Transcription
Archer was amazing. Then you jump into arguably what could be described as one of the most difficult businesses to get into. Why? You start to figure the humanoid robot is like the ultimate deployment vector for AGI.
It is truly my honor and pleasure to introduce to you Brett Adcock, founder and CEO of Figure. You went from a cold start in 31 months to shipping your first robot. We are designing a new hardware platform every 12 to 18 months. Like, by the time I filed the C Corp, we had the robot walking in under 12 months. I think you're going to see it in the coming years being put into homes. Just through speech, be able to do like very long horizon hours of work without any problems. There's like an iPhone moment happening with humanoids. Like it's going to be, this is going to happen right now, now. That's a moonshot, ladies and gentlemen.
I think most of you know that the news media is delivering negative news to us all the time because we pay 10 times more attention to negative news than positive news. For me, the only news worthwhile that's true and impacting humanity is the news of science and technology. That's what I pay attention to. And every week I put out two blogs, one on AI and exponential tech and one on longevity. If this is of interest to you and it's available totally for free, please join me; subscribe at Damandis.com. Subscribe. That's DMandis.com, subscribe. All right, let's go back to the episode. Thank you for being here.
Yeah, thanks for having me. I know with three young kids and a robot factory in production and an incredible team of engineers, you're really busy, and I don't take it for granted that you joined us here.
Yeah, my only request is next time I want a Figure robot with you.
Loud and clear. I, I, I, I begged him. Um, and uh, BMW's been taking the lion's share of them.
Yep. We, we do have a lot. We actually have them running every, every day now, so like they're there today, running and uh, in their largest plant.
Why'd you start Figure? I mean, you had this incredible, a few incredible successes, uh, and Archer was amazing, and then you jump into arguably what could be described as one of the most difficult businesses to get into.
Yeah, I think, um, we really need to figure out a way to give like AGI a body here. Um, I think it's like a really negative or like almost like dystopian future if we figure out how to solve AGI and it lives in a server somewhere and it's like, you know, more intelligent than the, than all of human, like, like everybody. And ultimately, if it wants to do something in the physical world, it'll ask, it'll have to ask or boss a human to do it. And um, the humanoid robot is like the ultimate deployment vector for AGI. It's, um, you can't solve this with anything else besides a human, like a mechanical human. You need, um, you need something that is a single platform that, with no hardware changes, can do everything a human can. And you need something that can also be good for the neural nets. Like the neural net here in a humanoid can basically learn, like, basically learn from transfer learning. Um, it can basically multitask across a variety of different, uh, applications, which is like really good for a neural net. So we basically can build like one single neural net, like foundation model that can empower the whole robot to do everything end-to-end.
I mean, you know, massive congrats. You went from a cold start in 31 months to shipping your first robot, uh, which is, which is extraordinary. I mean, a lot of companies get their PowerPoint decks ready and raise their first capital in that period of time. And uh, we're going to be seeing some of the robots in back here. Uh, when I visited you up north, you know, you showed me around. Um, we did a podcast together, and you showed me Figure one, and here's Figure two, and here's the designs of Figure three. Uh, one of the things I truly find amazing is the speed of your iteration. Can you speak to that and how important rapid iteration in hardware is, because hardware is hard?
Yeah, this is a hard problem. We have to figure out how to do something that's never been done before, uh, and it's like a very complex system, like definitely more complex from an engineering perspective than Archer was, like building an air electric aircraft. Um, so yeah, my rule of thumb is like the first or second-generation hardware is always going to suck. You know, like the first iPhone was not great. Like the first, first time you make something, like you're never going to get it right in hardware. You have to do that, like, um, you have to see like five years in the future. You have to know exactly what the product does, and then you have to clean sheet design it for that exact thing day one. And if you mess up any of those, you can't go back and fix it through the design process. You have like long lead time supply chain, everything else. So we are designing a new hardware platform every 12 to 18 months.
Um, by the way, that's pretty amazing just to hear that, right? Every 12 to 18 months a brand new iteration.
I mean, yeah, we had a Figure one walking, like, by, by the time I filed, filed the C Corp, we had the robot walking in under 12 months. Another thing you've done is you've completely vertically integrated. Yeah, that was, that a necessity? Like there was, there's no supply chain for humanoid robots. Like there's no like motor vendors, actuator vendor, sensors, battery systems, structures, like kinematics, like all the software, which is like pretty vast. It's like firmware, embedded systems, operating systems, middleware, controls, AI. So walk me through your fa, your factory. You walk me through it before, but like what are the different segments of what's going on there?
Yeah, in terms of like design for, how? Yeah, I mean, you've got, you've got component building, testing, integration, all those things. So we, um, so we clean, clean sheet design everything from basically the ground up. Uh, like all the hardware is like clean sheet design. We look at like ultimately what does the product need to do? The product needs to, um, you, you basically want to talk to a robot and you want it to just do things without any human intervention. You just wanted to go out and do stuff in the world. Um, so we're designing for a capable robot that can go out and do everything from putting robots in a home to walk your dog, make coffee, um, do the laundry. And then the commercial workforce, which is like roughly half of GDP, is human labor, so it's like the largest market in the world.
Yeah, 1101, 1220 trillion dollars, the global GDP. Your TAM is like 50, 50 to 60 trillion. That's pretty good.
Yeah, it's like, it's going to, it's going to build the biggest business in the world by a long shot. Like this, in, in our lifetime, like the, this, the space, uh, yeah. So we have basically, we, um, so we're looking at like the in-markets where the robot needs to go. We do all the hardware design, which is like kinematic design, joints, motors, uh, battery systems, sensors, um, we do all the software, firmware, embedded systems, controls, all the AI work end-to-end, um, and then we do all the testing and manufacturing and integration and fleet operations and deliver those to the CL. So we have robots now. We have two commercial customers. Um, the first was BMW. We have robots there that are operating every single day. Um, they're in Spartanburg, South Carolina. They're helping to build cars. Uh, we got some video I think from the, from the uh, BMW plant if we can roll in background or repeat that video. Yeah, we'll show that. And we have a second customer we just signed, and then within 30 days of like starting the work, we were doing the work all end-to-end with neural nets. And um, this is like one of the largest logistics companies in the world. And then we're, um, we're also pushing really hard on the home. So, um, yeah, here's a quick update for uh, for BMW. Um, so we have just robots here, um, that are basically doing, like, basically putting sheet metal on fixtures. Um, this is a job that every major manufacturing company in the world does. Our robots are doing that fully autonomously at, at the speeds we need to basically, um, hit high performance, um, with no human intervention, no faults, no failures, and no drug testing, no days, no days off, no days off.
Yeah, 24/7 totally. I mean, it's a, it's an interesting thing, right? Think about this. Um, let me jump into one thing. In volume in the future, I believe I heard you say you'll see these at a price point of 20 to $30,000. You still hold that?
Yeah, we like done a lot of work on the build materials. Like if you start breaking this down, like to the bare, like you kind of just basically look at it line item by line item of what it really looks like and what basically what it looks like in like high, high-rate manufacturing. There's really nothing in the system right now that would show that this product should be very like extremely like expensive. The calculation I do is if I, if I was going to lease a $30,000 car, it's about 300 bucks a month, which is, by the way, $10 a day and 40 cents an hour. So here's my question: how many of these humanoid robots would you own at 300 bucks a month operating 24/7, no complaints, no fights with the girlfriends or boyfriends?
I mean, the number could well be multiple per human.
Yeah, you're going to want one. They're going to see like, I woke up, like I wake up every morning and help unload the dishwasher and pick up kids' toys. Like I never want to do any of that ever again. Like I, you know, it's just like not like something I need to be doing when I get home or I'm, I'm at the house. Uh, we really haven't had a lot of innovation in the home for like almost 50, 70 years. We like same appliances, same stuff. Like we need old, we had old robots, we called them dishwashers, now they just like been around for a long time.
Yeah, and US humans are having to like work with it, right? Like we have to work with that machine every day, and it's just like not something you'll do anymore in the future. You'll just like talk to the robot and have it do it. It'll be on a schedule. Any moment you can just call it, text it, talk to it, and it's asking to do stuff, and it'll just go do it. It'll know you better than it'll know you just like yourself.
I remember a couple of years ago, I'm very proud, uh, Bold is an early investor in, uh, in Figure, and I, I brought Tort to, to meet you, and I said, listen, the thing, first of all, uh, Brett's an incredible operator, multiple successes. What's one of the best predictors of the future? It's what a person's done in their past, right? It is very much one of the best predictors, uh, but what I found amazing, uh, that sold me instantly beyond your charm, uh, is the team you pull together. Can you talk about that because it's, I think a lot of people in the audience here are focused on their moonshots? This very much is a moonshot.
Yeah, um, you, you exit Archer. How did you capitalize? What did you start? How do you pull your team together? You describe that early moment.
Yeah, like, um, so you know, I haven't founded a lot of companies in my lifetime. I get to like go back every time and like, what did I mess up on? What did I get right? Try to make things better. Um, fundamentally, the things that I spend a lot of my time is just like building. Basically, in order to build like one of the world's greatest products, you need like the, like one of the world's greatest teams, and then you need, um, you need to align that team with like what the shared vision is, and everybody needs to be accountable for that and understand it. And then you got to figure out how to hit the gas pedal, like, really hard. So the entire culture at Figure, even at Archer, when I built, um, initial team was, um, was like very deliberate. And even at Figure, if you go to the website now, we have like the culture deck, we have the master plan, we have like things laid out that are like really unique. We're in Silicon Valley, but almost like the anti-Silicon Valley. You have to work every day in the office. We work five to seven days a week. We work really hard, um, and not a lot of people want to do that, and that's fine. It's just not the right people for us. Um, we've assembled now a couple, like hundreds of, like the best engineers and AI robotics in the world. There's just like no, nobody even close to what we've done.
Seriously, like incredible.
Yeah, like it's unbelievable. Like my whole business team has been with me at Veret, Archer, now Figure. They're just, I mean, we've spent 15 years, years together. There's unbelievable operators. They give me the ability to like spend basically all my time on product engineering, uh, to basically build the best product possible, and they help scale the business, which is great. Um, hiring, just, recruiting, HR, like, uh, legal, just Finance, across the board, uh, they're great. So yeah, the team's insane, uh, but what's even better is like the culture is just absolutely like dialed in. Like everybody knows what they should be doing. Uh, I don't do 101s, things like that. Um, we have like a shared vision, what to do, and we work really hard to go get there. And the dopamine that we all get is the same. Like we, um, we want to ship product. That's what we're aligned to, like that's what everybody, like, basically, yeah, gets their dopamine, which is really great. So it's like this shared, uh, fuel that we have to ship product. And this is such a hard thing. Like this humanoid stuff is like, it's like a maybe one of the most complex things I could have worked on. And um, you just, you have to have that fundamentally or there's literally zero shot this is gonna work.
You know, we're going to hear from Travis Kalanick tomorrow, who's going to say very much the same thing, that your, your what we call your massive transformative purpose, that that clear mission, vision, and then aligning your team and culture around that, when it starts with you. So you made a commitment of your own capital to get it going, and then you start calling people at other companies and, uh, what was your pitch to raise capital? What's that to raise capital or recruit?
No, no, no, to, to get those employees on board.
Oh, um, the pitch in 2022 was, I'm going to fund this whole thing for many years, um, you know, we, we, and it was expensive. Like we got to a million a month of burn in six months, so it wasn't like, but I was like full pedal to the metal from day one. I just like knew exactly what to do. I mean, Archer is kind of like a flying robot in a lot of ways, um, so I, I, I knew how to build teams. I know how to, like, we knew the product, what to do. I knew the technical understanding of like the powertrain and control systems and beta software and sensors, um, so it's was like, you know, we just like went really quickly out of there. The pitch was like, Hey, I'm going to fund it, so there's like no funding risk at least in the near-term, like next couple years, uh, there's a good chance for us to build like the next, I like an iPhone moment happening with humanoids. Like it's going to be, this is going to happen right now.
And what did you tell them the probability of success was?
Uh, pretty low. Like, uh, the thing, the thing that we had to do was like, we had to do, we, we needed to prove like three things that have never been done before that you had to go get all three of those right in the next like sub, like, you know, sub five years or you fail for sure. You have to build like incredible hardware for humanoids that's like extremely complex. It can never fail. It's always got to work, and it's got to work at human speeds with human range of motion. Nobody's ever done that before. Like most robots that walk around can't even walk right, like they fall over all the time. It's very complex, like maybe like rocket turbo fan level complexity in terms of hardware systems. Um, the second is you need to be a, this is a neural net problem, not a control problem. You can't write code your way out of this. You can't hire PhDs with a robot and solve every problem. You have to basically ingest like, uh, human-like data in the robot through a neural net, and it's got to be able to then imitate what the humans do. So you have to solve that, which has never been solved on a humanoid system of like, um, you know, it's like a high dimensionality system, not like a robot arm on a table, which most of, none of those have AI. And then the third thing you have to do is you have to figure out how to generalize. You have to do something that's a holy grail of robotics. You have to figure out how to look at something you've never seen before, through speech, tell the robot how to do it, and then be able to execute that task fully end-to-end just with one neural net. So the, the, the, you know, and I wrote about this in the master plan in 2022, it's like we need to solve those. If you can solve those, you're in the right decade. You're going to go build the iPhone moment for this whole space, and we're in full lift-off. But like, but those looked pretty dire at the time in 2022. There was just nothing out there. I mean, you had Boston Dynamics that was like leaping around and doing back flips and parkour and stuff, but like nowhere near the level of like manipulation and dexterity you needed for humanoid robots to enter the home. So, so, um, I think we can confidently say now we've, like, we have solved or we're making substantial progress on all of those.
Amazing. So which is great. So like I think like [Music] yes, there was a pivotal moment, uh, late last year where you said OpenAI was a large investor, and you were baselining OpenAI's AI systems, and you made a critical decision. Say, nope, we have to build our own AI internally, uh, Helix, can you speak to that moment? And I'd like to show the video of Figure at home along that lines.
Yeah, that'd be great. Um, okay. So what you're seeing is Helix. This is our like, um, this is our like a large-scale AI internally, um, it's like a basically a large-scale like vision, language, action model. And this is public. It's on our YouTube. Um, so the prompt here that Cory gave, he leads the Helix team, was, um, putting groceries on the table. And the prompt was just, put the groceries away. Not telling you where they go, not telling you what they are, just put them away. And the trick here, like the, the, the, the tricky part for the robots, they never have seen any of the groceries before in training. We purposely withheld all of these items, so it's like the first time the robot has ever seen these in its life with its own cameras and sensors. And so you basically have to solve like the generalization problem in a home. Every home is different, like, you know, we all have different like toaster ovens, we have different appliances, we have different like spatulas and silverware, like every, and it's located differently, and things are changing throughout the day. So you really have to solve this, like I call like semantic intelligence, but like it's like a semantic grounding that's needed from a human world to robot world. And um, Helix, um, we can talk about why I was able to do that, um, is able to communicate on a single neural net on each robot and collectively together able to put these all away, um, with just a single English PL. And um, so I think this, this is like the first signs of life. I, I think I will go even like more, maybe a bolder claim. I think this, this is probably the most important AI update for robotics in human history. Everything in the future that moves will be a robot, and it will be powered by AI agents like this. Um, this was trained on also very little data, like 500 hours of data trained in this. I love the way they're like looking at each other to confirm, like, yes, I get it. Like, oh, where are you putting that thing? Yeah, I think that's a good, a good idea to put it up there. Yeah, actually, it's human. Is that, is that a created, you know, like they're about to look at each other here as he passes it over, like, I get, listen, a part of this was like, uh, that's funny, uh, part of this was like, U emerging from training. So when the robots are doing handovers, they actually look at each other. There's actually a very split second where like one robot needs to release the package of the item, other robot needs to grab it, so it doesn't lose, like, basically, like hold of the item and fall down. So what happened emerging from training is the robots actually look at each other as the clearing, clearing way signal for like, we should be releasing the item into each other's hands, U, which was like really interesting. Uh, the other stuff of like robots looking at each other and moving around, like, um, I think it's just overall important. There's like a certain level of communication that needs to happen from a robot in terms of like, uh, interaction design with humans. So you don't want, like, you know, you don't want to walk in a room and have a robot just like not move and like not look at you. Like humans look and like do nods and gestures, like all of this is extremely important to learn. Like we need to learn these expressions of humans, uh, just like we need to learn how to grab items. Um, it's going to be super important as we at scale integrate robots into the entire world that this happens.
I have a thousand questions for you. Let me hit a few rapid-style here. Okay.
Yeah, let's do it. So Figure three, when do I get to see? I saw the designs. When does Figure three get shown?
Yeah, you keep asking this. You like this one. You saw it. It was a, it was a, I mean, you know, degree of beauty was increasing.
Yeah, I don't think people understand this, how like incredible.
Well, they don't, cuz we haven't showed it, but we, um, so we like, we're on, this is like the ones robots you saw here on the videos on stage where Figure two. It's our second-generation robot. Um, you can like kind of, I guess Figure one's like online a little bit, but it's like, it's a little bit more gnarly. It's like got wires outside of it, and it's a little bit more fast, and it was a much more, um, quicker design cycle to get this to our engineers to start doing real use-case work. Um, the Figure two was like a feature-complete robot that was supposed to be, is able to do almost anything a human, or vast majority of it. Um, you know, we haven't talked about this publicly a lot, but we, we're done now with Figure three design. Um, I think we'll, we'll probably show an update next week, just a quick minor, like a minor update, not, not, not anything material, um, as it relates to how like what we're going about for, uh, that process. Um, Figure three is like, you look at like Figure one to Figure two, and it's like a huge step up. You're like, wow, this looks from a college dorm room project to a real, like, like pretty decent robot. And like the magnitude of the step-up was pretty material. That same magnitude happened again on Figure three. So if you were to see it, it's just unbelievable. We spent like 18 months designing it from scratch. Uh, the high level, it's just like 90% cheaper, uh, it's smaller, it's less mass, um, it's got better sensors. It's hands, head, and feet were designed for neural nets. It's a completely, uh, I would say like, you know, Figure two is probably the best humanoid on the market, maybe, you know, probably not by a ton, but like I think it's the best 10%, 20%. Figure three is just like next-level design. Like we've spent, um, it's definitely like the most, like for me, like the most proud moment I've had in engineering in my career, like looking at that robot. And so, uh, we're going into production, uh, manufacturing with that this year. We'll have some more updates on that in soon. Um, that's the robot that we want to send everywhere into the world. Uh, we want to make it a low-cost, very high rate. Um, it's even better just on like, so many dimensions. Um, like me about production rates over the next three, four years, and when I'm going to see it in the home.
Yeah, so we have like two tracks. We have this like workforce track, which is like, um, and then we have the home track. Like the, what, what most people don't get is like the workforce is the big business. Like it's half of GDP. We can charge meaningfully more per robot than the house, um, and it's also easier. The, the, the things that the robot does is just like the same things on almost on a repeat. Uh, the home is like the Wild West. It's like extremely hard. Uh, we have a huge safety area of like not falling on like any human or hurting people. There's a semantic in safety of like not knocking over the candle and burning the house down. There's like, the home is just like vastly harder, like, um, maybe in self-driving, it's like driving on the highway is like workforce for us, and driving into the city is like the home. It's just like unbelievably difficult. Um, between our two first commercial customers, which are very large businesses, uh, we have demand like if we had 100,000 robots today that all worked, they would take 100,000 robots today. And so, and, and then we have like 50 customers I could sign by the weekend that are all Fortune 100 companies that we've like literally visited. We know them. We just like, we can't, I've, you know, done a bunch of meetings today at lunch. Everybody's like, what do you think about helping out here in healthcare, construction? All sound great. Like we're just like bombarded with the amount of demand here. You're thinking about like the workforce, you have like a certain, certain number of supply of humans. It's literally going down demographically. Baby boomers are retiring, so you have less humans in the workforce. There's labor pains everywhere, and you know, like there's a lot of job shortages. Like we can, so anyway, we see like just unbounded demand. I, I think we could ship a million robots this month if we like had them all working and they're ready to go. And I think one thing that we're going to maybe add before you go, sorry, I know you want to rocket fire, but, um, you guys saw BMW and you saw our second commercial customer. It took us a year to do BMW fully end-to-end at high, like high speeds. Like last summer, if you look at Figure one, it was four minutes. Now we got down like 40 seconds, and just a lot of great engineering work into it. We started working on Helix. It was just completely transformative, like completely. And then we said, okay, well, what if we use Helix for this next use case for the new second customer? And we did that whole thing in, in, in under 30 days from scratch, had nothing. And I think if we had to do it all over again, we could maybe do it in less than 48 hours. And so the robots are going to learn how to do something in like the matter of hours here, not like 10 years from now, like this year. And I think that has pushed our timeline left multiple years for the home. Like the, the, the hardest thing, like the long poll in the T for the home is like, is like, um, semantic intelligence, like can understand what the hell is going on anywhere it goes. So, um, under, over on the home is what we'll...
We start Alpha Testing in the home this year. Which means, like, our we'll be doing internal work on the home, like my home, or like, like engineers' homes. You want to get rid of that dishwashing—doity dude, I can't do it anymore. Just like, what am I doing? Uh, it's just like not something I want to do. Like, I want to spend time with the family and kids and wife. You know, it's like just no bueno. So yeah, we got to fix that.
Um, I feel—I mean, at this point we just feel data-bound in the home. Like we think if we just like increase the data set that we trained Helix with by like a couple orders of magnitude, it would probably—right now Helix, we put in like a we put a little note on the website about Helix, and one of the things we put in is you just drop like small household objects in front of it; it can pick, pick up almost every object we put in front of it. Like, we put up this like weird cactus-like toy like from one of the kids' rooms, and it was like, scene, and we're like, "Pick up the desert item," and it's got of like it's got to relate like a cactus to a desert-like, you know, plant, and it was a toy and it was singing, it was moving, and it picked it up. So like all of that is like in the weights, and it has like a very large like LM backbone to it, so it really understands the world—semantic grounding. So we think just like we just need more data now—like, basically data-bound for it.
So I guess there's a lot of confidence that um you're seeing a sign of life now that you haven't seen in history—that a robot, intelligent robot in the in the world can be built. And the question is, we just got to keep extrapolating that on like the curve far enough to where it's entering. And I think it's like this decade. I think you're going to see it in the coming years being put into homes, just through speech, be able to do like very long horizon hours of work without any prompt, with any fix. Everybody, thanks for listening to Moonshots. You know, this is the content I love sharing with the world. Every week I put out two blogs—a lot of it from the content here—but these are my personal journals, the things that I'm learning, the conversations I'm having about AI, about longevity, about the important technology transforming all of our worlds. If you're interested again, please join me, subscribe at dmandis.com/subscribe. That's dmandis.com/subscribe. See you next week on Moonshots. [Music]