Latest / Cheeky Pint / The Bot Company founder and CEO Kyle Vogt on home robots and why he’ll never sell another company
Transcript
- John Collision: Okay, so let's back up. What is the pitch for The Bot Company? Kyle Vogt: Well, I don't like doing chores, and I think five to 10 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. I think it's been the holy grail of robotics since I was a kid doing robotics, to have the home robot that does everything. And it was very clearly not possible up until very recently with LLMs, and then end-to-end 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. John Collision: Okay, so this is vacuuming the floor, ironing the clothes, cleaning up after the pets, that kind of stuff? Kyle Vogt: 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. In the same way that it 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— John Collision: Yeah, we had a big spurt for a while. Kyle Vogt: And people were excited about it. There's all these great ads of the person in the kitchen being like, "Look at how much time I've saved with the microwave," and we've gone stagnant since then. John Collision: 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. 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 have those in our homes. 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? Kyle Vogt: All of the things you described—the toaster, the dishwasher—these are 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 $30 for a toaster, maybe not $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… 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 "Do my dishes" or "Do my laundry." I think those are great tasks to automate. I think they are very poor tasks to start with, counterintuitively. And the reason for that is, laundry and dishes are things which people are very particular about, and the cost of making a mistake is very high. You don't want to ruin— John Collision: So you don't want to start with those? Kyle Vogt: We will not start with those, and that's because we have existing machines that do these things. So you're competing with the dishwasher, you're competing with the laundry machine. But in between those tasks are 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, 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 humanoids, promising they’ll be a drop-in replacement for human labor and a drop-in replacement for a housekeeper on day one— John Collision: Seems really hard. Kyle Vogt: I think that's reaching, perhaps. We'll see. John Collision: 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. Kyle Vogt: I was going to ask, when did we cross that threshold? It's hard to say. John Collision: I feel like we clearly crossed it— Kyle Vogt: With no celebration or fanfare. John Collision: Exactly. With no celebration, no fanfare, in the past few years. So, what is the Turing test for robotics? Kyle Vogt: Well, where my head is drawn to is some of the toy problems in academia, like T-shirt folding. There are robots and neural networks now that can fold T-shirts. So perhaps in the same way that the Turing test was crossed at some point in time, we're not sure exactly when, we just know it's behind us, already for robotics— John Collision: Yes. If you have a set of robot arms that can fold T-shirts, it feels like we have crossed some level of— Kyle Vogt: That used to be the holy grail of manipulation, because 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. Clothes are hard because you pick them up and they collapse and wrinkle and fold over on themselves. So to get a machine that thinks in this very rigid world to work with such malleable items has been 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 than "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 we've made it. John Collision: You've talked in the past about how homes are just a horribly difficult environment for robots. 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 for the robots. They have stairs, they have rugs, they have kids and pets running around. How do you solve for this? Kyle Vogt: I mean, when you put it that way, it sounds like the Ninja Warrior obstacle course… John Collision: Yes, exactly. Kyle Vogt: …of robotics, basically, and robots have to get through all this stuff. I think what's different today, and actually, if there's one takeaway for today in our conversation, I think it would be that robotics today is a completely different field and a different industry than it was five years ago. All of the things we thought we knew, all of the businesses that were tried and failed, all of the tools that have become best practices and standard are now either worthless or completely different. Today, you don't need a robot that's repeatable, you need a robot that's adaptable, powered by neural networks. And if it makes a mistake or doesn't approach this object exactly at 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. Those systems are very fragile and easy to break. If you're willing to completely let go of that and embrace today's tools—so, you don't have to program anything, you have to show a robot how to do a task and it can mimic that, and let essentially ChatGPT-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. John Collision: For a device to save you work, it probably needs to be able to charge itself, clean itself, manipulate stairs. You probably need some minimum set of functionality. The early versions of Roombas, I think part of what frustrated people is they probably spent more time cleaning the dog poop they spread, or it would've been faster for me to just vacuum it myself. Kyle Vogt: I think it is dangerous to fall below the threshold of creating more work than you, or more value. I think that's probably the difference between a cool product and a delightful product that everyone loves. I think some of the things you mentioned are what make this problem, as with many AI-powered problems, deceptively simple-looking from the outside. John Collision: 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? Kyle Vogt: Yeah. And I mean, I worked on self-driving cars for a long time, and we saw this too. It's like, "Well, how hard 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, and you start making a list, and then you have three pages of stuff. Each one of those is a big technical problem to solve. I think we're already seeing this, but a similar thing is true for a home robot that truly creates more value and frees you of work rather than consuming all your time or asking for help every five minutes. John Collision: Is there a risk that home robotics have a similar character, where that final 1% actually turns out to take a pretty long time? Kyle Vogt: 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 environment, it's a very capital-intensive thing, it's 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, robot, clean this up." There’s 49 toys on the ground. And over the course of 30 minutes—it took it a long time, this is a prototype—it cleaned up all the toys but one. And my thought in that moment was, "What percentage of success is that? That’s like 95%, one 9 of reliability." Yet everyone who was watching that was just like, "Where do I buy this? I need it now." The takeaway for me is: The bar for commercial success for self-driving was five, six 9s of reliability. And understand that each extra 9 of reliability you add, so 10 times better, takes probably 10 times more engineering work. John Collision: Robotics is perfect for getting over-hyped on social media, because it's very easy to have a compelling demo that does the numbers on a tweet, and all the work is in getting from that demo to actually working reliably enough to sell as a product. It feels like we're almost inevitably in for a hype cycle in robotics. Kyle Vogt: 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 even as an industry, if the robotics industry sets the expectations too high on the whole for what the next generation of robots will do, everyone's going to be disappointed. And I think it without a doubt will happen in robotics, and not because of any one bad player, more just the natural way that these things go. John Collision: What is your iteration loop for working on robotics? Kyle Vogt: Well, I mean, if you have a weekly release schedule or monthly release schedule, what you're really doing is just withholding all that useful feedback for an arbitrary number of days 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. 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. Bringing those together can be tricky, but if you set it up right, you can actually have hardware development feel like software development. Simple hacks: 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 as frictionless as possible for people to try out new stuff on a real machine. I don't think you have to walk more than 10 feet in our office to go from your desk to running code on a robot. John Collision: 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 worked their way up to higher-volume cars. With SpaceX, they were selling launches to orbit from a very early stage and then progressively moved from Falcon to Falcon Heavy to Starship. How do you pull the revenue forward as early as possible in a robotics company so that you're not doing 10 years of R&D and then eventually selling a product? Kyle Vogt: 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. There's usually a gap. It can be a small gap or a large gap. So if a 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 one year or 10 years. John Collision: Let's talk about self-driving. You co-founded Cruise, which was acquired by General Motors. Is self-driving the most capital-intensive pre-revenue product ever? It's hard to think of a counterexample. Kyle Vogt: I don't have a good one either. I think it is insanely capital-intensive, and notably, the companies who were making these investments were not startups that were just doing this by raising venture capital rounds. 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. John Collision: Yeah. Previously, it was probably just governmental entities, and only as of recently do we have companies that are going to spend that much money pre-revenue. Kyle Vogt: The numbers coming out around large language models on the frontier, though, are getting up into that territory. John Collision: 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— Kyle Vogt: Healthier business for sure. John Collision: Exactly. Self-driving is interesting because it was so unproven when all the CapEx was required. 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? Kyle Vogt: One is, I think, on the regulatory side, and what it will take to truly reach large scale for these businesses. Right now there's a handful of players who are actually doing robotaxis or driverless trucking. And then the other is these seemingly diametrically opposed strategies of Tesla and Waymo, which everyone likes to talk about. So, the less interesting regulatory one first, and get it out of the way. In the US, it is still very much a patchwork of legislation. What probably most people don't see Waymo or someone doing is all the groundwork in each new city they do. The groundwork they're doing is because they don't know which small special interest group, or union, or local government, or city council, or state, 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, 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. So if we think that's important as a government, maybe there should be federal preemption, and we should ensure that this is open for everyone in the US. If that happens, I think we'll see more self-driving cars. Absent that, I think it's going to continue this really slow, city-by-city thing. And in the interim, a lot of people are going to get hurt because these aren't rolling out faster. The other big, perhaps false dichotomy that people create is LIDAR versus cameras. What I see is really Tesla, as a company who pioneered the end-to-end 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, 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 one-tenth as much as the guys down the street who are doing this." They put some crazy constraints on that. So the right technical vector, but really being held back by the weight of all these constraints that were put on the system. But 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. John Collision: It's the wrong technical approach because it's too expensive? Kyle Vogt: 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 it’s also probably unrealistic to assume that every car built in the future is going to have these giant spinning KFC buckets on the roof. To Waymo's 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. John Collision: Now that you have a little bit of distance from the Cruise experience, what are your reflections? Kyle Vogt: Oh, well, many. I think a lot of people over-rotate on things they would change the next time around. So, The Bot 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. John Collision: What else? Kyle Vogt: 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— John Collision: That pairs also with the small company thing, 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 that the diminishing marginal returns to being together— Kyle Vogt: Well, not to harp on this one thing, but there's so many dimensions of it. I don't think most people building companies today have a conscious decision and say, "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 many people like talking about it. I think that's a big one. John Collision: You said you're never going to sell a company again? Kyle Vogt: Yeah. John Collision: Why? Kyle Vogt: Well, it's— John Collision: 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? Kyle Vogt: 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 to work with you on this thing, why would you stop or give up control of that thing? I think that maybe part of the dogma of Silicon Valley is: You 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. I think if financial outcomes are your reward function, or fame, or whatever it is, then that's great. But I’ve talked to a lot of people who have gone through that, and they miss building the company. They would've preferred to keep it going. John Collision: The robots don't want you to sell. Kyle Vogt: Maybe, if they're sentient. I convinced myself when selling Cruise that it's the… At the time, it was North America's largest automaker, and if our vision is to get self-driving cars everywhere, isn't that true to the vision? I think my heart was in the right place, but I was naive about the ability to get 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 hitch a ride on that scale and make this thing happen. Experience has told me now that that is not the path to make the thing happen. John Collision: How generalizable is this view? Do you think fewer founders should sell their companies, or is this a Kyle-specific thing? Kyle Vogt: I think selling a company basically means “I'm done working on the problem.” There probably are cases where a founder is tired of it, their personal relationships are falling apart, whatever. There's an external reason to stop going forward. Absent that, and if the intrinsic pull is still there, then I think it's a bad idea. John Collision: In three years’ time, how many people is The Bot Company, and what percentage are engineers? Kyle Vogt: Less than 100 [people], 95 [percent engineers]. John Collision: Oh my god. So you're very serious about the small team, all engineers? Kyle Vogt: I think the next $100 billion company that's created in 2025, 2026 will be under 100 people. John Collision: That's quite provocative. How many people are there that have created three billion-dollar companies? Not that many. Kyle Vogt: I have been very lucky. Like I said, good people, good timing. John Collision: But still, if your view is that it's a much smaller headcount, we might be in for a new way of building companies. Kyle Vogt: I hope so. John Collision: Yeah. Okay. Thank you. Kyle Vogt: Thanks for having me.