
And it's like toggles.
Toggles. And when you build it that way, I think that we did this, I don't know when I did this, at some point in 2022 as we were sort of trying to keep everyone patient and make sure the team remained patient to ship this through so that let's look at the top 10 companies by market cap in the world today. I think 8 or 9 are hardware companies, right? From Apple to SpaceX to Tesla to TSMC to Nvidia. These are all hardware companies. Even Microsoft, you can say it's a software company, but has some hardware components associated with it. So it was really fascinating to see so many hardware companies. And what we realized is that in hardware space, compounding takes a long time, but once you get there, you genuinely have an opportunity to build impactful, transcendent company and you can get there versus in software, ups and downs are much faster.
Yeah, your moat is a lot deeper.
Correct.
Yeah, I think one of the best moats is just the required pain that must be experienced before the thing can work. And if you have 7 years of pain as a moat, that's a pretty good moat.
That's pretty good. It's just, so I'll give you an example and this was a really big one as well. So the Nest camera for, I mean, when I was at Nest, most of the time I worked on Nest Nest camera. In fact, I was a product lead for Nest camera, so that was my primary product, even though I give you the thermostat example. Nest cameras became hypercompetitive very, very fast. In 2015, when I went to Nest and started, took over as a product manager, Nest camera was the only product in the market. By December, there were at least competitive security camera products, 20. And I remember the joke was that cameras flying out of Santa's ears because every week new camera company would launch. But then I remember thinking even back then that, wait a minute, there is literally no competition in Nest Thermostat in terms of smart thermostat. Even today, 15 years or 14 years after, or 15 years after its launch now, if you want to buy a decent thermostat for your home, Nest is still it.
Yep. Right?
And that sort of perplexed me because the microeconomic principle says that Wherever there is a profit, competition will enter. And Nest Thermostat, or thermostat if you think about it, is a— no, you don't have to go and explain to any user why you need a thermostat. You don't even have to explain why you need a smart one because everyone of us have used the dumb one where we forgot to turn it off or it was too hot or too cold or unconnected one. There are 130 million households in the United States and everyone legally, lawfully requires a thermostat. So here's a clear-cut market, clear-cut profitable category, yet no one is entering this space and Nest is still it. On a flip side, security cameras, most of us actually did not grow up with the security cameras in our homes. It's a new product coming into our home. Many customers are asking why. In fact, I live in a home in a cul-de-sac in Los Altos where previous owner hadn't locked it for 30 years. And I remember, and the reason I know this is I asked him, hey, how's secure is this home or is this area? And he's like, well, I haven't locked it for 30 years, so take it for what you will. That was his answer. And I remember thinking that, okay, good luck to me trying to sell this guy a security camera because he doesn't even think security is a concern, right? So it was a new product yet everyone and their uncle was building security cameras. So it just bothered us for a long, long while. And then we realized that it has to do with the tediousness, unsexiness, and also which acts as a barrier to entry. So it turns out that if you want to build a thermostat, everything is hard. The hardware, software, the screen, making it beautiful, all that stuff is hard. But still, the most tedious part is making it compatible with last 60 years of HVAC system. That's a whack-a-mole. That's just pure grind, unsexy work. No one wants to do that. Similarly, when we were building smoke alarm, which I told you about earlier, 800 pages of regulations. No one wants to deal with that.
800 pages of regulations?
Yes, because each state has a different law. Each state has a different And then the type of alarm that sounds when you have a fire is very different than type of alarm that sounds when you have a carbon monoxide and how you turn it off and how it's supposed to work and how the batteries are supposed to last and what you're supposed to do to notify the customers. All those rules are preset and they're preset per county and states and stuff. So you sort of have to go through these regulations and make sure you're sticking by it. If you're building security camera, you can go to Shenzhen, you can buy a camera, it'll come with embedded software. You can use OpenCV to do person detection and voilà, you've got a camera. There is no barriers to entry. There's nothing tedious about it. So security cameras was a space that was easy to enter versus here, it was just really tedious.
It just sucks.
It just sucks. So when we did start building in this Category 27, one of the reasons to build this product was also that we just looked at Roomba and we're like, look, they haven't innovated, or iRobot, they haven't innovated for 15 years. Now it's 23 years. And yet, and even though iRobot went bankrupt, up last year in December. In 2024, they did $700 million in revenue. So if you tell me that I don't have to innovate on a product for 23 years and generate $700 million in revenue, well, sign me up. Most companies in Silicon Valley will die to have that sort of revenue. So it was really obvious to us that this was in a very interesting category where no one was coming except for sort of Chinese copycats. But what was even more interesting is that we knew from day one in 2017 that Google wasn't going to enter this category. Apple wasn't going to enter this category. Amazon won't. It's just too unsexy and tedious of a category. And even startups won't get excited about it because if you graduate with a computer vision PhD from Stanford and tell your mom, I'm working on self-driving cars and lane detection, she'll probably say, oh, my son is amazing and cool and look at this world-changing problem, it's— he's working on it. But then if you go over to Stanford, graduate with computer vision PhD, and tell your mom, I'm working on robot vacuums, most likely she'll say, what the fuck is wrong with you? Right? So we knew that it was such an unsexy space that no one was going to come, and that would allow us to build this amazing product that we want to build. And then it's a product that's needed by everyone, and we'll have a chance to build a sustainable business, which allows us to build Second, third, fourth product.
Let's end it on what's the hardest thing you've overcome?
I think my own patience. This is the longest job I've held in my career. I think prior to that, because of acquisitions and everything, the job that I've always held was 3 years at most. So it's really, really hard to learn to be patient. I'm by definition, very, very impatient. So just being patient with Malik and continuously preaching that, hey, we're making progress, stay alive. And then keeping that faith was absolutely hard. It's not easy. And then we've lost some people along the way. Some people who thought, you know, they'd be with— that I thought would be with us forever and ever. People lose faith. It's hard.
