Latest / In the Loop AI / How AI Brings Manufacturing Back to America | Aaron Slodov, Atomic Industries

Transcript

How AI Brings Manufacturing Back to America | Aaron Slodov, Atomic Industries · In the Loop AI

• Josh Kale: [0:03] If you ask the average person about the rate of innovation over the last 30 Josh Kale: [0:06] years, they'd probably say, yes, it's super impressive. We went from computers that were the size of a building to supercomputers in my pocket. And while that's true, that only covers half of the question. That's the world of bits, the digital cyberspace of ones and zeros. The world of atoms is actually a totally different story. If you look out your Josh Kale: [0:22] window, if you look in your living room, not a whole lot has changed over the last 30 years. And I think that's what I want to ask you, Aaron, is the world of atoms actually making a comeback? Aaron Slodov: [0:30] Similar to computing. I do believe that, you know, we've obviously made some great strides over the last few decades. But I do think that this idea of stagnation in the world of atoms, right, attributed to Peter Thiel, does have a lot of legitimacy, right? We've kind of neglected a lot of things on this front. And whether it's through regulation, through offshoring, right, through different labor tendencies and trends, there are a lot of very interesting forward momentum carrying kind of concepts that we see right now. And I believe that like the next few decades is going to shift from manipulating information and, you know, bits to manipulating matter, basically. So bits obviously gave us cloud apps. Atoms gave us semiconductors, EVs, vaccines, right? Like all kinds of really civilization critical things. But one is margin rich and obviously allows you to grow and recycle capital very fast. I think that, you know, AI is going to help be the bridge that allows us to kind of bring software scale productivity and like cycles to hardware scale problems. Awesome. Josh Kale: [1:46] Okay. So for the people who are just tuning in, we're talking to Aaron Slodoff, who is the CEO and co-founder of Atomic Industries. And I think a lot of people, Aaron, would refer to you as a physicist, but I kind of want to refer to you as a teacher, because what you're doing is you're kind of teaching machines to build the machines. And And it's this whole Josh Kale: [2:03] new way of revolutionizing manufacturing. So what I first wanted to find for everyone who's listening is why is manufacturing important in the first place? Why is it important that we get this right? And why is it so important to you to build this yourself? Aaron Slodov: [2:14] If this is just my handle on Twitter, but... And yeah, I did an undergrad in it, but I really believe that, you know, manufacturing is, it's so important because it actually converts ideas into power, right? This is how a country converts its ideas into power. And if you can't build, you can't defend and you can't lead. And eventually you just can't survive because you're dependent on somebody else. So investing in manufacturing and really pushing the boundaries is kind of... When you look at like on a historical civilizational scale, this is what civilizations are built on. Like being able to build and produce things specifically. And I don't think that investing in manufacturing is not just like subsidization, even though that's how some of our rivals have, you know. Gotten to the place that they are in right now, but it's about capability building, Aaron Slodov: [3:13] like training equipment, software standards. There's locating the full supply chain for various things within your countries. Josh Kale: [3:21] So when you say investing in manufacturing, what does that actually look like? Are you investing in the factory space? Are you investing in the human capital? Are you investing in the resources that are required, the raw materials that Josh Kale: [3:32] are required to manufacture? What does it actually look like to invest in manufacturing as a category. Aaron Slodov: [3:36] The way that I kind of view this is like a layer cake, right? And so on the top, you have all the really flashy stuff. You've got like the application layer, right? So drones, submarines, tanks, whatever you want, drugs, that's all on the top. That's what gets all the press and the news. People love the toys. But below that to even make that stuff, right? Like you have a layer of execution, which is the people and the know-how and the labor. I think about that as kind of like this precious pool of its own type of resource, right? So like, if you don't know how to do this, somebody else has to do it. And then below that, at the fundamental like low level is the capacity. So these are actually like the machines and the, you know, the actual factories in space and raw material. Everything kind of like goes in as an input to that process. So in my brain, at least, I see the industrial base as like a layer cake that's kind of divided that way. and investing in it, right? Like very similar to computing when you have this really high throughput of being able to execute a computation somehow, right? Like we measure things in flops. We have nothing like that for the industrial base. It's like how many atoms per unit time can you process, right? Like something like that. That's why I think the idea of the next few decades or beyond just being conceptually Aaron Slodov: [4:56] about how we manipulate that matter and like what the throughput looks like really. David Hoffman: [5:00] Downstream of Trump's whole like Liberation Day tariffs, I think everyone's just gotten a little bit more educated about the nature of manufacturing, David Hoffman: [5:07] Especially also after the whole like supply chain issues during COVID, which kind of just revealed how globally interconnected we all were and globally dependent we all were and some of the costs associated with that. So it's pretty interesting to see that the average bankless listener and the average person in the modern world who follows news is now just a little bit more informed about the nature of all of these things. So that's kind of the base of understanding that I think our listeners comes with. And also due to bankless's emphasis in the world of macro, we also understand the Triffin dilemma. If you have the global reserve currency, you have to export dollars and you need to buy things, which naturally exports your manufacturing base. And so this is kind of like all the education that I think the globe has really gotten informed on in late years, which brings up like two main conversations, which is, okay, we now want to reinvest in manufacturing domestically. And I think maybe all countries want to do this, United States specifically as well, because we once had this and now we no longer have this. We want it back. But also getting it back looks different than just investing in it in the first place in the 1970s, right? We're not just building out car manufacturing facilities. Whatever manufacturing is in 2025, 2030, 2040 is going to be different than David Hoffman: [6:23] building out manufacturing as it was. So maybe you can talk about that difference. So like now that we are trying to restore manufacturing domestically, what is different doing it today than is doing it like 30, 40, 50 years ago? Aaron Slodov: [6:35] I'll kind of start from the macro side, right? Because I think a lot of people push back on this stuff because they believe we don't have that comparative advantage anymore. And somebody else has built it. So why should we bother to rebuild our own comparative advantage? It's like, yeah, we don't want to do t shirts and toasters here and somebody else somewhere in the world is going to do it for for cheaper, basically. And people applaud that for some reason. And like. Aaron Slodov: [7:02] They've never been to a factory overseas, I guess, right? You can look at this any way you want, but making people slave away in a factory for like 18 hours a day for less than $3 a day is kind of insane. Like nobody should have to really do something like that. Did we just become so wealthy and so, you know, intelligent that we decided to stop innovating and building a comparative advantage, right, on like anything that we wanted to? Aaron Slodov: [7:29] And being a service-based economy, I think a A lot of people believe in that orthodoxy, that a civilization is going to trade off these things, you know, like over a long time scale. But something that I don't think people consider is that maybe that is just completely incorrect or that the economists that actually built those models, they were never necessarily against the idea of being able to reclaim comparative advantage in something like production. Or, you know, like, if you look at the, you know, the first and second and third order tertiary pieces of like an economy, agriculture and mining and inputs, right? Like, those can all still be a thing. But to answer your actual question, right? Like, I think that, you know, the distance from a supply chain, which I don't think anybody disagrees with this, but like distance kills iteration, right? Right. Like when your supply chain is halfway across the planet, you slow down, you lose quality, you bleed, you know, IP and domestic manufacturing isn't necessarily like nationalism. It's operational efficiency. Aaron Slodov: [8:35] It's just like being vertically integrated. Right. Like these are these are very straightforward concepts. And the question, you know, is like, do we want old timey, dark, dirty, dangerous manufacturing? Like, obviously not. Right. Like all of this stuff has to be rethought. It has to be rebuilt, reinvented and reimagined in a more like sci-fi kind of way. And when you, you know, when you co-locate design, production, and like all these things, you compress lead times. There's just huge compounding, you know, advantages to doing that, obviously. David Hoffman: [9:07] Yeah, is that co-locating, that co-location, the domestic supply chains, owning your own supply chains, is that kind of the predominant feature of modern manufacturing? Are there other features to also bring into the conversation too? David Hoffman: [9:19] Or is that, would you say that's the main one? Aaron Slodov: [9:20] I would say that that's a very heavy piece of it, yeah. Yeah, because being able to figure out how to do something viably, no matter how small it may seem in terms of the overall global supply chain, right, like the T-shirts and toasters kind of idea, you still need those feeder industries to do anything, right? So when we talk about like drones, for example, being able to make the motors and the propellers and the airframes and like the RF chips and the cameras and everything that go inside of these things, having the feeder industries to actually manufacture those domestically is important. And you can obviously, you know, big shocker, use that capacity and talent for manufacturing other stuff. But right now we're kind of like hamstrung, right? Because we don't have that here. David Hoffman: [10:09] So when it looks like, you know, rebuilding our manufacturing base, do we have to start with those feeder industries? You have to start with the small basic widgets in order to build up to the more advanced widgets? Can we shortcut our way straight to the top? How lengthy, how much work is there left to do to build out our manufacturing base? Aaron Slodov: [10:29] A lot. I would say, and there's, there's people on both sides of this, right? Like America is still the second place industrial base, right? Like, and first in a lot of things, but also lags way behind second in a number of things. You can go look at, you know, the, the NIST manufacturing report that comes out almost annually and, you know, skim through that and look at it. But I guess you have to kind of go back and look at like why we may or may not have outsource something right and trying to understand how to actually bring something like that back with the amount of right like input material and energy and like space and skill required for doing that kind of stuff right like and the machines so I just go back to the layer cake analogy basically so if you can figure out a way to short circuit something like that and do it more efficiently domestically, that's kind of the big question, right? So when you look at, you know, Elon producing Starlink terminals down in Texas, or just rockets or cars, right? Like there is a way to do this stuff. Do you need that kind of scale? It doesn't hurt, that's for sure, but, rebuilding these things from the ground up is a different game. And that's kind of like what my company does, but other people are figuring Aaron Slodov: [11:57] this out as time is moving on. Yeah. David Hoffman: [11:59] Yeah. Maybe you could actually just take us into the world of atomic industries. What would you say is your strategy for doing what you want to do? Maybe you could actually kind of give us the general pitch of what you're trying to do and then also the strategy of how you're trying to get there. Aaron Slodov: [12:13] Yeah. So, I mean, we're trying to build a 21st century mass production company, right? And if you guys are familiar with this, the underpinning of industrial society is that mass production helps us, you know, build a lot of these things. And to mass produce something, you generally need to, you know, design your product, and then you take snapshots of each different component inside of there. And then you make a mold or some kind of manufacturing tool that pumps out you know each of those widgets in a factory somewhere and the ability to make a mold or a manufacturing tool for a given widget is this like very precious tiny evaporating domain of trade knowledge right like that we don't have anymore and we've outsourced forever so the whole concept of what we're doing is trying to teach ai right like look at this widget now build the mold for it and move us from concept to production faster than anybody's ever been able to do. Right. And like we are vertically integrated and the idea of being able to design and build tooling or molds for anything and then put it into production and start pumping stuff out fast. Aaron Slodov: [13:24] And making sure that these designs are, you know, because they're computational in nature, they're kind of like physically optimal, right? So they're going to run super efficiently, they're going to be designed near perfectly. And the whole process that, you know, goes from start to finish is all done in a factory that we control, basically. So it's a super interesting way to basically crank the flywheel on the entire physical economy. And that's one of the things that we believe, right, like, is a way to short Aaron Slodov: [13:53] circuit that problem and do it here or do it anywhere, honestly. Could you give us. David Hoffman: [13:57] Just like the one-on-one of the whole manufacturing process? Because I don't think I've ever really heard that before. So you talked about molds. You talked about building the molds. So I think the molds are the thing that makes the thing, but you guys are also building the things that makes the thing that makes the thing. Can you just kind of walk us through so like you have input materials raw materials go into a factory you're building out some of that middle that middle processes and on on the out on the outside the output is the products can you just kind of walk us through the step-by-step how how raw materials turns into widgets on the other side and the parts that you are kind of like engineering in the middle we Aaron Slodov: [14:33] Basically like end up selling parts to people right like our customers want parts the input materials are basically like steel and plastic resin so they're like little you know pellets and we take giant steel blocks and cut them up into molds basically right so like airpod case this this came out of a mold somewhere and the design on this is very crazy but if you look around your desk right like most of the stuff probably came out of a mold somewhere and And so ultimately, yeah, we have to orchestrate a bunch of machines in a factory to basically go from like the design. Aaron Slodov: [15:10] You know, chop up the steel, make the mold, put it together, and then ultimately put that mold in another machine. And it's like cycling, right? So every time you open and close the mold, the parts are falling out of it and being transported into a supply chain somewhere. So this is like the pain of production, right? Because that barrier of saying like, all right, here's my product. I love it. This design is perfect. Snapshot. Now I want to go produce like a million of those a year, right? That whole process takes months and months to actually go from start to finish. And we're trying to reduce that, you know, as dramatically as possible. And the orchestration that happens in between, this may not be a shocker, but it's like, it's kind of similar to a self-driving car, right? Like you take the car as a system and you load it up with software and sensors or you give it vision and you're trying to teach it about driving around in the world. And what it's really doing is just controlling all the aspects of the car, right? It's very similar. So it's like a self-driving factory almost, and you just feed it designs, and then out the other side are being pumped out Aaron Slodov: [16:15] the parts, basically, that people want. So we're basically trying to obfuscate away all the horrible pain and suffering in between design and production. David Hoffman: [16:27] And so what you're talking about is, I'm assuming, what China has gotten very good at over the last 30 years. They have figured out how to build out this process efficiently without error, I'm just guessing. What are the skills that make somebody good or bad at this? Like, what are the pain points that you are trying to optimize? Is it something like when you're making the mold, maybe there's maybe you're imprecise in your tolerance. A product comes out the wrong shape. Now you have to build start from scratch and build a whole new mold again. Maybe that's a problem. Illustrate some of like what are the skills that go into actually doing this well? Aaron Slodov: [17:00] Well, this so this is great because this is like why it's a trade, right? So the humans that do this stuff and that are the best in the world at this, to Tim Cook's point, there's a famous video of him talking about a football stadium full of tooling engineers, like in China, and how America couldn't fill maybe the front row of one section or something like that. It's crazy. Aaron Slodov: [17:25] Basically what you have is people putting in the reps to solve and figure out these problems. And it's super, it's super interesting, right? Because every, every new shape that you want to ultimately like, have a manufacturing tool for is a new problem, right? So when you think about like language models, and if you guys keep up on this stuff, right, like, problems that lie with, you know, they're outside the distribution of the training data are generally like pretty hard for these models to do efficiently, right? So imagine that same kind of concept, but for a trade like mold making, where you're getting customers every day, every week with new widgets that they wanna build a mold for. So the more reps that you have, right? And going from that like end-to-end process from design and concept to final production and like parts in hand, the better. And this is why we, like in the real world, what we see is people specializing in certain industry verticals. So you might have a shop that's very good at medical devices or automotive components or aerospace parts, and it's because they've seen the same kind of domain or category of part over and over again. So they're familiar at least with, like how those molds ultimately kind of like function, how the materials that are used in that industry like work. Aaron Slodov: [18:50] It's a lot, right? Like you're juggling a ton of variables. So the people that are the best at this stuff, they've just put the reps in, right? Like they've seen hundreds or thousands of these problems over decades. And you go to them because they are super efficient at that problem. And, you know, our hypothesis is that capturing a very similar level of explicit, you know, knowledge around these problems is something that we can use as a general model to be able to do this stuff. Aaron Slodov: [19:18] And then, you know, short circuit, anything that gets like thrown at us, basically. So yeah, I'm. Josh Kale: [19:24] Curious to understand how you guys specifically are putting in the reps. I love the visual of the stadium full of engineers who are very good versus our front row. I'm curious, what is putting in the reps look like for you? Because I understand AI is at the frontier of a lot of what you guys do. And AI has been a tool for leverage. So is it possible for these new leveraged engineers who are just sitting in the front row to eventually acquire enough bandwidth as the stadium full of these Aaron Slodov: [19:47] People yeah and that's that's very very spot on so basically what's really interesting about this fundamentally at like a low level is ultimately like you're presented with a shape right and you have to figure out how the mold for this thing works and basically it's it's kind of mechanics in a lot of way right so it's it's a very physics driven problem and it's kind of a deterministic problem but the you want to think about it like the solution space for finding you know the best mold for the for that shape this is what these human trades people do right like they use all of their tribal knowledge with a handful of tools right like cad software and some simulation tool to kind of zero in on a good enough design and the problem with this in the real world is that everybody's on a timeline so these people don't have forever to sit there and like iterate and iterate and iterate on on designs so if you take a physics-based approach to solving this problem right like you can shrink down that solution. Aaron Slodov: [20:50] Space pretty rapidly and then iterate through it with software basically right so we we can use you know deep learning and a bunch of other very interesting machine learning techniques in ai right to zero in on these more optimal designs right and you're just like hey I want it to be like this price and then it goes and finds that design right and like oh by the way now that it's on a computer you can do thousands or tens of thousands of these things like every day. Aaron Slodov: [21:20] And like in the real world, it's one person at a time until they're done. It's not like a Google Doc where, you know, 100 people can get on there and make it go faster. So what does that look like? Josh Kale: [21:30] You mentioned there are some problems that are kind of outside of the context space. Surely you're doing more than just interfacing with ChatGPT and saying, hey, like, how do I how do I create this mold? Is there is there like this reinforcement learning that's happening where where you kind of teach the AI as it goes, what's good and what's bad? Like, how does how does AI actually play a role in in creating these new tooling? Aaron Slodov: [21:48] Yeah so this is this is where it comes down to kind of being like a full stack vertically integrated problem like self-driving cars right so you have to be able to teach it when i'm designing like you know a water channel inside of the mold to cool off the mold where like how do i actually drill that hole or 3d print that hole so you you have to take into consideration the manufacturability of like a feature or something. And it's like, once you, once you build up enough knowledge around, if I design it this way, I violate the physics of that problem or like the drill breaks or something. Right. So you, you do, you kind of have to, you have to start at like a first principles approach where you're taking heuristics and tribal knowledge from people that have done this for a long time. Plus, right. Like the physics of the actual problem. And it's not, it's not today, at least like the domain of a language model. So we don't technically use, you know, these foundational models from the big AI labs to do this kind of stuff. We have to build it from scratch and kind of like from the ground up, teach it like where things are violated, basically. So if you're, you know, you're going through the process and like. Aaron Slodov: [23:03] Dropping it, dropping in this thing as like a CAD file, you have to be able to, you know, understand, know, like the drill breaks here or like the metal won't move this way or go this way or it gets too hot. So there's a lot, there's a lot, right? Like there's a lot of different weird problems that like very analogous to how humans learn it. We're teaching the software at the same kind of rate, but it's a, it's a single training loop, right? So we, we kind of like teach teach our software one time and that's going to be good enough for the next you know 500 years kind of thing and it will only improve the more of these jobs that we we do ultimately because we're collecting data from the factory floor right like the cnc machines and the drills and the 3d printers that we use we take all that data and feed it back into the design software and then ultimately you know when the mold is running in a factory and spitting out parts we can collect that data too right like what do the times look like what are the temperatures and all this stuff like all those things can be collected and fed right back in so it's that's our version of like the self-driving car model it's. Josh Kale: [24:09] Funny as i'm hearing you describing this i'm drawing ties to this video that i saw last week with tesla's optimist robot and it was this crazy robot that was running around dancing but it was the first time it was ever actually dancing because it had never been taught this in the real world it was all trained on this synthetic data in this virtual reality world that they created to train it and i imagine there's there's a ton of cost efficiency there so when i'm hearing you talking about putting in reps and and getting as many iterations as possible is that something that you guys also do are you able to kind of get these reps in in cyberspace versus the real physical world to save a lot of money and increase that that Aaron Slodov: [24:44] Leverage you kind of mentioned something similar to this a little bit ago but like you can but there's a there's a catch the thing that matters the most specifically for this because we're talking about production engineering right where the tolerances actually matter right so you you like the one thing about language models and like rl for robots and stuff like that is that currently there's room for error right like if the robot just moves like an extra millimeter in like one direction like that's okay you're not gonna like the whole factory doesn't go down but when you're when When you're actually like designing and fabricating one of these, you know, manufacturing tools or molds. Yeah, it's like sub five micron tolerance levels. Right. So everything that gets spit out of the software, there's absolutely no room for error at all. David Hoffman: [25:37] How big is the micron? Aaron Slodov: [25:38] It's a millionth of a meter. So, you know. I don't know, in widths of hair, I guess it's probably like, what, half of a hair? David Hoffman: [25:50] Yeah, I think so. Half of a hair, very small. Aaron Slodov: [25:52] It's very small. I could be wrong about that. I think I'm going to get so grilled for being a physicist and saying this stuff. Yeah, cut that out. Yeah, I think it's, I mean, it's tiny, right? And like, the problem that you're dealing with, especially for something like this, like you guys probably have AirPod cases, but when you look at them right here, yeah the design and just like how crazy the angles on these things are which apple is famous for, imagine this coming out of you know like two pieces of metal like that has to mate together, perfectly and there can't there can't be like a seam there and that like you guys have probably seen this on a lot of plastic stuff that has ever been made you see like that little seam or like the weird little like dot like that's where the plastic kind of you know entered into that mold it's it's it's crazy because you have to like hide it it's a it's aesthetic and it's a really interesting engineering problem so yeah I would say you can't really like compromise on stuff like this so having it be a physics driven approach is you know for the time being at least like how we are able to solve these problems and you know tying it into what actually happens in in the real world and building this like product, you're building predictive capability over time. It's the same thing as like driving the millions of miles in the car. Yeah. David Hoffman: [27:16] So we have this visual of a whole entire football field of Chinese like manufacturing engineers. Is that the right title, manufacturing engineers? Tooling engineers. Tooling engineers, yeah. That was taught under their ships were sharpened with the iPhone, which is a very precise, very good thing, I would guess, to get like trained on. And now this is kind of what we are competing with here domestically inside the United States. But the cool new trick that we have is we have AI. And so instead of having to humans learn, at least I learn over multiple iterations, like I need to learn the same thing like four times before it becomes knowledge in my brain. And that's just experience. That's just kind of how it goes. Trial and error. AI has a much faster cycle. Like it doesn't have to fail four times. It can it can collect data and adapt its behavior much more quickly. And so I guess that's one of the big advantages that we have in restoring manufacturing in 2025 and beyond. And then also just like I would guess the toolings are just more advanced these days. But that's really the huge competitive advantage is we can just learn faster. Maybe we don't have the people, but we can just learn faster using AI in order to produce more dependable, more solid, more precise outputs. David Hoffman: [28:32] Where are we on that arc of training AI? Is AI a burden for you because you are still training it? Is it actually starting to become helpful for you now? Like, where are we in that journey? Aaron Slodov: [28:43] It is kind of a burden because people can still do many of these things like, order of magnitude a little bit faster. So we want to augment, you know, people, basically. And the idea for the next like couple years, at least, is can can the AI that we're building. Aaron Slodov: [29:00] Make, you know, like one American mold designer, right, like as effective as 100 Chinese mold designers, right, like, or productive as one. So we we have this very interesting in between moment where it can do a lot of the unnecessary kind of repetitive, you know, work that designers don't really want to do. And like, even internally at our company, they're like, can you guys just like finish this magical AI? Aaron Slodov: [29:28] We can design more. We're like, yes, we're working on that. So it is it's super interesting, right? Because like, I think, as soon as you play with one of these things, like if it was a chat GPT, or, you know, like what we do, there is that kind of like, that chat gbt moment where you're like this is crazy you know like i can be orders of magnitude more productive using something like this ultimately the faster we can build and train this thing the better and i think that that's not just for like what we do but for for a lot of things right like taking this very precious tacit knowledge that people have and being able to preserve it replicate it scale it and then like you know building a new workforce on top of that I think is, it's unbelievable and it's super powerful. Aaron Slodov: [30:15] And I think that's kind of another way that we can heavily incentivize, right? Like people going back to work in this stuff, because now you're kind of like an orchestrator of AI and there's people that'll push back and be like, yeah, well, you're destroying, you know, all the underlying skill and the knowledge. Aaron Slodov: [30:31] That people used to have to put the reps in to build. And I would also push back on that too, and just say, right, like, there's nothing stopping them from also learning the fundamentals while they're, you know, learning and being augmented through an AI like that. There's no reason you can't do both, right? Just because I use ChatGPT to write an essay doesn't mean I can't also learn how to write. Aaron Slodov: [30:55] So I think that a lot of these things can go hand in hand to train a workforce, you know, way faster. So instead of 20 years, how about just like two years, right? Like teach people the fundamentals like let them be augmented by these things and grow and scale really fast and now that you're kind of like a superhuman, able to do this stuff you get paid more right it's amazing there's tons of benefits across the board and you get to work in like a sci-fi factory too so yeah I think it's a really powerful thing because people love actually making things right like there's a very I don't know natural implicit kind of, idea of building something with your hands and seeing it like pop out the other side so. David Hoffman: [31:38] So i guess that layers on top of the the previous benefit that we talked about which is the the supply chain so if you can eliminate the global supply chain and all you have to do is transport materials domestically that's already a huge cost savings and then the second layered on top of that cost savings would be the classic example of just like well it's not going to be somebody it's not ai is not going to take your job it's going to be some high performer who's using ai who's going to take your job. And then you layer those two things on and then we have a domestic manufacturing revolution here. I mean, like this is this is the thing that we talk about at Bankless with our with our writers. It's like we using AI is an expectation. You are expected to use AI in order to be the best performing person at your job. And this is like going to become like a 100 to one like ratio in terms of leverage on your on your work. And this all got started, of course, when Garry Kasparov lost to Deep Blue in like the early 90s or something. And then what happened next was the birth of the AI plus grandmaster versus AI plus grandmaster chess world. It's always AI with a human rather than just AI. And so we're taking that, we're applying it to manufacturing. And so like, I would guess like, what kind of products are you making today? Like, where is this AI process engineer, this manufacturing engineer, plus the human, what are the materials that you are building today that are being manufactured onshore? What are the levels of sophistication that we're doing right now? Aaron Slodov: [33:04] Yeah. A lot of the stuff specifically, I can't really get into too much detail about, but if you think about different industry verticals, right? Like automotive parts and components, stuff for consumer across consumer electronics, I mean, CPG goods, right? like everyday kind of household items. There's a ton. Industrial components, medical devices, and then, you know, defense, obviously. And a lot of these things are... Standard components that go into products that you use every day, right? You just don't know where these things actually come from ultimately. So what we're seeing right now is even before the election, right? And to your very early points, right? COVID set a lot of this stuff off. I think when supply chain became a household term, right, when people were running Aaron Slodov: [33:56] out of toilet paper and like basic stuff, it freaked a lot of people out. And it really did put a huge spotlight on the sensitivities of the global supply chain, right? Aaron Slodov: [34:07] And do we actually need to eliminate the global supply chain and just move it all domestically? Like, no, not necessarily. But I think that having an approach like this allows us to reselect, right, like a level of globalization that we feel more comfortable with moving forward. Aaron Slodov: [34:25] Because I think a lot of people did not expect this type of outcome that we are finding ourselves in, right? Like, you see these like massive sensitivities in the supply chain, but then at the same time, we have enriched our ideological rival, right? Like over the last four decades, and now they are, you know, knocking at our door, or at least saying that they're going to just do whatever they want. And that's kind of interesting and frightening to a lot of people. But yeah, I do think that this kind of stuff moving forward at least is going to be super powerful for the country to like reclaim you know more of its sovereignty and its ability to to also produce what it needs and wants and be a little bit more independent and yeah i mean the opportunity is ridiculous i don't know it's huge manufacturing as a as a total of gdp is like yeah 14 15 trillion dollars right technically larger than i think tech as a whole but yeah these markets are gargantuan and if you do something like this where you're compressing you know the. Aaron Slodov: [35:39] The dev cycles, basically from, you know, months and years to days and weeks, you basically create a new economy. It's very similar to what AI is going to be doing in the digital world. Same thing for the physical. David Hoffman: [35:50] I definitely want to talk about the TAM of that and really how big all of this is. Before we get there, though, I do kind of want to understand like kind of where we are in being able to like access that TAM. So I think we're still we're still like early in this whole combination of AI and manufacturing. We're doing some of the high value stuff the national security stuff the the things that are non-negotiably must be done inside the united states the high the high value stuff that's important and then i think as you build up the skills and increase the capacity we'll start to go down the tail and into the long tail as this thing kind of scales out and i key i think i see like two different vectors in how that happens you get better at what you do and then you also do more of that so like qualitatively and quantitatively where where are we on those two kind of like spectrums here like how how quality like if we're working on up the stack of qualitative measures how far up are we and then where are you in trying to like multiply your work horizontally yeah Aaron Slodov: [36:47] I mean we're we're just in the beginning of this stuff right now right and this is this is something that like we we have to toe the line on very carefully. Aaron Slodov: [36:57] When when we're actually building one of these companies or one of the first of these companies it is very very early days so you have to do exactly what you're talking about right like you have to show people with the high high value stuff and then you can kind of work your way down the value chain into these much much larger opportunities but greater in number but the idea of being able to use your capability as a way to you know extract more margin out of something that typically in an inefficient market, right, like is hard to extract margin from. And so like a lot of the processes and, you know, shops and people that are doing stuff in America today are still working more on those like, you know, those easier targets, or more complex things that they can extract more margin out of. And I think, you know, throwing a layer of tech on top of that. Aaron Slodov: [37:52] Allows us to kind of like scoot down the the value chain a little bit more and open up that tam like what you're what you're talking about basically so i think you'll you'll kind of see more and more people taking these like vertically integrated you know whether it's with ai or just like being software defined you know from the ground up approach to you know the the thousands of different industrial processes that make up the industrial base and like in that 15 trillion dollar market, I think you'll see it like start to saturate more and more over time. And the idea of being able to, you know, demonstrate it on like a small scale and then go to people and, you know, try to get their help to scale it. Aaron Slodov: [38:33] I think the optionality will open up there on like how... Unfolds and who you know is ultimately backing and helping to build those companies and yeah, doing that right now is it's tricky right because like we have to use venture for that because you can't just walk into the bank and be like yeah give me give me a 10 million dollar factory and let me try out this like ai thing so it's tricky for sure i'd. Josh Kale: [38:57] Love for you to walk us through the actual specifics of this addressable market because i think a lot of people listening probably don't understand quite how large this thing can get an example i love is is apple when they released the iPhone, I think was worth like $70 billion. And at the time, I think it was Exxon, that was maybe at $350 billion. And if you were looking at Apple at that time, that was probably the perceived cap is about a 5X because surely it's not more valuable than oil. That was the most valuable thing in the world. Fast forward, they reached $4 trillion in market cap. So I'm curious the downstream effects and the size of the market as we get another one of these large unlocks, as we get humanoid robots at scale that can replace the productive workforce of the United States or as we get drones and delivery vehicles and just a lot of new ways of moving information and moving value around. How large does that market get? Are we facing a similar thing where maybe the ceiling was three to four trillion and now this next ceiling becomes even higher? How big can companies and can GDPs grow from the downstream effects of this? Aaron Slodov: [39:53] One thing to look at, right? Like look at the market cap of Foxconn. Right as an example here and i think i don't know what is it 75 billion maybe 100 billion, it's and they almost have like a million employees now or or byd even right like look at byd xiaomi any of these companies so foxconn's an interesting example because i think more than 50 of their revenues come from apple so when when you are basically the backbone of apple what is that what does that ratio look like i guess so i do i do think that that's a it's a fascinating concept but even foxconn right foxconn can only produce as much as you know like infrastructure they have in human and machine capital right so when you when you see these announcements of like byd trying to build a 50 square mile city that's basically just like a huge factory city. Aaron Slodov: [40:55] That's crazy, right? So you need some kind of lever there where the human capital is getting traded off for, you know, software automation. But in general, I think China looks at the ability to use its human capital as a huge lever, whereas like we can use it, you know, with an augmentation of AI to be an even bigger lever. Aaron Slodov: [41:19] Obviously, we don't have, you know, a billion people in America. But I do think that ultimately, yeah, the market caps and the opportunities for these companies are kind of like a function of their machines and their people, right? So like, if Foxconn could produce everything for Apple, Samsung, and like, I don't know, what's another gigantic brand, like, how big would it be? And how many people and machines what they need and could you offset that somehow by like you know an ai that knew how to do production and could you make those cycles faster and i don't know would you just be like spewing and pumping out like consumer goods to the world like everybody can now have an iphone because it's that cheap right like maybe an iphone is i don't know like 100 bucks in 20 years from now like those devices will become commodity or something right it's i don't know you could you can take this stuff to the extreme if you really want to. But I do think that manufacturing as like a concept is kind of a race to the bottom always, right? And that's why we kind of find ourselves in this situation. Aaron Slodov: [42:33] So it'll be a really, really interesting thing to see unfold, you know, like this new wave of industrial companies, because I think that the more they can handle and then produce, you get this weird, you know, Jevin's paradox of like, now we can produce more, people are going to consume more, yada, yada, yada. And like when when ai is happening on the digital side of this too right so like does the economy actually get bigger and keep growing at this explosive pace do we actually leave the planet right because now you have to produce stuff for space and like other other worlds it i don't know it's crazy just like the the compounding multiplier effect of all these things and then you need energy for that obviously energy might be like the next limiting factor ultimately Mm-hmm. Josh Kale: [43:19] That's what it seems like. So as we wrap up this section, I'm curious to hear a little bit about the reasons why we haven't quite been accelerating as fast as we like in terms of manufacturing. I have this question where if Congress handed you a one-time red button, you've raised or rewrite one federal rule, and in doing so, you can unlock a trillion dollars of factory output within five years. How would you address this? What would you create? What are the bottlenecks and the thresholds that are kind of in the way that if we're out of the way could accelerate this even faster. Aaron Slodov: [43:48] Do I have to only have one bottleneck? Josh Kale: [43:52] No, please walk us through however many you have. Aaron Slodov: [43:55] So, all right, I'm going to go back to the layer cake example, right? I think at the very fundamental low level, we need, obviously, like, we need energy. So energy has to be commoditized, or at least we need to have a surplus of energy. And, you know, it would be great to have economic zones with, you know, production capacity and they're all just, you know, directly like tied to a really heavy grid. Aaron Slodov: [44:25] We can draw on that power, whether it's like having a, you know, a nuclear reactor, like in your factory, just powering that factory, like a data center, right? Like the same thing, but just for mass scale production of whatever. And then obviously like raw materials. So we need, you know, we need an explosion of mining again. And I do think that that's super important, but that like, yeah, energy, raw materials, huge. Aaron Slodov: [44:49] And then the other two pieces of that layer cake. So on the capacity side, we don't really have the ability to make machine tools here. So like CNC machines and lathes, right? And like the actual tools and machines that we need, those basically all come from somewhere else today. So I would love to see somebody retaking you know that problem and blasting it into high gear like I'd love to see basically like Apple applied to industrial machinery like all software driven super modern flexible machines and I'd love to see an operation warp speed for machine tools basically. And then on the human capital, like, you know, upper level, upper, upper level on the upper level there, I'd like to see being able to incentivize people to go back into this stuff. Right. So rethinking labor laws, right? Like the way it is now federally. I mean, if you touch a machine, you basically have to be like an hourly employee somewhere. Basically, there's a lot of really arcane rules that were meant to protect people, you know, during the last industrial revolution that we need to like rethink. So I think that there's, yeah, regulatory stuff that we can work on in terms of like labor laws. Aaron Slodov: [46:11] Also, plenty of other stuff on the regulatory side too, but I do think that, you know, export controls are something that's really interesting. Certifications, you know, requirements and quality, like stuff like ITAR, right? That kind of stuff. And then stuff that's not necessarily like subsidy, but just rethinking like tax laws around standing up capacity and capability. So incentives, not just for like buying machines, but maybe, I don't know, like remember opportunity zones, maybe like opportunity zones, but for manufacturing, very specific ones, right? Like stuff like this. So there's a lot of, I think there's a lot of really smart things that we can do that don't necessarily mean that like DuPont is going to be poisoning your backyard again, right? Like for the next 50 years, but allow us to like move faster as well. So yeah, there's a lot there, but that's kind of how I would wave my magic wand. David Hoffman: [47:10] Yeah, yeah, there is a lot there. And it comes from all over. There's some regulations and bureaucracy stuff that you talked about. There's just like the whole need to bootstrap our own companies and manufacturing base that bootstraps the rest of manufacturing. So there's like second order bootstrapping that we have to tackle. It seems like a lot. It seems like it's very hard. Are you optimistic about the United States ability to do all of this? Because it seems to be that there's so many things to do. And a lot of they're all hard. They're all hard. It seems like every single frontier that we need to go in in order to do this is a hard frontier to push. So what gets you motivated? Are you excited to do this? Do you dread it? Like, how optimistic are you that we can even do this? Like, share us a little bit of your sentiments about this. Aaron Slodov: [47:59] I don't think there's an option, first of all, right? And secondly, I think you need people like me and other people like us that, you know, like Elon has a quote from somebody I think famously about building companies, right? It's like chewing glass. You want people that like to chew glass to do this stuff. Because to your point, right, it's hard. Aaron Slodov: [48:21] It's not fast. It's not easy. And I am super optimistic about it, right? Like nobody has come to the table with this much force about, you know, bringing American manufacturing back as a whole for a while. And I do think it's like re-industrialization is one of the most interesting, like unifying fronts politically that we all have. And yeah, sure, people have different ways of thinking about doing it. Aaron Slodov: [48:47] But I don't think that people necessarily hands down disagree with this concept. So I do think that like, I derive a lot of optimism from, you know, talking to people every day about this stuff, right? And like, you know, building this company, seeing the support that we have seen out in the public sphere, you know, a lot of what we do in, in this realm, at least so far has been super exciting and seeing all the companies that are popping up and trying to, you know, add momentum. I think like the more of these vectors that we like throw down and add, it's just like, it's not going to stop and I love that so I think yeah I'm super thrilled about the future of what this looks like and I think you know building one of these companies firsthand too gives me a lot of promise because I see it every day you know when I first started I didn't know that this was going to be possible I mean like I was going to make it possible somehow but yeah making it making it happen every day is like super super exciting and yeah. Aaron Slodov: [49:55] I don't know. Yeah, we can get into more of the stuff that we're doing on the back end. It's not part of the company, but yeah, I'm stoked. David Hoffman: [50:03] Yeah, I think we, I know what you're talking about. And I also, I think we want to go there next, but I do want to kind of paint a picture for what this would look like when this all kind of comes into formation and it looks like a reality. Like what lifestyle would the average American be living in 20, like 20 years? Is everything on Amazon is like cheap or free? We can build things with after drawing them in microsoft paint after two seconds like what is what does the lifestyle look like once this like domestic ai automated manufacturing base becomes extremely sophisticated in very high capacity Aaron Slodov: [50:37] Yeah i mean i i do think it i mean it's i don't know if it'd be like paint to you know your hand in two seconds but it's like low fidelity yeah i do think that the the entire pace of our world you know dramatically increases and i think that our ability to you know to have more comfortable lives and like this idea of abundance right like in that new book but even before that right like a lot of what sam altman says like having an abundant future relies on something like this right yes like the language models are great And they kind of act as this digital brain that with enough data, you can kind of teach it to do any kind of digital task, right? But most of what we ultimately have is an industrial problem. Aaron Slodov: [51:26] So being able to live in this new world of abundance is kind of what you're talking about, right? Like, the cycle of going from A to B on any industrial process is like dramatically shortened. We don't have to worry about these things anymore or at least like worry about them as much and it really opens up right like a lot of new possibility right, Because I think we can actually rebuild, you know, the middle class with a lot of this stuff. And I don't think people realize it because the focus is so much on like the factory job, right? Like maybe a factory doesn't need a thousand people in it anymore. Maybe it needs like a hundred. But, right, like the up and downstream effect of that factory job has as much or more impact on the economy as it used to, right? I think a pretty standard, well-understood multiplier effect of manufacturing in the economy was that A, it has the highest ROI of dollars invested back to the economy, almost three to one, which is crazy. And then in terms of productivity and labor multiplier, one factory job created seven other jobs, up and downstream, whether it was logistics, the retail, quality inspection, whatever. Aaron Slodov: [52:45] It's very interesting. right so like do do sci-fi manufacturing jobs multiply that even more right because if we bring back raw material processing energy like all this other stuff and we start exporting goods too potentially right there's there's a lot there's a lot of a lot of work to do and i think yeah it is it is going to be hard but i i think that looking forward that the country in the middle class and a lot of america could you know could really stand on this as kind of a new a new accessible industry for a lot of people that doesn't have like a super high barrier right like the technology does and if you guys I don't know how old you guys are but like you know growing up from the 90s to now and seeing just like how big tech and technology in general has like transformed the economy right like people being able to study computer science and like go off and, get a crazy paying job out on the coast somewhere, or just like go into a non-traditional industry that didn't have technology and applying that there, right? Like in finance or something. It's, I think that's going to happen again if we are staying on this trajectory. Josh Kale: [53:56] Yeah, we've, we've, uh, I've seen that optimism a lot from you and I really believe it. And, and, and one of the things I admired when we were looking for someone to find as this categorical defining episode, uh, we landed on you because of the, the optimism and drive That seems very genuine and relentless towards pushing American manufacturing forward. So first, thank you for that. And second, it led me down this rabbit hole where I found this thing called the New American Industrial Alliance. And that is something you created. So I'm curious if you could just tell us Josh Kale: [54:23] a little bit more about that. Aaron Slodov: [54:24] Sure. So last year, we put together a summit in Detroit. And the idea of this thing was answering this same kind of question, right, that I've answered probably hundreds of times over the course of building my company, which was what does this future look like, right? Like people want to know across the military, the government, industry, finance, tech, everywhere. And I figured it'd be a good time to take all of these people at kind of the stakeholder level or above and, you know, smash them all into a room together for a couple days and start building community around this and common understanding of where the future is pointing. Aaron Slodov: [55:06] So last year, I was lucky enough to have like a crack team of people, you know, join me on this crazy journey. And like in less than three months, we put together a grassroots, you know, event in Detroit, had thousands of people like RSVP to it had, you know, we brought 800 people in had amazing people sponsor it. But that's what it turned out to be, right? It was this really awesome Big Ten thing that we put together on short notice, and people loved it. And one of the takeaways, you know, coming out of that was, hey, we want to actually double down on this, you know, momentum, on this moment, and build something around it that's, you know, durable, and can endure, you know, more than just like a political cycle. And the idea of what NIA, or yeah, the New American Industrial Alliance was, is that, you know, we wanted to establish publish, a coalition of, you know, founders, builders, operators, you know, brothers in arm trying to basically like unfuck the American industrial base. Aaron Slodov: [56:12] I mean, I don't know if we can blot that out, but we want to, we want to put everything forward. Right. And we, we don't want to, we don't want to wait for bureaucrats. I mean, like even where we have to, but we want to do our best to have a big tent effort and do it from like the bottom up and yeah have have like everybody on board so like having the startups is kind of the the tip of the spear and then all the kind of like the heritage and smbs and you know the rest of the industrial base along with us so it's something that we started kind of on the back end of of that summit last year and in in you know a year's time, we've definitely made some great progress and looking forward to, you know, kind of unveiling what we've done over the last year because we're doing that summit again, in Detroit in July. And it's going to be, it's going to be great. Very much looking forward to it. Josh Kale: [57:07] That's super exciting for the, for the people who are curious about this, who, who want to kind of get involved, what type of skillset is required for manufacturing? Because you mentioned we have a shortage of tooling engineers, but perhaps software engineers now can kind of fill that void by using these tools. Like, what is the skill set that someone like you would be looking for in a Josh Kale: [57:27] person? If they wanted to get involved in manufacturing, what do we need more of? Aaron Slodov: [57:30] I kind of like to explain this in two different buckets. One is basically, you know, I need to make a factory to make something like now. So the people required to do that obviously have that skill already, and can be kind of like plugged in to be able to do that, right? And the second bucket is I want to reinvent the factory, right? So yes, we obviously need people that have some of that skill of the old factory, right? And like old process, but we need to bring in the new as well. So having a fresh set of, you know, eyes and skill from software engineering, because I think these cultures are, they're very interesting. They have a lot of commonality, but they're also very different, right? I think a very interesting common characteristic of either people from like manufacturing or software engineering, or tech, are people that just like want to grind, right? And like mission driven people, especially ones that like to get their hands dirty. I think those people thrive in something like this. And it's honestly crazy, because you have people that, you know, might've worked at Google or something. And maybe they had like, you know, a CNC machine in their garage or they had like a whole woodshop or something. Right. Cause, but having, having people that are super psyched about, you know. Aaron Slodov: [58:55] Actually getting their hands on steel and seeing like how how their engineering capability can transform a process like this is super powerful and on the other side right like people in manufacturing they work their ass off so there's there's no shortage of people that you know grind really hard there it's more just like the people that want to see and embrace technology kind of like, because for a very long time, these worlds have been like oil and water, you know, it's really, it's really, really hard to, to get technology to make a big impact on manufacturing, at least traditionally, right? Like we've just been, we've been like layering in software, just like slapping software onto a factory or something. But it's, I think we are, you know, like cresting that wave of, of showing people what like software defined manufacturing and like AI powered manufacturing is going to look like in the future. Josh Kale: [59:54] I'm curious if you do you see a shift of trend in the workforce because as I'm hearing you talk about this like earlier on childhood days I was building little rocket ships the model rockets and doing a lot of things in my hands and then over time the technology just kind of got so interesting that it was hard to pull myself away from it where there was a lot of really cool things happening in the convergence of like software and hardware now we have AI and it's it's kind of nerd sniped me in a way that was all encompassing but now that i'm hearing this the software is being applied to the hardware world in a way that's kind of accessible and exciting it feels like not listening to you speak i'm like man going to a factory and building really hard things sounds pretty cool and i'm curious if you if you see this shift happening are there people who have sat behind a desk their whole life who are now really excited to actually get in a factory is that something because traditionally the like the the big tech has this connotation where people aren't really working super hard they're just kind of getting by and they're shipping code and and how does that shift now yeah Aaron Slodov: [1:00:47] I mean that's 100 right i mean it's what when you when you can start demonstrating to people that you know this type of tech has a direct impact on you know you walk into a factory and see these huge machines just like chewing through metal it's it is kind of crazy because it's stuff where it's like yeah i never even knew that this happened and now now you can basically go from like you know your laptop in another room and then like go walk out on the factory floor and see like your stuff getting spit out the other side and you're like wow okay this is like this is a real a real thing and i think that these these factories are the more and more of them that we see they do kind of turn into these like kind of cathedrals in some way it's like it's a it's a very interesting experience you know because you kind of like you're taken aback by the the scale and the size and the awe of it, right? Like, I can't even imagine going into, like, the Tesla factory, you know, like the Gigafactory now in Texas. That thing is, it's absurd. Aaron Slodov: [1:01:52] I do think that I've seen a huge, you know, trend in this stuff over the past like two years. And the more, this is going to sound silly, but I mean, I do think the more, you know, more content that's out there, the better. And I think that, you know, showing people, because this stuff is so visual, right? And like, in a lot of ways, it's one thing to talk about it. But when you actually see it and like get to put your hands on it, it's just, it's a totally different, like visceral kind of experience. And i do think it captures people it's it's really easy to snipe people at this time it's like well. Josh Kale: [1:02:27] And to your point the scale like we have gigafactory and then just recently starbase is now an official city so you can go to texas and you can visit a city that is built around a rocket making facility so it's the scale is growing really rapidly and when when you hear things like this like oh now there's a whole city just dedicated to building rockets that's pretty badass that just seems really cool something people can get really excited Aaron Slodov: [1:02:46] About yeah i think so i think obviously there's some kind of you know like hedonistic treadmill for all people right so if you get exposed to enough of this stuff you just kind of get burnt out and you're like yeah whatever but i do think that like this is one of the really rare exceptions that it just keeps giving you know like from little kid all the way to full-grown adult that's i don't know 80 years old like seeing a giant machine it's like such a cool thing and i don't think yeah the novelty of that never really wears off and it just i don't know i think it helps show us that we are capable of building these like crazy things and that capability is like only going to get Aaron Slodov: [1:03:30] stronger and stronger over time yeah and. David Hoffman: [1:03:33] The timing of this episode i think is pretty good because the episode that we did right before you was with isaiah taylor talking all about nuclear energy and the prospects of all having a nuclear energy box inside of our homes to plug in. And I think it's pretty easy to visualize the idea that one of your factories or one of the factories that you help spawn is powered by nuclear energy on one side as the energy input to allow for us to build anything. And then on the other side, the output of that, we haven't done this episode yet, but we're doing an episode with the founder of Zipline, who's trying to just allow drones to take and transport all goods all around America. So you could just totally imagine a nuclear energy powered mass 3D printing, 3D printer machine that's output hops into a little drone and then comes to my doorstep after I purchased buy on Amazon like three hours earlier. So you can kind of see the whole supply chain starting to come into existence here. So thank you for joining us on Limitless today and help illuminate your part of that domestic supply chain that is helping us bring forward the future. Aaron Slodov: [1:04:33] Yeah, for sure. Thank you guys for having me. It's awesome. Josh Kale: [1:04:36] And also, where can people find and support? Either you or the Naya or Atomic Industries? Where should people who are interested try to find you at? Aaron Slodov: [1:04:45] I'm on X. My handle is aphysicist. And Naya is that newindustrials.org. And Atomic is atomic.industries. And yeah, we'll be out there. All right, everyone. David Hoffman: [1:04:59] Thank you for listening to the Limitless Podcast. And now we know a little bit more about what the future of domestic manufacturing here onshore. Aaron, thank you so much for joining us here today. Aaron Slodov: [1:05:09] Thanks, Dave. Thanks, Josh. Thank you. Music: [1:05:11] Music