The Exchanges

Q Do you have a prediction for, um, either because of the distribution of, like, um, skill level or the, uh, measurability? Like, where you should expect that models are better at evaluation or identification of talent beyond, ah, you know, human data first.

A Yeah, so it's really everything that you can measure with text, the models are really good at. Like if you can ask questions in an interview and read through the transcript, the models are superhuman at that, uh, across many more domains than one would think. Like it's not, it's more domain agnostic than I would have initially anticipated. I think the things where models are going to be slower is on the multimodal signals and understanding, like, How passionate is this person about what they're working on, right? Like how persuasive are they or good at sales? And those capabilities will come, but they'll just take a little bit more time. Um, so that's my mental model for thinking about it right now.

AI assessment note: “everything that you can measure with text, the models are really good at.”

Q Why'd you make the move now? So for, you know, Arm, I believe existed for a few decades now. The focus was always on IP, which is effectively like designing the way that different chip components are put together. And then you license that out to other people to actually manufacture and incorporate into their designs. Why did you decide to start making some of your own CPUs?

A Yeah, it was, it was an evolution from the early days of where we just supplied simply the IP Components, the pieces, the, the CPU IP, the GPU IP, the system IP, et cetera, et cetera. Few years ago, what we were starting to see was that product cycle times aren't slowing down, uh, chip manufacturing times are extending, uh, the ability to get solutions out faster was becoming more and more important. So we moved from these individual components into what we called compute subsystems. I used, when we went, did the roadshow a few years ago, I used the, the Lego analogy where essentially We're providing the blueprint on here's how you stitch it all together. Demand for that was, was insane. Uh, and what we were finding was we were, and we initially people thought, well, people aren't going to want these subsystems because that's what a chip designer does. Why are you providing that piece? But it saved time to market and it saved a whole lot of things in terms of cost, speed, et cetera, et cetera. The physical product was sort of the next, the next leap, if you will. And There are certain sets of customers that will license IP to, and they've got all the capability in the world to, uh, to build chips based on ARM. There's a lot of companies who want to have product based on ARM. Not all of our customers build products that serve those markets. So Meta was that first example. They w…

AI assessment note: “Meta was that first example. They wanted a general purpose agentic CPU.”

Q Mm-hmm. How does, um, how has, uh, being part of the SoftBank group or working with all these different companies or even SP Energy and the, and, um, the broader ecosystem changed your point of view on what you can do with Arm?

A Well, one thing it does, it gives us a huge, uh, bird's eye view relative to, uh, where the broader industry is going, whether it's around infrastructure, whether it's around capital, whether it's around energy, but also you can imagine it could provide a home. For, for our products, right? So it, it doesn't need to be the home, but it certainly can be a home, uh, which is also a big, you know, a big, a big help. Uh, when we think about the verticals that soft banks involved with, Robotics, energy, data center infrastructure, and then you look at the products that Arm has. The only one we've announced so far is the Arm AGI CPU. You can start to connect the dots and say, gosh, there could be some very interesting opportunities that, uh, that could be an opportunity for Arm, which necessarily doesn't mean that we're getting into the broad merchant ship business. We could be just doing products simply back for, uh, for SoftBank.

AI assessment note: “We could be just doing products simply back for, uh, for SoftBank.”

Q For people who are interested in working at or investing in science, if you are successful, you know, what will be the change to the human experience 20 years from now besides you not worrying about your pancreas as much?

A Yeah, I mean, that's it. Like, that's the, there's the, There's a sense, like a fragility that we all live, like there's this jeopardy that we all live under as part of the human condition. And I think that if we're successful, what will happen is that sense of jeopardy will fade. Like we will be, we will just become much less fragile. Um, we will have the ability to upgrade and replace parts of ourselves. So neurodegeneration, we don't know about that one still seems that's still difficult. That still needs like real investment. Um, the two leading causes of death though, are cardiovascular disease and cancer is not metastasized to the brain. And I think both of those are going to be really attackable through this type of work. Um, the other extreme is if we are serious about exploring the universe and going to the stars, we are going to have to adapt ourselves to that environment. We're not going to export earth with us everywhere we go. And these bodies are great, but they're designed for this planet. And it is going to be adapting ourselves to the hard vacuum of space is definitely going to be, um, I think the thing that we want to do in the long run, and ultimately those are the same, those are the same project.

AI assessment note: “if we're successful, what will happen is that sense of jeopardy will fade.”

Q I want to ask about two more, uh, businessy topics and then talk about AI in the context of chess and sort of your, your reflections on that. And the first would just be, how do you like the Queen's Gambit and sort of the, you know, increasing cultural relevance of chess? Like what's your, what's your reflection of that?

A Yeah, honestly, it's like, it's incredible, because when I was, um, you know, I got into chess a little bit late in life, I was 18, um, you know, so going to college and getting into the game, and I'm like, nobody plays chess, no one talks about chess, like, It's almost like a secret hobby. You know, I was kind of, I was kind of like a skater snowboarder who like kept chess as like the secret thing I did. You know, when I was in high school, it was like for nerds, no one would talk about it. Like, why would you, why would you spend time on that? Um, so that's what I grew up with. And then now to have my own kids where, you know, all their friends are playing chess and like chess is cool. And like the jocks play chess and the nerds play chess and like everybody plays chess. And then you've got like, you know, Louis Vuitton centers their whole, you know, fashion shoot around chess. And like, it just blows my mind. It's like how, you know, and I do feel partly just along for that ride, but I also feel that we've been partly responsible for that, that we've kind of revolutionized what it means to be a chess player. Chess can be in your pocket. You can do it any time. The content is funny and relevant and interesting. And we also just redefined what it meant to be a chess player. Cause previously, You know, it was like, you weren't really a chess player until you were 2000 or above …

AI assessment note: “honestly, it's like, it's incredible, because when I was... nobody plays chess”

Q gonna ask kind of a, a funny question. Feel free to not answer. Like, how do you guys think about cheating at chess.com? Because you, um, you have also dealt with an issue that I think a lot of the world is dealing with now. Maybe they will call it cheating or not, but it is machines can create superhuman output and it can pass off as human performance, right?

A It's tough, right? I, I, I wish humans didn't have this drive that like to take advantage of someone else or, you know, put themselves above somebody or steal from them, but there's some segment of the population where, you know, either for, you know, financial gain where they're using AI tools and, and, and technology to steal from people, or even if they're just getting onto online chess and use, you know, loading up a computer and, To steal rating points or boost their ego or whatever it is, it is sad, you know, it's, um, and it bums me out that, that humans can't just, you know, get in there and, and, and be honest and full of integrity at all times. Um, that said, like chess has been dealing with cheating, uh, for, for a very long time, um, much, much longer than, than AI, um, as we've had these chess computers. And so, you know, a lot of people thought it was an existential, you know, threat to the game and that it would, it would end the game, but, You know, unfortunately, we also have technology to fight the cheating. Um, and so, you know, we spend a lot of time, a lot of money, a lot of resources on anti-cheating basically, and we have a whole set of tools, which I'm not going to get into because I don't like to give ideas to people on how we do it, but we track a lot of stuff. We have more data. We have more statistical models, machine learning models, and that really…

AI assessment note: “we spend a lot of time, a lot of money, a lot of resources on anti-cheating”

Q Yeah, it seems like a lot of, um, private equity shops as well have kind of shifted from the eighties style, you know, come in, do big layoffs, take apart a conglomerate, and much more to your point, how do you optimize value or increase the value of something?

A I will say, though, it's still, they are always the first conversations, very focused on cost cutting, because I think they don't see a lot of products or platforms like ours I'm not really there to cut your costs. Sure, that is happening in this way, but I'm really interested in this is how you're going to make net new revenue. So that is new. You have to actually start that conversation. You have to show them intangible examples because otherwise it still focuses on how do we get to the bottom line and cut some costs, right? But they have to almost expand their horizon on thinking about what else is possible with AI, right? It would be pretty sad if we used AI only for cost cutting.

AI assessment note: “I will say, though, it's still, they are always the first conversations, very focused”

Q you guys are doing, uh, a whole bunch of things on the autonomy and robotics side as well. Like, you, you're, clearly your view of DoorDash as founders is broader and more ambitious than I don't know, maybe just like the surface level view of it's a food delivery network or whatever the first, you know, one-liner for the company was. Um, how long ago did the robotics efforts start?

A Yeah. We've actually been looking into robotics and autonomy probably much longer than people thought, like since 2018, actually, uh, back when it wasn't obvious autonomy and robotics was going to be a thing. Uh, but we felt like this was going to be a technology that was going to be transform, formative to our space and potentially disruptive. And I think, I think that's the nice thing about being a founder led company is like, we are, we get to think about kind of much more Future speculative things are on the horizon and, and, and constantly think about like, how do we make sure we don't get disrupted by the next one? I think like, like Andy said, like the next DoorDash that comes along is not going to be someone that builds the exact same version of DoorDash, but maybe with a better UI is going to be.

AI assessment note: “like since 2018, actually, uh, back when it wasn't obvious”

Q Alex, you were, you started at MetaFair, um, but you were on the path to, you know, you'd assemble the team at evolutionary scale, and you'd raise venture, and you were making progress in your models. What was the pitch from Mark and Priscilla where you said, like, that's actually the right way to go after the mission?

A Well, I think for me, it was really kind of the moment when I understood that, um, you know, they, they really saw this as, as an integration of frontier AI and frontier biology. And I think, um, I had developed conviction that, you know, this is really a new era of science that's, that's just beginning kind of what's going to be possible with artificial intelligence. And, you know, we're, we're in the age of information theory at scale, and we have these systems that can basically kind of predict the next token and they can, you know, learn, World models from that. They can learn biology from the data. And so, you know, I, I think that it just, it was really clear that, you know, to build kind of that next, that next kind of institution for the next era, you would really need to have frontier artificial intelligence. You would have to have frontier biology. You would need to start to put those things in feedback and really have models that are learning from the biology. And I think, you know, just And you need the right scale on the right people. And so this, this just really felt, I think, like the way to do that.

AI assessment note: “they really saw this as, as an integration of frontier AI and frontier biology.”

Q starting with building blocks and building up, but modeling cellular behavior is very different from modeling protein folding. The data is very different. The modeling is different. I'm just curious, like, do you think it's all, uh, similar in terms of it's just data and you train stuff, or do you think it's actually, uh, there's some differences in terms of how you actually have to deal with these systems?

A I mean, there are probably some differences. I mean, you can probably talk more to the specifics around this, but like, I mean, I think each layer is going to end up being somewhat qualitatively different, right? I mean, the, the, but you need to be able to understand the protein interactions in order to be able to understand how cells work. So you can't just go straight to cells in a way without understanding the protein modeling. And then if you're trying to understand something like the, you know, the way the immune system works or a bunch of cells interact together, um, then. Um, you know, it's tough to do that without first understanding cells. I mean, you might be able to at like a very high level of abstraction, simulate a system, but if you really want to like understand how it's going to work, you kind of want to build the simulations at each level hierarchically. So that's basically the approach that we're going through, starting with the, um, the building blocks and the, and the protein. But yeah, I mean, I think that there's going to be different types of data that you want to collect for each, um, the modeling techniques, I think we'll see. I mean, that'll all keep on advancing across the board, but I do think that like A big part of the strategy is this view that you need to build it up hierarchically.

AI assessment note: “I think each layer is going to end up being somewhat qualitatively different”

Q I don't believe in, like, permanent business models for any of these domains, but in the near term, do you have a prediction between, uh, you know, outcomes-based pricing, token-based pricing, enterprise bundles?

A Yeah, the way I think about this is always we've had, like, let's even take the per-user pricing. The per-user pricing is really an artifact of someone creating a budget needing certainty. Right, because it's the most important thing. Like, somebody wants a budget, they need a per user. And, and per user is just a set of entitlements to usage. Right, that's kind of what it is. And so the way is, if the first bundling will be, take some usage, bundle it into per user stacks, and, you know, then sell subscriptions. So subscriptions, I think, are going to be there, per user is going to be there. Then the next big thing will be consumption. So people will say, I want consumption. And it's also possible that people will say, I don't even want to pay for any of the subscriptions or the consumptions outcome. But remember, most people love outcomes until they have an outcome, because once you have an outcome, it's like giving away royalty, right? I mean, I've talked to customers who love, you know, outcome based pricing, and I say, I'm all in until they, oh my God, like, what are you talking about? You're sharing in my outcome? No, no, no. I want you to go back to per user pricing, and I want you to consumption price, right? So I think that debate will go on. Uh, and all, all, all of these business models have a particular time and a place versus one to rule them all, and if anything, …

AI assessment note: “all of these business models have a particular time and a place”

Q I asked you the same question. I said, is anyone going to do this before you run out of money?

A And, and I think there was a good chance that, uh, I would have run out of money before, because I think you were right. Like, I think there was an element of chance here, but then I think the market did happen. So we had suddenly reasoning models that could do long horizon tasks. We had a cloud code, which became like the really first, uh, widely used autonomous agent. And then we had co-work and open claw. And, and I think We're starting to see now that these types of agents that are very autonomous, even though they're like, uh, everyone was afraid to build them. So everyone started building these low code platforms that were much more limited, much more based on connectors. And those platforms ended up being quite limited so that we didn't get the productivity gains from those limited platforms. But when we started getting the crazy benefits from these very unleashed agents that could do everything, that had much less controls baked into them, And even very large enterprises decided they're going to adopt it, you know, like Anthropix revenue is coming from enterprises that are paying for cloud code to do a lot of the work that developers used to do. That was a bit of a kind of how we started, and we definitely were in luck that very autonomous agents appeared before it was too late.

AI assessment note: “I think there was a good chance that, uh, I would have run out of money”

Q Our level, like human intelligence or our level of your models? Okay. Human intelligence.

A Yeah, I think like, yeah, exactly. I think as, as humans, it might still be very difficult to understand what weights and activations mean, and maybe mechanistic interpretability, it seems like, oh, maybe that's too hard or shouldn't be possible. But as we're starting to have models that are much smarter than us, at least in some important ways, we think that, uh, we'll be able to start tracking mechanistic capability much more effectively. Um, And I, and I think it's gonna be extremely rewarding, by the way, long term for understanding intelligence in general, like not just overseeing, but just understanding what intelligence is, how it works, what's the difference between the smarter model and the less smart model.

AI assessment note: “Yeah, I think like, yeah, exactly. I think as, as humans”

Q Do you have a point of view on the phased rollout or controlled rollout with Glasswing and Daybreak from, from Ant and OpenAI in this area?

A I don't have a strong opinion, but I think it's, uh, On the one hand, like, if we knew that there's not going to be anyone who's going to release a Mythos-level model soon, I think that would be great, because it gives enough time for everyone to prepare, to build the know-how, to build the playbooks, to share that around in the community, and to make sure that we're not starting to see airlines go down and power plants go down, and really, like, disastrous effects that could happen. The problem is that if anyone gets to a mythos level model earlier, then in retrospect, it would look like a huge mistake because we could have at least given companies the choice to start moving very quickly and give more companies access to mythos. Now they're all vulnerable because, you know, there's a Chinese model that's mythos level and there's nothing they can do about it. So I think hopefully we'll manage to do the gradual rollout correctly. I would really encourage that we expand The amount of companies that get access to this and make it much easier for people to get. I would advise everyone to assume that these models are coming anyway. The only thing you can do right now is to invest in these foundational controls that will stop the downstream effects of these vulnerabilities are going to be found in their systems.

AI assessment note: “I don't have a strong opinion, but I think it's, uh, On the one hand”

Q labs could ever do? Or do you think it's a structural thing that I, I ask because the number one question amongst the startup ecosystem in the Bay area today is, you know, if you assume capability improves or, you know, when the labs just gets hungrier from their already currently ambitious stance, Uh, why wouldn't they do this too? And, and so I, I ask you the same question.

A Today, if you're, if you're a private person or if you're a security buyer, there are some places where you don't want to trust the same person that you're buying it from. So, you know, maybe, you know, if you're buying a car, you're not going to have the same guy that you're buying it from certified that the car is good, right? And maybe you're going to have someone else do it. And if you're a security, You're not going to trust the vendor of a product to tell you that this product is not going to mess your environment. You're going to want to have an independent party whose whole business depends on telling you that this thing is correct and being right. This, this thing is legitimate and being right. So that's like, there's the buyer psychology in this space that I think really goes in our favor. And then I think there's the core problems, like why are models even making mistakes? Why are agents even making mistakes? Right? So I would broadly Categorize it into two things. One is You know, there's the jagged intelligence of these models and there's like sometimes kind of very silly mistakes that they make. And I think that problem will go away. I think we're heading for much smarter models that make less silly mistakes and, and our role is not going to be to prevent silly mistakes. That will be taken care of by the, the model vendors because they're very incentivized to do i…

AI assessment note: “You're going to want to have an independent party whose whole business depends on telling”

Q When you think about the 14 countries that are already part of PAC silica or, uh, you know, this as a potential blueprint for others to be involved in American for deployed industrial bases, like, what's the value prop for them?

A So the value prop is, um, one of the amazing things that we're seeing is obviously the AI revolution is leading to huge growth. I mean, despite the volatility in the energy markets, um, You know, the American economy has been proven incredibly resilient, and a big part of that is AI being this incredible, you know, incredibly strong economic force that is already fueling over a third of our economic GDP growth right here in the US. Overseas, we're leading, we're seeing that growth translate to record demands for copper, record demands for cobalt, record demands for, you know, lots of different inputs that go into data centers and, Um, and, you know, record demand for electricians and all the rest. And so the takeaway for a lot of these countries is, um, if they find ways of actually, um, having a bigger part and a bigger stake in that supply chain at different layers, layers that make sense for their companies and their economy, they can actually derive a lot of economic growth from that revolution. Because, you know, the amazing thing about the tech industry, especially when we go through these inflection points, Um, as you guys know, is the pie grows really fast, and so it's really not zero sum, which actually makes it, uh, incredibly conducive to forge very mutually beneficial partnerships because we're not approaching it as, you know, what I gain, someone else loses. It's a…

AI assessment note: “they can actually derive a lot of economic growth from that revolution.”

Q When we last spoke to you on NoPriors, you were just beginning your tenure at the State Department. What has the biggest surprise been? You've been very active since you started.

A Well, I, the biggest surprise is honestly how entrepreneurial, um, it's the, the Trump administration has been, because I think the stereotype for people who work outside of the government is usually really focuses on, you know, this idea that the government is super bureaucratic, and obviously that's not to say that the government is a massive enterprise, um, and lots and lots of, you know, uh, people involved, lots of processes, Um, but the, the really extraordinary thing is we have a president who, you know, also spent most of his life in the private sector and who really likes speed. I mean, you know, the running joke, um, especially, you know, with our current counterparts overseas that deal with the Trump administration is that we like to move in Trump time, um, because, you know, the president, when he likes something, he wants it yesterday. And, and so that part's really been amazing because The appetite to try new things and the appetite for risk and, and to move really, really fast is highly unusual, um, by government standards and really speak for the philosophy that President Trump brings to, uh, brings to bear in the Oval Office. And so that part's really been extraordinary. That combined with the, the really great leadership that we've been able to benefit from at the cabinet level from Secretary Besant, Secretary Lutnik, Secretary Rubio, um, Secretary Burgum, who…

AI assessment note: “the biggest surprise is honestly how entrepreneurial, um, it's the, the Trump administration has been”

Q I want to talk a little bit about the impact on the business, if you're right as well. Um, I don't actually know how SAP prices broadly today, but the question would be like, how do you price? And if you are, you know, delivering more outcomes for customers or serving them, you know, services, software in a different way, do you think that changes the business model for SAP?

A It does. Absolutely. I mean, there's no, there's no question. And we have prepared for this already. So, For me, it was always very clear. I mean, for the most part, SAP software is seat based, licensed, uh, uh, uh, today with a few exceptions like a conquer or a field glass, for example, or the business network. Um, but you know, Very clearly with AI, it was very clear for us that, you know, step by step, it will go towards this consumptive world, right? First consumptive, and then maybe in the next step, once we have more verifiability in the system, then also towards maybe an outcome-based license model to, for example, what Sierra is doing and so on and so forth. Um, but the reality is also, It is today for us. It's a hybrid model. It's consumptive, but it still has a certain element of seats in there and so on and so forth, because also it's a joint journey with the customer because the customer saying they are not yet ready in many cases, uh, for a purely consumptive model, right? Because they need one predictability, right? And then of course they are not yet fully also everywhere trusting the outcome, right? And well, no, then also of course, is the value already there, but then they are afraid of that the costs may Explode from a consumptive perspective, et cetera, et cetera. So what we, so what at the end of the day, what we have designed as a hybrid that is basically…

AI assessment note: “It does. Absolutely. I mean, there's no, there's no question.”

Q Um, if you were to fast forward five years, like, what do you think is different about the employee base at ServiceNow or even like the work being done?

A I think that in terms of employee base, I think you're going to see that the net new added headcount will be dramatically reduced. And that's because the company will be far more productive. The agents are real and they will take on a tremendous Workload. So where to keep up with growth in a growth company like ServiceNow, you would have to hire thousands of people in finance and HR and the supporting functions and services just to keep up with it all. And now the agents are going to be able to do a lot of that. So you're going to invest in things that really matter, like humans that engineer great innovations and that humans that actually manage the relationship. And the importance of human to human connection, building trust, um, making promises and keeping them, and enduring a relationship and the net present value of loyalty in that relationship, that's all going to be human. Um, but I, I can see a company where, you know, you don't have to really increase head count to achieve that goal. And I do think that the, uh, the bar got raised in terms of differentiating your skill set. And making sure what you're capable of doing can be easily replicated by an agent. Um, not just in service now, but in all businesses, if an agent can do it as good or better, that's an easily easy economic decision to make, which is why, you know, there'll be 2.2 billion of these agents entering th…

AI assessment note: “net new added headcount will be dramatically reduced. And that's because the company will be”

Q How did that lead you to materials and atoms and, you know, the physical world again? I know that was sort of your starting point in terms of academics, but what brought you back given how much is being transformed right now through language?

A I think just the inevitability of connecting these systems to the physical world. The opinion that I and others held as periodic was, you're not going to see the same kind of acceleration in science and technology unless you start connecting these things to the physical world. Science ultimately isn't sitting in a room thinking really hard. Um, you have to conduct experiments. You have to learn from them. You have to interface with reality. And the creation of ChatGPT in late, um, was a, you know, Important technology, but it's still far too weak. Like we couldn't have done periodic on technology of that era. I think over the next few years past that we saw ever improving models. Um, we saw reasoning. I think like test time inference became really important. Uh, that led to more reliable error correction, more reliable tool use. And we see like the rise of coding agents and other agents. And I think those were foundational technologies necessary To then connect these systems to the physical world. Like it was just not impossible, not possible with like the AI technology of.

AI assessment note: “I think just the inevitability of connecting these systems to the physical world.”

Q Oh, so interesting. Yeah, that's cool. And then from a architecture perspective, is there anything unique that you're doing or interesting? Or can you talk a little bit about how you're actually constructing some of these models on top?

A Yeah. So, uh, language models are incredibly powerful. It's a very natural interface. Uh, and so we continue to use these, um, but we think about them almost as like an orchestration layer. So that's sort of a, a co-pilot assistant, but also like a system that can direct, um, experiments. And it's almost, it's orchestrating other specialized models as well. So we do construct neural nets that, um, are specially designed for atomic systems where there's like some symmetry awareness, um, and those have much lower latency and they've been like fine tuned for that. And so basically you kind of think of this like orchestrating layer that can ingest literature. It can go through our experimental data. It can go through different modalities. But they can also use specialized neural nets as tools, as reward functions. So it's like an overall system.

AI assessment note: “we do construct neural nets that, um, are specially designed for atomic systems”

Q In other words, do you need something like pi or skills or something else to work in order for, uh, periodic to hit that escape velocity in terms of a closed loop system?

A No, but it's a huge accelerator. Um, the goal for periodic is to generate high quantity, high quality data, diverse data, and automation is assistance to that. So right now we employ people as well, and we have autonomous parts that are just, you know, very reliable. If you had a dexterous humanoid who could wander into an unstructured lab and make sense and follow instructions reliably, that would be a huge accelerator. Right now, the automation of physical systems is, requires a very careful design and it's slow, but I think with improvements in robotics, that's just going to accelerate this. But already the reliability of the sort of like hybrid systems is sufficient to produce Huge amounts of, um, reliable data, but it's just gonna accelerate us forever.

AI assessment note: “No, but it's a huge accelerator.”

Q Right. Um, because I, I would argue like, well, you have the hardware, but you've now thrown away the software or the UX layer of it. Um, do you think that's what people want?

A Yeah, I think there's this, like, there's this sense that these apps that are in the app store for using these smart home devices, et cetera, uh, these shouldn't even exist kind of in a certain sense. Like, shouldn't it just be APIs and shouldn't agents be just using it directly? And, um, wouldn't it, like, I can do all kinds of home automation stuff that, uh, in any individual app will not be able to do, right? Um, and an LLM can actually drive the tools and call all the right tools and do, uh, do pretty complicated things. Um, and so, In a certain sense, it does point to this, like maybe there's like an overproduction of lots of custom bespoke apps that shouldn't exist because agents kind of like crumble them up and everything should be a lot more just like exposed API endpoints and agents are the glue of the intelligence that actually like tool calls all the, all the parts. Um, another example is like my treadmill. Uh, there's an app for my treadmill and I wanted to like keep track of how often I do my cardio. Uh, but like, I don't want to like log into a web UI and go through a flow and et cetera. Like, All this should just be like, make APIs available. And this is kind of, you know, going towards the agentic, um, sort of web or like agent first, uh, tools and all this kind of stuff. So I think the industry just has to reconfigure in so many ways that it's like the customer…

AI assessment note: “Yeah, I think there's this, like, there's this sense that these apps... shouldn't even exist”

Q Ivan said, I should ask you, um, how many times you've rebuilt Notion and rebuilt your harnesses?

A Yeah, yeah, it's kind of a running joke almost. I mean, we, we, we rewrite our AI harness probably every six months or so, and, and the time to rewrite has kind of been, been decreasing just because, I mean, like, like, progress has been accelerating. I think this is honestly a, a really key thing and something that a lot of companies get wrong is just like doing one thing and then just like, like sticking with it. You really do have to keenly aware of what the current state of the models and the technology is, and then designing the harness, the system and the product deeply around that. And it basically means you have to rewrite it every six months. And, um, I find it pretty fun. It's part of the process. Um, you know, you get to, you get to restart and, and, and, and rethink it. You know, we're working on Uh, we're about to release a new version of our harness, like in the next week or two. Uh, and then, and then we're already thinking about the one after that as well.

AI assessment note: “we rewrite our AI harness probably every six months or so”

Q What do you think has most dramatically changed in how you think about how, um, engineering and product should work at Notion over the last two, three years?

A Yeah, I mean, it's, it's definitely changed multiple times. I mean, in terms of the coding agents, we kind of went through multiple eras. There was kind of like the tab autocomplete era, and then we, and then we got into sort of inserting, rewriting some code, uh, but, but it wasn't really until the, the agents started working. I would, I would say like early last year, uh, we started to Adopt the agents. Like, I started using ClogCode, I think, around April of last year. That was a huge unlock. Like, I would say the, the, the big shift there is that, you know, you can really push on getting these agents to end-to-end, you know, implement and, and verify and maintain stuff, but it, but it requires pretty significant thought in terms of how you architect things and what is the verification loop. Um, but, but, but the upshot is I think if you do it well, you can be much more ambitious about what you're building. And also make it much more robust than you could have done, uh, with, with, with humans writing it. And then the flip side is if you do it badly, it's all slop.

AI assessment note: “it wasn't really until the, the agents started working... That was a huge unlock.”

Q I've definitely experienced that with our portfolio of companies that are building large trading clusters. Uh, it, uh, CoreWeave has a reputation for reliability that not everyone has reached. Can you just help characterize, if you fast forward, like, two and a half, three years now, like, what is the scale of the problem today?

A Yeah, so if you look at, um, kind of CapEx, right, let's starting with that. So CapEx for AI compute and infrastructure in twenty-twenty-six, you know, at least from the hyperscalers is projected to be between 666 190. Uh, billion dollars. And over the next several years, um, you know, that scales to trillions of dollars, right? And so the, the scale of the problem is how do you build, um, you know, that size of CapEx efficiently? And I think a lot of that has to do with not only, you know, your ability to have access to, you know, those core elements, um, energy, power, you know, uh, and, and your ability to have data center space, et cetera, But I think one of the things that's not talked about as much is capital, and access to capital, and how is capital structured. Um, and what I mean by that is, this is, you know, billions to trillions of dollars of CapEx, and just using equity dollars alone is not an efficient way to scale this. That's obviously massive dilution, you know, there's, there's, it's not an easy problem to solve.

AI assessment note: “CapEx for AI compute and infrastructure in twenty-twenty-six... scales to trillions of dollars”

Q Couple topics to hit before we lose you. Um, uh, new players. How do you think about the sovereigns and what they're doing in their build outs? Yeah, I think, um, they seem to be able to fund themselves to some degree. Exactly, right.

A Um, you know, you saw the news from India last week, uh, obviously a lot of the news in the Mideast, Southeast Asia. I think, you know, we're continuing to see that sovereigns view, compute, and AI, you know, as, uh, and even we do here in the, in the United States as, as, as a matter of national security. Um, and obviously the funding of those clusters is, is very different than funding like a private cluster, and so you've got, you know, Government capital that can be used for that. I, so I think there's two things that, you know, I find interesting in that space. I think one is who are the partners, um, that are going to build those, that capacity and what are the cybersecurity kind of implications and environments for that? And so those are, those are the two nuances I think with Sovereigns is they need to find players that can rapidly scale compute, um, In the, in their countries, and oftentimes they don't necessarily have these players that know how to build and scale GPU compute. And I think that's a great place for the United States to lean in and help build, you know, sovereign ecosystems around the world. And then there's a matter of cybersecurity and how do you make it into a, a truly, um, you know, safe ecosystem for, for those sovereigns. And so I think there's a lot of work to do still on the cyber side, um, especially as you look at, you know, scaling sovereign A…

AI assessment note: “I think there's two things that, you know, I find interesting in that space.”

Q So, I'm gonna fast forward through a bunch of, like, lab work you did, and going to MIT, and being a Thiel Fellow, like, why a longevity fund?

A Me just being very literal, like, at the time that I was interested in longevity, I think if you ask the average person in the field, like, what's the big problem? Most people would say, well, we just can't get enough funding for our projects, and so that's the big problem, and I just took that literally, like, when I was a teenager, so. I'll solve that problem. Yeah, I was just like, literally, like, I should just get a lot of money to, like, help push longevity drugs forward, and the name for that happened in venture capital fund, but it definitely wasn't working downstream of the idea of venture capital, like, that came after this idea of, like, Um, just getting money for projects that should have money and didn't.

AI assessment note: “we just can't get enough funding for our projects... I'll solve that problem”

Q related to moving fast is prioritization. And I think in another interview in the past, you said something like, 14 priorities mean no priorities at all. And you've kind of really honed down into a handful of key areas. Do you mind walking through sort of those key areas from an overall broader innovation focus? And then maybe we can touch on one or two of them as we go.

A Yeah. I mean, so just to analogize to Uber or any company that, you know, we worked on together, Imagine you went to a management team and you're like, what's your, how many product, you know, how's your product line going? You're a series B company. You're like, well, I've got 14 product lines. You'd say, well, what about sort of uninvestable, right? So, um, when I got here, there were 14 critical technology areas. So critical, critical to our national security. So I looked at them. I said, well, yeah, number one, that if, if there's that many, they can't be critical, but number two, Hard for people to hold in their head that many priorities and wake up every morning knowing that they have to get up and execute against those priorities and make progress because you get distracted, you get too much, uh, dispar, you know, uh, separation and therefore not enough resource to any one thing. So I cut them down to six after studying and I made it more action oriented. So we put sprints behind them like you would an engineering team. So, uh, you know, off the top of my head, the, the number one is applied AI. Because we're not building a foundation model at the DOW because there's being hundreds of billions of dollars being spent by the private sector on this. So how do I adapt or use what's being developed in the, in the private sector and apply it to the Department of War use cases?…

AI assessment note: “So I cut them down to six... number one is applied AI... number two is”

Q You mentioned that one thing you have to have, uh, to get these changes to happen is like some talent from Silicon Valley or just talent that is excited about this pace. Like who are you recruiting? Who do you want to be part of the Department of War?

A Big untold stories about Doge, which is, yeah, it did some, some aggressive actions in the beginning that were, um, controversial and all that, but the talent they brought into every department, um, could then be deployed on other projects like the AI action plan, uh, that we did with Gen AI. So, because they were already, like, Elon chose technically sophisticated people. There weren't all engineers, some were lawyers, some were Had from, but they all had a mentality of, of a fixer builder mentality. So I borrowed a lot of those people and then they attract more people on the way in. And then we have this U S tech force that, that Scott Kapoor from Andreessen Horowitz is deployed. And we were hoping to get thousands of people out of college for a two year stint, sort of make it, um, you know, this is your service to the country as a technologist rather than as a soldier. Um, so now we want both and we're going to try to, you know, make, try to make that a badge of honor so that you go back in the industry with some credit, you know, so additional credibility and be proud of your service here. So those are the ways I'm attracting talent and I'm dialing for dollars. I call everyone who's had just left a job and I'm like, Hey, do you have a year to spare doing the coolest stuff you could possibly imagine? Recruiting Tuesdays, I call them and I call my friends like you guys and sa…

AI assessment note: “we were hoping to get thousands of people out of college for a two year stint”