BUILD SCALE

“Historically, models would be hosted with a single inference engine, and that inference engine would ping pong between two phases. There's pre fill where you're reading the sequence, generating kv cache, and then using that kv cache to generate new tokens, which is called decode. Some brilliant researchers across multiple different papers essentially made the realization that if you separate these two phases, you actually gain some benefits. You don't have to worry about step synchronous scheduling, and you allow yourself to split the work into two different types of pools.”

“Well, so the goal was we need to come up with some productionization of GPT four. And there's questions about, like, how do we turn this incredibly powerful model into products? And we're all spitballing ideas like writing bot, coding bot, you know, very natural at the time. Some of our least interesting ideas were a meeting bot. So it would just sit in a Google Meet, take notes, send and send out, like, to dos after. But John Shulman was very opinionated. He's like, we think we should keep it very general. Let's do a chatbot.”

“I don't want to spend my time on a small business anymore. If I'm going to put in the same amount of effort and stress, I want to make sure the ceiling is ten billion dollars, not ten million. You have to look at categories that are already massive but haven't been disrupted by a modern brand or a better user experience in decades.”

“Now we're in a position to use that core technology to apply it to different form factors and use cases. And that's where the indoor drone comes into place, because a lot of dangerous work happens indoors. It's where the fixed wing drone that can cover much longer ranges comes into play. And it's a very exciting, fun time to basically take these mature technology building blocks to be able to pretty quickly assemble them into a fundamentally new capability.”

“There's a massive opportunity in building an ACH distribution layer. Right now, moving money is still slower and more expensive than it should be for most businesses. If you can create a seamless way for companies to handle those high-volume transactions without the typical banking friction, you're sitting on a gold mine that every enterprise will want.”

“Dr. John Pemberton, a Confederate war veteran who had not only been stabbed, he had also been shot during the war and got army disease just like all these other soldiers and was addicted to morphine for the rest of his life. So after the war, he moves to Atlanta and as part of his entrepreneurial aspirations in this new patent medicine consumer economy, and also to probably solve his own problem, he starts casting about for other drugs that could cure him and others of army disease. And that is how in the mid-1880s, he learns about a new miracle drug, sweeping America, promising to cure all ill including army disease, cocaine.”

“Availability is the best ability. If you're available constantly, always able to work, and that includes, you know, your quality days, your easy days. If you're available constantly, then you can stack good day on top of good day on top of good week, good months, good years, and over time, that's when you see the consistency really start to pay off.”

“Brev was just it's a developer tool that makes it really easy to get a GPU. It was actually Evan Conrad, SF Compute, who was just like, you guys are two dudes in the room. Why are you pretending that you're not? And so then we were like, okay. Let's make the logo of Shaka. We brought surfboards to our booth to GTC, and the energy was great. My wife was, at the time, fiancee, helping me put these vinyl stickers on and she goes, you son of a, if you pull this off. And so, pretty much after the acquisition, I stitched that with the acquisition. I sent it to our family group chat.”

“For the design team in particular, the thing that is probably most important is the velocity of design work. So I typically meet with our designers for a couple hours every week, and we just look at work. I mean, new work every week, you know, hundreds of ideas, I would guess, you know, on a weekly basis.”

“Kimi two comes out, right. And it's an interesting model. The creators of Kimi k two actually talked about it in a blog post. Attention scales to the number of heads. They made a very specific barter in their architecture. They basically said, hey. What if we give it more experts? So we're gonna use more memory capacity, but we keep the amount of activated experts the same. We increase the experts' sparsity, and we decrease the number of attention heads.”

“And I think we, more and more, you know, we're starting to say, look, we know what the gold standard for safety is. We're building it. We're pretty far along. We're actually starting now to see other companies say, oh, maybe we should do it. Roblox has done we're seeing more and more governments say we like where you're going with this. Like, this is really cool.”

“I really think the most significant fact in this whole thing is how small it is. We're in the very rare and harder position amongst robotics and AI companies of actually having rapidly declining capital needs. And it's a testament, I think, to the strength of the core business, the demand for these products, having a really elite team that's capable of operating extremely efficiently, that we actually don't need that money to keep scaling and even make more and more aggressive bets in building new products.”

“It also informed a lot of our thinking about investing in other places that are really hard to copy, including hardware, where it's really, really challenging, you know, to copy our our fully vertically integrated stack around augmented reality. So I'm certain certain there's a lot more to talk about, there. But I think, you know, we learned early on that software is easy to copy, and so it's really important to build more durable modes.”

“The one of the reasons I ask is, I used to run this company, Color, and we built our own liquid handling robotic systems. We buy liquid handling robots, but then we have to adjust them dramatically. We had, like, cameras that would use ML to monitor the system and sort of make adjustments. We had to three d print parts to decrease vibrations on the platform because we were dealing with such small, volumes of liquid. And so there's enormous amounts of customization versus just having and the firmware for it was awful and writing against that was painful.”

“The philosophy for the company is that can we do a smartphone that is much better quality and we can offer to users at much more affordable prices and that's something that you see perminated in the history of our products, right? We try to offer, you know, very high quality solutions to our users at affordable prices.”

“The last thing about everyone coming out of Tesla, you're obsessed with vertical integration, you want to build everything in-house. We haven't done that at Exowatt to that extent yet. We've started working with contract manufacturers because we know we need to scale fast quickly, so they have that infrastructure. But Elon really taught us how to vertically integrate everything and build it yourself and build it better, faster, cheaper, and don't take no for an answer.”

“What really mattered was connecting you to the right people. And so if you could just connect someone not to all their friends, but to their best friend, to their partner, to their spouse, the people that they cared most about in the world, that that that's where the majority of the value is in the network. And so that's what really allowed us to grow in those early days.”

“A year and a half ago, we had the great honor of being sanctioned by the Chinese government. And it was a pretty aggressive action. They announced the sanctions, and then they showed up at the suppliers that we still had in China, shut them down, stop them from doing business with us, really tried to kill us. So that was a fun adventure for our supply chain team. They've done incredible work, and we've been able to maintain supply. And I think the good news piece of this is that we now have by far the most secure drone supply chain in the world, independent from China.”

“I have a twenty four seven agent running. I hooked up to run pod. It doesn't shut down instances, and I've tried prompting it. I've given the instructions, shut down when you're done. It's like, I need to keep it warm. I'll need it soon. It's horrible on time estimates too because it's like, yeah. I'll need it in forty five minutes. Forty five minutes of human time is actually three minutes of agent time, so it's like, I'm booting it up. I'm waiting. I'll just leave it on all night.”

“It depends like where you sit in the stack, right? Like um, now for Meta, it's like all it's all like creatives. Um, a lot of people are going to cycle into like AI creatives and then that's going to burn out. And then it'll be the next thing. Might be UGC creative uh, creatives again. Maybe affiliates. But it really like any good marketer uh, is going to figure out what the next arb is and they're going to maximize it. So, if you're like a real marketer, not just like one shot one shot wonder to understand Meta, then yeah, you're going to be able to you're going to be able to sell and you're going to be able to sell a lot faster in this new era.”

“For the design team in particular, the thing that is probably most important is the velocity of design work. So I typically meet with our designers for a couple hours every week, and we just look at work. I mean, new work every week, you know, hundreds of ideas, I would guess, you know, on a weekly basis.”

“You should know how a workout is going to affect them mentally and emotionally. Has anybody, and I'm going to speak from experience here, has anybody ever gone through a very hard, long workout and just afterwards you're emotionally drained and you just maybe cry a little bit? Like, your athletes will do that too. And if that happens, how do you get them from there into the next workout?”

“I built a business called Gruns. We make gummy vitamins. We started it about three years ago and just recently crossed that billion-dollar valuation mark in the exit. It was a fast ride, but we found a hole in the market where people wanted the benefits of greens powders without the terrible taste and the mess of a shaker bottle.”

“This is why you keep saying that it's a it's like a little secret that Roblox is actually an infrastructure company? That's right. Okay. You have not yet figured out how to create the technology to go with to do what we are describing, what you want to do. Correct? And that's what you're doing every day? I would say part of the job being interesting and fun is I think we have a reasonable idea of how we're gonna get there.”

“And, we we have a saying in the company, like, per performance is a growth feature. And, we put an enormous amount of work on raw performance features, scale features, those kind of things. That takes a lot of hubris. So What's a raw performance feature? We watch, how long it takes on a wide range of devices when someone clicks, I wanna play that experience, to the time till they're interacting.”

“One of the engineers on our team was looking at a reported material, property, and it was just sort of extracted values from literature. And It was really interesting to see the reported value spanned many orders of magnitude. And so you train a ML system on that and it's like, well, the best you can do is model this distribution, but you're no closer to, like, a ground truth. And that's where experimental data comes in, where you now have a grounding in this.”

“At a certain level of business, the actual product is almost secondary to the access you have. Being able to get into the right distribution channels or get on the phone with the right retail buyers is what separates the billion-dollar brands from the ones that stay stuck in Shopify land forever. Access is the ultimate unfair advantage.”

“Sam Briggs mentioned this about when younger athletes were coming in to CrossFit. And I have the utmost respect for all the young athletes, but I remember her saying this is that they just don't know how bad it's supposed to hurt yet. They don't know. They haven't experienced it. They haven't hit it. And so when they do, they just think that's right. Whereas as we age and we get older, and we know how bad that hurts, sometimes we're like, okay, I'm gonna go almost up to that.”

“McCann does when it comes on board with the Coke account, is it starts conducting scientific market research. And one of the first things they do, is they run a blind taste test between Coke and Pepsi. Consumers, when presented with the two drinks, Coke and Pepsi, in a blind taste test, a statistically significant number of people prefer the taste of Pepsi. They come, they present it to Woodruff. They present the findings, and his response is, do not ever share this with anyone, and do not ever run this test again.”

“I 100% agree that we are a few breakthroughs away from general purpose robotics, you know, that it's the dream that we are working so hard for. I think, again, if you want something commercially viable, something that will maybe make money or help some people in the world, I think a lot of those ingredients are already ready to have a larger impact than maybe even just a few short months or years ago. But for the true full vision of embodied, you know, AGI, I do think there is still fundamentally a few open research challenges left.”

“I program for their group, I always also follow, it is program for the best, scale for the rest. So I tell everybody on it that like, listen, this is a lot. When they are following it, I tell them, start out doing about 70% of each piece. That means that if I have in there a 60-minute run, go do 45 minutes, 42 minutes, whatever it is, the first week that you are on it. 70% of everything. If there is a four-round workout, do three rounds.”

“You could use like a like a drop shipping model to test out an idea first. So, I think the best way to test an idea to see if it works and to see if you have the right unit economics, it's like yeah, drop ship it first, run some ads to it, see what the customer acquisition cost is going to look like. Um, and scale it up. You could you could basically launch a site within like like under a week. You probably depends on how cracked you are. With AI for sure. Like website, you can AI with it. You can get a bunch of like fast creatives, static using AI or or UGC.”

“I think one fallacy is thinking about intelligence as a scaler. We've consistently seen these systems have a very odd spikiness. And it's actually possible to architect a system that is world class on some math domain, but then you could do some perturbations to the questions and actually degrade it substantially. So it's like a bad high school student. And so there's this, like, odd spikiness to these systems.”

“I don't think people contextualize how much energy that is. So if you talk about a gigawatt data center, that's almost a million US households of energy. So if a data center says, I'm building a five gigawatt data center, I'm basically building a city that's five million households. And they're building in a very tiny footprint, although not so tiny. The funny thing about data centers is, when they start talking about buildings in acres, you can start realizing how big these buildings are.”

“Everyone was doing powders and pills, but the compliance on those is terrible. People buy them and they sit in the cupboard. Gummies change the habit because they actually taste good. We took the most nutrient-dense profile we could find and put it into a format that people actually look forward to eating every single morning.”

“New Coke. The company immediately starts getting thousands of letters and phone calls every single day. One of my favorites is a letter that reads, My dearest Coke, you have betrayed me. We went out just last week as we had so often. And when we kissed, I knew our love affair was over. But last week, I tasted betrayal on your lips. You had the smooth, seductive, sweet taste of a lie. You have become corrupted by money, denying your ideals.”

“At a certain level of business, the actual product is almost secondary to the access you have. Being able to get into the right distribution channels or get on the phone with the right retail buyers is what separates the billion-dollar brands from the ones that stay stuck in Shopify land forever. Access is the ultimate unfair advantage.”

“Everyone was doing powders and pills, but the compliance on those is terrible. People buy them and they sit in the cupboard. Gummies change the habit because they actually taste good. We took the most nutrient-dense profile we could find and put it into a format that people actually look forward to eating every single morning.”

“I think it's one of the most exciting takeaways for me, at least, was the fact that the line, the boundary between what are perception problems, what are open world object recognition, and what is robot control. This line starts to blur, right? We do not have a pipeline system where you first take care of perception and you solve that and then you solve control after. We're literally just treating both of these problems as a single VQA kind of instantiation.”

“I think one of our chairman saying many years ago which resonate a lot with me was 99% of the problems there was already a solution out there. You just need to find it and maybe with AI you can find it fast enough.”

“You have to be willing to fail on ninety percent of your ads. We were pumping out fifty to a hundred different creative variations a week just to find the one or two winners that would actually scale. If you're not testing at that volume, you're basically just guessing with your marketing budget and hoping for a miracle.”

“Fifteen years ago, we essentially learned that software is not a moat, right, which is something that everyone is discovering today with AI. But fifteen years ago, because all the software features, that we could create were so easily cloned by our competitors, we started to think about how to build a more durable business, how to build a business, you know, that had bigger and more effective modes.”

“In 1934, Pepsi, in almost a last ditch effort to try and just do something to stay alive and save the company, tests using recycled beer bottles, which are 12 ounce bottles, to sell Pepsi, also for a nickel. The amount of liquid in the bottle is approximately free. Whether you're serving six ounces of liquid per bottle or 12 ounces of liquid per bottle, not going to impact your margins that much. Pepsi starts selling 12-ounce bottles also for a nickel. Twice as much cola for the same price.”

“Well, one of the aspects is, as you mentioned, rules are sort of subject to interpretation. And even if you have the same language, there are multiple ways to interpret it. So here's an example. So we said, well, don't do things that or don't interact with anything that's harmful. And I think there was something in the data set which like it's it's all a cigarette. And then it was like, well, I'm not going to pick up a cigarette because it's going to be harmful. Currently, I think our robots are more the problems don't come from the fact that they are too smart to work around the rules. It's just that I think they are too incapable of doing zero-sharp things in the real world.”

“What's new about our approach and our hypothesis is making this in a modular fashion, making this in a factory, and then scaling the production to millions and ultimately billions of these, which ultimately allows us to go down the cost curve and generate electricity at our ultimate goal of one cent per kilowatt hour. And so that's why we started the company. That's kind of been the goal North Star from the beginning.”

“We are making these human robots to enhance our own manufacturing capability and efficiency. We haven't launched any 2C robots. So all the robots that we are developing, all the humanoid robots we are developing right now are used in our own manufacturing scenario. I think the video already showed that there were two robots that are working consecutively for three hours with a very low margin of error.”

“In 2017, we started to think very hard about, first of all, how do we really achieve the mission? What will that look like? And we start to do the math on compute and you start to realize that it's gonna take a big computer. Elon, Sam, Ilya, and I all agreed that the only path forward for OpenAI, the only path to achieve the mission was to create a for profit entity associated with OpenAI of some form.”

“There are many moments along the way where you feel like it's real now, it's going to really happen. The economy is going to transform into this compute powered world. And I think that those moments are not yet at the end. I think that we have many more breakthrough moments where you realize that the next stage is possible.”

“And we just kept scaling PPO and we exceeded the performance of the best humans. And that itself was the finding, right, that actually massive compute with simple algorithms, right, that that is something where we cannot just—it doesn't just work in theory, it works in practice. We can really make it happen and in this incredibly messy environment where you cannot program it, you just need this almost human like intuition.”

“And what is a perpetual motion machine? It's something that can keep going, get better and better. That's what kind of the the notion of building a cloud three d UGC system. We keep building that system. Creators are gonna make more and more amazing content. We can keep tuning the system, and we'll get kind of that perpetual motion machine.”

“We came up with what I would really say is almost the technical plan that we have pursued for the past ten years. Number one, solve reinforcement learning. Number two, solve unsupervised learning. And number three, was gradually learn more complicated things. After that off-site, I sent offers to everyone and said, hey, we wanna get started in the next two to three weeks.”

“The rise of the forward deployed engineers tells us a lot about how AI adoption is actually going to happen in the enterprise. It’s not just about shipping code; it’s about having engineers who sit with the customer to understand the specific edge cases of their business. We are applying that same philosophy to IT services, where the AI handles the standard stuff and our best people focus on the complex, bespoke problems.”

“From a product perspective, I think the course that we're charting is towards our drones being like flying agentic AI, just like you have an agent that you interact with on your computer or in the cloud, this thing is an agent that can move and do more for you in the physical world and you should interact with it in similar ways. Like it should have the intelligence and domain expertise to be useful to you in that way.”

“And in line with kind of system thinking, we think of our company as the system. And the company, Roblox is really running almost as if it's nine separate companies. They are all very well connected. We all get together once a week and connect all these companies together. There is a three d cloud simulation and toolset company running within Roblox.”

“The main arguments would still stand for humanoids. One is that our world is sort of designed for humans. So one hypothesis is that if you design policies for like, they single out mobile managers, then once you solve a lot of tasks in that environment, then you see that it's limiting because many tasks in our world are like opening a bottle, or like opening a fridge and then taking something from it. So you have to keep the door open. Or even, I think some people say, well, you don't need wheels, but then what if you solve a lot of tasks on a wheeled platform and then there's a little curb on your floor or by a street side and then the robot is like stopped there. So I do think that ultimately, if you want to do a lot of tasks and be useful in environments where humans operate, you need to go to a human or as close to a human embodiment as possible.”

“You have agents that are infinitely patient, and what happens is you have these triggers that are firing at any given time. You have cron jobs that are firing at 2 a.m., which is what we have, that will ingest all of my content from my podcast, my YouTube channels. We have Cold Outbound launching on Saturday, sending thousands of emails before Monday, right? You can do that. We have Deal Resurrection crons that will find the right time to reach out to people.”

“I think for me, the understanding was like people used to think that all the robots are so different. All of their data is like so different. And every lab has or like they invest in like a couple of embodiments. It was just I think, post RTX, the idea was that people moved in the direction of thinking that all robotics, all robots are kind of similar. It's like, it's only as different as like English and Chinese or something. And the concepts are similar. It's just the manner of expression that's different.”

“This essentially focuses the light coming in from the sun onto the battery material. It gets very hot and then stores that energy. So, essentially, our battery is a heat battery, so what that means is we're heating up rocks, right? And then storing that energy in formal heat, which is very cheap as compared to doing lithium ion or any other type of electrochemical battery, and you don't have to get lithium or magnesium or cobalt or any of these other fancy chemicals.”

“One of the things that I always have to do is actually pull them back. Hey, let's pull back for a minute. You're running too fast right now, you're getting burned out, you're not able to hit a good day after a good day. So before we get hurt, be the brakes, not their breaking point.”

“Perhaps recently, you know, you know, for example, with this work, Pivot, maybe the answer is that actually there is some very good amount of physical intelligence already contained in these like internet trained models by themselves without any robot data pre-training or fine tuning. Again, I don't, I also don't think that like internet data alone, just watching, you know, Reddit threads and Wikipedia is enough to solve contact rich robotics. But I do think that we've so far just been like seeing the tip of the iceberg for the knowledge that is already contained in these, you know, large VLMs.”

“Mr. Lee himself is a entrepreneur, serial entrepreneur, but he's also a product person, right? I think if you look at when we get into the car industry, he personally has personally driven 150 different models of car himself and take detailed notes so that he understand what's on offer in the market and how can he improve on it. He make us all take professional racing car licenses in order to know how to drive properly.”

“Coding agents have been so much more effective than general purpose agents. And I think a large part of that is it just has access to the terminal, and that means it has access to everything that you've installed into your terminal. It can write code, and it can compile the code. And if there are errors, it can fix it. It can run your suite of tests because that's all just in your terminal. Computing began with a terminal with a shell, but we said that it's not empathetic to humans, so we built these nice user interfaces. And then now we have LLMs navigating our user interfaces, and ironically, we're not empathetic to the machine anymore.”

“Pure play software often struggles in these services categories because the friction of adoption is just too high for the end user. You can't just hand a traditional IT shop a new AI tool and expect them to re-engineer their entire workflow overnight. By being the service provider ourselves, we can bake the software directly into the delivery model and prove the value through better margins and faster response times.”

“So much of consumer technology focuses on, you know, am I building the right product? Do I have product market fit? Have I, you know, built something that's really gonna resonate, with customers that they're really gonna wanna use all the time? And I think people don't spend nearly enough time thinking about, you know, distribution and figuring out distribution. And that seems to me, to be a huge differentiator.”

“So these systems have become so incredibly impressive on this on this domain as a result of huge amounts of data, really cheap verifiable environments, like, you know, you can check Unitesco from failing the passing with just a few CPUs. It's basically instantaneous. There's no domain expertise gap between an AI researcher or software engineer. And, obviously, this will become and is becoming a larger contributor to the next generation of the system.”

“Yeah. So, language models are incredibly powerful. It's a very natural interface, and so we continue to use these. But we think about them almost as like an orchestration layer. So that's sort of a a copilot assistant, but also like a system that can direct, experiments. And it's almost it's orchestrating other specialized models as well. So we do construct neural nets that are specially designed for atomic systems where there's, like, some symmetry awareness, and those have much lower latency and they've been, like, fine tuned for that.”

“SOL is actually I think of all the lessons I've learned, that one's definitely my favorite. The speed of light moves at a certain speed. So if light's moving slower, then you know something's in the way. So before trying to layer reality back in of why can't this be delivered at some date, let's just understand the physics. What is the theoretical limit to how fast this can go? And then start to tell me why. Because otherwise, people will start telling you why something can't be done.”

“There's the other concept that is explored a lot at NVIDIA, which is this idea of a $0 business. Market creation is a big thing at NVIDIA. Jensen says we are completely happy investing in $0 markets. We don't care if this creates revenue. It's important for us to know about this market. We think it will be important in the future. It can be $0 for a while. An org doesn't have to ruthlessly find revenue very quickly to justify their existence.”

“Fifteen years ago, we essentially learned that software is not a moat, right, which is something that everyone is discovering today with AI. But fifteen years ago, because all the software features, that we could create were so easily cloned by our competitors, we started to think about how to build a more durable business, how to build a business, you know, that had bigger and more effective modes.”

“I think a default expectation in five years is if there's an emergency, you call 911, a drone shows up in a few seconds, and that's going to be everywhere in the US, hopefully everywhere in the world. And that's going to change the way policing works. It's going to get better outcomes, you're going to have fewer officer involved shootings, faster response times. And I think you can also do that while protecting privacy and transparency. Like these things are essentially flying body cameras.”

“We don't test. What we do is we race, and the race is our test. We use that as a performance metric. If I want to run a test on somebody, especially one where I want to get really good data, usually you want to like take a day or two off in front of it, make sure the test is their best ability, and then they're going to have to recover from it, and it could be up to like five, six day process, and then you just lost a week of training.”

“There's a massive opportunity in building an ACH distribution layer. Right now, moving money is still slower and more expensive than it should be for most businesses. If you can create a seamless way for companies to handle those high-volume transactions without the typical banking friction, you're sitting on a gold mine that every enterprise will want.”

“A lot of the money is moving out of software into fiscal goods. So, I'm like pretty bullish on physical products right now. Unlike software, there's like software really has no moat right now. I've seen Jackie Vibecode like absolute trash. Um, Yo, chill. If and if Jackie can do it and if I can do it, uh, there's going to be like a like 10 million other degenerates that are going to code up the same [ __ ] So, I think like at least for venture and like private equity and growth equity firms to invest in software is going to be very challenging moving forward. It's an unknown space and money's going to money's going to rotate to the fiscal.”

“I think, for example, right now, people are massively underestimating the role that human adoption and human comfort, you know, with advances in artificial intelligence will determine its deployment. I think technology leaders think that folks will just blindly adopt, new technology as it comes out, and I think we're gonna enter a period of time where there's gonna be a huge amount of societal pushback on a lot of the changes that are coming, with AI.”

“The local partners are willing to work with you developing products that are more customized to what we want and as a result it's not just faster but it's also our ability to offer much more customized solutions that fits better into our smartphone, that fits better into what we want and as a result, you know, we go to market we are able to offer, you know, products to our customers at some offer something that they really want.”

“And I would even emphasize that to expect such a result where the generalist outperforms specialists on the very niche domains that, you know, the specialists have kind of been overfit to, this was actually quite shocking to me. You know, like, I think there's been so many examples over the past years where people have tried to scale single task methods to multitask methods. And you definitely get a lot, you know, maybe you learn faster, you learn a more robust policy that's less brittle to small perturbations. But oftentimes, you have to give up raw performance, right? Generally, in a lot of cases, the only way to max out your performance on this one narrow regime that you care about is to train a specialist and overfit to that domain. And so it was really exciting here to kind of see positive transfer, where the generalist outperforms even this presumably very tuned baseline from the individual labs on their setups themselves.”

“In 1899, two guys from Chattanooga, Tennessee, named Benjamin Thomas and Joseph Whitehead, come to Candler with a proposal. They want to bottle Coca-Cola. Candler thinks it over and he's like, that's a pretty good deal. I've got nothing to lose here. So in July of 1899, the three of them sign a contract that includes the following terms for a token contract price of $1, which Candler never collects. The Coca-Cola Company will sell syrup to Thomas and Whitehead at a volume discount price of $1 per gallon. There is no term length on the contract.”

“I think marketers are primed to win the biggest in this space. Cuz developers are obviously clapped. Um, to give you guys some insight, we recently stumbled upon a tech walk in Vancouver. And half these guys are looking for a job. The other half is saying they like to think with their brain um, rather than use AI tokens. And I think I think Tony said something like if you're thinking, you're not using tokens, which is like NGMI, like automatically, right? Um, Yeah, I mean, there's a there's a thinking machine for us. Like why use that type of like deep thought, deep thinking?”

“We didn't reinvent the wheel with the ingredients. We just remixed the format. It's the same thing as taking a successful software product and moving it to mobile, or taking a successful supplement and making it a gummy. You take a proven value proposition and deliver it in a way that fits better into a person's daily routine.”

“We need to manufacture domestically. We need to source domestically. We need to make this using raw materials that don't come from China, that are not rare of minerals. And all of that is what we continue to iterate on. So every configuration that we design, we looked at the bill of materials or the BOM. And we asked ourselves, what can we eliminate from this? How can we make this simpler? And the same approach you see at SpaceX, right? Or at Tesla.”

“I do think designers feel vindicated in a lot of ways. Right? You know, a lot of designers had parents who were saying, why aren't you studying computer science? You know, what are you gonna do with this skill set, drawing things? You know, this doesn't make any sense. And I think today, you know, a lot of our designers are now shipping code, which is extraordinary.”

“The three elements are the optical table, so these lenses, and then you have the heat battery and the PC, or the power conversion unit, all in this kind of container. And the idea is that we capture energy from the sun throughout the day. So think of this as like a solar panel, but that has a built-in battery, and that essentially you can scale linearly to any project size. If a datacenter customer needs 100 megawatts or a gigawatt, you just stack these next to each other.”

“We started out kind of as training partners, and it didn't go into coach and athlete until I had back surgery. And then after that, it was more like I was just watching her more, but she doesn't like me watching her work out. So we go from there, but then she started to hand it over and she would trust me with it. And that is a heavy load for me that like, look, I don't want to break her trust. She will ask me every single workout, why am I doing this? And I have to have a reason because if there's not a reason for something that I write in her program, it doesn't get done.”

“So they create this design brief and they send it around to 10 different glass companies around the country that says, we want to develop a bottle so distinct that you would recognize it by feel in the dark or lying broken on the ground. The Root Glass Company of Terry Hote, Indiana, designs the bottle that goes on to win the contest. Something got lost in translation and the bottle was designed to look like the cocoa plant. The cocoa pod that you smash open to get out cocoa beans. This is a whole different thing called the coca plant, not the cocoa plant.”

“It also informed a lot of our thinking about investing in other places that are really hard to copy, including hardware, where it's really, really challenging, you know, to copy our our fully vertically integrated stack around augmented reality. So I'm certain certain there's a lot more to talk about, there. But I think, you know, we learned early on that software is easy to copy, and so it's really important to build more durable modes.”

“GPUs are so extraordinarily expensive. And what's interesting is just the compute cost relative to physical infrastructure is actually surprising where, you know, so much money is spent on the compute, that the physical infrastructure sometimes is actually lower, but, you know, has very large lead times and there's intrinsic difficulty of having these well calibrated, well functioning physical systems. But from a capital perspective, it's primarily a compute cost.”

“The AI race is so intense. It's now time to power, right? I need power, I don't care how you make it. If you talk about a gigawatt data center, that's almost a million US households of energy. The grid is just really not built for that. What that means is the data centers now have to bring their own power online because it's also important that when you plug in a data center, it consumes a lot of power from the grid.”

“Understand the difference and how they relate to correlation and causation. What is actually causing something to happen versus what just is happening at the same time as something else? If I program something or a week and the quality day of the next week, the athlete has a great fantastic day. Is it something that I did with the week before? Or is it something, did they get better sleep? Did they get better recovery?”

“Yeah, so we're very excited to be announcing our Series F. This was $110 million fundraise, $4.4 billion post-money valuation led by insider investors. And I really think the most significant fact in this whole thing is how small it is. We're in the very rare and harder position amongst robotics and AI companies of actually having rapidly declining capital needs. It's a testament to the strength of the core business, the demand for these products, having a really elite team that's capable of operating extremely efficiently, that we actually don't need that money to keep scaling.”

“You're gonna have a token budget for each of your employees, because if they're gonna be managing a fleet, they're gonna be consuming tokens, right? So for us, we're thinking about, how are we, what percent of our token budget is gonna be frontier models? And then what percent is going to be local models where we're running open source models? And those are obviously gonna be a lot cheaper. You wanna think of this as maybe the frontier models, that's like your Ferrari.”

“The ones pulling ahead already have agents doing real work, real systems that do real tasks with credit cards and everything. I have 12 agents running inside my company right now. They handle sales, content, SEO, and recruiting without my team touching it. One of the agents, the finance agent, even saved me 500 grand the first time I used it. The gap between who gets this and who doesn't is opening fast.”

“Have you ever heard David that Coke tastes better when you get it from McDonald's? The thing that is definitely true is Coca-Cola ships the formula to McDonald's in stainless steel tanks, instead of being delivered in bags. McDonald's does some stuff. They pre-chill the water and they make sure that the hoses are chilled all the way up in the dispenser. McDonald's apparently actually has a different syrup to water ratio that accounts for ice melt. They add a little bit more syrup than the standard recipe, which sounds like it would be heresy.”

“So much of consumer technology focuses on, you know, am I building the right product? Do I have product market fit? Have I, you know, built something that's really gonna resonate, with customers that they're really gonna wanna use all the time? And I think people don't spend nearly enough time thinking about, you know, distribution and figuring out distribution. And that seems to me, to be a huge differentiator.”

“We didn't reinvent the wheel with the ingredients. We just remixed the format. It's the same thing as taking a successful software product and moving it to mobile, or taking a successful supplement and making it a gummy. You take a proven value proposition and deliver it in a way that fits better into a person's daily routine.”

“What really mattered was connecting you to the right people. And so if you could just connect someone not to all their friends, but to their best friend, to their partner, to their spouse, the people that they cared most about in the world, that that that's where the majority of the value is in the network. And so that's what really allowed us to grow in those early days.”

“We had a mission, a vision of saying, we think that we can build human level AI, make it be something positive for the world, make the benefits be something that are distributed broadly, but how? And how do you get people to actually leave their jobs to come and join this thing? Initially, the set of people that I narrowed down to were actually Ilya, Dario, Amadai, Chrysola, and myself.”

“We have an agent known as Flash, which handles content repurposing. This piece of content over here, I talked about how to practically deploy Jack Dorsey's World Intelligence, which is exactly what we do at Single Brain with these revenue agents, 348,000 views on it. This entire piece over here, I might have helped edit it for 15 minutes or so, and you should have a human in a loop here, but this is a task that continually repeats.”

“I think there's a lot of headlines about data center projects being planned or initiated. I don't think all of them are actually happening. So there's a lot of like phantom data centers that are being announced. Let's call it that way. I think data centers are definitely in a rush to get online fast. They do have massive supply chain constraints. Two years ago, it was maybe NVIDIA GPUs. Now, it's power. Now, it's actually not even hardware. It's actually labor.”

“I'm actually personally bullish right now on um, like one-person agencies. Like if if if I had to make my first freaking 10k a month, I'd just be one guy with AI, armed with AI, probably like a Claude Max. And then just try to figure it out from there. Um, and I'd post every day on socials, like one shorts one short a day, and just go that that route. It's probably cuz I know that route. Like I know that method pretty well, so I feel like I'm I could run it up. Like I don't know, gun to my head type of thing.”

“A little bit over less than three years we launched our first car, we designed it and then we built a factory as well. So China speed. If you believe that a car a EV is going to be another piece of consumer electronics and that's something that we have a lot of experience in whether it is managing the software hardware integration whether it is managing the supply chain. I think those are the stuff that we have experienced working with so many consumer electronics products before.”

“Literally, what you're looking at on my screen is me asking it to pull up information on a cold emails campaign. It will actually create all the cold email sequences, it'll come up with all the leads, it'll scrub the leads, it'll de-duplicate the leads as well, and it will send it on a sequence, and it constantly iterates over time. It will constantly self-improve. And so this agent we have, that's in our stack, we have a stack called Single Brain, which is all of our revenue agents.”

“Because of how heavy and important the running is in this sport, we don't stop. Yes, if you stop running, you would get stronger faster, but just because you're running doesn't mean you can't keep getting stronger. When I go into a strength block, it usually involves more specific sets and more touches to any type of resistance than it does mean eliminating my running completely. That volume of it goes from maybe 60 miles, 55 miles a week, down to 45.”

“Bill Gates lets it slip on a panel with Warren and Roberto and Don Keough that Warren has always told Bill that Coca-Cola could be run by a ham sandwich. While Roberto is sitting right there. Roberto gets very offended and apparently never talks to Bill Gates again. Now, given what's about to happen here, it's actually highly debatable whether Roberto was a ham sandwich or not.”

“I think it's, in a way, owning our destiny, in a way. So I would say building data centers is... data center cost performance control. Yes. What other areas of the business are you like that with? Our game engine, for example. We said we we always imagine we need a multiplayer three d engine that does 10% of everything really well.”

“Be an expert in the rule book. You should know the rule book as better or as well as or better than the athletes themselves. You need to be able to tell them what they are and are not allowed to use. They're going to come to you and say, hey, can I bring this with me? Can I take this gel? Whatever it is, you need to be an expert in how the rule book is set up.”

“I think, for example, right now, people are massively underestimating the role that human adoption and human comfort, you know, with advances in artificial intelligence will determine its deployment. I think technology leaders think that folks will just blindly adopt, new technology as it comes out, and I think we're gonna enter a period of time where there's gonna be a huge amount of societal pushback on a lot of the changes that are coming, with AI.”

“At my company, for whether it's SingleGrain or SingleBrain, we want people that have good judgment, because people that have good judgment, they're using AI right now, and they're becoming turbo brains. They're moving very quickly, and they're making the right moves. If you have someone that has bad judgment, and you're using AI, they become slot cannons. How do you decide what to automate, and what to never let AI touch? Well, it depends on your business.”

“By the time America enters World War II in 1941, the military and the US government realize, hey, Coke may actually be one of America's best weapons in this war. The military under Eisenhower grants Coca-Cola employees quote unquote, technical observer status, meaning that they can participate in the supply and infrastructure build out of the military around the world just the same as military infrastructure people. As the American military is advancing in the global theater all around the world, Coca-Cola is right there with them setting up bottling plants and production lines to supply the troops.”

“Agents can do three things. They can access your files. They can access the Internet, and then now they can write custom code and execute it. You should really only let an agent do two of those three things. If you can access your files and you can write custom code, you don't want Internet access because that's one issue. Vulnerability. Right? If you have access to Internet and your file system, you should know the full scope of what that agent's capable of doing. Otherwise, malware can get injected or something that can happen.”

“The most important thing is when you're setting up these agents, you want to set its own soul.md, so give it a personality, talk about what you like, what you don't like. Memory.md, so for persistence, obviously you want it to remember things as well. And lessons.md, too, you don't want self-correction after mistakes, right? You want lessons.md to remember the mistakes and also help you not just call out the mistakes but cover them.”

“I think my one-sentence explanation is that with the era of internet scale foundation models, things that used to work maybe 20, 30 percent of the time are now working 60 to 70 percent of the time. And in robotics, right, as a very complicated, dynamic, engineered system with many pieces, in the past, if every small component of your entire system only worked 30 percent of the time, it would take many, many iterations to get a whole performance system working at scale. But now when every single part of the entire stack just works that much better, from the research iteration process to the engineering scaling process to the data collection engines, I think you can really just see the pace increase when you just have many more successes and a much higher hit rate when you're going about and scaling up your research.”

“At about a year in, I had a bit of a it's almost like a vision where I was saying, woah. You can't be logical on this. You have to be intuitive and go back to some of the roots of Knowledge Revolution, which is all about fun and about play and about building something very innovative. So instead of this logical track, me and some several people actually from Knowledge Revolution said we're gonna do this very unorthodox thing and build this wacky new product.”

“In 1931, Coke commissions the artist Haddon Sunbloom to create Christmas ad imagery for Coca-Cola featuring Santa Claus. So this being a Coke ad, Sunbloom is like, well, I'm going to make Santa as red as possible in Coca-Cola red, and then I'm going to make him as big as possible to get as much Coca-Cola red in the picture. Before this point in time, you couldn't really get mass-produced color images out to the public. So Santa didn't have a color. Nobody really thought about what color Santa was.”

“You have to be willing to fail on ninety percent of your ads. We were pumping out fifty to a hundred different creative variations a week just to find the one or two winners that would actually scale. If you're not testing at that volume, you're basically just guessing with your marketing budget and hoping for a miracle.”

“It's very intuitive. So if you like, try to learn new, new sports, like do you go surfing or skiing? I feel like during the day, like when you started, it's really hard. But I found that like once you, once you like, if you go surfing for like two days or skiing for two days, like initially it's like really hard. And then you go, you sleep overnight and then you come back. And then you're immediately much better. And I like that in some way, the learning to learn faster paper has sort of mapped it into like, as Ted said, the day cycles and the night cycles, where the day cycle is sort of like in context learning, where you collect more examples, but then it's in context. And then the night cycle is like where you go retrain or find you change the weights of the model.”

“There's a summarization of that trend that I like to say to my team. This is the year system as model. Where, instead of having a single model be a thing, you have a system of models and components that are working together to emulate the black box model. So when you make an API call to something that's like a multi agent in the background, it still looks like an API call to a model. Under the hood, it's like a billion different models.”

“The managed service provider market is essentially a hundred billion dollar industry that is stuck a decade behind modern technology. When you look at how most of these MSPs operate today, they are still using legacy ticketing systems and manual processes that haven't changed since the early 2010s. We saw an opportunity to come in and rebuild that stack from the ground up using AI to handle the rote tasks that consume so much human time.”

“In terms of the material we use for that rear floor we generated over we stimulated over 100 plus formulas and use AI to predict the performance of each of these formulas and ultimately we picked two to be the material for that. You don't use humans to do it anymore. You just use AI to take pictures very quickly. They can diagnose whether this piece is good or not. And then you can use AI in terms of material generations and a lot of those things.”

“Like literally and maybe to kind of just put this a bit more concretely, you know, if you have your robot in some given initial condition and, you know, you try something with RT1, RT2, it doesn't work. Well, you're kind of out of luck. You can try the same thing over and over again. You can slightly maybe rewrite the language instruction, like instead of, you know, pick up the cocaine, you can write like maybe like lift the cocaine, but you don't really have the granularity you need to be like, actually, you are two centimeters, you know, too low. You missed the table because it's at a new height. It's kind of obscured by shadows. So you want to like be more gentle and approach more from the left. There's no really way to do that right now with the interfaces, the language interfaces that we train RT1 and RT2 on. But with RT trajectory, the idea is maybe if you have this kind of like line sketch of a course trajectory of how the robot should do the task, you could, under the same initial conditions, just change the prompt a little bit, do some prompt engineering and actually see qualitatively different behavior from the robot.”

“I do think designers feel vindicated in a lot of ways. Right? You know, a lot of designers had parents who were saying, why aren't you studying computer science? You know, what are you gonna do with this skill set, drawing things? You know, this doesn't make any sense. And I think today, you know, a lot of our designers are now shipping code, which is extraordinary.”

“Vertical integration is really the secret sauce when you are trying to disrupt a legacy services industry. By owning the full stack—from the software that routes the tickets to the technicians who actually fix the servers—you eliminate the finger-pointing that happens between vendors. This allows us to capture the full economic benefit of the automation we build, rather than just selling a license and hoping someone uses it correctly.”

“This couponing strategy aligns incentives for everybody in the value chain in a way that had never been done before. Consumers, they love it. They get free drinks of this great tasting beverage. Drugstores and soda fountains, they super love it, because now they're getting more foot traffic. And then once consumers come back and start buying their second, third, fourth, you know, four hundredth drinks, is a highly profitable drink for them to sell.”

“I think it's a great way to think about the world. It's, like, very principled, very, like, hard nosed scientists, very careful. And I don't know. I think it's just it's such an incredible field. You have such high leverage in computer science, in AI. And so I think a lot of physicists were seeing that, particularly in, like, high energy physics. After the discovery of the Higgs, I think a lot of high energy physicists were sort of looking for what's next.”

“We have a couple of people at NVIDIA. We've been working with security to bring agents really close to compute. So we now have stuff where you can tell Dynamo, like, go run some experience with Dynamo on x cluster and just try it right now. We've actually been able to one shot problems. We used to have this problem where, with Dynamo, you have to find the right configurations. We've just had an agent just completely one shot that. It goes. It gets the compute. It runs a couple experiments. It's like, this is the best. Go run this. And then we just give that to people, and it's faster than anything that they have.”

“I think one of the things that was certainly true at Apple then is just this insistence on their leaders being exceptional engineers themselves. I remember a conversation that made a big impression on me. I was talking to one of our advisors who was a key senior executive at Apple at the time of like, how do you think about prioritizing sort of like management and leadership skills versus technical ability? And he just said, you need both. Like, you can't compromise. The best people are able to do both.”

“I built a business called Gruns. We make gummy vitamins. We started it about three years ago and just recently crossed that billion-dollar valuation mark in the exit. It was a fast ride, but we found a hole in the market where people wanted the benefits of greens powders without the terrible taste and the mess of a shaker bottle.”

“Berkshire owns about 9.5% of Coca-Cola today. That stake is worth about $28 billion, which is a 22, 23x gross return on the $1.3 billion investment over the course of 40 years, which equates to only just over about an 8% IRR. But Coca-Cola stock kicks off, these days, about $1 billion a year in dividends to Berkshire. So a $40 billion total return on $1.3 billion invested. Including dividends over that same time period, the S&P 500 is up about 11% annually. So the famous Berkshire Hathaway Coca-Cola investment today is actually underperforming the market.”

“I don't want to spend my time on a small business anymore. If I'm going to put in the same amount of effort and stress, I want to make sure the ceiling is ten billion dollars, not ten million. You have to look at categories that are already massive but haven't been disrupted by a modern brand or a better user experience in decades.”

“The curve is even flattening more. The growth rate in bookings per year for creator number a thousand, ranked by how much they're making, they are growing faster than creator number one. Wait. Say it again. The curve is flattening, wider, use around the world, more opportunity for vertical content, more opportunity for some content for older people. So that is a a flattening of the curve, which bodes well for Creator a Thousand.”

“The number one constraint we're facing right now is building more drones faster. The demand for these products has really just exploded over the last couple of years. We a couple of weeks ago announced a $50 million, 3,000 drone order from the US Army. That size of contract is actually not that much of an outlier for us these days. So it's a good problem to have, but we will be tripling production over the course of this year.”