Building X With Astro Teller: Monkeys, Card Counting, And The Power Of Failure

Speaker A: You take this moonshot story hypothesis, huge problem, science fiction sounding product or service, breakthrough technology. As soon as you have one of those high five, good for you. You did it. We have another one of our investigations. Here's $20,000. Here's $50,000. Go see how fast you can kill it because there is a 99% chance you're wrong. That's just our batting average.

Speaker B: So go kill your idea because if you're not able to kill it, it's a great. What if I told you that making the impossible possible can not only be systematized, but also productized. My guest today is Astro Teller, the co founder and captain of Moonshots at Google X. Astro and his team have been behind some of the most ambitious technology projects in the world. From self driving cars with Waymo to AI with Google Brain to drone delivery with Ring to Google Glass to Loon to Tara and many other breakthrough products. But what I found most interesting in this conversation was wasn't just Astro's ambition, which is crazy high, it was the how he builds the discipline, the process, the principles he brings to turning impossible ideas into reality. Astro has helped build a system that takes ideas that sound like science fiction, tests them against reality, and then learns as quickly as possible whether they deserve to live or die. And that process is both incredibly unique and surprisingly clear. As you'll hear in this episode, many of these ideas might sound obvious in theory, but actually require a very unique crafting practice. Astro unpacks how X turns moonshot ideas into real world projects, products and companies. There's so much to learn from this episode, so let's get into it. Astro, it's a pleasure to have you here. Thank you so much for joining me.

Speaker A: Thank you for having me. I'm really looking forward to this.

Speaker B: Thank you. So we always start on a personal journey because I think it gives the audience the mindset of behind how you build. You've spent your life moving between science, technology, literature, business, so many disciplines. You come from a family heritage of groundbreaking science. Folks can research you after and your uh, lineage there. I, uh, want to start when you were younger, is this the career you imagined you would have or has it turned out very different than what you actually expected?

Speaker A: I'm not sure I conceived of it as a career, so I won't take credit, uh, in that sense. But starting when I was about eight years old, I had this sort of not very well formed fantasy that just stuck with me. Secret lab, really special people who are all working on like robots and like crazy Things for the future, for humanity that we could, like, be somewhat sequestered away. And as you were mentioning, my father's father, my mother's father's a Nobel Laureate, but my father's father was Edward Teller. And so I grew up on this diet of understanding about the Manhattan Project, taking this exceptional group of people who were really expert at all these very different things, sequestering them away somewhere where they could work really hard, really intensely on something that really mattered to them, that had never been done before. Uh, I wasn't interested in making a bomb, but the process of sequestering people away, creating a special microcosm where the culture could be different and extraordinary things could be created. I've wanted to do that since I was 8.

Speaker B: And is this a small sense at home where like, hey, groundbreaking is just in the air, like thinking beyond the obvious, or did you feel it was more, um, kind of disciplined into you in a way, which was, you know,

Speaker A: weirdly, I think it was a little different than that. So my both my grandfathers were kind of titans of the 20th century, which it was almost impossible to live up to. And it happened that my younger brother is just clearly smarter than me. And from a very young age, when he was maybe five and I was seven, it was clear that he was smarter than me. And I think in a weird way, that was one of the best things that's ever happened to me because really early on I'm also mildly dyslexic. I had to start thinking about, well, what am I good at? And how can I add value in the world without being the smartest person in the room? And I think it developed me into a much more three dimensional person where I could build skills on purpose and pay attention to things about myself or I was really interested in them or really good at them that maybe I would have missed if I'd been the smart one in the family.

Speaker B: It's, uh, interesting, sometimes they talk about a family is like a theater, uh, and everybody takes a role. So the fact that your younger brother took the role of this is the smart kid and we expect him to go to academia, kind of unleashed you to do whatever you want.

Speaker A: Yeah.

Speaker B: Uh, that's amazing. Uh, now you briefly told me about how you got into X, and I'm really curious about the beginning of it. And as I understand it, from what I researched, it was less about, hey, Google has some issues, Astro. We need you to help us solve them. And it was more about the world has some issues. And we want to see if we can solve them with some groundbreaking technology. Um, that sounds like an amazing mandate. It also sounds like a recipe for disaster a little bit, uh, at least in other companies.

Speaker A: Why disaster?

Speaker B: Uh, I think that it's almost too big for a commercial company. You don't see many companies outside of Google. In fact, Google is the only one doing it. So partly it's what's unique about Google that allows it to do so.

Speaker A: I think the founders of Google, Larry and Sergey, have always been really interested in helping the world as broadly as they could. And they were smart in focusing to just organizing the world's information first, but they had an appetite to go find a lot more problems. So you're very right. At the beginning of what became known as Google X, the mandate was do not go help Google. Google has lots of smart people helping Google get better at what Google's trying to do. But there are a ton of other problems in the world that Google is not naturally going to care about the problem or naturally find solutions for. That's what this place will be for. So think of it as like a 21st century Bell Labs, except in the 20th century Bell Labs, you had these super bright academics looking over the fence at the world's biggest network, Western Electric, at the time, solving their problems. So in our version in Google X, we had to come up, we have to come up with the problem itself and the solution. And you can see that as harder. I guess in some ways it is, but it's also an incredible opportunity to find and do things that matter.

Speaker B: Were you skeptic of. Of that being real? Did you like. You know, I think in some companies it would be, uh, you're telling me that, but then, you know, shit hits the fan and then you're going to ask me to do something for the company.

Speaker A: I took the founders at their word and I told them from early on cultural and operational independence are going to have to stick because this place will have to operate differently. It's just inherent in the innovator's dilemma that you can't do this in the middle of any large company. It's not an attack on Google. That's just kind of the physics of companies that you have to sequester this somewhat separately and build a very different culture if you want to explore efficiently, which is arguably what our job is.

Speaker B: So for anyone considering this for their company, this is not about establishing research or labs. That's different.

Speaker A: Yeah, I would say research is depending on how it's done, about somewhat stockpiling humans who you hope will be valuable to you in the future as a company and possibly having them spend their research time just sort of the intellectual building of knowledge in directions or on subjects that are relevant to the business. That is not what X does. So X's job is to make enduring businesses that are Alphabet scale and that can be phenomenally good for the world. Now in order to get there we have to try a lot of things, each of which has a low chance of success and that would be ruinously expensive if we did all of those five or ten years down the road. Most basic part of X to understand X is our job is to start on these unlikely journeys and then to filter really hard, really fast, really efficiently on the basis of evidence and to throw away most of our own efforts as fast as we can for good reasons.

Speaker B: The idea of taking people with great ideas and funding them and kind of scaling them to enduring businesses and you want them on the Alphabet scale, it's kind of a little bit in the air here. So how did you think about differentiating from how the VC community thinks about funding its portfolio companies to what you are doing?

Speaker A: You will probably get people who are in different parts of the space describing it to you differently. But this is how I would describe it to you. Taking moonshots is incredibly easy. It's absolutely trivial. You find really smart people, you tell them to be very audacious and you pour a lot of money on them. It works, you'll get some moonshots. It's just not efficient. I would argue that a lot of the world participates in that process, which I would roughly describe as gambling. You can go to Vegas and you can come back with more money than you went there with. But you just got lucky. And many people think they have a system for doing it, but most of their systems are a, uh, lucky rabbit's foot of one kind or another. We're trying to be the card counters of innovation. And so there's a lot that people would like to do because they have a conviction about flying cars or whatever. And we will consider at uh, the very wide part of our funnel almost anything which we can afford to do because we viciously attack our own ideas on day two over and over again. On the technology front, the feasibility, the techno economics. Does anyone want this? Why would they pay for this? And we're very careful and not to sub optimize for things like product market fit on not only month one but even year two. And I think we're going to get into this, but if you want to solve really Large problems, they tend to be systems problems. And there's a class of efforts for which the standard Silicon Valley playbook is exactly the right thing to do. When you can see just on this side of the horizon where your goal is, then just like maniacally running towards that thing is the right thing to do. But as soon as your goal is over the horizon, you don't even know what it is. You don't know if it exists, and you don't know which direction to run. All of a sudden, speed is not your friend. That's the thing X is really great at.

Speaker B: Uh, people describe X as the moonshine factory. It's even in your website. I love that. But there's something else that stuck with me. I think you said once that if you get obsessed about being in the learning game, things are going to get better. And I am a massive believer in learning as a skill. People ask me much. People should learn in the future, learn how to learn the biggest skill you can have. So I was thinking, is this the deep operating principle of X.

Speaker A: Everything is wired around that. So let's play a game for a second. I want you to think about something really hard that you and some team did. It doesn't matter what it is, just think of it.

Speaker B: Okay?

Speaker A: Now imagine that you lost all of the software, every single printout, any hardware that was associated with it. You have the humans still and what's in their heads, and you have to rebuild the solution. Once you've succeeded, what percentage of the time would it take in this thing you're remembering, for you and your team to rebuild the solution that it took from when you started trying to solve the problem?

Speaker B: Give me your number in this specific example. Very quick.

Speaker A: 5%. 10%.

Speaker B: Yeah.

Speaker A: So let's call it, I don't know, 5, 7%. If 7% of the time is what it would take for you to rebuild the solution, then 93% of the work was figuring out what the solution was and 7% was building it. 93% in that specific case you're imagining was learning, only 7% was building. Now imagine it's a moonshot, that 7% is going to get smaller, that 93% is going to get larger.

Speaker B: And you compound this against other, uh, projects.

Speaker A: It's only learning. Everything is about learning. And so if you work at X and you say, I want to make progress, I say, I don't. I mean, of course, conceptually, you and I both want you to make progress. The reason I refuse to talk about progress with you is because you Believe you know which direction you should go. You know more about the teleporter if you're building the teleporter at X than I do. But I'm telling you, from having watched hundreds of these, you're wrong. You're just wrong because everybody's wrong the second time, the fifth time. So you making progress is mostly wasting time. What I want you to do is see how fast you can discover that you're wrong and why you're wrong so we can actually get going in a slightly better direction. Is that how you're spending your time? Because that's what I'm judging you on. And that's the difference between at least at X, progress in learning.

Speaker B: Yeah. I don't know if this is the most inspiring motto, but what I'm thinking in my head is like, we're going to be less wrong over time and that's going to be our biggest.

Speaker A: And there turn out to be a lot of things where figuring it out early matters a lot. Let me give you an example. So first, maybe let me take a step back so that your audience understands some of the things that have come from X. Google Brain, which was one of the epicenters of what is now the modern explosion of machine learning. The Transformers came from Google Brain.

Speaker B: Transformers is the technology that basically started LLMs M at scale.

Speaker A: Uh, TPUs came from Google Brain, so Google Brain came from X. Um, Waymo, the self driving cars came from. They came from X Wing. Uh, the drones for package delivery came from X. I was going to use an example from Wing 12 years ago when they started doing demos that they were lowering something on a string. People thought that was like the goofiest thing they had ever seen. Every drone company in the world is starting to converge now to lowering something on a string because we spent the time to sort out partly by looking at some pretty ridiculous stuff. Oh, there's some fundamental reasons you have to do it that way. It meant that we ended up with technology lock in to the right solution instead of technology lock into the wrong solution. And you know as a product person how painful it is once you've locked in on the wrong solution to like reorient sometimes, sometimes impossible. And so that's another way to understand what we're doing is if this is going to change the world. Our job right now is not to get a bunch of customers. Our job is to lay the foundation properly so that we don't have to take big steps backwards. We're trying to figure out these fundamental ahas, whether it's self driving cars or drones for package delivery or whatever.

Speaker B: And Astro, if you were to critique the focus on learning just in kind of in a intellectually honest way because everything has a shadow effect, do you see times when it becomes an excuse? Like it's kind of like, oh, I'm learning, I'm um, not pushing through yet.

Speaker A: Uh, let me give you an example. But I actually really like intellectual honesty, so I'll go as deep on this as you want. But here's an example way that it happens is I'm always asking people basically the process at X and is the scientific method mildly changed to make it appropriate for innovation. But it's not that different from the scientific method. It's a very structured, habit oriented way of approaching, uh, learning and understanding, getting to a great solution in the long run, which is centers on experiments. Build a hypothesis that matters, then test the hypothesis as efficiently as you can and then be intellectually honest about the results of the experiment you just ran. Okay, makes sense. That's the learning loop, right? But sometimes people pick something they wanted to do anyway and they just call it an experiment because they figure, hey, I'm um, being performative, but whatever. This way if it goes well, I get credit. And if it doesn't go well, I can be like, yeah, it was an experiment, but that's not an experiment. The person who has weaponized our word didn't have a hypothesis. There's not actually a well formed statement about what we're trying to learn here. They just wanted to do that thing for whatever reason. So that's an example way it can be weaponized.

Speaker B: Let's go to the card counting, which is I think in many ways about building a system that basically improves the odds.

Speaker A: So let's start by making sure the audience knows the basics of card counting. So if you go to Vegas and you play blackjack, you lose maybe 1.5% of your money on every hand that you play on average, right? So it's like 49.51. It's around there depending on how you play, which is actually pretty good odds. Or a lot of things in Vegas that are worse odds than that. But they're draining your bank account slowly. But if you watch. So there often are a set of decks. So think of it as like they call them a shoe, but you see them, they're like sloped like this. Maybe it has six decks of cards in it. So that'd be a six shoe deck. So they shuffle six decks of cards all together. They're in there and then they sort of swipe them out. The dealer swipes them out one at a time. They're dealing them out. But over time, you can pay attention to how many face cards have gone by. The face cards are all worth 10, how many ACEs have gone by, they're worth 1 or 11, how many tens have gone by. And the more accurately you can remember all of the different cards that have gone by or the buckets of cards that have gone by by amount. And you know the statistics of blackjack. The odds can shift in your favorite depending on how you marry up the statistics relative to how the statistics are changing because of how these particular shoe, this six, uh, deck, um, rack of cards is being used up.

Speaker B: Then they kick you out of the casino.

Speaker A: If they notice, they kick you out of the casino because you have now moved from 4,951 against the house to 5,149 against the house. You can make money with every hand on average. That's what we're trying to do as innovators. We're saying, number one, you have to have a moonshot story hypothesis which has three basic components. You have to tell me about a huge problem with the world you want to solve. If you don't have that, it's an academic exercise. Two, there has to be some kind of science fiction sounding product or service that no matter how unlikely it is we could make it, we can agree that, that if you made it, it would resolve that huge problem with the world. And then there has to be some kind of breakthrough technology that gives us at least a glimmer of hope that we could make that science fiction sounding product or service that would resolve that huge problem with the world. As soon as you have one of those, it's a moonshot story hypothesis. High five. Good for you. You did it. We have another one of our investigations. You are almost certainly wrong. Here's $20,000. Here's $50,000. Go see how fast you can kill it. Because there is a 99% chance you're wrong. That's just our batting average.

Speaker B: So go kill your idea, because if you're not able to get. It's a great sign.

Speaker A: Exactly. And if you tie your sense of self worth to having your idea survive, you'll be miserable at X. We have a 1% hit rate. So your sense of self worth has to be repurposed to how good you are at paying attention to the evidence, like the counting of the cards in blackjack. And then how you pay attention to that evidence and use that Evidence to either kill your project or repurpose it in a way which gives it a better chance of success? That's card counting in X.

Speaker B: And this is true. I'm uh, thinking for myself, you don't need to be at Google X to understand this is just the right way to build product. Because I think in so many cases we want our beliefs to be right, so we get vested in the outcome and we almost get like detached from the process. We're just really the opposite.

Speaker A: Right. And so here's the secret. Our thesis of innovation, including everything I've just said to you, as you just said. Yeah, that sounds right. That's not the secret part. That's not the hard part. The hard part is actually creating a culture that supports that. Have you ever worked anywhere where there was a 1% chance of success and people accepted that as true and worked on things like that was true?

Speaker B: That would not be surprising to anybody.

Speaker A: Exactly. Uh, making a place where that's true and where they don't feel stupid. Because if you came and worked at X, you would be very suspicious for months, maybe years. Are you guys for real? If I kill my project, am I going to get fired? Am I going to get a bonus? Can I actually get promoted if I'm killing my projects? Just feels wrong. Even though intellectually you know that that's right. It just feels like the wrong way to climb the corporate ladder. So everything you've learned since you were 5 years old screams out to you in your gut and your heart, don't do that. Don't do what Astro's saying. So the real magic of X is actually the cultural engineering necessary to support people doing what we're talking about.

Speaker B: So this leads me into like talent. Now people cannot just move from Alphabet into X and vice versa. It feels like it's almost like there's a chasm between not the culture, as there's still Google and there's still desperation on the ambition. But just the way you work is so different. Do you see that flow of talent moving back and forth? Or is it. No, no, we're actually. It's almost like a separate entity. It's very hard.

Speaker A: I mean, we do have people who have worked at Google and then turn out to be great at X, but by and large they are quite different cultures. Think of it this way being clear. There are some incredibly innovative people at Google for sure. I don't mean to suggest, but by and large Google's main effort is to, on like 15 or more platforms, service multiple billion people every Day with like five or six nines of reliability and

Speaker B: report to the market quarterly how they're doing.

Speaker A: That is a world class skill and Google is world class at those things, so massive respect for that. But that is a very different way of being in the world than being an explorer. And so that's why you have to create a culturally separate place.

Speaker B: Now another thing you said which I thought was super interesting and I'm curious always about is this a Google X thing or can I extrapolate it to a broader takeaway for builders in every company, people should push on that. I think getting vested in the process and not, uh, the outcome is true no matter where you work. You mentioned the 10% versus the 10x and I think you said that it's easier to make something 10 times better than to make it 10% better. And people hearing us. That sounds very counterintuitive is the point that 10% thinking keeps you locked into the current assumptions.

Speaker A: If I ask you to do something to make a phone that runs on the same battery the same number of milliamp hours for 10% longer, at some unconscious level, you will start with the phone that already exists.

Speaker B: Yeah, I'll optimize.

Speaker A: You are now in a smartness contest with everyone else in the world. Good luck with that. But if I tell you it has to be 10 times better, you might well not succeed. But what's for sure, what you will know, not just in your head, but your heart and your gut immediately is you have to start over. Now you're in a creativity contest. You have to change the game or you have no chance of winning. And uh, being in a creativity contest with the rest of the world is still hard, but it sounds a lot less impossible to me than being in a smartness contest with the rest of the world.

Speaker B: Now I go back to the story with Joe Butter.

Speaker A: Exactly, exactly.

Speaker B: So how do I apply this? Because then this is where people would listen and say, yeah, I want to work 10x. But I also know that there are some very successful companies that just do 10%. 10% and they compound over time. Amazon meta.

Speaker A: Uh, yeah, I mean, to be fair, some of those, uh, some very successful companies have done a few fairly standout things. Uh, Amazon is famous for having had Jeff Bezos's early list of things when he was only shipping books of all these other things that he wanted to do. Maybe some of them weren't right, but they have taken some sizable risks, to be fair to Amazon. But I want to come back to at the very basic level, I want you to play a game with me. I've done this the world over with CXOs from large and small companies the world over. And I get the same result from everybody. Choice A. Choice B.

Speaker B: Choice A.

Speaker A: You can give a million dollars of value to your business this year, guaranteed. Choice B. You can give $1 billion of value to your business this year, but it's not guaranteed. It's one chance in 100. A million. Guaranteed B billion. One chance in 100. Who's choosing choice A? Nobody in the room raises their hand, who's choosing choice B? Everybody in the room raised their hand, big smile on their face because they've all passed the math test. Choice B has 10 times the expected utility of choice A. And they've all come to me hoping for a lecture on innovation. And then I say, now leave your hand up. If on their best days, in your most generous understanding of them, your manager, your CEO, your board of directors supports you taking Choice B bets and every hand in the room goes down. And then I say, you don't need a lecture on innovation, you need a new manager. The problem is not that people don't understand innovation, it's that organizations are not actually seriously setting themselves up in an organized and efficient way to take those Choice B bets. That's the problem.

Speaker B: Yeah, I was thinking for this checklist within a, let's call it a regular uh, tech company and uh, best you can get two out of the four. I think maybe it's the factory part that actually gets the underscoring. It's not the moonshot.

Speaker A: Can I, since you're a product guy, can I pick product as an issue for a moment about. Because this is a real tension that comes up all the time in what we're doing. I would describe it this way. There's like a spectrum on one end of the spectrum is industrialize and sell the bleep. And out of the first thing you can find that people want and are willing to pay for, the other end of the spectrum is experiment semi randomly forever. I would consider both of those complete failure modes at X. This one's more obvious. This one, this spectrum, this side of the spectrum, the industrialize and sell. A lot of the first thing you can find that people want. People have been entrained in Silicon Valley in particular to do this because of the live to fight another day, go prove that you have something, and that ends up with a very greedy local optimization kind of thing. What we're trying to do is spend the time pressure testing your idea, working on the riskiest parts of the problem first. But when it becomes the biggest remaining risks. Does anyone want this? Why do they want this? What would they do with this? If you gave it to them, how much will it cost to get it to them? How much will they give you in exchange? And is that the subtraction ah of those two numbers? A positive number. These are fundamentally product questions and go to market questions. If, when that becomes the biggest risk, you have to run experiments and get the answer to do we understand how to make a once in a generation business here? And ultimately you have to do things that look more and more like building a product in order to ask and answer those questions.

Speaker B: So this is very unique. This is what I, so I want to, I want to unpack this a little bit. So an idea comes to the factory. Let's put the factory aside now.

Speaker A: Idea comes to the teleporter, let's say,

Speaker B: and then somebody pitches really well. You feel like they're, you know, this is associated from that personality. They found something really massive. You give them a little money and then you ask them to focus on the monkey. And I want you to expand the monkey for the audience. So that's how it starts.

Speaker A: Yes, it is. So you take this moonshot story hypothesis. What you just described is what I would call the moonshot story hypothesis. Huge problem, science fiction sounding product or service, breakthrough technology. Once they come up with it and I say, hey, you're probably cool. High five. You're probably wrong. Go find out how fast and efficiently you could kill it. What I want you to work on is not the thing that will make you look cool. Uh, because what's cool at X is the intellectual honesty and the efficiency with which we can learn. It's the information gain per dollar spent. So we have this funny story which is, imagine if the project we were working on together was to get a monkey to stand on the top of a 10 foot pedestal and recite Shakespeare. Which should we do first? Train the monkey or build the pedestal? If we're really honest. Most businesses most of the time are building the pedestal because you can say to your boss, look, I'm half done. And there's like a sense in which that's, there's progress, I'm making progress, I'm progress. And like, hey, can I have a bonus? I want to be a, uh, vice president now. Can I have that? Look how good my pedestal is. But in that extreme situation, we can see since 100% of the risk is piled up on whether or not we can Train the monkey. There is absolutely no reason to build the pedestal first. The only thing you should spend your time on is training the monkey. If you can train the monkey, great, we'll get a pedestal. And if you can't, thank God we didn't spend time on the pedestal first. So people have really deeply understood this EDX to the point where they just talk about, hey, in my project, this is the monkey. They'll just say, this is the monkey, which is shorthand for this is the riskiest part of the problem. This is where the information gain is the highest. And that's why we're focusing on this. So I want you to spend your time on that. And then if it turns out that the result is, hey, the monkey's not speaking Shakespeare, but we got a sonnet out of him, awesome. All right, let's keep going. And if it's like, nope, uh, the monkeys, mute after six months, you know what? Let's move on to another project.

Speaker B: So literally, you would take the highest risk, hardest problem to solve. You would say, start here, prove to me that this is doable.

Speaker A: I mean, yes, absolutely on that. But also imagine you and I were working on a project, and now a new project has come up over here which has twice the risk but four times the upside. Technically, that thing has twice the expected utility of the thing we're working on, even though it has double the risk.

Speaker B: And this is where you come into play.

Speaker A: And, uh, yeah, I would say, hey, guys, we should seriously consider having you stop doing this and go help them on that. If that actually has a higher expected utility. Can I give you an extreme version of this?

Speaker B: Sure.

Speaker A: So we spent four and a half years on a fusion project at X. And most of the people in the fusion space. Nuclear fusion, nuclear fusion, uh, are doing something where the physics risk is not low, but it's lower because they're doing something. Many of the best ones are doing something which is relatively well trodden physics in the fusion space called a tokamak. Uh, sort of surrounding a thing that you want to keep at very high temperature and pressure with a lot of magnets to sort of keep it in, like a tiny little sun almost.

Speaker B: I'm nodding as if I understand.

Speaker A: Uh, and so you have call it medium risk and crazy high expense on the engineering side. Someone came to X and said they were already at X and they said, here is a crazy idea. There's this weird bit of physics that back in the 1980s, people had some good experience with, and then they couldn't keep it going, so it's kind of been discredited by the physics world. But if it was true, if they misunderstood in the 80s, why that good news sort of couldn't be repeated. This way of doing fusion would be wildly cheaper to do afterwards. So the physics risk is much higher, but the engineering risk is much lower. And their argument was it's, I don't know, call it a trillion dollar opportunity if it works, maybe a 1% chance that we actually get there. So let's call it an expected utility of $10 billion if we have to spend $10 billion to find out if we're right or not. Not a good lottery ticket. But these people said, we can ask and answer this question for like under $10 million. And I said, hell yes. If we can take that kind of risk with that kind of in the money upside on a lottery ticket, like, that's the kind of thing we should be doing. They did it for about four or five years. It was a tiny number of people. We actually did almost 1000 plasma shots before we shut the experiment down. They shut their own project down because it turned out that the physics risk wasn't playing out. And, uh, we've moved on to other things. I'm incredibly proud of the fact that we did that. Not because it didn't work, of course, I'd rather it had worked, but because that's choice Bs. That's exactly choice Bs.

Speaker B: And is that very often that happens? More often than not. The comparative, like, hey, I'm finding a different problem with, uh, a better risk to benefit.

Speaker A: Yes. I mean, I think it is rare that we're taking physics risk. Often the risks are much more system integration oriented, uh, or there are issues around skating to where a puck might be. Um, so I don't know, maybe you think you're at X and you have a theory about some really crazy stuff we could do with carbon nanotubes, but right now carbon nanotubes are a thousand times too expensive to do that thing with. And you have a theory that other people are going to solve the cost problem and you want to build these cool new things with carbon nanotubes. Um, that's not exactly a physics risk, but that's a risk we would have to accept in that project.

Speaker B: And Aster, how do you think about risk reduction as part of the process? So I started my career in semiconductors. And I used to say, in semiconductors, you can design a chip, build it, take it out to market. In the meantime, you can have a baby, get married it takes a while for the process. You do a lot of work to take the risk out of the design. You don't want to find out you have an issue when you go to production. That's the worst outcome possible. Um, so there's a lot of exit points. There's an exit point. This could be just, uh, an incremental chip, the previous chip, not a 10x chip we were hoping for, but there's an exit point. So they kind of built exit points along the way that basically de risks the project. That's one way to take something really big, high investment, and kind of DE risk it 100%.

Speaker A: So think above the eponymous moonshot, right? It wasn't like the first thing they did was try to go to the moon, right? It started with, hey, can we have a rocket just go up and come down again. Can we have a rocket go up and get to the place where if someone was inside, they would have experienced zero g, but there's no one inside and come down again? Can we have a rocket go up, go around the world once and come down again? Can we have the rocket go up with a dog inside, go around the world once and come down and the dog's still alive? There was this whole story series of things where they were asking and answering questions and learning what it would take to take a person to the moon. That's exactly what our moonshots are like. Hopefully they're cheaper than that, but they have that same kind of structured theory, uh, design of experiments where we're working on the riskiest parts first, but we're trying to figure out how fast to retire, risk per dollar spent. And we keep going when the risks are getting retired really quickly relative to the money we're spending.

Speaker B: So is this like a template that somebody produces? Once you said okay to the project, then the next step is, you know, figure out the riskiest part and then how do you derisk it? Is this so systematized to that point or.

Speaker A: Well, again, imagine that you worked at X and you've. You're proposing the teleporter not for the first $50,000, but if in a few months, whiteboards, talking to some experts. We haven't been able to easily kill off your idea. Maybe we're getting more excited about it. You and two or three people are going to actually start being full time on the teleporter. It's really two or three people because you should be able to run some experiments without some huge team. It has to work that way in order for us to be efficient. What we want, uh, you to do like right then is set kill criteria. We call them kill criteria. You can't even become a project at X without kill criteria. And those should be married against the monkey. What are the things that if we could agree now, if you can't get good evidence over the next 12 months that the following is true, we should just call it a day. Now we have definitely gotten to kill criteria and missed or sorry, we've missed our kill criteria and still continued projects sometimes, but at least we can agree ahead of time. The default is to turn it off. So there has to be something else we've learned that gives us a different reason to go forward.

Speaker B: I want to take three projects, products that you guys took out. Waymo Loon and Google Glass, they're all very different. Uh, and then one became a major company. One, the original, uh, project died, but seems like the learning lived on. And one seems like it was ahead of the market in a way. So Waymo, everybody knows right now. Maybe two years ago, it was hard to know the name, but now it's the self driving car. Um, actually, funny story. I took it from Palo Alto to sf, uh, this week. And uh, I was coming off the highway and was supposed to merge into the lane, but the lane had a traffic jam and the women was trying to get in and nobody was giving it the space to come in. And I'm like, oh my God, I'm stuck right now. Like, because it's not gonna, you know, human being was just kind of nudge their way in. And now this thing is gonna take me to the next interchange and I'm gonna be late to my meeting. And then it nudged and I was amazed. It kind of felt like a, you know, New York taxi driver just finding their way in. And I was like, there was like a moment of, uh, light bulb for me. This is.

Speaker A: We've just learned, we call it body language. That it's turned out to be a whole next wave of things that Waymos had to get good at over the last five years is the body language as a car and how other drivers read the body language. And these movements of your thinking about

Speaker B: how do you solve the left turn, which in the US is you can do, uh, as well for the traffic coming from the other side in a jammed, um, that sounds like a really tough problem. But then you could see that aspect coming through. Why was the monkey in the Waymo project? Because there's so many decisions. I'm sure maybe the monkey changes shape all the Time.

Speaker A: But it did. I mean, in the early days it was, how too early are we? Can you get this to work even sometimes? So we said, here are 10 different 100 mile routes around the Bay Area that were picked to be very different from each other and all very hard in order for this project to continue. The kill criteria, we hadn't named them. Kill criteria was, um, for you to do each of these routes at least once without a human touching the wheel or the brake of the gas pedal. Uh, these were Priuses that had been sort of. I remember those fly by wire. And you can try as many times as you want, but you have to do each of them at least once with nobody driving the car. And that was a question that had nothing to do with feasibility or like nines reliability or anything. It was just like, is it even possible? Later, there were very different questions. We're like three different epochs for Waymo. We thought we were making a car, we thought we were going to sell it to people. And about 2015, it was so good on the highways, uh, and city streets, but including the highways, that we said, okay, we're going to let Googlers commute with them. And we gave, I don't know, 10, 20 Googlers these cars, and said, you can commute to work with these, but you have to promise to keep your hands right by the steering wheel. We're going to have cameras watching you. And these Googlers all said, oh, yeah, I promise. They were like, putting on their makeup, falling asleep in the backseat, eating their lunch. It was horrible. And we had to call after a week or something. We just called off the experiment. It was dangerous. Humans are not a good backup for machines. They're just terrible at it. And so there was this we trust too quickly. Yeah. And there was this crisis moment where we said, do we have to kill the project? And what came out of that crisis was a decision. It was a big pivot. Oh, our job is not to sell a car. Our job is to sell mobility. We're transforming mobility. That's what we're doing. And for, uh, at least five years afterwards, that was the rally in cryo. Waymo, we're transforming mobility. And even that turned out to be wrong, because transforming mobility includes charging the car and the app for calling the car and all these things that Waymo is happy to work on, but that's not their job. So now if you ask Waymo, what are you doing? What's your job? They say, we're making the world's safest driver. That's what they're doing. And that architecture for understanding what you're really up to as a team is part of what comes out of really deeply asking, what are we great at, what's important about this and what's secondary.

Speaker B: So interesting. My experience as a user of Waymo is that it's transforming time for me because I was debating this week they could take an Uber or Waymo. Granted, Uber is quicker. Waymo can take the other route a little bit. Uh, but in a Waymo, I can do whatever I want. That firm was like, okay, this is my meeting area. It's where I practice. I was giving a talk as I was practicing my talk, and an Uber, I wouldn't feel as comfortable. So that's from Time for me. Are you surprised sometimes how trusting human beings are? Like, it's. I remember, like, my first ride, I had to convince my wife to go on away, my ride with me. And she was, you know, looking out the whole time and freaking out and saying, this is the last time I'm doing it. And then literally the next ride, she

Speaker A: was on her phone, she wasn't even paying attention.

Speaker B: And then I'm thinking about.

Speaker A: Everybody's like, that but.

Speaker B: It's a but. For me, that was mind blowing because then I'm like, hey, robots. We talk about robots in my home doing chores. I'm like, oh, this is going to be creepy. But then my sense is it's going to be creepy for an hour. And then you're like, you're going to get upset that the revenue is not doing more for you.

Speaker A: So people are at least as much like that with Wing the drones for package delivery. It seriously is like Harry Potter, owl post. The experience of just asking for something. Like, imagine anything you could get at Walmart that fits in a bread box and it just literally comes out of the sky, like, drops into your hands a couple minutes later. It's kind of magical the first couple of times. And then you're like, hey, where's my, you know, I need a monkey wrench. It's been, you know, 90 seconds. Where's my monkey wrench?

Speaker B: It's amazing how quickly we move from. There's no way to. I don't know which. There's a. I don't know if there is a. If there is a scientific term for it, but that at least a version

Speaker A: of it is called hedonic adaptation.

Speaker B: Oh, that's good for me. I'm just. I'm just going for it.

Speaker A: Um, you mentioned lune and glass. Lune is an interesting Story for us. It's a particularly good example of what we call moonshot compost.

Speaker B: Did loon become Tara?

Speaker A: Yes. Okay, so loon was our first big moonshot in connectivity. And there were two problems. One is could you make this ad hoc mesh network of balloons in the stratosphere that somehow magically all talk to each other, move the data around, beam it onto the ground and solve like rural connectivity? That part worked amazingly. Hats off to the team. You also have to make it into a business. And we were doing it, but it was going too slow. Our ability to get there with the telcos, it wasn't what we wanted it to be. So we shut it down. But in the process of that ending, we had been having these balloons talk to each other and the ground using lasers. And a few of the laser people said, I know this isn't supposed to work on the ground, but will you just like let us take these lasers and start a new project at X? Just seeing if we could solve the world's connectivity problems using lasers on the ground. And eight years later, the Tara team, which has now graduated from EX and is an independent company, is moving more data every like two or three minutes in 20 countries around the world than Loon did in its entire nine year history. And so it's really upsetting when something ends, but as long as you're in love with the problem and not in love with the current solution. Again, if you worked at X, I would say, I know this is upsetting, but your people don't have to leave X. The patents don't have to leave, the code doesn't have to leave, the hardware doesn't have to leave. Even the partnerships that you've built, we can repurpose those. If you're passionate about solving the problem. It all goes back into the dirt and you'd be surprised it comes back.

Speaker B: That's amazing. And Google Glass in many ways I remember we talked about, you know, LinkedIn used to be in Sterling Court and we were like, you know, doing the round around the campuses and seeing this is like a decade ago now, people walking around with Google Glass and like, you know, using it and like saying this is cool but you know, it's actually painful a little bit more than it's cool. And then now you see it's becoming more and more mainstream.

Speaker A: I mean, I think increasingly it's obvious we were right and we were way too early. That would be my summary of the situation as glasses come.

Speaker B: So what did you under appreciate about that moment? That uh, you know, you So I

Speaker A: think we got something incredibly right about pre product and then something incredibly wrong about product. And here it is. We called the early versions of Glass the Explorer edition. And at first we saw it for ourselves and we were telling you, did you have an early version of Glass? Yeah, yeah. So then you remember you are an explorer, and if you buy this, we need you to think of yourself as an explorer. It's literally what we called you when you bought it. And so we were trying to train you that this is not a product, this is a learning platform. And in the early days of Glass, people were kind of crazy about it.

Speaker B: Yeah, you saw it, you saw it.

Speaker A: Everyone loved it. And there were so people were flooding back to us, filmmakers and doctors. People were chefs.

Speaker B: They wanted to remember, like I wanted to love it.

Speaker A: Right. And as a learning platform, I think it was pretty exceptional. So that's what we got. Right. Really right. Then somewhere along the way, we lost the thread and we started to believe people's excitement and started to think of it as a product when it was not a product yet. And partly because we were charging as much for it as we were, I think that encouraged people to think of it as a product. It ended up on some Runway models in New York. There were a few things like that that coincided to leave people feeling like we were promising them an exceptional product.

Speaker B: They were buying the future in a way.

Speaker A: Right. And then, uh, once their expectation had risen up to here, and it was still a great learning platform, but it was not a product, but they were now expecting a great product, then we deeply disappointing a lot of people. And then I think that's what caused Glass to regress in terms of people's experiences.

Speaker B: So what would you have done differently in that respect?

Speaker A: I think we should have started with, uh, professionals and prosumer experiences and spent longer with doctors, people on oil rigs.

Speaker B: So narrow the use cases.

Speaker A: Narrow the use cases. And for longer said, we're listening to you, but this is not a product. We're glad that you find some use in it, but you have to keep telling us what you wish this did. Where do you find this useful? We're going to keep trying to improve this, but do not expect this to be a polished, clear value proposition. You know, solve for something in your life, because it isn't yet, uh, even though doctors like, oh, I could use this as a checklist for the things I have to go through every day. The mechanics for airplanes were like, oh, I could look at a million pages of manuals while I'm like fixing? Yes, conceivably, kind of today. But it wasn't optimized for any of those things. And you know what it takes to make a product. The difference between proof of concept and a really thoughtful capital P product is often years.

Speaker B: Do you see augmented reality in the same category or you see it different?

Speaker A: Yeah, there was a decent amount of augmented reality in what.

Speaker B: But even kind of the whole augmented reality glasses for doctors to try and pull it out, a big project at Microsoft, and then it was not.

Speaker A: Oh, do I think that they suffered some of the same problems? Yeah, absolutely. I think they might have been even less successful at flagging it as a learning experience, but they fundamentally suffered from the same problem, which is it wasn't a product. They were kind of making it sound like it was. Developers don't develop, at least in large amounts for a learning platform. They'll experiment with a learning platform, but developers go crazy once it's actually stable enough that.

Speaker B: I don't recall any of the learning platform narrative when I was using it. Uh, because I think actually even if it was out there, maybe people didn't want to listen to it or didn't want to. I don't recall that narrative coming out.

Speaker A: But you remember being called an explorer?

Speaker B: Yeah, to an extent. Yeah.

Speaker A: Yeah. So that was.

Speaker B: I thought it was cool. Okay.

Speaker A: But I think this disconnect where we were seeing it that way, we were calling an explorer. But then you even. And you're like much more thoughtful and sensitive than probably most of our customers were. We're still experiencing it like we were making a product promise to you. This is exactly what we got wrong.

Speaker B: So you gave some pretty amazing principles, which I think people can take away. I think people kind of want to get all the use cases out. So, like, what's a good way to get all the use cases out from this? Amazing years of learnings you have.

Speaker A: So we started a podcast a little over a year ago. We did a first season which is audio only, though we have some deep dives which are video. Uh, those are more like an hour. The audio ones are also an hour, but they're more of a smorgasbord. And the hour long deep dives are individual conversations like this about a specific project. We have a second season which is a little more than half out now. It's still coming out every Wednesday. And each of these episodes is an introduction to usually two of our different projects that have something in common. So this is me as the host and interviewer. So this is not me saying how it is. It's me asking people who've gone through X, how did we find you? What was it like to start a moonshot? Where did you get lost? What was hard about it? What was the breakthrough? What happened after you graduated? And we are having these conversations with the people who started Google Brain, with the people who started Waymo, with the people who started Wing and Loon and Tara and others. I think there's a ton to learn from it. I think it's really fun for people to listen to.

Speaker B: Love it.

Speaker A: And the name is called the Moonshot Podcast. If you just search for Moonshot Factory Moonshot Podcast, you'll find it.

Speaker B: So, uh, we talked a lot about what you're doing, your team is doing is just unbelievable now. Just like, if you were to zoom out, I think that would be. Just getting to know you through this time would be actually easier for you than most people. Uh, if you look at the next decade, you zoom, um, out of Google X and Google in general, and you think about the next generation of builders. What do you want them to be brave enough to try and build, and in a way also wise enough to not touch?

Speaker A: One of the things I would like builders, especially people in their teens and twenties, to hear. Every generation feels like all the good ideas have been used up. They have. Absolutely not. We are at the beginning of a Cambrian explosion of innovation. I really believe that. And I can feel, and I've heard many times, people, especially younger people's fear that they sort of somehow showed up on the scene at the wrong time, that they sort of like musical chairs ended just before they got to the next open chair. Everyone sat down and they're sort of left hanging. I don't think that's true. I'm not saying it's easy. It's all rainbows and unicorns, but I really do think that innovation is going to ramp up. We're going to see more positive change in what's possible for humanity in the next 20 years than we did in the last 50. And I hope builders can feel that opportunity. There is so much temptation to just vibe code your way to something that's mildly better, that people already kind of understand that's like a super comfort feature, but won't fundamentally make things different than they are today. I just would encourage people to go get their boots a little bit dirty and be open to the idea of trying things that sound uncomfortably exciting. One of my mantras, or maybe metrics, is if you ask a hundred experts and even 10 of them tell you what you're talking about sounds exciting or reasonable, you're too late. If 0 of 100 tell you that what you're talking about sounds interesting or exciting or reasonable, you're too early or it's just not possible. You want like three or four maybe of 100. I, uh, just would encourage people to take a little bit more risk.

Speaker B: Anything you want them to notify to build.

Speaker A: I mean, I have a bias, which is I think the pursuit of knowledge, understanding how to, how the universe works and how to engineer things, Almost nothing is off limits there. But there are things that are not the right product to build, not the right service to build. And unfortunately, I think we live in a time where we have over lauded people for going extremely fast and not caring very much about the damage that's done on the way to making your first billion. Um, I'm not saying that money isn't important, but I would love to see people have a little bit more of a principled approach that doesn't necessarily mean go slower or take less risk. But think of your bank account as not just the number of dollars that you have, but the pride in yourself that you can have and the legacy that you're creating. If we could reconceive of what we're building in those broader terms, I think people would build somewhat different things.

Speaker B: You know, we have not said the word AI, I think once in this session.

Speaker A: I'm so proud about that. Every single thing we do at EXP essentially has AI in it.

Speaker B: Sure.

Speaker A: Um, but it drives me crazy when people talk about AI.

Speaker B: Well, you know, now it's kind of hard because there's like AI since 2022 and then there's AI before. Why do people get wrong about AI in your mind when they talk about AI?

Speaker A: So I have a PhD in artificial intelligence from almost 30 years ago. So I mean, first it's a 70 year old field, it did not start five years ago. Two, it's not actually a monolithic domain. There's a lot of different things happening in AI and mostly when people say AI, they mean LLMs. But AI is a much more complex world of learnings and experiences than that. I guess here's maybe what I would encourage people to take away. We say that hot air balloons fly, that dragonflies fly, that 747s fly. We use the word fly for all of them. And yet at some level we understand that they fly in incredibly different ways. Intelligence is similarly multi dimensional in its characteristics. And are we smarter than machines? Like that's just not helpful. That's just not uh, what I think of as an interesting topic. It's like saying, do dragonflies fly less well than 747s? I mean, kind of yes. Kind of no. Kind of, who cares? I would just encourage people to think of intelligence as a much wrinklier and more high dimensional subject than intelligence.

Speaker B: Uh, if you can snap your fingers and solve a product issue or a tech issue like right now, what would that be for you?

Speaker A: I mean, I have some particular fondness for room temperature superconductors and a number of other things. But I guess if I was trying to. I'm going to turn your question into what One underlying technology change would unlock the most new moonshots. I'll tell you that way because that's what's on my mind. I think having energy density 10 to 100 times higher than what we currently have in lithium ion, uh, would change the world in a lot of different spaces. I can't tell you the number of dreams we've had at X that we've just had to put in the freezer because we need a battery that's 20 times the energy density of what they are right now.

Speaker B: What's, uh, a product that you did not think would be successful but surprised you?

Speaker A: One of the really mature projects at X is a synthetic genomics effort. It is being able to simulate small cell biology so well in the computer that you can reprogram the biology to ask it to manufacture new things for you. And that involves a level of being able to understand, have the computer understand biology well enough that you can simulate it. When we started that nine years, I was incredibly sure that was not going to work. And I was wrong. And I'm really excited that I was wrong.

Speaker B: What's your favorite, uh, non digital product?

Speaker A: I mean I'm constantly grateful for my glasses. I think glasses are one of the best user interfaces uh, that humanity has come up with. You forget that they're there. You use them in actually quite dynamic ways. These are progressive lenses. So I can see up close, I can see far away. Um, they never run out of batteries. They actually change. These are progressives so they turn into sunglasses when I go outside. That's a bit of a miracle, but. Can I give you a second one?

Speaker B: Yeah.

Speaker A: I think we should be trying to get technology to be like glasses sometimes, but also like anti lock brakes on your car when you are about like something bad happens and you smash your brake. That's great. Like there's a very kinesthetic experience to smash the brake. But when you smash the brake that does not actually break. Like, put the pads, um, against the brake. Um, part of the car. You are talking to a robot, and the robot is like, I hear you, and I will now try to stop as fast as I can. Subject to not putting us into a skid.

Speaker B: Yeah.

Speaker A: What an incredible user interface where you don't have to think about that at all. You can just be like, stop. And then all this complexity happens behind the scene. And it's a very kinesthetic thing. There's no button pushing. I wish more of technology was like that.

Speaker B: For young, uh, aspiring builders who are seeing you, they're inspired by you. Um, they want to be you. What do you think they should learn or go do? Like, what would you advise somebody who is now 17, 18?

Speaker A: I mean, I definitely would say you can always get a soulless job later. Find something you're passionate about and go do it. Uh, I think people are playing it safe way too early in life. 1, 2. Surround yourself with good human beings who you can learn a ton from. It doesn't matter what they're doing. Who cares what your degree was in? Who cares what they're currently doing? If they will hire you? If you can find fantastic human beings to be around and a great manager to learn from, that's what to optimize for, not the subject they're working on.

Speaker B: And if they want to get into X, what's the recipe for that?

Speaker A: I mean, we're looking for people. We are looking for people we are hiring, but we're looking for people who are weirdos, who have never quite fit in other places. Um, we hire a lot of people with no job description, even quite senior people. And if that sounds crazy, like, hey, I'd like to hire you, I'm probably going to pay you half of what you're worth for right now, and then we'll ratchet it up when we figure out what your actual job is. I want you to spend the next six months here just wandering around X and figuring out what your job should be. Most people wouldn't take that deal, but that's actually good because the people who wouldn't take that deal would be miserable at X. Anyway. Um, I just hired someone to a deal, sort of like that. And she had just given most of the money that she had raised back to her VCs. Does not happen very often. That wasn't the reason we hired her, but it was like one of 10 reasons we hired her. Because that's showing a kind of intellectual honesty and not just Playing the system and spend every dime that you got that signaled to us. Oh yeah, she needs to be with us at X. Astro, this feels wonderful.

Speaker B: Thank you so much, Homer.

Speaker A: Thank you for doing this with me.

Speaker B: Thank you. I love this conversation with Astro. What struck me the most is how he took something that can easily sound very abstract. Moonshots, innovation, impossible ideas. But then he made the day to day practice sound incredibly concrete. Here are some of my main takeaways from the conversation. First, when it comes to solving really hard problems, the best builders are not trying to be right. They're trying to become less wrong as fast as possible. Astra Strata the operating principle at X is not simply dream big. It's actually to become the best in the world at Ah, the learning game. In moonshot projects, the hard part is not building the solution. The hard part is figuring out what a solution should be. Because once you know the answer, building it is much easier. That mindset completely changes how you think about progress. It's not about shipping more, it's about learning faster. And sometimes the best way to do it is by quickly proving that your current direction is wrong. That's why Astro wants his teams to try and kill their own projects early and discover a score quickly and honestly as possible. Why? Their idea might not work. If they cannot kill it, that is a good sign to keep going and invest more. And if they can kill it, that's also a good outcome because they'll learn quickly, they saved years of wasted effort and they can move on to the next idea. Second, don't build the pedestal before you train the monkey. I love this metaphor. If your project is to get a monkey to stand on a pedestal and recite Shakespeare, most teams start by building the pedestal because they know it's doable. It feels like progress, it feels good. But the real risk is not getting the monkey on the pedestal, it's getting the monkey to recite Shakespeare. Can you even do that with moonshots? You need to attack the hardest, riskiest assumptions first. Because until the risk is addressed, everything else is just theater. Third, going for 10x improvement is not just about being more ambitious. It actually changes the game you are playing. Astra framed this. If you try to make something 10% better, then you start with the existing system and you enter a smartness contest with everybody else. But if you try to make it 10 times better, you are forced into a creativity contest. Now, this thinking doesn't apply to every problem because not every problem deserves a moonshot. But it does mean that sometimes incremental thinking traps you inside of the box. Fourth, innovation is not gambling. It's actually card counting. Astro wants his team to be card counters of innovation. In blackjack, card counters do not know the next card. They are not predicting the future with certainty, but they are paying super close attention to the evidence. And then when the odds improve, they increase the bet. That's how X thinks about moonshots. At the beginning, most ideas are probably wrong. So you start small and you observe them very closely. You give them some time, some capital and a clear question to answer. But then you watch the evidence. And if the odds improve, you invest more. If they get worse, you fold. The same lesson goes for builders. Let conviction guide your path at the beginning, but then let evidence tell you when to double down and when to fold. And last but not least, real innovation requires culture engineering. This might be the deepest lesson from this episode. Astro's process sounds obvious. Run experiments, follow evidence, kill what doesn't work, double down with actually improves. But most organizations are not actually built to do that because they say they want learning, but they reward certainty. They say they want risk taking, but then they punish failure. They say they want bold ideas, but then they ask every team to show predictable progress. And that is the part of what makes XTO unique. It was built as a separate environment with operational independence, very strong founder support, and a culture designed specifically for this work. So now think about your own company. Would it tolerate such a process? Would it actually reward it? And if the answer is no, then the bottleneck for innovation is not the lack of ideas, it's the culture. I'm Tomer Cohen and this is Building One. I hope you enjoyed this one.

Speaker A: You've been watching Building One. Our show is hosted by Tomer Cohen. Building One is produced and edited by Mason Cohn and the team at Coastal Production Works. This episode was mixed by Tim Boland at LinkedIn. Our team includes Rachel Karp, Sarah Storm, Dave Pond and Alicia Mann, with support from Alex Kuznetsova and Mujeeb Mehrdad. Until next time, keep building.