在艰难市场打造硬科技：Kyle Vogt谈Cruise、Twitch与The Bot Company

Today, we're joined by Kyle Vogt, a serial entrepreneur who has helped build some of the most influential tech companies. He co-founded Twitch, shaping live streaming, and Cruise, the autonomous vehicle company acquired by GM for $1 billion. Now Kyle has launched The Bot Company, a startup focused on building consumer robots. Kyle, welcome.

Awesome. Let's get going.

Obviously, you've done a variety of different things over time. Everything from co-founding Twitch to starting Cruise, and now you're working on a new startup. Can you tell us a little bit more about your Cruise experience? I think that whole era was incredibly formative for everything that's happening today, and I'd love to get your perspective on why you started Cruise when you did, how that all evolved, and how that's informing what you're doing now.

1. Cruise Starts With A Retrofit

Sure. We can go back to the beginning. This is 2013, and back then, there wasn't really self-driving car technology like there is today. There was just Google working on its self-driving car project. Rumor had it that they had spent about $100 million, and they had the world's best engineers, so going after something like that was a little bit crazy.

Even after having worked on Twitch, you'd think that would be enough credibility that, as a repeat founder, I could go back and raise money. But it turned out that even this was a crazy enough idea, and Twitch hadn't been acquired yet, that I had a hard time. I had to scrape the bottom of the barrel to raise money. I think I pitched 120 investors over the course of probably a couple of years to raise all the money we needed.

Our thesis back then was very simple. Instead of going directly after what Google was doing with self-driving cars—they were trying to make the ultimate self-driving car, I think, as a moonshot—we took the lean startup approach. What's the lean startup approach to this? Can you build something that has the minimum quantum of utility, is maybe lower-cost or easier to execute, so you can get to market more quickly and move from there?

We started with a retrofit system where we would take a regular car, put some sensors on it, put a computer in the back, and get it to drive. We got that working pretty quickly, like an early version of Tesla Full Self-Driving.

I think that was for just 1 car model, too, right? That was a BMW or something at the time.

That is the challenge with a retrofit business. Without the blessing of the carmakers, you have to reverse-engineer protocols and figure out how to attach motors to steering wheels. So it wasn't necessarily sustainable, but we were still going to try to figure that out.

I'd say that peaked around the time we went to YC Demo Day. Sam Altman was in the car, actually, and we turned it on and rode it to Demo Day. We worked on that product for about a year and a half and then realized that we had done enough technically that maybe we didn't have to take the lean startup approach. Maybe we could just go straight after the big fish, which would be building robotaxis.

Around that time, Uber and Lyft had risen in popularity and were becoming household names, with talk of going public and all this kind of stuff. They had this big hole in their unit economics, which was paying the drivers. Suddenly, there was a strong market pull for self-driving technology, whereas before it had been seen as just a cool sci-fi thing.

We were able to raise some money from Spark Capital and go straight into that. Within a year of that, I think we were acquired by GM. We had working prototypes driving around San Francisco, obeying traffic lights, changing lanes, and going from point A to point B with an iPhone app back in 2015.

That's pretty amazing. How do you think about the different approaches that people are taking today? There's Tesla on one side, with a very specific approach, moving everything toward things that are more camera-centric but training on a richer set of sensors and approaches. There's the Waymo approach, which is much heavier on the hardware side in terms of what's actually on the vehicle.

Both seem to be doing very interesting things. One is robotaxis, and one is still mainly building cars. How do you think about the different approaches, both from a business model perspective and from a technology perspective?

2. Commodity Sensors Change Autonomous Driving

To be fair, Elon nailed it from a business model perspective. He's been making billions of dollars of profit while developing self-driving cars, whereas everyone else has been burning billions of dollars to try to get to basically the same point.

In the end, when you start with custom vehicles with lots of sensors that are really expensive and make them work in a constrained environment, while Tesla starts with an unconstrained environment and low-cost sensors but doesn't quite work without a driver, they're all trying to get to the same spot: low cost, working everywhere, with commodity sensors. Those are different paths to get there. I think Elon won that hands down.

What do you think of the criticism that you can't get there—that you can't get to full self-driving from mostly self-driving?

That statement is almost certainly wrong given a long enough time span. Again, going back to Elon's approach, he doesn't have to finish by a certain date or run out of money. He's making money along the way. I think the only risk is that customers get fed up and rage-quit his program, but they're getting something they like along the way: Tesla Full Self-Driving. So I think that's the right approach.

In 2013 for sure, in 2015, and even in 2018, it really wasn't viable to have a full driverless car that just used cameras and low-cost sensors. It just wasn't. The technology was not there.

I think now, if you take a fresh look at where we are today, with large language models, generative models, and other things, that class of technology applied to the classical challenges of perception for autonomous driving, and even motion planning for autonomous driving, has completely changed the game in terms of the magnitude of compute that you need and the expense of that.

I think now, with cameras, as long as you have sufficient redundancy, low-light sensitivity, and some robustness there, you can extract beautiful, really accurate depth data from a single camera image—not even stereo. Those models are getting better every day.

If you're making a bet on the right technical approach in 2025, it does not involve a bunch of expensive lidars or exotic sensors. It involves the most commodity, high-volume, readily available sensors you can get, and probably just several more of them than you'd find on a typical driver-assistance system. So I think that's the path from here on out.

Is there anything else you think is lacking from a technology perspective, either in terms of hardware or just scaling models? As you know better than anyone, everybody started moving toward end-to-end deep learning over the last year or two, and that's really made a big difference. But is it just scaling that up, or is something else lacking?

In the end, the approach will be an end-to-end-type model. It's hard to put it in a bucket of end-to-end or smaller models because there's such a spectrum in between, and everything I've seen is a mix and match of various technologies.

If I look at the limiting factors, at least on the hardware side, previously it had been hard to get high-performance compute in an automotive, high-temperature-range, safety-critical environment. Cruise made custom chips. I'm sure Waymo makes custom chips, and piecing together things from the supply chain to solve that is a little challenging. So there is room for more high-performance compute automotive silicon, and I've seen some things happening in that space. That's one.

I'd say the other piece is that most robotaxi deployments I've seen rely on some form of remote assistance. There's a question of how you get reliable connectivity to a vehicle from anywhere using multiple cellular networks, as Cruise has done—and Waymo, I'm sure, and others do. That works, provided you have cell phone coverage.

I think the missing piece may be Starlink or something similar, where you can have always-on connectivity between Starlink and maybe a cell phone and some other fallback. I think that really opens up the opportunity in terms of the number of places you can deploy robotaxis.

Whereas before, it was an open question what you do if you're driving down Highway 1 in California and there's no cell coverage there. Should you still have an AV on that road, given that if there's an issue or a customer needs some help, you literally can't get in touch with them?

In retrospect, was there a right year to start a self-driving car company? You were so ahead of the ball on this.

It takes a long time to spin up the automotive pipeline and everything.

So probably circa 2020 or so would have been the right time to get started. Around now, I think you'd have a combination of hardware and software that's mature enough. If you have a nimble enough engineering team that's able to adopt new technologies when they pop up and quickly pull them into the pipeline, you're actually well positioned even if you started a while ago and your tech stack was based on some older technologies. If you have all the infrastructure in place for validation and testing, training models, and deploying on public roads with test drivers, I think you can go a lot faster, even if you have to rip out and change some of your tech stack to adapt with the times.

One thing I've heard opposing viewpoints on is the autonomous vehicle market in China. One point of view is, well, it's not that real, and it's mainly teleoperation, and it's a little bit more sizzle than steak. The other opposing view is, well, actually, they've advanced dramatically, really rapidly. There are fewer safety constraints, so you can do more, try more, et cetera. The models and approaches there are at least at parity with the leading contenders in the Western world. Which of those two views do you subscribe to, or how do you think that market will evolve?

From what I've seen so far—and I don't have a lot of inside information, just from what I've seen in videos online and other things—it does still seem like there's a lot of teleoperation. I think even someone like Tesla may start off with a 1:1 ratio of remote operators to vehicles. Cruise and Waymo probably started off pretty close to that, and then I think over time moved to a smaller ratio.

In the interim, to get the deployment numbers up, get more experience, and accelerate data collection, people are brute-forcing it, which means there's probably a lot of remote operation. I actually think that's fine because it doesn't take much. Once you get to 50% remote assistance, or even 25%, you've already reduced the labor costs by 75%. And so you're almost already at the diminishing returns point.

It sounds kind of crazy to say, "Oh, if there's 1,000 AVs out there, there may be 250 people monitoring them." But that's actually not crazy from a cost and unit economics standpoint; it actually makes a ton of sense. A ratio of 1:4 is trivial, I think, with today's technology. Over time, you could get to 20:1 or 50:1, but you're just talking about single-digit points of margin at that point.

The real benefit is that getting to anywhere below 1 full human per car makes the economics of this really good and, I think, puts you on a pathway toward better safety for the vehicles, because they're primarily driven by a robot that has great reflexes and is going to avoid situations, but then also lower costs to consumers over time.

You did amazing work on Cruise, and then you decided to start another company, which I always think is a really brave endeavor, because anybody who's been through multiple startups knows how painful and terrible it is. Could you tell us a little bit more about the impetus behind The Bot Company and what you're doing there?

3. The Bot Company Builds Home Robots

Yeah. We talked about this a little bit when I was making that decision—what to do next. I did some soul-searching and determined that I'm just a builder. I like building things, and sitting on the sidelines or helping other entrepreneurs or doing something else, I think, would be fun, but not quite scratch that same itch. I'm 39. I feel like I got at least one more startup in the tank. So the question became what to do.

I look back on my career. This is my, I guess, depending on how you count it, third major startup. The first one was Twitch and Justin.tv, straight out of college, and that was just doing anything. Doing a startup and trying to make it work was the priority, and that ended up being video games and entertainment.

The second time around, for Cruise, after doing entertainment, I decided I wanted to focus on impact. So what's something where we can use technology to meaningfully improve people's lives? Self-driving cars: they save lives and give you tons of time back. That was squarely in the impact category.

Third time around, I definitely care about impact, but also fun. So it's working with people I like on problems I like, really challenging technical problems, and building amazing products. And so we're building home robots.

The impact side of that is one of those things that's hidden in plain sight. There's only 24 hours in a day, and if you're sleeping for 8 hours and working for 8 to 10, there are precious few hours left that are actually your time. People spend a surprising amount of that remaining time doing essentially unskilled labor, acting like robots every day: making the bed, doing the dishes, folding the laundry, and picking up toys after your kids. These are not things that make us human. These are actually things that detract from our humanity, and they're the perfect criteria for that reason to be automated by machines.

I think when you describe that to people—"Oh, you don't have to do all those things anymore. There's a machine that could do this for you"—it clicks instantly. People are like, "That is so obvious." To the point where I think in 5 years, maybe 10 years, it will seem as insane to have a house without multiple home robots as it would be to have a house without a sink, a laundry machine, or a toilet. These are going to be critical things that, if you can afford them—and we want to make them really affordable—are just going to seem like extreme common sense: Why wouldn't I want to have the time in my home be my time, not consumed by these chores?

I just thought that analogy was really interesting because we never really think about plumbing as a technology, and it is. And to your point, up until recently, for most of human history, we had no running water. You'd walk down a hill with a bucket, and you'd bring it into the house. I actually think that's a fascinating analogy, because nobody really talks about some of these things that are actually technology, and the degree to which we now just take it all for granted.

Yeah. Plumbing and electricity were sort of turn-of-the-century things. And then I'd say in the 1950s and 1960s there was a resurgence around home appliances. There are some great advertisements from the 1950s and 1960s. If you look back, it's the 1950s: the housewife is standing in the kitchen, and there's the microwave, the dishwasher, and all these new appliances that make it so they have more time and can do more and be more productive.

The last time we had a surge of excitement and progress in our own homes was 50 to 70 years ago. So I think it's time to revisit that. Going back to the robotics side, actually pulling this off has basically been the dream for people working on robots since the dawn of robotics: to build a robot that can go to the fridge and get you a beer or something like that. You sit on the couch. It's the dream for nerds working on robots, and that's obviously a tiny subset of what you'd want a household robot to do.

But that's really hard. Why is it so hard to just have a robot open a fridge, get a drink, and bring it to you? The reality is that it's similar to self-driving cars in some regard, where it's a very unstructured environment. Every home is different. Everybody organizes their home in a different way. The layout is different, the objects in the home are different, and how they live in their home is different.

Having a robot that lives in this unstructured environment is the polar opposite of a factory assembly line, where everything is rigid, repetitive, and precise. In a home, it's sloppy and changes every day. Using classical approaches, where you have computer vision and you're trying to reconstruct 3D objects or fit to a map, would make this a really, really challenging and computationally intensive problem.

Moving to more modern techniques like end-to-end learning or imitation learning, even reinforcement learning, now, if you can teleoperate a robot and demonstrate how to do something, or collect data from humans in some way or from internet videos, you can imbue a robot with a sense of common sense and an ability to make sense of these unstructured environments. On top of that, you can talk to the robot in natural language using your voice, rather than typing into an app or on a keyboard.

So I think you asked when the time is to start a robotics or a self-driving company. Maybe that was 2020. I think for home robots it feels a little bit early. So now is definitely the time, in my view.

How much of what you did at—or that people have learned at places like Cruise or Waymo or others—is also useful in the context of home robots? In other words, what sorts of things overlap, and then what things are just completely different or new? People would often talk about driving environments as similarly chaotic and messy, and the canonical example is always a kid suddenly chasing a ball across the road, or things like that.

Is it even more difficult in the home? Is it less difficult, or is it more structured? I'm curious about the analogies that could be drawn there, if any.

To start with, the big difference between the two is that, for a driverless robotaxi, you basically have no product until it achieves superhuman safety performance. Whatever you establish that as—just slightly better than humans or 10 times better—I think most driverless cars that are on the road today fall somewhere in that category. And to get there means there's no MVP; there's no launching with something that's partially useful. It's like you have to reach that human safety performance.

On public roads, it's hard to constrain the environment to the point where you make the problem much easier. You can operate at night or in sparsely populated areas, but the reality is, just like you said, at any moment, anywhere, a kid could dart out in front of that vehicle. And so you need a high number of nines of reliability to have any sort of product.

I think in the home and most consumer applications, and even most industrial applications, safety is still critically important, but the bar that you need to reach, the functionality that you need to reach, or the constraints you can put on the system enable you to launch a product much more quickly. And so I think that's one big difference.

On this topic, how do you imagine deployment to work? You're obviously saying, "Hey, Tesla had the right path here." Is there a Tesla analogy where you make billions along the way? It's not obvious. Are there constraints you can put around it where you have teleoperation, or just a couple of tasks, or a more constrained environment in the messiness of a home?

There's a number of ways to attack that. Approaches I've seen are like, you sell a really high-priced robot today, like a humanoid or something resembling a human that's fully teleoperated, and you just tell someone, "This is going to cost, I don't know, something crazy like $50,000 and $1,000 a month." But it's the first robot you can buy that will do stuff in your house. I think that's one approach to try to make money along the way.

I think your market size is pretty small doing that, but that's a viable approach. And the other side would be to sell robots that don't fulfill the promise of a household robot that does all your chores, but do little bits of useful things.

I just saw at CES this year that they have little iRobot Roomba-type things with a tiny little hand that could come out and pick up a sock that's in the way. These are incremental approaches to sell products, get data, and hopefully learn what it would take, train models, or try things to work up that ladder, I guess, to the holy grail. The holy grail would be a robot that takes the place of your butler and your housekeeper and just about anything else that you would ever want, if you could have an infinite staff of people or robots doing things in your home.

I guess while we're on the analogy to self-driving, if you look at what happened from a market-structure perspective, there were originally dozens of startups that raised collectively billions of dollars. And one could argue that the end winners, or the things that actually somehow worked in the market, were 2 incumbents: Tesla and Waymo; Cruise/GM; and then, maybe to a secondary extent, Applied Intuition, which is building more general software for cars and things like that. Most of that market didn't end up with the outcomes one would've hoped for.

Do you think there's going to be a similar sort of shakeout here in robotics, and do you think there will be incumbent bias? Do you think there's a lot of room for startups? How do you think about how that market will evolve?

4. Robotics Faces A Startup Shakeout

I think for sure, both in AI generally, like pure software companies, and also in robotics as the next wave is starting, there will be that bubble effect. This is just how the Silicon Valley ecosystem and venture-capital market works. There are either a couple of big rounds that get everyone excited, and then other investors start throwing money into the same space because they see the markups happening quickly, and that follow-on effect kind of floods the market. When there are a lot of investors talking about funding these companies, more people drop out of their PhD programs or quit their jobs to start a company.

And I think the majority of those companies, I would say, are low quality in that they're founded by someone who's only half into it, or by a founding team that's half into it and half hedging—going back to work or whatever. Maybe they're hoping it's a get-rich-quick thing, or the founder-market fit or founder-product fit isn't there, even though they're technically smart.

I saw this in the self-driving wave: people who are really brilliant academically but have the wrong mentality—not a product-centric mentality. They will leave their academic program because they want to commercialize their research, which to me is a huge red flag because that means you're saying, "I'm not going to be flexible on how I solve the problem. I'm going to force my solution, you know, like a square peg into a round hole." And I think that can be very problematic for a startup when you're constantly wrong and need to adapt to whatever you see.

So most of those companies will be low quality, and as a result, we'll say that the bubble popped and there's a huge wipeout in the industry inevitably, whether it's robotics or AI. But I think really what it was is that there were a handful of good companies that were started during that time and before, and those companies did just fine. I don't think they'll be affected by the collapse. It's just all the noise, the follow-on, and the hype and mania that follows that sort of gives the impression that these things are collapsing or not viable, when in reality, I think there are a handful of companies doing really good work.

I don't know if they're necessarily limited to the incumbents, but that is possible, especially in hardware. It's really hard. But there were companies like Aurora and Zoox: one of them went public, and one of them was acquired by Amazon and so has the resources to keep going. Cruise fell into that category. So these were not incumbents. These were companies that were started from scratch during the beginning of that self-driving car cycle and are enduring and hopefully do well.

Thanks for that overview and explanation of what happened in the industry. You made a great point about Zoox and others also being among the companies that either had an exit, worked in different ways over time, or are continuing to be built.

One other thing that a lot of people do in the context of robotics—and you can see this maybe even being accentuated more in the context of a home robot—is they ascribe personality, or they project personhood, onto these machines. Is that something that you think is worth leaning into? Is it something that's worth avoiding—the anthropomorphization, I can never say that word, the humanization of these devices? How do you think about that as somebody who's actually building things that will be in the home with consumers and may get interpreted in different ways by the customer?

To start with, the analogy in the self-driving-car industry was interesting because we named our cars. Every car had a name, and people would personify it when it came up. A car in itself doesn't look like a creature; it looks like a car, something that you drive. But once it starts moving on its own, your brain plays some tricks on you and starts treating it like it's an entity or a creature of some kind.

And so you can try to pretend that doesn't exist, and then you have this weird cognitive dissonance where you're saying it's a machine, but it seems like it has its own consciousness or life force in some way. Or you can lean into it and acknowledge that and find a way to integrate it in the right way. The challenge, I think, with anthropomorphism—I think I said that word right—

You're just showing off now.

Yes, seriously.

That was the hard part of The Bot Company.

Yeah, but too much anthropomorphism can imply a set of human-like behaviors that don't exist in that product. And I think Rodney Brooks wrote an essay about this, basically saying that with robots in particular, their appearance sets the expectation for what the product will do.

If you make something that looks exactly like a human—and in fact, the more human-like you make it, the higher the expectations I think the average person will have for that machine—they'll say, "It looks like me, it walks like me, it has a face and talks like me, so it must be capable of doing all the things that I can do."

And today, in 2025, I think it would be a leap of faith for any company to sell a humanoid robot or something like that and imply that it can do all of these things, because we're still, I think, many years out from that, at least from meeting those expectations.

And so I think there's a lot of thought that can go into the design of a robot, the form of a robot, and other things to try to match the expectations you have for a robot when you see it to what it can actually do. You can even go the other direction. Instead of overpromising by showing a humanoid, maybe you can do something in the other direction and surprise people with how much it can do. That's how I think about it personally. I like to surprise and delight customers rather than set them up for disappointment.

On this front of consumer acceptance and expectations, are there lessons that transfer from self-driving to home robots?

One thing I saw in self-driving, which I guess you could say is intuitive, but it was still very striking, was that most people, on the whole, were very skeptical of self-driving cars. About 75% to 80% of people were like, “I’m never going to trust one of those things.” That dropped to about 20% or 30% after 1 ride.

Amazing.

And so it's one of those things where you simply do not believe it. The more transformative and the more like science fiction a technology feels, the higher the skepticism will be for that kind of thing. Anytime you're doing something new, whether it's self-driving or a home robot—which, let's be honest, sounds like science fiction—I’d love to have that, but can this be real? That's the question.

I think there will be a barrier. There will always be skepticism, and people will say this is impossible, or it's never going to scale, or whatever it is. Maybe that's the introduction with any new technology. What I would say, looking back, is that the most powerful thing to overcome that is people using the product and telling other people about the product, saying, “I rode in this thing,” or, “I tried this thing, and it’s real. You’ve got to try it.”

And so I think leaning too much into classical marketing and trying to tell people what this thing will do, what its specs are, and all that is very different from hearing from someone you trust: “I have this thing in my home,” one of your most intimate spaces, “and it’s working. I love it.” That's how I think about it for something like this, where it's just hard to go straight at people who are skeptical and just don't believe that a science-fictional thing can exist and try to convince them through any medium other than having them try it themselves.

What's the timeline for that? You mentioned in passing a pretty important claim: maybe 5, 10, or 20 years until everyone who can afford them expects robots in the house like they expect home appliances. What changes that timeline between the 5-to-20-year span, or whatever it ends up being?

Well, I'm in my office, so basically, when I get off this podcast and go back to work, that'll—

Okay, so we should let you go, is what you're saying. Yeah.

No, not at all. To me—and I felt this way in 2013 when I started Cruise—it seems like the basic building blocks are there. I can point to all the challenges for a low-cost home robot, and low cost is important to me because I want a lot of people to have this. I can point to all the technical challenges that, at least today, I think we're going to face, and I can point to a technology we've either built in the last year, some research that came out, a product, or a chip that's coming in the pipeline—whatever it is.

I can see, from where I sit today, the path to put these things together and assemble a great product and a great product experience. So I think it comes down to execution: how quickly and how well those things are put together. And then the big question, which you always face in something like this, is: What are the unknown unknowns? What can I sit here today and simply not see because we haven't put enough robots in homes and haven't tried this out?

I think there could be some new discovery that happens. Maybe it turns out that people are not happy with a home robot unless it does X, or whenever a robot does this other thing, it makes people never want to use it again. And so we're kind of early in our own journey to figure out what those things are. As far as I've seen, there's no one else really doing this. There's a large cloud of uncertainty in front of us, and that's why I can't be more specific on the timeline.

It could be 1 year, or it could be 20 years. The best way to figure that out is just charge forward and try to discover it as quickly as possible.

All of those are really exciting as timelines. We talked about Chinese AVs. How do you think about manufacturing and supply chain, given competition with China and Chinese robotics companies?

5. Home Robotics Faces Global Competition

Yeah, I've spent a lot of time thinking about that. The sentiment I've heard in robotics—or, I'd say, in the hardware company space—is almost that you shouldn't even try.

Yeah.

Because if you're just making a widget, you'll make it using U.S. engineering, which is $100,000-plus a year for an engineer. You're going to go to U.S.-based machine shops and U.S.-based tooling, all this kind of stuff, and it's going to be slower. You're not going to iterate as fast, or you're going to pay more for it, and the quality may or may not be as good. So don't even try.

And I think it is possible to be a global company today and have a manufacturing footprint in another country, use contract manufacturers, and tap into existing supply bases. It takes more work, and you have to be willing to travel and pound the ground and make these connections and get access to these things, but it's not impossible. I think probably, for a U.S.-based company, you're going to have a hard time competing on pure engineering services. If that's all you've got—if you're just doing mechanical engineering and cranking out products—you’re going to have a hard time on the margin side and have to build a brand instead in order to create that margin.

I think when you get into more complex machines, and I saw this with self-driving, even though they may be doing a lot of teleoperation and other things there, I do think that China is still years behind the best U.S. companies for self-driving. It's been my experience that anytime you have a sufficiently complicated technical problem on the software side where you can't just copy it by measuring something and then recreating it in CAD, or it has to do with taste—the product experience is more than just a light switch where you flip it on and off—then it becomes a little bit harder to quickly copy that and commoditize it.

Even if you're a fast follower, if you are aggressive enough as a company and can maintain a lead and keep innovating and keep building new products, I think there's room to be a hardware company in the U.S., provided that you take steps to ensure that if your cost is higher than a potential Chinese competitor or somewhere in Asia, it's not that much higher, to the point where you can win on the merits of your product and brand and other things.

But in a place like home robots and other things, this is the Wild West. There's no established stuff to copy. We've got to build a lot of stuff ourselves, and I think it'll be interesting to see how this plays out too, especially if these end up being connected devices and they're constantly getting new software updates with better models on them or whatever it is. I can see a bunch of angles in which these companies are very durable, whether it's us or another one.

You've already been through the wringer once on the regulatory front with AV. If you could wave a magic wand—we'd just call Trump right now—what do you think is the right policy approach to make sure we have a competitive domestic robotics industry, if that's relevant?

First of all, I think there should be tons of regulation on AV. It reminds me of what's happening in the AV space, and companies like Cruise got wiped out. Companies like Waymo are growing or expanding much more slowly than they should based on the merits and the safety of their technology.

It reminds me of when the first airlines were formed in the U.S. and there was no FAA, there was no regulation. If you had basically any kind of plane crash, you would get sued out of existence, and you'd just be wiped out. Many of the first airlines are no longer in existence because of this.

And so the FAA was created because the government decided that we should have airlines, and if they kept going out of business, no one would start an airline anymore. The approach was to create the FAA, monitor airplane manufacturers and airlines, make sure they meet safety criteria, and, in exchange, give them reasonable protections and limits on liability so they can actually operate in a society like the U.S. That hasn't happened for self-driving cars.

And so I think the only companies that stand a chance today are the ones that can afford to take on that liability because they're a giant tech company that makes money in other places. Otherwise, it's very bleak.

So I would recommend that for AVs and for home robots, a similar thing is true. There are no regulations right now on cybersecurity, for example. You can have a Chinese-manufactured robot in your home with cameras and a microphone running and sending that data who knows where. In fact, many of us do, and that's not regulated. That's not inspected by any government agency, and I think that's a major concern.

On the safety side, I would love to see that, too. There are lots of best practices from the industrial robot industry, but they're not a great fit for home robots. There are lots of good best practices for other consumer products, but home robotics is a bit of a vacuum. I think that generally, regulation is a very, very good thing for companies operating in environments like this, especially ones that are unpredictable.

It encourages discussion of best practices and oversight. All of these things lead to better outcomes for both companies and consumers, I think. So in terms of regulation, even though I think the Trump administration is anti-regulation, it's actually a necessary enabler to get these industries going, as odd as that sounds.

It seems like there are some circumstances where your point—that a clear regulatory framework helps a lot—applies. For the crypto community during the Biden administration, a lot of what they wanted from the SEC was just guidance in terms of what to do, and then everybody was going to go do it. It was that ambiguity that really hurt.

The argument I've heard around drones in particular is that, because of the FAA, the US is now behind on the drone side, and China was really able to get a leg up. If you look at everything from drone shows to just what DJI drones can do, it's very low-cost, and they're being used for military and other applications in Ukraine and elsewhere.

There's also the argument that the FAA overregulated drones in airspace, and that prevented the US market from evolving down the proper route. So I'm curious how you think about that balance between too much and too little regulation, and how that may apply to the robotics world given what's happened with drones.

Yeah, it's a good point. I agree. I would love to see more transportation innovation generally. In aviation, there were tons of startups working on electric takeoff and landing airplanes, and they've all sort of fizzled out, it seems like, or bumped into these FAA certification challenges.

As far as I understand, there's still no pathway, even today, to make a pilotless plane of any kind. There are baby steps that we're taking, but I would love to see these 2-track approaches where you have a very mature track for existing industries and technologies, like passenger airlines today, and then an innovation track where you're almost encouraged to innovate. There could be grants or other programs to spur innovation in this category, and then maybe a phased release process to go from a working prototype to being regulated but able to operate at scale.

I think it would be okay for us to sit down and say, "Here's what we expect to see at each level of maturity. Provided you demonstrate to us that you meet that level of maturity, we'll progressively open up the regulations or the areas where you can operate."

This is a standard thing. This is what we did in self-driving cars, but also with new airplanes, like the Boom Supersonic jet that just made history. They started off with a low and slow flight and then, as they saw the data check out, ratcheted it up until they hit supersonic.

I think regulations are there to prevent irresponsible people from going zero to supersonic. But there are plenty of responsible people willing to take the stepped approach, provided that there is a pathway to do so. I think that's the right balance between regulation, overregulation, and none at all: having these phased approaches.

What's the number 1 thing that you do differently given the Cruise journey, besides, I guess, having more fun and not selling it to GM?

6. Cruise Lessons Favor Smaller Teams

Yeah, I learned a lot. Big companies that have their core business in another domain doing an acquisition that's not in that domain—selling cars to people who buy pickup trucks and SUVs in the Midwest versus robotaxis in urban environments—these are not compatible things. When push comes to shove, they're going to pick one over the other.

We got completely decimated by GM's lack of priority and then completely abandoning Cruise. So, lessons learned—plenty. First is, I'm never going to sell another company again, ever. Maybe it will IPO or something like that, but there will never be an acquisition in my life again.

The reality is that if I'm working on something today, I'm working on it because I think it's important and I care about it. The day you sell the company is the day that you have to let that go. If, by definition, this is something I care about, I'm not going to let it go.

The second thing is team size. Along with many other people who started companies around the same time, I fell victim to the Silicon Valley dogma of traditional engineering management for Silicon Valley companies. That's hiring the VPs first, then having them hire the directors, who hire the senior managers, who hire the managers. There can't be a ratio of more than 8 to 1 for fan-outs, and you do performance review cycles and all these things.

That creates bureaucracy and structure, and it creates communication gaps between the people actually doing the work and the people making the decisions. The solution, of course, is just to keep companies very small. Have fewer employees. Make every seat count. Get the absolute best person in every role, in every seat, and just never grow the company to be so large that it crumbles under all the structure, bureaucracy, and politics that seep in.

These are natural things that happen to large groups of people when they organize together in companies. Traditionally, that would mean you have to limit the scope of what you want to do. If you want to keep your company small, you have to have small ambitions. You need a large company to do large things.

I think now that is not true. With coding assistance that continues to get better and things like Deep Research, I've found that nearly every job function can be partially automated, and I think that trend is going to continue.

As someone who spends a lot of time programming, I feel that my ability to take on things where I would have had to hire a team of people—or a specialist in iOS development or a specialist in low-level Rust programming for motor drivers—has expanded. With a couple of good engineers sitting next to an LLM and using good coding tools, they can adapt and do all those things.

So I think it's actually viable to do what I want to do. The lesson learned is that I'm going to keep the company small—to build a company that has grand ambitions but a very tiny team. That's what we're going to try to do.

It's pretty amazing. I think you've had a really amazing career arc overall. You've taken on 3 high-risk, complex companies back-to-back, with little downtime in between. You've run 7 marathons on 7 continents in 3.5 days.

Where does your drive and stamina come from? Is there a supplement we should all be taking? Is there some Bryan Johnson-style treatment where we should inject ourselves with young blood? What's the deal?

I haven't tried that. If you do that, let me know how it goes.

I have a problem, which is that once I get an idea in my head, it just burns a hole in my brain. I cannot sleep, do anything, or focus until I see this idea through.

For whatever reason, it doesn't happen 20 times a day. I'll get this idea that the stars aligned: this thing should happen, now is the time, and I need to go for it. Once I'm on that track, I just cannot let go until it's done.

I think I latch onto these problems and have this sense of delayed gratification where I want to work on something for a long time and get the result. That's also really satisfying to me—the notion of putting in a ton of effort, building something really complicated and hard, taking on these difficult challenges, and making a little bit of progress each day.

That is motivating to me. You could say starting these companies and taking on high-risk things is difficult, and it is, but I enjoy every day. I can't imagine doing anything else.

That's awesome. I think that's what makes Silicon Valley so great. Thank you so much for joining us today.

Thanks for having me.

Find us on Twitter at nopriorspod. Subscribe to our YouTube channel if you wanna see our faces. Follow the show on Apple Podcasts, Spotify, or wherever you listen. That way, you get a new episode every week. And sign up for emails or find transcripts for every episode at no-priors.com.