Ep 142: Understanding the AI Wave with 8VC's Alex Kolicich & Jack Moshkovich

This has been the biggest surprise of having the AI wave. Technology is no longer the bottleneck. It's the process and organizational structure. How do we think about that as investors? Palantir Smart Enterprise pre-AI was like a $40 billion company, and Palantir AI is like a $400 billion company. Productivity is getting better in the economy, and it's going to keep getting better. So are these models commodities? There's a ton of divergence in business models. OpenAI will likely continue being a consumer business. Anthropic is clearly a developer infrastructure business. Yes. Grok today is very much trying to position itself as the most unbiased, truth-seeking model. You know, Meta is in the business of content and culture. Slop. Slop. Slop. Put another way, Slop. The human is still the bottleneck in any of these business process decisions, and that's what I think we're finally coming to solve with the AI wave. This is a generational opportunity for everybody. A lot of people are wondering what is going on in the AI wave. Is it a bubble? What do the next 10 years look like? How do we understand what the companies are actually doing? How are the investors and builders thinking about this? There's trillions of dollars being invested. And my two partners at AVC, Alex and Jack, are right on the ground. They're making dozens of investments, working with a lot of the very top companies in AI. Let's hear from them. How do we think about this world? How do we navigate it? And what does it mean for the United States? Welcome to American Optimist. Today, we are at the 8VC San Francisco office. I don't worry, I'll get out of here before they can tax me. We have two of our partners here at 8VC, and we're going to talk about the AI world. First of all, your guys' backgrounds. Alex, you were Peter's math guy, apparently, for a while. You were at Google, helped out of Palance here. How long have you been at 8VC now? More than 10 years. More than 10 years. And Jack, you came here right out of Stanford? That's right. When I was a junior in college, you and I met for the first time. I've spent my entire career working for you. All right. Well, I'm sorry about that. That's okay. We're going to talk about AI, but we're going to get some background on it first today. It's been probably our most exciting year we've ever had in the business world. A lot of cool stuff to go over. This firm really got going like about 15 years ago with Formation 8 and then turned into HVC 10 years ago. And early on, we had what we called our smart enterprise thesis. There's like a big paper we wrote on it called The Coming Transformation in 2013, about what's happening to the economy. And there's a lot of like echoes between this coming transformation, you know, 13 years ago and like what's happening in AI right now. So, Koli, first of all, like what was the smart enterprise thesis as far as you're concerned? Yeah, I think the most succinct way to talk about it was, you know, early on and you obviously we saw this at Palantir was we thought there would be a transition that happens from old enterprise software to new. And so enterprise software 1.0, we always thought would be these not dumb systems, but they're almost like databases. You know, they're they are innovative, but you store your data in them for your business. You're able to retrieve that data in the future. Think of an ERP, think of a CRM. They're organizing, but they don't help you really run your business or make decisions in your business. And so it was revolutionary, but we thought there was more that software could do. And so with the advent of A, the cloud, and B, the commoditization of big data, we thought there'd be a new generation of enterprise software we called Enterprise 2.0, where with this idea of man-machine symbiosis, you'd now have software systems that help you make decisions. And so Palantir was obviously, I think, the most quintessential version of that. But we thought there'd be many vertical software companies that would start around this. And there were several hundred winners overall. We got into a bunch of them. We maybe should talk about a few. I guess, to me, the big difference was always there was like very linear systems in the 1970s and 80s that were like very simple tasks, whereas now you were doing more complicated tasks. It's interesting because it's like kind of climbing the conceptual pyramid. And I guess kind of cool because today you're climbing even higher on the conceptual pyramid, obviously. Exactly. So, you know, we had Enterprise 2.0 where it would use data to help you run your business and make decisions. And now there's an AI wave, which we'll talk about more, where potentially the software can actually do the work for you, not even just decision on it. I think the word that everyone used around the office right as I joined was non-linear workflows, which is actual real-time operational decision making rather than just a data store. That's sort of like how it was framed, I think, back then. And it's interesting because it was obviously successful. We made a lot of money. Everyone made a lot of money who was involved in good companies there. But it probably didn't raise productivity as much as I would have hoped in the economy, right? This is like the knock on it. Is it productivity didn't go up that much? I feel like maybe just like did save people time and then they just screwed around. I don't know. Like what's the right way to think about this? Yeah, it could have been. There were many things happening. You know, the cloud wave was one of the core themes. And so a lot of the value could have been moving from on-prem to cloud. That's one potential. But does that make productivity go up and let you do some things better? It's efficiency, right? You don't have to manage servers. You don't have to worry about uptime. You don't have to do manual updates to software systems. There's less customization. That's all efficiency, roughly, but that's efficiency from delivering software and not necessarily running your business. And so, yeah, we did see improvements. These new cloud-based systems were a little better, but I don't think you saw step function improvements in business efficiency. So I think there's a handful of different interesting things. One is, the answer depends on how you measure productivity. So on one hand, TFP, total factor productivity, hasn't budged at all in certain areas of the economy, like healthcare and construction. It actually has improved pretty well in some areas, more like financial services. But the other thing is, if you look at both revenue and earnings per employee of S&P 500, it has continued just going up steadily over the last 25 years. And so I think it's probably not exactly right to say that there was no productivity improvement. But Alex's point, you have to close the loop. And if the human is still the bottleneck in any of these business process decisions, this is just always going to be the bottleneck. And that's what I think we're finally coming to solve with the AI wave is you can actually handle the process end to end. Because there were companies that were very successful. Like you basically be irresponsible to run like a big REA not using Adapar. And you'd probably be irresponsible to run a big title insurance office without using Qualia to the things we're in. And so it probably did make them more profitable. It probably made them a little more productive. But I guess the point is the overall economy had some areas where productivity just wasn't as much as we would have hoped. I think a big shift was actually adapting to evolving consumer preferences. Like if you think about the average home buying experience, it went from being fully on paper where you had to get on the phone with someone and then you had to go through this asynchronous process. A lot of the cloud software adoption, I think, was actually just on the back of there being a challenger company like a Rocket Mortgage or Quicken Loans. And all of a sudden you could apply for a loan online. And so on the back of that, there was this enabler opportunity for a business like Blend that said, okay, Wells Fargo, we will enable you to allow your customers and prospective customers to actually do the work online. And that was a big part of it. So productivity is one side, but serving the customer experience and actually allowing the company to be competitive in the marketplace was another big part. It's like the Red Queen thing where you have to run really fast just to stay in the same place in order to compete, basically. You got it. And there's some element where I think you're seeing this with the AI too, that a lot of tech diffusion is incremental. And so it's 2% a year, it's 3% a year. Maybe this year it's 6%, percent a year. But it's like small changes that compound over very long periods of time. And how much of it was just going to the cloud? That was probably a big part. We invested in Zen Payroll, became Gusto. It's doing very well. People just want to do these things on the cloud. It's easier, like you said. So part of it was just delivering software. I mean, obviously, I'm sure it makes the experience better for everyone. If we look at Qualia, it was the first cloud software. It was technically the second, but the first real modern, serious title insurance software product that was cloud first. And you'd see that in a lot of these vertical software companies. I think that's probably the biggest discrepancy between the white paper as was written versus the practical reality. There was more about the transition to the cloud and less about end-to-end intelligence. And I think that's sort of the promise of the new wave that we're finally seeing, like realizing. Maybe there's like a Palantir analog, like Palantir Smart Enterprise pre-AI was like a $40 billion company and Palantir AI is like a $400 billion company. I think it's worth like 10 times as much. And I think that's true of this wave in general. There's some like huge multiple difference right now. So whatever we did the last 10 years is like exponential. So let's go to the AI wave. And so there's lots of different ways of measuring how much faster this is. I guess there's been a lot more SaaS companies, obviously, but if you take the top hundred SaaS companies and the top hundred of these new AI companies, and you look at the median, the median rate of growth, I guess I saw the Stripe report was like 3.6 times faster as of a few months ago, basically of AI versus SaaS. And it feels even more than 3.6, but 3.6 times is a lot because if you do over two years, that's what's 11 times. So 11 times in two years is like a pretty big difference. These things are like growing really fast. And all these like young kids are making me feel really slow these days because like I never built anything that grew that fast. So I got to step it up here. You know, we talk about investing in AI. We talk about the six tiers in computer science. You start counting from zero. Kind of like, I always thought it was stupid when European buildings had a zero floor. I'm like, come on, guys. But that is also how we do it in computer science. It's just different. And so in the spirit of the stupid European buildings, like level zero is energy. Level one is chips. Obviously, NVIDIA is dominant, but there's been a lot of new things there we're excited about. Level two is data centers, which all of our friends, family offices invested in. Level three is where we put on top of those is the actual LLM companies. It's actually Anthropic and Gemini and OpenAI and XAI and a bunch of stuff there. And then level four is software infrastructure. It's how you deploy these things in the economy. And the level five is the apps and services actually going out there and delivering the value, actually performing the value themselves in many cases. And so obviously, we have some bets in different parts of the stack. I'd say we're probably most focused on level five. And Koli, a lot of our level five companies, whether they're doing healthcare billing or logistics workflows or – gosh, there's a whole lot of lists of them we can go through. They're using whatever models are best at the time, but they're swapping between models. Are these models commodities? I mean, obviously people think they're worth trillions of dollars. Like, where are they commodities? Where are they not commodities? It's a tough decision or a tough thing to discuss because there's areas where there are commodities. There's areas where there aren't. So if you're doing coding, you know, Anthropic is frontier at coding. And so you're going to pay a premium on that. And they probably have fine gross margins on doing that. They also seem to be ahead on certain computer use things. I think although Opening Eye claims differently right now, but I think Anthropic's ahead. That's new and frontier. And they're both actually very good at it. But if you look at like a document processing use case or a structured output or conversation, you could use an open source model where the tokens are really cheap. So I think you're going to see this trend where in at least certain domains, the big model companies are pushing their frontier in certain areas and then everything in the wake gets commoditized. And there's a long term question, I think, where does the open source Chinese models or otherwise, do they catch up and they're close enough where everything is commoditized? Is there a sustaining advantage where you'll pay a premium on like frontier, some sort of frontier capability? Will there be memory? Like, will the big models become better so that over time as you use them, they learn about your use cases and thus there are switching costs? There's a lot of questions. That's the thing with open AI is they have all these consumers. They got to be able to lock them in based on what they're learning and doing. Jack, I'm curious how you think about this. I tend to think that because people have put tens of billions of dollars into the model companies, they're a little bit naive about how much of a commodity they are in certain areas. I'm not saying they're all not worth a lot of money. I just think that they're more commoditized, people realize. How do you think about it? I think the word commodity is like a polarizing word these days. Be more polite here. I think everyone says it. This is San Francisco. We have to not use my phone. Everyone says it, but no one I actually don't think knows what they mean by that. So I think there's two interesting things. One is, if you look at true commodity businesses, whether it be mass metal suppliers or mass agriculture producers, over the course of any given cycle, they will run somewhere between 5% and 15% gross margins. If you look at the model companies today, look, the information is not fully public, but they run somewhere between 50% and 60% margins on their API businesses. Well, of course, they're not currently commodities. And so clearly, currently, that's not the case. Yeah. So then to Alex's point, the question is, how does it evolve over time? And what do you have to do to continue increasing the switching costs of these things? And I think the interesting parallel that people are talking about, which I do subscribe to, is like, what are the analogs here to the hyperscaler businesses? So if you look at the hyperscaler business, the primitives of it, EC2 for compute and S3 for storage. What does that mean? A lot of people don't know. So if you go to Amazon Web Services, there's two primitives. S3 is like a storage primitive. Literally, you can put data in it. And then EC2 is a compute primitive, which is like you will run some operation on top of your data store. So those two primitives are in fact commodities. However, all of those companies, Amazon, Microsoft, and Google primarily, have been able to move up the stack into managed services. And so now you don't just go to Amazon to use S3 and EC2. You also use RDBMS and you use their managed database service. And so actually migrating off of those platforms is a gigantic pain. They've just built a lot of other things everyone needs. Well, isn't that what we're calling level four in a way though? It's the infrastructure to deploy things. So again, you're totally right. And the question is, what are value-added services these businesses can add on top of the primitive to increase the switching cost? So how much are they going to own level four that's built for them, built specifically? And how much should level four be outside? Because sometimes there are certain things you want outside. For example, the tool that lets you switch between models as prices change, that's going to be a level four thing. It's always going to be outside, they might assume. There's also this idea in economics of minimum economy of scale where you had to be a hyperscaler to build the clouds. There's only three of them, maybe four. And so they have some pricing power. Another minimum economy of scale business used to be telco companies or cable companies. Only one company is going to wire the whole city. It's like a network effect basically. Exactly. We just dominate. And then so is there that dynamic here? and I think that really depends on the capabilities. If you can commoditize it in eight months, there is no economy. It's interesting because like, I feel like everyone has different answers to this. Like one answer to this, which sounds silly to me, is that nothing really matters. It's just who gets first to like AGI or ASI and the whole world hits like a singularity where there's like this godlike intelligence and everything changes. So that's like the one like really weird, like almost like religious kind of thing going on out here, which with a lot of our friends, And I worry some of the very top people subscribe to it. So that's like option one, and that's what they're doing. They're racing towards that. Option two, which is like, is like that they're just going to keep being ways that the frontier stays ahead of everything else for a very long time. So the frontier is significantly ahead in 2032 and it still ahead and you still paying for it to be ahead Okay that like a really good business for the next six years It going to make a lot of money Whereas intuitively in these things to me it like eventually when I say commoditized eventually like everyone else catches up. It doesn't matter that you're only, that you're three months or six months, nine months ahead because everyone's so much better and it's good enough for everything you need, in which case it's bad for them. But that would be my intuition, you know? Look, I think the interesting question is like, why are there so many diverging opinions on this? And I think it's actually, Well, it's actually going back to like, show me the incentive, I'll show you the outcome. There's never been as much vested economic interest in a wave as with this stuff. This is why I think people are naive to this question, because they all put all my friends here. I go walk on walks with the billionaires who haven't been chased out yet by the new taxes here. And they or they're just visiting like me. Sorry, I try to avoid this. These little snipes against California. But I go on walks with these guys and they put like $3 billion or $5 billion in these companies. So, of course, their whole mind is like, these things are worth trillions. And by the way, hold on. They're also the fastest growing companies of all time. For sure. You know. Yes. Which is true. Which is true. Then you will listen to, you know, the CEO of a data warehouse business for whom the benefits would obviously be if the model primitives were, in fact, commodities. And he will very confidently publicly be speaking about that and calling them, as he would, commodities. Because they own the other infrastructure. Exactly. And that's their vested interest. They would be best off if they were in fact. Well, NVIDIA, that's what NVIDIA wants. NVIDIA actually wants both though, because NVIDIA wants people to think they're effectively commodities. So they're going to capture all by having the best chips. But at the same time, they don't want them thinking that too hard because they want them to raise lots of money to keep buying more of their chips. So it's kind of a funny situation. Look, I think the other interesting thing here is that people miss, actually, there's a ton of divergence in business models between these different labs that are chasing AGI. OpenAI is consumer. It seems like. It seems like. OpenAI today and will likely continue being a consumer business. I think they're bad at enterprise. I think they don't have the, I think they're keep trying and they're not succeeding. Maybe they'll, maybe they'll get it though. I don't know. But focus matters, right? And so, okay. They're a consumer business. Anthropic is clearly a developer infrastructure business. Yes. Google historically and likely will continue being the productivity business. What do you mean by productivity in Google's case? I think Google is actually good at infusing the suite of products they have with the latest technological investments. And if you are a G Suite user, or if you're a Gmail user, chances are these products are going to continue getting smarter and smarter. My email keeps trying to write itself for me. It's a little annoying. Personal preferences. Okay. So then what is meta? You know, meta is in the business of content and culture. Slop. Slop. Slop. Put another way. That is a slop. I did not want to say that. It's a metaverse. Look, all of these things. It might work. All of these things are a function of what data asset you're sitting on top of. And if your data asset is meta, is all the content that people are making and consuming. Then you're going to be a content and culture and you got it. And slop business. What is X then? Because X is a lot of things. And so I like to say that X is in the truth business. And Grok today is very much trying to position itself as the most unbiased, truth-seeking model. But how does that make money? There's a lot of ways to make money, but how does it make money than you think? And there's customer support stuff it's trying to do. There's macro hard, which is exciting. They're trying to be the best at reasoning, right? And so I think I kind of feel like they went after the hardest thing that might be the most valuable in the end, where if you look at like most workflows in the world, you know, you're not necessarily reasoning, but the highest value workflows might be reasoning. So it takes a while to build up to that. They also think they're the best at building infrastructure. And so it's like, not only is Elon building these giant things faster than everyone, but he owns SpaceX. And so you can actually just like, if you're going to use the moon or whatever to make these things go a thousand X, like he's the only one who could do that. So the question, if that's the most viable thing, then X is going to be the most viable, right? It's interesting. Well, going back to the framework of like what your data assets is, is what business you're in. They're also the most real physical world focused business as well. Yeah, exactly. I shouldn't leave out Tesla. You have these amazing robots coming. You got it. And he may end up having world models there, which by the way, a lot of people are working on world models, but none of the other model companies seem to be. So maybe Elon surprises everyone with that too then. Look, I think all of this goes back to the thing where building the model by itself is actually not enough. And just being good at building models is not enough in a world where the frontier capability, yes, changes consistently, but it's becoming easier and easier to catch up to the frontier. And so you actually need to have a business around the core capability. And that's where we're seeing the divergence between these jobs. And I think if you look in the past, right, how did software companies have defensibility? In my view, it was always you either had product, you had data gravity. Those were kind of the two things. And so how far up product do they go? And then how much data gravity, which is probably memory, do they get? So we're obviously investing in a lot of these level five companies on top of these things. And it's an interesting question because a lot of these things are growing really fast, faster than anything we've ever been involved in before. And they're solving really hard problems with big teams. And I guess it's like an interesting question just on the model thing for a little bit longer. Like obviously, if you've raised tens of billions of dollars and you focus on an area, you're going to be very competitive in that area. But there's like a thousand of these areas to go after. Why should these areas exist either inside or outside of the models? How do we think about that? Because it's like, should you be afraid that all the areas get eaten by the models, even though there's a thousand of them? Or how do we think about that as investors? You know, this is all of a sudden probably a discussion topic where we have a vested interest in a fund that's investing in a bunch of these level five companies. But I think our take, which is slowly becoming consensus, is the last mile of delivery of the customer experience and product in any given enterprise use case is actually a ton of work. And turns out just delivering the raw primitive or the raw intelligence is not enough. And so focus in these things actually really matters. Customer understanding in these things really matters. Business development and the network effect of you becoming the standard akin to quality becoming the standard in title insurance or ad about becoming the standard in RA world actually really matters. And customer set gravitates towards the best in class solutions. That's why software are going to take most dynamics. And so just empirically today, best in class use case specific application layer software companies are outcompeting the lab businesses in any given use cases. Now, the only outlier example with this, of course, is code generation and developer tooling. That's so close to the decority and what these guys do. You got it. And so this just kind of goes back to focus, I think. Anthropic has made code gen and developer infrastructure their main focus. Therefore, Cloud Code is doing really well. Now, actually, it turns out some of these markets are so big that Cloud Code is a great product and Cognition is growing really well that we're obviously big investors in. and Cursor is doing really well, and GitHub Copilot is doing really well. And Cognition shows you that last mile gap where a lot of the revenue is enterprise, and servicing large enterprises and doing that final step of delivery and orchestration and all that stuff. I mean, Cognition has, you know, obviously dozens of these global champions in math and physics and programming. There's all these really smart guys solving really hard problems, but specifically around enterprise deployment. So I guess the question is, if we're going to impact that market, there's like $3 trillion a year spent paying programmers around the world. And if we're going to make that market twice is a lot. If you're making it even 30%, that's almost a trillion dollars, give or take. And so if you can even capture 5% of that new market, you effectively have something that's worth about $40, $50 billion revenue a year, which is probably a half trillion dollar company. And so there's massive companies going to be made around that. So much of the excitement, I think, around these AI markets is also that the math is not just existing market share capture. The math is actually also market expansion. Yeah, there's a need for a lot more programmers. That's why you need like another trillion or two. And so you're actually bottlenecked today, right? And so what happens if the average programmer becomes a really good programmer and person who didn't have programming skills before all of a sudden can actually generate high quality code? Well, that's a ton of market expansion. what happens if it was not economical for you to go and talk to every single one of your bottom-up PLG customers, but now you can have a computer go and interview them, collect their feedback? Well, that's a ton of market expansion, right? And so many of these things, the cost curve is just being dramatically shifted with AI. And I think that's actually a way more positive sum, exciting way to look at this stuff. This is the higher productivity thing. I mean, literally, I have a cousin who's a bright guy, but he's not technical. He's a baseball coach. And he is building a company right now using VibeCoding. And I love it. And he's doing it on the side using AI. Totally. And it's amazing, right? It's like this is something you can never have imagined. So hopefully, this is a positive story that brings a lot more productivity and upside to everyone overall. How does Palantir fit into all this? Let's just go to the Palantir thing for a second. Obviously, we're really proud of how Palantir, like you said, it probably would have been like a $30, $40 billion company in the pre-AI world. is now obviously like over 10x that, which is awesome. And I wish I still had all my shares, but I'm very happy with where we are. But it makes me really happy to see everyone there just crushing it. And so Palatir had a lot of things that we figured out, obviously, that apply in this world. We see a lot of our network around Palatir coming out and really crushing with our new companies. Why is that in your guys' minds? What did they figure out that's relevant here? I think they did a good job of delivering to enterprise. They're very good at mapping organizational workflows internally. You have Foundry, which builds your ontology, so gives you a unified pane of glass to view all your data and actually be able to implement changes in your systems of record or systems of action in the business. And then because they have that platform basis, now with AIP, you can build and use AI sort of in your new enterprise workflows. And so it's kind of the perfect insertion point for using AI in any organizational process. or any organizational product you want to build. Yeah, I guess Paldry got really good at understanding what is valuable to the business that they need to, what's the hard problem they need to solve, and how do you map out all the processes around that. And then because they have that really well, now the AI could do some of the things for you. It's a little bit like magic there. It's one more tool in their quiver or whatever, but it's an incredibly powerful tool. I think there's two funny evolutions on consensus and all this stuff, right? One is, if you talk to Palantir employees 10 years ago, AI actually used to be a dirty word. And the whole premise was- Well, it didn't really work. Exactly. This is really funny, right? And it's a sign of the times, AIP is now front and center. But the other thing that has almost become memetic to an extent, right, is this notion of forward deployed engineers. Everyone has an FDE now. Everyone has an FDE. And on one hand, you look at the stock and you think, well, how much of that is sort of driving it? On the other hand, there's actually practical reason behind it, which is so many of the CEOs of the biggest companies in the world are setting top-down mandates to do something with AI. But no one in practice actually fully understands how to use this stuff today. And so you have this unique demand sort of structural moment where if you go to big companies and tell them, hey, we'll future-proof your business, we'll be your trusted AI vendor for the next decade, we'll hold your hand and everything is going to be okay. You can actually get these engagements that feel consulting-y, service-y at the outset, but actually generates a ton of lock-in over time. And I think that's like a practical reason as to why this concept has grown so much. I think it's a big part of it where basically they're helping them access the AI world. I think there's something else though about the motion that's like a motion you need for how you even do AI now, which is kind of what Coley was saying. You need the training to understand how you go into a business and map out its processes and map out its business goals and then figure out how to connect what your technology could do to accelerate the most important processes using the data really quickly. So I think all of our companies now that are coming in and deploying and solving these problems actually do need FTE. So basically the Palantir model was the model that was actually, it's like the platonic model that should exist for how you do AI, right? And so that model is now being kind of like learned by others as well. And I think it's fascinating to me because most people, they used to make fun of us at Palantir. I'm sure they still do, but they used to make fun of us and think it wasn't worth anything, which is at least they make fun of us and now we're worth a lot more than them, which is great. But I think despite the fact that that Palantir is valued really highly, I still think most people haven't done the hard work to learn about this model and why it works because there's a lot of kind of core insights there. And I'm not even Claire Palantir wants to share with everyone why it works. It's just because everyone else is so arrogant. They're just like, we're just going to do it ourselves. We're not going to learn what Palantir does, which I think is a big mistake. By the way, one of the reasons why it works so well is once you go for an engagement, the average Palantir contract is $10 million a year. You can basically make anything work if on the other side of it, you're going to get paid $10 million a year. But you have to find value you're adding that's high enough for that though. My point is much simpler. You will now see companies in Silicon Valley that run 100, 200, 300 K ACVs and think they can get away with the forward deployed motion. And fundamentally, your sales economics just don't work if you have to spend a bunch of hands-on building time to service any given customer. To sell something for $200,000, you've got to build a machine that just runs. But I do think there's something early on in an AI company where you probably do need an FTE-like motion to iterate and figure it out before you can build that machine sometimes. So there might be an in-between. And Kola doesn't like this because he always wants all the metrics to be perfect right away. And I'm like, no, you have to let them have some time to figure it out. I think there's a bold case, actually, where you only need FTEs. Right. Like if AI, if code, if AI generates all the code, then the only thing that matters is connectivity to data, you know, your organizational understanding, your standard operating procedure at a company. And so an FDE, well, OK, you have a platform that connects your organization's data. You map the organizational structure and then it's about mapping, documenting that company's standard operating procedure. that an input to an LLM to either generate code or run your business logic or whatever right That the AGI version where if code is literally free then every deployment everywhere can be fully accessible It just all people working with people to implement it for them And it's just English. I tell the AI what I want my system to do, and it probably generates code and runs that code. And if I change my operating procedure, I'm changing English, and it translates it. And so the super long-term world is it's all context to the LLM. It's all requirements. It's all standard operating procedures, which is what FDEs do. It's all changing the way people of the business operate, which is an FDE job. And that's the future of software. One last anecdote on this, which I think is interesting. So you keep going back to what is the hard thing and why is Palantir working so well and like lots of these things that are attempting to replicate the model or not. And it sort of goes back to the business model of the labs question. So what is Palantir really good at? It's understanding both the specific customer problem, as well as all the underlying data assets, mapping them so that you can actually deliver value. So OpenAI has an enterprise business where they go and sell access to the underlying model capability. And one of the customer case studies that they love talking about is their engagement with T-Mobile, where they automated a bunch of the customer support queries. And it sounds really cool until you realize the entity that was responsible for the implementation of that engagement actually wasn't open ai it was an entirely different company started by a former palantir guy it's gonna come and it's called distil and so doing that stuff is really hard the technology by itself is not enough you need to understand the business process and the business logic and the data ontology and all these things around it which is why these use case specific companies are so valuable yep that's right well A lot of stuff is working amazingly well and growing really fast. I want to give you some examples, just like maybe kind of more rapid fire to listeners. Like what is working here in San Francisco? And obviously, there's some of this stuff's going on around the world. We're going to start with one that's based partially in Texas with our former founder of Deliver, Harish. He's running Augment. Like, Cole, what's Augment doing with AI? Give us an example of that. There's this idea of what does the future of software look like? And in many ways, I think we have a company, Cognition, which is Devin, the AI software engineer. And Augment is along the lines of that type of product where it's a logistics employee. So they work a lot with freight brokers in particular. So the product itself is an employee to help freight brokers. Exactly. So the employee will help augment. That's why it's called Augie, Augment. Augie is the agent, will help augment certain procedures that they have to run in their business. So, for example, when you have a delivery of a truck or some package or whatever to a warehouse, in order to get paid for that delivery, you need to collect what's called proof of delivery. And so this is a common task that happens in freight brokerages where if it doesn't get sent in by the driver or the carrier or the shipper or whatever, they need to collect it. And so it's a manual task today where you have an employee calling one person, calling another person, trying to get this proof of delivery. That's something that Augie can do very easily. Augie could actually also call to find someone to drive a load and can negotiate with them on the price too, right? And it can go all the way up and down the stack. And so, you know, what we find is these agents are not perfect. They can do tasks like that. They can help with negotiations for pricing. They can help with collecting proof of delivery. You can do all sorts of different things. You can ask questions. You know, you can give a status of a load, all these things. And what ends up happening is Augie can handle maybe 95%, 90% of cases. And when things get challenging or it doesn't know what to do, it can hand off to a human. The negotiation would be pretty hard against the AI because it could basically tell you, like, you could take this price or I'm going to call 15 others in real time. And I'm starting to dial right now. Let me know. You have two seconds. It's going to be tough because it's so easy to talk to others to get the price. And you know what the price is if you're a computer much better. I'd imagine it's a pretty good negotiation. It's pretty good. And usually the best humans are better, but the AI is very better on average. And in tough cases, you can elevate to a human. And so that's what I think the future of work is. So if I start swearing at Augie, it's just going to put me on with someone else or something. Exactly. So I think the future there is actually optimistic. Rather than having the humans just doing these pretty boilerplate tasks of collecting proof of delivery, the human is the sort of manager of the AIs. And they only handle the challenging kind of fun cases on it. So I think there's two really cool things about it. One is going back to the positive sum version of AI. Today, if you're a freight broker, how many calls do you think you're getting that you're not even fielding? Yeah. What happens if you have a 24-7 concierge that actually does pick up and does provide a quote to now 100% of the loads? And coordinates everything. Right? And then the other thing that I think is really cool that you love talking about, and I think it's actually spot on. You know, we used to live in a paradigm where people drove the software, but now software actually drives the people. And these agents and these digital employees are really good at either doing the tasks themselves or creating the highest priority work queue for any given person on the overspill of the tasks. Yep. Let's go over some more of them more quickly. These are all really impressive. They each deserve a lot of time, but we'll go fast. Glimpse, Jack. We'll do three with you. Glimpse, what's Glimpse do? Glimpse solves this really annoying problem for consumer brands called deductions management, where if you have a retailer or distributor partner and you send them $100 worth of stuff and you get $70 back, you have to figure out where $30 went. And usually it'd be a person going for every single invoice and trying to come up with supporting evidence for why a deduction might be invalid. And we provide that entire service end-to-end for you. This is like a multi-billion dollar spend right now. And these guys are growing crazy fast because You're just saving all this time, huh? Yes. And not only is this like a huge issue in itself, the way the company is being built is by having really tight integrations over the retail and distributor portals, which gives you access to a bunch of underlying data on which a bunch of other retail operation, back office, CPG workflows can be built. Guys, we're just going to jump into a bunch of other stuff there too. You got it. Maybe we should just keep doing the next round ourselves and letting someone else in. Let's talk about it. Outset, what's that do? This is, I think, sort of the canonical example of what is possible if you dramatically shift the cost curve on something. So, you know, five years ago, going and interviewing a user to do market research or user research would be pretty expensive. You'd need a person to get on the phone with them. And so you'd have this trade-off, right, where you'd have the expensive, high-fidelity, but low-scale option in manual interviews. or you'd have the opposite, which is high scale, low cost, but low fidelity option in surveys. And Outset solves the trade-off. It is now a computer agent that can go and interview users on your behalf with a predefined guide, synthesize all the learnings and then deliver the insight. So instead of doing surveys to everyone, you actually do tailored interviews at scale. So you're getting way better data about your stuff. But you can do this stuff because the cost curve gets bent. Like it was prohibitively expensive to go and interview every single one of your users at scale. Now you can do that. This is working. It's growing fast. I think it just raised another up round. We did a round in March of last year. And then the business closed the Series B another round in November of last year. And this is happening very commonly. Like Cognition, which we talked about earlier, you know, raised three rounds last year. Each like more than, you know, each of big, big markups. People are just realizing they're working. They're realizing they're going to take over. multi-billion dollar markets. And so their valuations are going up, but they're still very low compared to where they're going, I think. I think that's totally right. One last one, Field Guide. What's that to you? I think this is sort of the canonical AI productivity story. It's a business that sells CPAs and over time has evolved from a business that owns domain-specific data ontology to then owned workflow enablement for the CPA practitioner to now just does a bunch of the work that any given CPA would have to do on any given accounting engagement. And all of a sudden, because you're getting paid a fixed fee, you have 50% more capacity to go and serve 50% more engagements in a labor supply constrained market, because the number of CPAs in the country only keeps going down. You're saying there's like, it's accounting, but isn't there auditing stuff going on? That's right. The customer is an accounting firm that is doing either a financial audit for a customer or risk audit for a customer. And field guide is a system that both the practitioner uses to do the work and the field guide agents do the work for the practitioner. And they just also, I mean, it's not announced, but they're raising lots of money on high valuations too. That's right. Not announced. And we're totally not supposed to talk. All right, good. But these guys are crushing it and they keep growing. So you usually have all of these kinds of apps just crushing it. There's a lot more that we're involved in on that side. We can't talk it off today. There's also these acts to companies. So in my view, like a lot of the companies, these are all doing kind of like new types of services. Sometimes they're existing services in the economy. Sometimes there's new services you could do inside of a company that wouldn't even have been possible before, like the extra service or whatever. I think a lot of the value is also going to accrue to existing SaaS companies that can now do a lot of the agent work on top of them with AI. And so one great example of that is Qualia. Qualia was started out of our firm over 10 years ago when it was Formation 8. Nate started it. We backed it at the very beginning after he worked for us. And he's dominant kind of in the whole US thing. More than half of the house sales go through his platform, right? So it's completely dominant in this area. And dozens of companies plug into him. And he just released a product, I guess it was like four months ago now, that's basically gone from zero to almost 10 million run rate. That's actually doing the work for a lot of title insurers. How does it go? You're on the board there. How does it work? Quality are clear. Yeah, exactly. I mean, we break down, you know, a lot of people ask us, where can AI fit in? And we generally break it down. There's five key things that AI is good at. Like these are the tasks. You can add one more now, computer use, but you think AI is good at multi-turn conversations, document understanding, structured output, search, and reasoning. If you can break a workflow down into any of those five skills, AI would be pretty good at it. It turns out in title, like a lot of what you're doing is document understanding, search for title search and structured output and some reasoning. And so it fits really well into the title workflow. So they launched a product called Qualia Clear. We're able to streamline and automate a lot of the parts of the title search and analysis process, which is a lot of what title agents do. And so you could reduce labor by we're still doing the case studies, but like large percentages. And so actually right now, as more interest rates come down, title is going to have to scale up a lot. That's going to save them a lot of money, basically, and make them way more efficient. Yeah, in most cases, we find it's better to get operating leverage. Like a business has people. The people are the human capital of the business. You don't want the people to go away. You want them to be able to do more. And so it's usually positive sum. It lets you do more business at your company with the same number of people. A quick, I think Nate is one of the most impressive CEOs we work with and was a speaker at our annual meeting last November and had a quote that I think every CEO of an at-scale software business should really internalize. He had an all hands where he told people, the span of outcomes for Qualia as a business has never been wider. In two years, we can be a half a billion dollar ARR business, or we can be a zero dollar ARR business, depending on how right we get this AI thing. And I think that is just so illustratory of the state of the world. These existing businesses have all the inherent advantages to them to crush this. How many of them are actually going to do this is very much an open question. If I could say this has been the biggest surprise I've had in the AI wave. It hasn't been that SaaS is under attack or anything like that. That's a common narrative. I think it's a very finance bro narrative without understanding what's hard about building software, but we don't have to get into that. But if you look at the public markets, the AI bubble, it's not a bubble, but the AI, let's say euphoria has been in hardware, in data, in infrastructure, things like that. And what's been left out has been the application layer. I've thought for a long time that would rally because you'd have existing incumbent businesses launching AI products, and you'll see that they have data gravity and they have workflow gravity. They're a great channel to sell AI to their businesses. And so there'll be huge beneficiary of distributing AI to their customers. And 95% of them have not done that. Like, it's actually kind of shocking to me knowing how easy it can be if you have really the right data and some of the right processes, how easily you can create, you know, AI driven application software for customers that you could deliver and how few companies have done that. It's like truly perplexing. I don't know what's going It's a very entrepreneurial motion right now still, right? And the talent that's good at this, like we're all fighting really hard for it. So I think companies like Adapar and Qualia that still have really great tech cultures, they're going to do it. But imagine if those companies are like 10 years older and like I'm not involved and no other top VC guy who's tied to this is involved and they don't have the entrepreneurial energy still there. It's tough. That's been the case. And yeah, I mean, I think they might go away and it's because they can't, they don't have that spark. It's not because they're well positioned. It's because they can't execute. You know, public applicationless software companies are the antithesis to Alex's motto for 2026, which is you can just do things. Turns out doing things as an at-scale business when using AI is going to eat into your gross margin profile and changing the pricing model might actually cannibalize some of your existing revenue is really, really hard. Yeah. Most leaders are not bold enough to figure it out. And so most of them are not. And I think even the ones that are bold, you've got to be really bold and persistent and fight like hell for the talent. Other stuff, there's too many things to go over in AI that are really exciting. I think one that I'm particularly excited about, and we're going to probably have him on the show soon, so I won't go too long into it, is Kevin Mandia, who did Mandia. And I think this is a pretty crazy situation, just to put it out there, where it turns out that you have the very best cyber guys in the world that advise the top companies and understand how it works to protect them, how all the hacking works. and you like feed all that knowledge and train like the new agents. The new agents actually, it turns out, are now much better hackers than people. They manage other agents. They do it. We'll ask him more about it. This is kind of terrifying to me. You have any thoughts on this area? No, I mean, I think it's a great use case of AI, right? And this is an area where reasoning starts to matter a lot more. You need to start red teaming is all about planning and, you know, trying to understand systems and trying to find exploits to them. And so I think it's a great use case. And Kevin is world class on it. Yeah, his new company is called Armiden. and we were obviously involved in the seed. And like when he showed these amazing results a couple months ago I think I got like nine calls from all the top funds all fighting to get in and a bunch of the good ones did a round with us there So it gonna there a lot of this stuff that just seems like it a whole new wave you have to be involved in to protect yourself like what honestly should we expect over the next decade for you know for the stuff that's the business side i guess you said i should take a break from talking about our civilization in my session with that uh in terms of saving it but just on the business side like like i mean are the markets going up a ton the next decade because productivity shooting upwards and this is just like a golden age as long as we don't ban AI? Where are things going? Yeah, I think, you know, it was a good week because this week was the JP Morgan Healthcare Conference. I'll do 20% of my investments in healthcare. So I had a chance of talking to people kind of outside of the AI bubble ecosystem. And the number one question was, is AI a bubble, right? Is it overvalued? And I think it's an emphatic, no, it's not a bubble. Probably it's still a little underhyped. Like there's always questions of, are the models going to get better. And I think the reality today is if you look at the capabilities of the model, you could stop all development. We could tax the billionaires and everyone can leave. And even if we do that, you'll still have a revolution in the economy. You'll still have a revolution in the way things happen because the limiting factors now are not the technology, it's diffusion of the technology into the economy. That means knowing how to build product on the models, knowing how to deliver product with the models, like new product. And it's even to the point of like, you have to reform some institutions. So imagine law, right? Law fits into four of those core skills that LLMs are good at today. It's like reading documents, searching, writing documents, some reasoning. That's going to be, law is going to be fully transformed from really the bottom up. Like their whole business model will have to shift. So you'll have to move from today, law is sold per hour. You're going to have, hours don't make any sense when it's agents doing work or when it's LLM's doing work, you have to shift now to like pricing outcomes. That's a whole business model transition of a huge part of the economy. That's another bottleneck, right? And so these are the challenge. These are the things that we're going to overcome every year, 4% growth and sort of productivity in these spaces. And there will be a slow revolution that we'll see. It's about people and institutions adapting to the new possibilities. So basically the possibilities are already there and now you have to like roll it out. Exactly. And it's the rolling out where there's friction. the tech is way ahead of the rollout today. Yeah. And we didn't talk much about healthcare and AI, but of course, like healthcare itself could be, I think we believe like a tiny fraction of the cost and better at the same time for many areas of healthcare, maybe not certain surgeries. Yeah. All of those are going to be impacted too, but like so many areas. And, but you're right. This is going to be special interests going to war. I guess, I guess the people who are trying to tax the billionaires here in California are the healthcare workers unions, which is like ironic because it's like, that's the area we need to make efficient, you know? Just start the counter. I think that's mentioned number four during the podcast. Look, I think the key point here is technology is no longer the bottleneck. It's the process and organizational structure. And these things are really hard to predict and really hard to measure, which is why I think with all of these cycles, the direction is correct, but the timing is off. And so this stuff is going to happen. It's just a question of if, sorry, not if, but when. And people can disagree on that. I think it's just probably slowly over 10 years. And so if you look at a lot of these public companies, no earnings multiples are dislocated or anything like that, really, if you look at them. NVIDIA is like 25 times forward earnings. That's right. It's not crazy. And so I'm actually surprised how little of a bubble there is, to be honest with you. These things are just making so much money, but the multiples are not that high, you're saying, basically. And it's going to continue, and it'll probably get more crazy, to be honest with you, because we're going to – every year, there's going to be more inference. And these companies are growing quickly, which means more inference. I think as productivity clearly grows, because we've already seen examples in trillions of dollars in the economy where productivity can double or triple. So as it clearly grows, I think the bubble does – at some point, you actually do get a bubble because people get so excited. They put a lot more money into it. We're not there. And this is what – I mean, this is not a macro podcast, but this is what a lot of the macro financial commentators don't get is you could look at like money supply. You could look at interest rates. But everyone is missing like actually productivity is getting better in the economy. and it's going to keep getting better. You know, one of the reasons why I think that's the case, by the way, is the macro people are so far away from the practical reality on the ground. They're not talking to five, 10, 15 person teams who are actually doing this stuff. Like it's actually really difficult to internalize how impactful this stuff is, unless you're literally using code code and seeing what it can actually do for you. Like it's really hard because of how drastic of a change it is. It's something where I think you have to be doing it yourself. And it's like I think young people who are kind of native to this and learning and like focus on it, like they're adapting really, really quickly. Whereas I think most of the macro people that I know anyway tend to be on the older side. And most of them is just like – it's like there's a famous book called This Time is Different in Macro. And you're supposed to like make fun of the fact. And I think, by the way, there are parallels to the past when productivity grows. We can probably learn a lot from the late 19th century, by the way. It's a really good thing to study. But it is probably very different than the last few generations in their lifetime, because we haven't had something like this in their lifetime. Totally. Look, not to nerd out about public stock investing, but to Alex's point, Micron, which is obviously a key participator in the AI wave so far and produces high bandwidth memory, the stock is up 250% over the last 12 months. And it was a nine times PE forward multiple stock 12 months ago. And it's still a nine times forward PE stock today. Like, where's the bubble in that? And to give you a sense of CapEx, like, what are we spending now on AI CapEx? Like $500 or $1 trillion? If you look at the Industrial Revolution in Russia in the early 20th century, I think investment as a percentage of GDP got up to 20 or 25%. And then China was investing 40 or 50 percent in their industrial revolution of GDP. And now in AI, we're investing like 3 percent or 4 percent. It's like it's not even yet on the scale of the industrial revolution. Yeah. And so if it is an industrial revolution, this thing just gets a lot crazier the next few years. That's the analog people are missing is it could be that. Sure, it is digital, right? It's more digital work than physical work, but it could be on that scale. And if you think about in those historical parallels. And obviously in the US, investment's always been less than in the Soviet Union and China, of course. But it shows you things can get a lot crazier if you really pushed it. And in the US, you'd still have investment of percentage of GDP in the 20s. Before I ask you at the end, we always try to see what makes people optimistic. What worries you the most? Because we're being pretty optimistic about this giant important wave and how great it is. What's a big concern over the next 10 years? I think it's the social dynamics. I mean, you have people who are always worried about job displacement. And when I went home for the holidays, you know, I come from more of a working class area. So I mentioned a few of our companies, some of the autonomy companies were invested in. And they were like, oh, so who's going to drive the taxis, right? If you're thinking about a Waymo, like where are those jobs going to go? And I said, OK, look, why aren't we all tilling wheat in the field, right? Like I could have said, well, who's going to, you know, scythe the wheat or do the harvest this year? And that's because we have machines to do that. And that's the reality of it is that it's better. It's positive some will have more productivity in the economy, but there is always friction when that happens. In the Industrial Revolution, you had the Luddites who, in some sense, have a point where they're like, I like the way my life is. This is going to change. We got rid of Coopers. The people have last name of Cooper because they made barrels, right? We got rid of the Boyers. We got rid of like all these jobs that you go back and it's tied to people's names, you know? And it's like we don't have the Smiths anymore in the local town. So I think it's very unintuitive to people that that was like a really good thing. And it should happen again because it's scary. And I think you need to manage it though, right? Because I do think there's some right people have to the way they – You say you need to manage it though. That's a really tough thing is who should manage it because what's going to happen is some crazy populists, both on the right and left, this is like non-political comment, crazy populists want to manage it because they want the power. And then they're just going to screw it up, right? And they're going to steal. Well, that's the risk you're going to have. But you'd want to make sure people have access to retraining, things like that. So you've got to have competent training. And by the way, we have $40 billion a year of training programs, and they're just completely incompetent. We should make them accountable. So there's stuff like that that we're working on. So I agree. There's policy that makes it work better. What do you think, Jack? I think the societal stuff is totally right. And I agree with everything that was said. I think on the more financial side, two things are true. One is, as with any cycle, I have no doubt that at some point, we're going to overshoot what the economy can actually efficiently absorb. And there's going to be a correction. There's a reason semis manufacturers are cyclical businesses. The psychology takes over at some point. And then two, I do think it's extremely reasonable to say that there are pockets of the market that have very bubbly behavior. I was working with you guys during the peak ZERP 21, 22 cycle. And one of the big behavior patterns back then was a company would go out to raise money. They would get a really good round done and then be three or four investors around the table who unfortunately didn't get into the round. And two months later, they would be investing in the same business with two months of progress at 5x the valuation. So we're starting to see that in Silicon Valley. And these are just like inherently really difficult things to sort of balance. You can be long the trend and you can be rationally optimistic about both the timeline and the magnitude of the change that's coming. And you can still not make any money doing it if you're off on some of these things. And the infrastructure is hard historically to invest in. I think the American railroads, we famously have all these bonds and certificates of all the British guys who lost everything betting on various crazy railroads around the country. This tends to happen in every cycle where the infrastructure gets overdone at some point. But I think this is, again, speaking to our book, is I think some of the apps and services layer companies are a lot safer when that happens. But I don't know. So I think that's the question. You know, one of the interesting Zerp lessons that I think we learned is technology as an industry is actually more cyclical than people gave it credit to be. And a lot of the startups in that cohort were indexed to serving other technology startups. And when inevitably the venture industry went for a bit of a correction and there was less money going on those companies' balance sheets and they had less money to actually spend on other software products, the companies serving the customer segment actually really struggled. Like there were lots of companies selling HR tools to other SaaS companies. Like you have to be careful what you're indexed to. It makes it more up and down. I mean, the famous quote from Eugene Kleiner, who obviously was a famous semiconductor leader who started Kleiner Perkins, was like the time to take the vignettes when they're being passed. He was Austrian. So I guess they're vignettes. I think that's what he said. But I guess the point is like there are times when there's money available and there's times when it's not. It's definitely available for great companies right now. It's a great time for builders. If you look at the aggregates, like we are well below 2021 in terms of like raised money in venture capital and deployed in venture capital. Oh, yeah. I think the funds overall raised like even a lot less money last year. I think only like the funds doing really well are crushing it. And then all these small funds that probably shouldn't have been around anyway, in our view, like are struggling. There's too many funds started, obviously, back in the easy money time. So, you know, we started American Optimist to push back on a lot of like cynicism and nihilism and to kind of show people really positive things going on in our country. I think this is like a really exciting, really positive sector right now. It really is centered here still in Northern California, even though it's happening all over the country and the world. What makes you the most optimistic for the next 10 years? Well, I just think the AI wave, honestly. I think it's going to be incredibly disruptive but positive, create incredible amounts of wealth. A lot of sectors are going to get much more efficient. There's very bold cases where it's massively deflationary. That's my hope is if you bring down costs for everyone, that'd be really great. It could be the marginal cost of like intelligence goes way down, for example. And these are pretty high cost parts of our economy. You guys are expensive. We are very expensive. And so like it's unintuitive people. But when you make scarce assets cheaper, like that increases the wealth, purchasing power, everything of everybody. Jack, I think Russians are famously not supposed to be too optimistic. But can you give us your best take for a positive scenario here? First of all, just so that you don't have any questions about it, both of our jobs are AI proof. Okay. Nothing can replace us. So let's just get that right here. It's more of a craft. Exactly. It's not reasoning. Look, I think the technology argument is totally right. and then not to sound like a techno maxi, but having spent eight years working with you, it's so obvious just how much technology pushes everything forward. And Silicon Valley is obviously at the center of it. And one of the things we're seeing is just how much sophistication and the level of ambition in the Valley just only continues going up and up. Like I'm pretty young, I'm 30 this year. and I'm meeting these 20, 21, 22 year olds who started programming when they were 15. And oh, by the way, all the materials from YC and other successful entrepreneurs were online. The digital distribution rails were built for them to consume all of that information. And so that's kind of the most optimistic take, I think. There's one camp that says, social media is ruining the next generation and everyone is losing their attention span and yada, yada, yada. On the other hand, if you have agency, you literally have every single last resource you need to be successful in this day and age. And we're just seeing people take unbelievably advantage of it. If I could add one thing on top of that, since it's a podcast, not just for Silicon Valley, is I think this is a generational opportunity for everybody, right? Like you have incredible tools at your fingertips from these AI products. not every industry will be transformed, but many industries will be transformed. So as a young person, no matter what it is you're doing, now's your chance to start a company in that domain and make it AI first. And maybe even for some people who are in their 40s, 50s, and 60s too. Anybody, anybody. Anybody who has agency and is willing to learn. We absolutely did not discriminate based on age at this firm, okay? That's true. At all. That's true. We do have a lot of very difficult young people to deal with, which is part of my karma from having been one of those myself. But guys, that's a great optimistic note to end it on. Thank you very much. Thank you, Joe. Thanks, Joe.