Cheeky Pint - The Bot Company founder and CEO Kyle Vogt on home robots and why he’ll never sell another company

Episode Date: June 25, 2025

The Bot Company founder and CEO Kyle Vogt—who also cofounded Twitch and Cruise—joins John Collison to talk about applying AI to home robots, the similarities between robotics and self-driving, and why the next $100 billion company will have fewer than 100 people.Full episode transcripthttps://cheekypint.transistor.fm/3/transcriptTimestamps(00:00) Intro(00:38) The Bot Company pitch(02:05) Single-task vs. multi-task robots(04:27) What is the Turing test for robotics?(05:52) Why this time is different for home robots(08:42) The last mile in robotics and self-driving(09:47) Viral demos and hype cycles(10:38) Commercializing frontier tech(13:06) Self-driving CapEx(14:15) Regulatory hurdles(16:18) Tesla vs. Waymo(19:21) Why Kyle regrets selling Cruise(21:39) The next $100 billion company

Homes are just horribly difficult environment for robots. I mean, when you put it that way, it sounds like the Ninja Warrior obstacle course. I think it will be strange to move into a home or apartment in five years that doesn't have a home robot. So you're very serious about the small team. I think the next $100 billion company that's created, you know, in 2025, 2025, 26 will be under 100 people. Kyle is one of the only entrepreneurs to have started three separate billion dollar companies. He started Twitch, Cruise. And now he just started the Bach Company, is trying to make house.

old robots finally happened. Cheers. Okay, so let's back up. Yeah. What is the pitch for the boss company? Well, I don't like doing chores. I think five to ten hours a week people spend doing essentially unpaid, unskilled labor in their own home.

Yet we all take that for granted and do it every day. And I think it's been the holy grail of robotics, you know, since I was a kid doing robotics to have the home robot that does everything. It was very clearly not possible up until very recently with LLMs and then in neural networks to control robots. I think just maybe, maybe this time is the right time to build this company and deliver this home robot that does all the things that you don't want to do. Okay, so this is vacuuming the floor, ironing the clothes, cleaning up after the pets, that kind of stuff. We're going to start very small, but it will continue to evolve as the state of AI evolves and be able to do more and more things in your house. I think it will be strange to move into a home or apartment in five years that doesn't have a home robot or you won't want to without one.

And, you know, in the same way, they would feel weird to not have plumbing in a home or a dishwasher if you can afford one or laundry machines. These are all machines that had a huge impact on our lives. And these are all from the era of, I don't know, the 50s and 60s. And we haven't really had that next. Yeah, we had a big spurs for a while. And people were excited about it. There's all these, like, great ads of, like, you know, the person in the kitchen being like, look at how much time I've saved with the microwave. wave and, you know, we've gone stagnant since then.

You could be bearish on home robotics, and an argument I think you could construct is we are currently so underperforming what we could have in terms of household appliances. And so dishwashers take forever to run and don't clean the dishes that well. And commercial dishwashers exist that are super fast and much more effective, but we don't do those in our homes. You know, the toaster can't tell when it's burning the toast, even though that seems fairly trivial to detect, should this make one worried about future household robots? All of the things you described, the toaster, the dishwasher, these are like single-function machines.

And I think everyone when they buy a single-function machine or any machine is thinking about the cost and value, how much is it worth to me to have toasted bread? Maybe like $30 for a toaster, maybe not like $2,000 for a multi-toaster that can do it perfectly. But the question becomes if you have a machine that is a multitask machine and can do lots of small things that you would maybe even pay $0 for if it was a standalone machine. But when it's bundled into this general purpose machine that can pick up all the kids' toys and clean the dishes off the table and put stuff in the sink and pick up your packages from the front door

and bring them to the kitchen, like I would never pay any money for a machine that just does one of those things. But when assembled, I think it becomes extremely valuable, or at least that is our theory. When you ask people, if you had a home robot, what would you want it to do? one of the top three things is like, you know, do my dishes or do my laundry. And I think those are great tasks to automate. I think there are very poor tasks to start with counterintuitively. And the reason for that is laundry and dishes are things for which people are very particular about, and the cost of making a mistake is very high. You don't want to like ruin. So you don't want to start with those. We will not start with those. And that's because, you know, we have existing machines

that do these things. And so you're competing with the dishwasher, you're competing with the laundry machine. But in between those tasks are like a thousand small things that we spend our time doing every day around the home. And it's our hope that solving those things really moves the needle for people. And then, of course, like over time, as the technology improves and I can confidently say to you, we can do the dishes in the exact way that you want, then we'll deliver that experience, but not before. There are people doing the other thing right now with like humanoids and promising there'll be a drop in replacement for human labor and a drop in replacement for like a housekeeper on day one. Seems really hard. That's reaching, that's reaching perhaps.

Yes, yes. We'll see. And, okay. People in AI talk about the Turing test where basically can you have a five-minute conversation with the AI over text and be able to tell that it's an AI is the loose meaning of the term. I was going to ask, when did we cross that threshold? I mean, I feel like we've clearly crossed it. With no celebration.

No celebration, no fanfare in the past few years. And so what is the Turing test for Robo? Well, where my head is drawn to is some of the toy problems in academia, like t-shirt folding, and there are robots and neural networks now can fold T-shirts. And so perhaps in the same way that the T-Ring test was crossed, and at some point in time, we're not sure exactly when. We just know it's behind us already for robotics. If you ever had a robot arms that can fold T-shirts, it feels like we have crossed. That used to be the holy grail of manipulation, because, you know, classically, robots are designed for repeatability and precision

and picking up the same thing in the same place every time. And for that to work, the thing you pick up also has to be rigid. And so clothes are hard because you pick them up and they collapse and wrinkle and fold over on themselves. And so to get a machine that thinks in this very rigid world to work with such malleable items has been like this tough research problem for a long time for anyone to be able to go buy a thing, put it in their home. And without any other instruction, then like, my clothes are in my bedroom. Please put them in the laundry machine and fold them and put them away.

that, to me, would signify, I think we've made it. Yeah. You've talked in the past about how homes are just a horribly difficult environment for robots. Like, you know, if you compare it to, say, a warehouse where there's a decent amount of robotics already, a warehouse is very standardized and standardized to be easy for the robots, whereas homes are an extremely difficult and non-standardized environment with the robots. They have stairs. They have rugs.

They have kids and pets running around. How do you solve for this? I mean, when you put it that way, it sounds like a lot of it. like the Ninja Warrior obstacle course of robotics, basically, and robots have to get through all this stuff. Well, I think what's different today, and actually, like, one of the, you know, there's one takeaway for today in our conversation, I think it would be that robotics today is a completely different field in a different industry than it was five years ago. Like, all of the things we thought, we knew, all the businesses that were tried and failed, all of the tools that have become

best practices and standard are now either, like, worthless or completely different. You know, today, you don't need a robot that's repeatable. You need a robot that's adaptable. Like, powered by neural networks, and if it makes a mistake, or doesn't approach this object to exactly the right angle, it doesn't matter, it can correct that. Like how a robot sees the world. Typically, it would have expensive laser scanners and try to perfectly reconstruct everything that we see

and then use very complicated and hard to tune algorithms to plan how this arm should move through space and time to accomplish a task, and those systems are very fragile and easy to break. If you're willing to completely let go of that and embrace today's tools, You don't have to programming anything. You have to show a robot how to do a task, and it can mimic that,

and let, you know, essentially chat GPT-like technologies instruct these things at a high level. With these tools, going into a home environment is no longer as much of a crazy obstacle course. And that, I think, makes it much more tractable than it was five years ago. For a device to save you work, it's like probably needs to be able to charge itself, clean itself, manipulate stairs. You probably need some minimum set of functionality. And the early versions of Rumba is, I think part about frustrated people is they probably spent more time cleaning the dog poop they spread or it would have been faster for me to just vacuum it myself.

I think it is dangerous to fall below the threshold. Exactly. Creating more work than you are more value. I think that's probably a difference between like a cool product and a delightful product that everyone loves. And I think some of the things you mentioned are what make this problem. As with many AI power problems, like deceptively simple looking from the outside. As in it's easy to have a cool demo and it's hard to have something that actually saves people time in their home. Yeah, yeah.

And I mean, I worked on self-driving cars for a long time. We saw this, too. It's like, well, how hard it could it be to keep the car between the two yellow lines on the road? And then you think about all the things that could go wrong or could happen while you're driving in it. And you start making a list. And then you have, like, three pages of stuff. Each one of those is a big technical problem to solve. And I think, you know, we're already seeing this, but a similar thing is true for a home robot that truly, you know, creates more value in, like, freeze you of work rather than consuming all your time or asking for help every five minutes.

Yeah. Is there a risk that home robotics have a similar character where that final 1% actually turns out to take a pretty long time? It's possible. I think what is different to me are a few things. I think, first of all, just to build a car in a highly regulated environments and very capital-intensive thing is very different than to build a small consumer product. But the other thing that I noticed is several months ago, one of our early prototypes, we do this thing where we just dump a basket full of kids' toys in. a room and say, hey, a robot clean this up. There's like 49 toys on the ground. And over the course of like 30 minutes, it took it a long time as a prototype. It cleaned up all the toys but one. And my thought in that moment was like, you know, what percentage success is that? Like 95%. One nine of reliability. Yet everyone who was watching that was just like, where do I buy this? I need it now.

And so the takeaway from me is like the bar for commercial success for self-driving was like five, six-nines of reliability. And understand that each extra nine of reliability you add, so 10 times better, takes probably 10 times more engineering work. Robotics is perfect for getting overhyped on social media because it's very easy to have a compelling demo that does the numbers on a tweet and all the work isn't getting from that demo to actually working reliably enough to sell as a product. And so it feels like we're almost inevitably in for a hype cycle in robotics. Well, look, I love the demos. I think they're inspirational. They get people excited. They get more people coming into the industry. They get

investment dollars. So I think they have a purpose. I think the problem is when you align customer expectations too squarely on what they see in a demo, or like even as an industry, if the robotics industry sets the expectations too high on a whole for what the next generation of robots will do, everyone's going to be disappointed. And I think it's without a doubt will happen in robotics and not because of any one bad player, more just like the natural way that these things go. What is your iteration loop for working on robotics? Well, I mean, if you have a weekly release schedule or a monthly release schedule, what you're really doing is just withholding all that useful feedback for an arbitrary number of days, right?

Yeah. Or weeks. Doing hardware often requires a lot more upfront thought and planning. There's lead times. There's manufacturing times, all that kind of stuff. And so you have to use one process for that, and it's more schedule-driven. And then another process for software, which is much. more iterative because you can make changes on the fly.

And so bringing those together can be tricky, but if you set it upright, you can actually have hardware development feel like software development. And like simple hacks is you have everyone work in person in an office, you have lots of robots available for developers to push code to in real time, and you make it like as frictionless as possible for people to like try out new stuff on a real machine. And so I don't think you have to walk more than 10 feet in our office to like go from your desk to, you know, running code on a robot. I feel like one of the underappreciated aspects of Elon's playbook for building companies is how much of a commercial thinker he is.

Elon's companies have actually always been surprisingly scrappy. I mean, famously with the Tesla Master Plan, they started with the Roadster, which was deliberately a low-volume car, and then kind of worked their way up to higher-volume cars with SpaceX. They were selling launches to orbit from a very early stage and then progressively, you know, move from Falcon to Falcon Heavy to, to Starship. And so how do you pull the revenue forward as early as possible in a robotics company so that you're not kind of doing 10 years of R&D and then eventually selling a product? I think the way that you do that is by understanding where the absolute frontier is for technology and then understanding what is commercializable in the near term. And there's usually a gap.

It can be a small gap or a large gap. And so if the technology has gone through enough cycles of investment by enough companies or you've done it in-house and it's at the point where now it's affordable, robust, and can work, then I think you can build a business and get to revenue quickly. The problem comes in when you have a business that is premised on or conditioned upon commercializing today's frontier of technology because that will just take time and we don't know if that's like one year or 10 years. So let's talk about self-driving. You co-founded Cruz, which was acquired by General Motors.

is self-driving the most capital-intensive pre-revenue product ever? It's hard to think of a counter-example. I don't have a good one either, I think. It's insanely capital-intensive, and notably the companies who were making these investments were not startups that were just doing this by raising venture capital around. They were large corporations with R&D budgets or basically the pockets that were deep enough, to make strategic long-term bets that could significantly move the needle for the company, knowing that there's a significant activation energy to unlock that future value.

Yes. Yeah, previously it was probably just governmental entities, and only as of recently do we have companies that are willing to spend that much money pre-revenue. The numbers coming out around large language models on the frontier, though, are getting up into that territory. They are getting up into that territory, but I think with pretty clear user economics, where they actually sell a lot of AI these days. Healthier business for sure.

Exactly. Self driving is interesting because it was so unproven when all the CAPEX was required. So self-driving is having a real moment right now as we finally see a lot of deployment on the streets in volume. You worked on this for 10 years. How do your industry views differ? One is, I think, you know, on the regulatory,

story side, and what it will take to truly reach large scale for these businesses. And right now, there's a handful of players who have actually doing robotaxies or driverless trucking. And then the other is, is like these seemingly diametrically opposed strategies of Tesla and Waymo, which everyone likes to talk about. Yes. So the less interesting regulatory one first and get it out of the way. You know, in the U.S., it is still very much a patchwork of legislation. And what probably most people don't see, you know, like Waymo or someone doing is all the groundwork in each new city. And the groundwork that they're doing is because they don't know which small special interest group or union or local governments or city council or state, you know, whatever it is. There's probably two dozen lists of organizations that could meaningfully bring the thing to a halt in that community because there is no federal preemption.

There's no real federal safety standards for autonomous vehicles. And so they have to win that battle with every single stakeholder in every single location. So I hope, and there's maybe some signs of this, that the federal government will get ahead of this and establish that it's pretty clear at this point the data shows that these cars are saving lives and reducing crashes. And so if we think that's important as a government, maybe there should be a federal preemption, and we should ensure that this is open for everyone in the U.S. if that happens, I think we'll see more self-driving cars. Absent that, I think it's going to continue this really slow sort of city-by-city thing.

And, you know, in the interim, a lot of people are going to get hurt because these aren't rolling out faster. And the other big, perhaps a false decadal—I mean that people create is like LiDAR versus cameras. Yeah. What I see is really Tesla as a company who kind of pioneered the into-in neural network approach to self-driving, which I think is the right technical bet. long term, but they put some constraints on it. They said, hey, engineers, like, you can't have the best sensors, like LIDARs and radars, and the sensors have to look good when we put them on the car. Oh, and by the way, they have to cost, like, one-tenth as much as, you know, the guys down the street

who are doing this. So they put some crazy constraints on that. So the right, like, technical vector, but, like, really being, like, held back by the, just the weight of all these constraints that put on the system. But they've been, all of their technical approach from day one seems to have been pointed in the right long-term direction. So that's good. With Waymo, they started off in the DARPA Grand Challenge era of self-driving, which is old school, classical computer vision, classical motion planning. And they built this highly validated, robust system that's now on public roads, and it's great. But they know that it's the wrong technical approach. And they need to move more in the direction of Tesla of more neural networks. And it's wrong technical.

approach because it's too expensive? Because it is just intractable to maintain a 3D map of every square inch of the planet and update it in real time and then expect that every time you go somewhere the map is still accurate on one hand. And also probably unrealistic to assume that every car belts in the future is going to have these giant spinning KFC buckets on the roof. To weigh most credit, I think they know this and they've started moving towards a Tesla-like approach.

The challenge is they've got a validated safety critical system on the road. And the last thing you want to do to a system like that is start changing stuff in it. because that introduces risk. Now that you've a little bit of distance from the cruise experience, what are your reflections? Oh, well, many. I think a lot of people overrotate on things they would change the next time around.

And so the bought company is a small company. I feel like I, like many of my peers, got swept into the dogma of building a Silicon Valley tech company, which is lots of people. And you have the manager, senior manager, director, senior director, VP, hierarchy, all these structures that are designed to get a lot of people to work well together.

And they become horribly inefficient, and it's very easy for them to become bloated. What else? I believe in the in-person environment. I think everyone ran various experiments of remote work during COVID and has ended up, depending on the company in terms of full return to work or remaining some. But that pairs also with the small company thing, right, where I think anyone would say that if you're hiring a small team, it's very attractive. Like, the reason companies tend to go remote or certainly go to multiple offices is just

ultimately you need to hire so many people. diminishing marginal returns to being together. Well, not to harp on this one thing, but there's so many dimensions of it. Like, I don't think most people building companies today have a conscious decision and say, like, well, when we go from 80 people to 400 people, our productivity per person is going to drop by 90%. And are we going to sign up for that and understand that it won't get better until we're past 400 people. I mean, that's the reality of the situation. Not making people like talking about it. And so, you know, I think that's a big one.

You said you're never going to sell a company again. Yeah. Why? Well, it's... Like, why is it not? I'm going to be very careful to sell a company to the right acquirer with the right vision, or a more nuanced statement. Let's flip this around. If you go through all the pain of starting a company,

and you do so knowing that you're going to spend 10-plus years of your life on something, and it's that important to you, and you've told everyone you know about this thing, and you've recruited all the best, the smartest people in the world that you know to work with you on this thing. Why would you stop? Or give up control of that thing. And so I think that there, you know, maybe part of the dogma of Silicon Valley is you like start a company. If you're lucky enough and it's growing fast enough, someone will make an offer to buy it and you sell it and that's victory.

And I think if, you know, financial outcomes are your reward function or fame or whatever it is, then that's great. but I think you know, you talk to a lot of people have gone through that and they miss building the company. They want, you know, they would prefer the robots don't want you to sell. Maybe they're sentient. But, yeah, and I convinced myself when selling crews that, you know, it's the time of North America's largest automaker. And if our vision is to get self-driving cars everywhere, isn't that true to the vision? And I think my heart was in the right place, but I was naive about the ability to get like a large corporation, which is like an aircraft carrier. You can't steer it. You can't get it to change

its focus. It's going to do what it wants to do or what it's already doing. It was naive of me to think that I could kind of hitch a right on that scale and make this thing happen. But, you know, experience has told me now that that is not the path to make the thing happen. How generalizable is this view? Like, do you think fewer founders should sell their companies, or is this a Kyle specific thing? I think selling a company basically means, like, I'm done working on the problem. And there are probably our cases where founders tired of it. They have, their personal relationships are falling apart, whatever. There's an external reason to stop going forward.

Absent that, and if the intrinsic poll is still there, then I think it's a bad idea. In three years' time, how many people is the Buc Company and the Pursensurate engineers? Less than $1195. Oh, my God. So you're very serious about the small team, all engineers. I think the next $100 billion company that's created in 2025, 2025, 26 will be under 100 people. That's quite provocative. How many people are there that have created three billion dollar companies? Not that many.

I've been very lucky. Like I said, good people, good timing. Yeah, yeah. Yeah, if your view is that it's much smaller headcount, we might be in for any. a new way of building companies. I hope so. Yeah. Okay. Thank you. Yeah, thanks for having me.