Why the VC Hype Cycle Always Gets It Wrong | VC Roundtable | E2307

25:00Yeah. You know, one thing I think that's important for founders who are listening or watching is like there was for enterprise software, this rule where one to X was like triple, triple, triple, double, double, right? That was best in class, right? You'd go one, three, nine, 25, 27, something like that, right? But that's because of what's going on in AI, the bar has really changed. So if you're a founder that wants to raise seed or ARB, you need to know that basically, I think right now it's probably one to five or one to four and a half, depending on what you're doing, and then probably five to 20.

25:36So you really have to be growing in to be able to raise venture capital from the folks who do A's and B's. The growth curve looks very different in 26 and 27 than it has before. So quintuple, quadruple, not triple, triple. That's a lot harder. What do you guys think? I agree. I don't think that it is overall a good thing for the ecosystem that we've moved into this sort of... uh like i think i think it is preventing capital from from flowing into certain kinds of businesses that are that want to do harder things and that want to that want to enter businesses where the moat is more difficult or where the sales cycle is more difficult and i think there's sort of it's i i honestly think one of the big problems is it's forcing money into companies that are solving problems for right now like problems for the next few months i see so many companies that are building based on what the models do today.

26:37And they're solving a problem that is like a problem that exists for the next 11 days, raising a bunch of money and going and chasing it and getting like very easy, calm, easy go revenue. And it's, I find it to be very frustrating. And, and, you know, we as a fund have a little bit of a sort of bias to being gluttons for punishment and liking things that feel a little harder. And the problem is right now, when I want to go be brave, bravery is maybe not the right word, but when I want to go sort of encourage something really difficult, but if it works, there's an actual moat around that business, I have to do so into the void.

27:18I sort of fund that business knowing that I can't take it, or there's a very low likelihood that Sand Hill Road is going to be interested in the next round, and that I think there's a big gap right now in where we can go take really interesting, great sort of hard things with teams that don't look like they're out of the central casting for following capital. I assume you're seeing the same thing, but you know, you guys are in San Francisco and closer to the sun and, you know, I'm here in New York and, you know, see some of what's going on and feel like, you know, I don't want to say bubble because I do think AI is the most fabulous innovation that I've seen in my career.

27:59But there's just some, this idea of consensus seed rounds getting done at 50 and 60 and 70 million valuation. I mean, we could go into the next topic, but this is why I come in grumpy today, Alex. No, I'm here for old man Ben stamping on the ground. Okay, cool. Thank you. Yes, good. Well, this is all insane to me. I love this. So Mike says rule of 70. When I was taught this by the guys who founded HubSpot, it was rule of 40, rule of 70. Dear God. Now it's no longer triple, triple, double, double, double. Now it's 5X, 4X. And Ben, you're saying that everyone's trying to solve the problems for this minute.

28:36So are we just essentially. Not everybody, but I think that's a general, I'm generalizing. Yeah, yeah, yeah. No, I'm not trying to, I'm just trying to summarize here. So essentially, it seems that only the things that have instant takeoff today are attractive to the multi-stage funds, which is changing what you can invest in. Going back to the top of the show, I didn't think it's going to be that pertinent, but why don't you then raise more money, Ben? I mean, if you're not going to be able to go get Sandhill to take the next round lead, why not do it yourself? I have had some of my LPs ask me that question recently.

29:03Um, uh, you know, I, I think that for me that has to do a little bit with, um, building, you know, each fund we've raised has been a little bigger than the one before. Our, our eighth fund was 145. Our ninth fund's 200. We have grown, but, uh, you know, I, I don't know that I can solve the problem for these businesses if I have a$300 million fund and I don't, you know, in our bones, we are an early stage group. We are really about talent. That is sort of like what we're built for. It's what we know how to do. It's what we love doing, frankly. I'm jealous of people that manage many billions of dollars because I think that like those fees are probably super sweet, but that's not, that's just not like who we are or what we aspire to.

29:47I want to do early stage. I don't suddenly want to be the like series B lead. I think it's a different skill set. Yeah, but I think Mike, you're famous for saying your fund size is your strategy. Yeah, I guess I, I'm, that's my story. and I'm sticking to it. Yeah, by the way, Mike, that's a huge inspiration for me. And I really, I think it's true and it's something that we try to live by here. One of the themes we talk about over and over again on This Week in Startups is making sure you do your chores. I'm no expert on these things. I have some experience. Steven Estes from CLA is an expert. Let's talk about being cash efficient.

30:25Tell us about efficiency and what you see in the top tier startups in your practice. We're seeing kind of an interesting trend out there where companies aren't needing to raise quite as much as they had in the past. You really have to be careful as a founder to only take on as much money as you really need. You've got to do the forecasting, you've got to do the modeling, and you've got to dial it in and get it right. Otherwise, you're going to end up either not raising enough capital to get to where you're going, and you're going to have to go get venture debt or go back, have an extender to the round, or you're going to give up too much of the company because you just didn't recognize how much money you actually needed.

31:00Yeah, very important to get this stuff right, folks. And that's really a bummer when startups don't do things in a button-up way. Always have a great partner, a good partner to have on this adventure. While things change, my friend Steven over at CLA. So if you want a trusted advisor by your side who will navigate you through taxes, accounting, and everything in between, it's time to take action. Visit claconnect.com slash with you. And don't forget to drop a mention that your boy Jake Al sent you. That's claconnect.com slash with you. Start today. Mike, same question to you. Why don't you just raise more money and solve the problem of backing outlier people and just take on those later rounds yourselves?

31:38I'm not sure if you have a different answer than Ben, but I want to get more than one perspective on this point. Yeah, well, so I do believe your fund size is your strategy. So, you know, and the reason is the power law is real. So to me, your fund size is a commitment to your LPs about what your largest exit will be. So let's suppose that you have a fund and your aspiration is to have a 5x fund. I believe that what you're really saying is your biggest exit will be 2.5x the size of the fund in terms of exit profit. So if you have a$100 million fund, your best exit needs to be$250 million if you're going to have a 5x fund.

32:18And so then it's just a function of what you loan and what you get in at and what you get out at. But I guess in today's world, I don't know if others would agree with this. I'm really seeing two kinds of projects. One is what I would call hot projects. And hot projects are, you know, the multistage firms love them. And they may not necessarily have any momentum at all. It may be somebody peeling out of Anthropic or OpenAI, Rockstar credentials and a few good white papers and stuff. and what I believe is the non-consensus play for hot deals is the upside's even bigger than you thought it could be and so it's it's not that nobody thinks it's exciting it's that it's even more exciting than you thought it was in spite of the fact that it's exciting so Anthropic was not done at a cheap price early but I don't think most people thought it was going to be worth a trillion dollars uh and then and then there's what i would call weird projects and those projects aren't even on the radar of the multi-stage firms ever uh so you know my colleague and miraco invested this company called smarter dx in 2022 and it exited for a billion dollars and uh you know that's that's good living for a seed fund but i i don't think it even matters if you're a multi-stage fund i don't I don't think even if you had that exit, it's interesting.

33:46And so I think that there is a different strategy for each type of project, right? If you're going to go after something hot, you've got to work your way in. You've got to build a relationship with a founder. You have to count on the fact that the multi-stage guys are going to de-risk it financially. And in the weird stuff, I think you've got to be prepared to go it alone or to get enough momentum for its own sake. but you can't count on the Silicon Valley echo chamber bailing you out. Right. You've got to, it's got to make it a, it's, it's almost like venture capital in the eighties or something or the nineties.

34:22Before we move on, Ben, I think letter hippo backed board, the board, the digital board game company, which I quite like. I have a picture here of it. Oh wait, no, I'm sorry. That's the original Microsoft surface table. Here it is. Okay. I see what we're doing here. i i did think that up before the show and thought i was brilliant uh no i brought it's cool i knew something snarky was coming there just i am i am a big lover but it is my job literally to sprinkle some some sprinkles on top of uh is this a physical product yeah it's a physical product it's super cool it is doing extremely well and people love it yeah i bought one for christmas have you been using a mic yeah cool it's um it's brin putnam who built mirror uh the workout device that she sold to Lululemon.

35:12And it's her next business. And it's a very, look, I love consumer. I've always loved consumer. I've done a bunch of consumer. I think so much of consumer over the last few years has just been very incremental. And so even though there was this Microsoft product from a long time ago, I think in general, there's a lot of novelty to the way that Bryn is building this and the way that over time the community will be building their own games and the physical pieces and the creation of new kinds of IP. And I think there's also some big sort of tailwinds to get people and kids off of screens, or at least the screens that we are currently addicted to and move to more collaborative play.

35:59And so, and by the way, the usage data, which you're always terrified when you invest in something like this pre-product and it goes out and you start to sell and you're like, is anyone going to actually use it? Or did people click on Instagram ads? The usage data is pretty stunning. People use it with extreme regularity and it's a fun one, but there's still plenty to go. When you combine board the product with the ability of people to now make their own software, which is still a little bit nascent today. We're working on making games with Cloud Code and so forth, but I can absolutely see in the future, my children, you know, thinking of a game idea or taking pictures and then having that kind of baked into it.

36:36So I think it's a really cool company, but Mike was just talking about, you know, stuff that doesn't really resonate on Sandtill. And if I was thinking about a category that's out of favor right now, it would be children's games. You know, I mean, like it's not agentic orchestration for the - USB did the last round there. It's not so wildly out of favor. You know, Mike - Where is USB based? I don't know. I'm not sure. I can't remember. Yeah. Is it on the West Coast or East Coast? I think there's probably a Union Square in San Francisco as well. Well, there is, but I mean. It's definitely based in New York.

37:17Anyway, I will say that this is the, Bryn is the kind of founder who, maybe to Mike's earlier point, Brynn has some of the qualities that I do think are very attractive to the sort of multi-stage Sandhill vibe. But the product itself is definitely, you have to be sort of a creative thinker to get your head around it. I'm just glad things like this are getting funded, straight up. Like, I mean, first of all, I loved the original Microsoft Surface. We had one in the old TechCrunch office right in the entryway, and it was fantastic. And I was a longtime Surface user. So I just like that some things are a bit early, and then when the technology is a lot better, this is not the size of a couch.

37:54it can really meet its mark. All right. Now we were talking earlier about enormous seed rounds. I also prepared some notes on what I'm calling the hundred million dollar series A round. So a couple of names recently, StarCloud, 170 million series A, General Intuition. I think that was 320, Scale Cognition, 100 million, Scout AI, 100 million, et cetera. Lots of these companies going on. How should founders think about these insanely large seed and series A rounds and what they say about the state of the company and its prospects, Aileen, because I don't even know how to describe them because the dollar amount so does not match the stage as I understand it.

38:30I don't even know what to tell people when I read these headlines. Yeah. I mean, I think they're calling it a C or an A because it's their first institutional round or their second, but it's not a C or an A in the sense of how much they're raising, the valuation, who's going to do the next round, and the metrics that people are eventually going to hold you accountable to when you go out again. So if you're raising at 400 or you're raising an eight, you know, you're raising a seed at, I mean, yeah. In 2026, I can't, it's of all the years I've been doing this, I have never seen more people who started a business three weeks ago and have decided that they're going to raise a $20 million seed down a hundred free.

39:04It's wild. And like Ben said, a lot of them are very tuned to like this point in time. And this point in time, generally every idea, every problem that people are facing has 15 or 20 competitors. and you know that like the ecosystem is evolving so quickly you don't know what free tools the hyperscalers or the the frontier models are going to give out and wipe you out quickly and i think when you raise some and i think people are also um planning on spending a lot of it on tokens which that's also a moving puck where i think the number of models that are coming out that are going to be like the open source models the open weight models like they're a lot cheaper and they're getting so much better that i think also token spend hopefully for startups will go down.

39:44So that won't be the reason why you need to raise 20 million bucks because you won't need 10 for tokens. But it just sets you up. If you're at 120 post after your seed, then you and your employees and your investors want to fill a markup. So you probably want to be at 200 or 300 for your next round. So you're calling on people who are basically doing 50 to$100 million round sizes, and they're going to be looking for traction and customers. And I think what Ben alluded to earlier about revenue quality versus just like everyone trying stuff and the renewal is not looking great. And a lot of people falling out because they're trying everything right now, like you just have to really be on your game and know what you're signing up for when you raise those prices.

40:27Yeah. I keep seeing companies doing like agentic security, like getting agents identities and so forth. And I'm like, this is cool. But I think I've lost track of the number of companies that have raised$50 million to do that. Well, a lot of these early rounds though, Alex, this is like, there's, there's king making going on in a way that I've never seen before or Or queen making. Or them making. They making, whatever you want. Yeah, they making. There is makings happening where, you know, like a company gets knighted as the one or a big multi-stage, you know, puts in the first check and sort of likes what they're seeing, but it's a competitive category.

41:04And they want to communicate something to the market, which is stay away, watch out, we've got this. and in some categories you have two or three kings or queens or whatever. But it doesn't always work. No, no, no, no. By the way, we are still in the part of the cycle where right now people, we don't have the blowups yet. We don't have the collapses of the companies that have raised the$100 million Series A's. So everything only goes up and to the right at this moment in time. At some point in the next 12 to 36 months, the rubber meets the road and we figure out what companies are real and what companies raised hundreds of millions of dollars and don't have anything.

41:44And if you're going to have companies that are worth tens of billions of dollars in a year, then you are going to have companies that are worth tens of billions of dollars that also go to zero, which traditionally like wouldn't happen on the speed. You know, it has to happen. Yeah. Yeah. You know, it's interesting. If you look back at the dot com days, where did the big exits happen? Everybody remembers Amazon, Google, companies like that. But most of the people who got really rich in that era are the people who got exited in a window of time at the end of 98 until early to mid 2000. You know, if Mark Cuban had raised$100 million for a series A broadcast dot com, we wouldn't know who Mark Cuban is today.

42:27Right. And so I think that what a lot of the founders are missing is that great. You can raise$100 million in your seed round. nobody that I can find in history has ever had a greater than$10 billion exit raising that much in their seed round. The biggest seed round I can find on record with the exit that size is whiz and it was$21 million. And so what happened in the dot-com era was people raised money at these crazy prices and then everything crashed and the venture firms are like, okay, I've got to figure out which companies are real and which aren't real. And a lot of these companies had to give the money back or pretty much had to shut down because there's just no way that even if they executed perfectly, that they could ever be what they raised their seed round at.

43:15And the venture firms are like trying to figure out who the winners are, who to stick with. And so I think that a lot of people lose sight of the fact that you lose an amazing amount of optionality by raising rounds this way. And if your goal is to create generational wealth and you believe we're in a bubble, this is the last thing you would do. You know, you would position yourself to profit in a wide variety of scenarios. People keep throwing the phrase generational wealth around. It feels like it's like a TikTok theme. I don't even know why people don't realize that 10 million is generational wealth.

43:53You don't need to have 500 ,000 trillion billion dollars. Anyways, Aileen, you mentioned, you know, raising 20, spending 10 on tokens. That to me implies that the startups that are raising relatively outsized C and Series A rounds are doing so not simply because they can, but because they have a relatively high cost basis. So do you think that startups are kind of forced into raising this type of capital early because they have expenses they need to meet, not just humans now, you also have your token budget, or are they just making a mistake and kind of just getting over their skis too soon? I think it's both.

44:21I mean, it's getting better. I think, I mean, if you chat with your portfolio companies right now, like, I think one of the interesting things is obviously, like, Cloud Cowork is amazing. And Tags is really fascinating. But it's really, and it's smart because it can create lock-in, right? But a lot of folks know that they need to build layers so that they can switch models, right? Because, I mean, most of the frontier model CEOs will say, like, you don't need to use the best, most expensive model for everything, right? So you actually have to build things that you can swap things in and out. And meanwhile, the labs are going to try and lock you into using their model as much as and using their tokens as much as they can.

44:59So I think that there is just what is exciting for investors and founders is like there's a lot to be built for this new ecosystem of AI. There's a lot of infrastructure. There's a lot of security. I mean, there's a reason why those are hot areas is because we need a lot of new stuff. Yeah. Just to throw some notes on that, Etch just announced that it's raised$800 million for its transformer specific ASICs. Light Matters raised$850 million. DG Matrix raised 20 million for solid state transformers, which none of us here ever thought about until like 20 minutes ago. There are even a number of companies, startups that are working on data center cooling alone.

45:31So Aileen, do you think that those companies fall under the companies for this moment? Yeah. My old portfolio company, Bloom Energy, which I worked on at Klanner is one of the beneficiaries of like this incredible data center because people need power. And so, yeah, there's a whole, yeah, you do a map of all the things you need in data center or for AI compute and like memory, like look at Micron's numbers. It's incredible. Go read Micron's earnings. I'm telling people, just go look at them. It'll blow your top. Now, Ben, you're slightly more consumer focused. So how much of this translates over into your world?

46:09I mean, I am more consumer focused. As a fund, we are probably, you know, 75%. Sure. Look, I'm very excited about consumer right now, albeit I'm still searching for the sort of like application layer boom on the consumer side that feels really differentiated. We see a bunch of lightweight wrappers that, you know, have a bunch of explanations for why they're not lightweight wrappers that are still lightweight wrappers. And, you know, it's obviously unclear what Chachi PT or Anthropic or Google or whoever will, where they will sort of extend their products. I am very open for business on the consumer side.

46:57I would love to find, but, but I, I really, to find companies. I really do think though that building a, you know, an, another agent to do something that is built on everybody else's infra is, is just like not that exciting. I, I, I keep getting to not the finish line on this. yeah now aileen you shared this back in may uh over on x back to your best share of the year aileen oh but it wasn't it wasn't my chart i think well then i give you credit credit for but i love this aileen uh for folks who are on the audio version can you just quickly sportscast what this is what it shows like there's this it's funny i would love to work with anyone who wants to build like a tech gestalt machine because every year there is like a hot theme right so So 2013, it was wearables.

47:522018, it was VR. 2019, it was scooters. But then the best companies that are actually founded in those years never match what the theme of the year is. So like Anthropic was when crypto was really hot in Web3. Wiz was when we were talking about future of work. So yeah, you just can't. First of all, like to Ben's point, it usually takes a long time for a great company to be built. And a lot of these companies, for the first three to five years, no one's heard of them. They're not cool. You know, they're kind of baking. And so you kind of have to have faith and take these leaps of these risks. We love backing kind of to the earlier point.

48:33We love pedigreed founders. We also love what we call off-Broadway founders, like people who don't have the perfect pedigree. And when you look at the list of the companies who founded, like the founders of a lot of those companies, they are off-Broadway founders. They are not perfect pedigree founders. The founders of Databricks were academics and open source software kids. You know, hardly your... I think in today's world, that's considered pedigreed. Okay, but at the time, that was kind of a non-consensus. I know we've raided academia and now there's like three people left. Yeah, so it's like a whole venture firm strategy is just spending time in university labs.

49:03Well, you've sold me on venture at last, Aileen. I volunteered. That sounds like a hell of a good time. So why is the conversation so wrong? And in this case, what are the categories that are hot now that are not going to manifest great companies later on? Because I know you guys place bets, but a lot of founders listen to this. And I just, if we can give people a way to not go down the wrong path, I think it'd be very helpful. So, Mike, what do you think is the most overhyped thing to build today, apart from agents, of course, that founders probably should stay away from? Oh, boy. Yeah, it's a tough one for me to answer because I root for all of them, you know.

49:44You know, I think that I would say that agents that improve productivity and the function of agentic workflows, I would stay away from. So I think that you've got to have something that's attached to it that creates some type of a path to network effects or some type of cumulative increasing returns power. So I would stay away from any type of AI productivity or workflows that doesn't embody some type of increasing returns mechanism at the core design. And most of them, unfortunately, don't. You look at it and you say, that's awesome. I can totally see why I would want that. But I don't know why Sam Altman is not going to have that in his next demo.

50:37So that's what I would look for. I just realized that I made my agentic trust point and Aileen's firm is back to Drata. So do you want to tell people why my slander was incorrect? Sorry. I was going to say, well, actually, I mean, I think that's an example of a company that was started before LLM's, but it's not that old of a company. It's grown really quickly, Drata. But trust is really important. So I think when you've got mid-market and enterprise relationships and you are helping them with compliance and trust and visibility when because so many companies are interconnected and you need to if you're going to poke a hole in some and pull data in or out of a company you need to make sure that they're doing it securely and um they've got the right business processes in place like that's not a burn the boats whoops we're replacing everything completely tomorrow because you need to have these are relationships where i think um consistency and trust is really important but drada obviously is going to be helping people monitor agents and the trustworthiness of agents.

51:39So that's kind of more of an evolution of the relationships they have with customers and what their customers want. So we're really happy to see where that's more of an evolution than a burn the boat situation. And I think it's a smart one. To prevent Ben from thinking I'm only picking on him, Aileen, tell me why that the either major AI labs themselves or the current owners of enterprise workflows won't do that themselves and consume what Jada is trying to build. I mean, I think like, I don't know, Mike is so good analogies. He'd be kind of like, that's like letting the fox watch the hen house.

52:08Is that the analogy? That's not bad. You want to have a trusted third party that isn't your current vendor that's trying to use all your data. It's funny because I've been doing a lot of thesis work lately and Drata was one of the companies that I kind of regret missing based on some of the work that I've been doing. Not too late, Mike. It's only a$2 billion valuation. Get your money in now. So the way I've internalized it is that AI creates abundance in terms of work products. It creates generative AI is generative. But what people are going to want to start having is what I would call acceptance AI.

52:49So like, for example, when you have financials, you have an audit firm audit your financials. It's not enough for you to just say, I'm really good at doing my financials. You have to have some trusted third party certify. And that credible neutrality is important. And I think what's going to happen is there's going to be a lot of AI generated slop across the board. And it's no longer going to be just the work that gets outputted. It's going to be the work that counts. It's going to be the work that is that there's a consensus mechanism for validating. and uh quite often you're not going to want to trust you know the the frontier labs to do that you're going to want a credibly neutral third party uh so i i agree with aileen it reminds me a little bit of why we invested in nocta back in the day we thought that uh you know identity management should have a neutral trusted uh person right yeah slop is a good one this is like to me if slop is abundant then you start to ask well what's scarce and I think what's scarce is correctness and proof of correctness.

53:57And if you can be seen as a credibly neutral network effects, scalable provider of that, I think that that's, in many ways, that to me is where the application layer is gonna come alive in a lot of these. So we're talking a lot about costs here, controlling them and kind of owning your own data and alpha. A lot of people in the last couple of weeks have been talking about moving to open-weight models, especially GLM 5.2 seems to be quite hot. And this is due to the government essentially precluding us from accessing the latest models from both Anthropic and OpenAI lately. Can we just kind of look through the headlines a little bit and tell me how prevalent is it that startups are actually either rolling their own models or simply turning to existing open-way models to either reduce costs or to ensure that their data doesn't go to training Anthropic into building what they've already put together?

54:49And Aileen, why don't we start with you? Then we'll go to Ben. I was just thinking, I keep thinking that GL1 is the new GLP. The gestalt of this year is like GLP1. And maybe we had this deep seek moment and now we have this GLM moment, right? Which is, it's quite good. A lot of our portfolio have been playing with it and they're saying they're getting equivalent results for a fraction of the cost. And I think that's why they're all getting ready if they're not like to be model agnostic and to not spend as much time fine tuning because you could spend a lot of time and money fine tuning something.

55:26And how long is that going to buy you? A month? It's not a great use of time and money because stuff is moving so fast. So unpack that for us. So essentially, if you train, if you fine tune Kimi K 2.5, as Cursor did to make Composer 2 and 2.5, by the time you're done with that, they'll have Kimi K 2.6 and then 2.7. And so you're always, you're chasing a ball that's going faster in front of you. Okay. That's almost dispiriting, Aileen, because I would love it if founders were able to take the best from the open way, open source world, and then really turn it into a weapon they can take to market.

55:59But it sounds like you have to kind of take what they offer, just the whole cloth. I don't know. Others should weigh in here. But from what we're hearing from portfolio companies, they're not, I think some people considered it and they started doing it and they're like, wait, this is not a great use of our time and money. There's a lot of other ways we can benefit customers. Ben, jump in here. from the firm enterprise perspective. And also, if you have any consumer notes on this particular topic, I would love to hear them. I don't know that I have consumer specific notes. I think in general, everybody is building everything multimodal and has to out of the box be able to say, I am not beholden to any one model and everyone's long-term business model is predicated on token prices going down and down and down and down and down.

56:38And so I think, you know, I mean, generalized models are going to get better and better. The fact that this conversation is happening is actually, though, like the flip of it is it's why Anthropic and OpenAI and Google are so scary because they are aware of the quality of open source and the fact that they're not going to be able to just go and endlessly charge more and more and more and that they're going to move further in the application layer. And they're going to do it too, which makes the application layer right now just like a finicky, weird space to invest in because it's not clear where infra and application layers sort of bump into each other.

57:20And I do think, though, we will move to – and actually a colleague of mine wrote an interesting little sort of substack yesterday about this. Yeah, I wrote it. I thought it was well written. I gave him a hug. You're talking about the end of decisions by Maurice Rousseau. By Maurice. And just the idea that we're sort of real high level decisions are starting to be more possible with with AI. And, you know, I don't know exactly. Like, I think that's where we end up with, like, the most expensive models having a real expansive market for a kind of decision making that I don't know is really that we're quite yet relying on AI for.

58:04To summarize what we're talking about here, and if you're watching this later on, it'll be in the show notes, a link to the post, the end of decisions. But what Maurice argues is that we've seen the effect of essentially computation in fields like chess and most recently in poker. If you play cards, you know about GTO and so forth. And he says that AI is, quote, the first general purpose reasoning layer that can start to function as a solver for domains that have historically been too qualitative for software. Now, if that's true, Ben, then to me, the actual incremental or marginal intelligence gain you can get from a new frontier model version is incredibly valuable because if you can literally have the brain that runs your entire business be smarter, that's quite useful.

58:44That's quite useful. And that's why I think that's where that business model makes more sense than doing the checking my inbox and preparing some drafts for me and whatever. Yeah, the way I've kind of internalized it, and I really owe this thinking to my partner, Ann Miraco. So Ann's been doing all this work with companies that she calls AI-pilled. And an AI-pilled company basically thinks in terms of what processes, what decisions, what mechanisms do they have that define competitive advantage that can be thought of as an ever-improving compounding loop. for that you want to use the best models and you're using those to gather customer feedback come up with new product ideas a b test different things you want the most intelligent models possible that i think is different from what goes in the bill of materials of the product you ship and so sometimes you know you can you can get by with not the very best model for certain fun you know sorting email or like you know performing certain functions within a product and so what I'm finding that once people create new knowledge with the frontier models, they capture and transfer that knowledge with the cheaper models.

1:00:06And so it's kind of like, how do you how do you turn a deep work discovery into a checklist manifesto deliverable? And, you know, you progress down the ladder of model expense as you do that. Do you think most startups are capable of building the routing mechanism and collecting the necessary context to actually enact something like that? Or is this only the companies that are the most AI-pilled, no one's sleeping, everyone looks frazzled, they've got Alex Carp hair going on, and they're just wizards, you know, at the top of the tower? Well, I think when you're an AI-pilled company, you don't mind spending money on the frontier models because you're token maxing as a person or as a C-level manager or as a team.

1:00:52to me that's a separate issue from you know I'm shipping an AI travel agent as a consumer app and I want to know what aspects of that travel agent need the frontier models versus what aspects of that can be adequately solved right with open source models and I think that that's where I think that's where the open source models really come in and there'll be a bunch of interesting companies helping with routing and evals and you know I don't know that I don't know that companies will have to build that infrastructure themselves versus buy that infrastructure and focus on their core value problem.

1:01:28What are they owning? If they don't own the routing, they didn't build the model, they're not doing the compute, they don't have the customer data, then what the f*** are they for? Well, hopefully they have the customer data. Yeah. Good job. They better have the customer data. They're just a bucket of data that isn't even theirs and they're just doing whiz-bang stuff with it and that's the whole jam. That does not sound defensible. That sounds like this is why the job is hard right now, dude. Okay. We're looking at so much non-defensible stuff every day that, by the way, then goes and gets done at 50 by someone who looks pretty smart.

1:02:02But like, let's just take an example, right? Like applied intuition. You know, they're they have a very differentiated product and they're doing very well. But they're AI-pilled in the sense that they've created a real-time performance feedback system where they can get input about which managers are most effective and what's working best and things like that. And they can implement these systems at enormously fine-grained detail that you could have never imagined doing in the past. And so they're not necessarily AI-pilled in the sense that they're doing all this to make their end products different, but the way they do business and compete is fundamentally impacted by it because they're embedding AI into just the lifeblood of how they force multiply every employee.

1:02:55Have you seen Bedrock Robotics? No, I haven't. Well, I've heard of them, but I haven't spent time there. We had them on the show the other week. They're doing something kind of related to this. And I really love them because what they're doing does seem defensible because they're going to a very specific part of the world where there's no other companies, maybe applied intuition. And they're building essentially Waymo for diggers. And it's a great idea. What an enormous industry that no one cares about because no one in Venture has ever held a shovel in their life or startups, you know, tech people.

1:03:21So I think it makes a lot of good sense. Now, okay, we're going a little bit long. So I want to do a couple of final questions for us. And Mike, we're going to start with you and then we're going to go around. uh give the trump administration a grade on how it handled mythos and fable and uh do you think that major ai labs have been actually harmed in the last couple of weeks or do you think this will blow over it's hard for me to grade the the trump administration because i just don't know all the things behind the scenes and so um i'm reluctant to i'm i uh i i do get nervous nervous about these frontier models are so important now that they've got the attention of the government.

1:04:03And that's always a very mixed blessing. And so I get nervous about the government sort of backdooring its way into regulating AI in the way that I was afraid that the Biden administration was going to do. And I think that would be very bad in terms of our competitive posture with China. So I do hope that we can resolve this, but it's hard for me to give a grade because I think in some ways it is a work in progress. And I don't think that the hyperscalers have done themselves any favors in the discussion either. And so I think this is an Amazon, but it was definitely Amazon. I think, unfortunately, this is an example of us muddling through a situation where you're seeing the sausage be made in real time.

1:04:49It's hard to say that there's an optimal strategy. you know. I don't think I can say it better than Mike. I mean, it's hard to know. Obviously, we're not behind closed doors. We don't know what those guys know about what can be done with the models. And, you know, there's, yeah, there's a lot of history here, right, about like whether it was developing nuclear weapons and the scientists who are building them having concerns and wanting to talk to people about what should we do about this? Should we build it? Should we not build it right there. So we obviously want the United States to maintain its edge, but we also have to be prepared for, there's a lot of nefarious actors who can do a lot of bad things.

1:05:26And a lot of our companies, both our federal institutions and companies are not ready for the fact that, you know, basically we can be trying to hack into systems 24 seven with agents. Yeah. The struggle that I have with that, and thank you both for answering that with such clarity and honesty, is that the rest of the world isn't stopping. And so unlike the Manhattan project when we were very much ahead of the Soviets, of the Germans, China's really banging on. I mean, we all saw the GLN 5.2 headline about how they think that it's going to be roughly commensurate with maybe fable, maybe mythos.

1:05:58I want to do you next. Your question is very simple. How's Crunchbase doing? I think great. We love Crunchbase. We're proud of them. Tell me more. I own a lot of shares of Crunchbase. So tell me about how it's doing. They have raised quite a few rounds and we were early investors. So I am not as close to the latest and greatest. But I think like you were saying before, having data is really important. Having proprietary data is really important. Crunch a lot of proprietary data and a lot of private company data, which is, as you know, like it's very valuable. A lot of VCs will use the models to ask for competitive intelligence and information, but private company data is one of the hardest things to find out about.

1:06:33I'm really hoping that comes good. And that way, my children can eventually go to school where I went to school. It'll be good. All right, Ben, to round us up here for you, I'm curious what you think we should do at the startup level, the venture level, the technology industry level, and maybe even the government level to ameliorate what I think we can all see as rising discontent amongst the populace against AI. And this is often seen in data center protests and so forth. But what are some proactive steps that the tech industry can do to get on the right side of public opinion before this becomes an electoral issue?

1:07:03My question is so much harder. This is like a setup. I'm like, I can't believe that You just laid that on me. It's like you're being punished. Like I came here. It's so fun when Ben's on the hot seat. Trying to be nice. Well, the cool thing is about me is that I don't want your money, so I don't have to be nice. Okay, great. Cool. Everyone else has to kiss her ass. Look, that is a great question to which I do not have a great answer. I think that I am always amazed by how negatively AI is viewed by people that don't work around this business. Like friends that I have that are one or two degrees removed are generally terrified.

1:07:49And I think think of, by the way, there's also this narrative that somehow like I think people actually think anthropic is like pretty good. But open AI is like totally the like, you know, the empire in Star Wars or something. You know how these stories get told. I think it's a I think it's a big issue. Maybe to touch on the question that the other two answered, I don't trust our government to know how to monitor this and to sit over and to figure out how we should or shouldn't use AI models or what the rules should be. At the same time, we're in a like we're in a cold war and have been for a while.

1:08:32And, you know, even though it's not called that and like this is national security, but against what's best for jobs and the economy. I mean, the wealth gap is only getting worse by the minute in a way that is like, I don't know how this ends anything other than terribly. It is, we're set up in a really unfair, awful way right now. And this is, and AI is not going to make this better in any short or medium term. Yeah. And if you're on the video version, I just pulled up a chart from our dear friends over at Fred, which shows the share of labor comp as a percentage of GDP. And if you go back to the 50s, the era that people like to kind of pine for, fairly or not, it was up in the high 60s.

1:09:20And it's fallen down very sharply lately, all the down to about 57, which I think is really the root of a lot of discontent. And I don't have a solution either. But I think it would also be incredibly sad if we ended up shooting our own feet or tying our own shoelaces and preventing a lot of future economic gains because we couldn't figure out a way to share the pie a bit more effectively now. That just seems to be a non-GDP accretive approach. But you guys are not - By the way, maybe, but like going in some of these, you know, I'm not in California, but the billionaire tax and some of these sort of, you know, very brute force measures feel like, you know, at best band-aids or punishments, they, that does not feel like the solution to figuring out how we fix this problem.

1:10:05No, because everyone's going to leave. I'm already talking to founders who live in Nevada just across the border because they want to get away from the tax. It's interesting because like when you think about it right now, if you're a free market capitalist wanting to make a pro common sense argument for it, you have no home. You know, the left is becoming Democrat socialists and the right is MAGA and saying we hate immigrants. And so, you know, like you can't even make a credible case for why AI is good because the people driving the discussion don't want to hear it from both ends. And so that part of it really bugs me is that there's no natural home for the adult in the room conversations about what the right answer is.

1:10:51Well, there shouldn't be no party for it and candidates for it at a, you know, like, I don't know, let's see where we're getting. Although, Ben, what you were saying about, I don't know if I, it's not about the trust or the effectiveness of the government, but I certainly wish, looking back, that we had had more regulation of social media. Social media, yep. And that we have a whole generation of kids that have been so negatively impacted by the fact that there was no oversight whatsoever for social media. And obviously, the ramifications from a security perspective are so much more grave with AI.

1:11:26If we weren't able to keep people safe. It's a fabulous point. It's a fabulous point. Having young kids, we've been left holding the bag to try and make them the only kid in the grade without access to Snapchat or something to punish them for the fact that nobody got in front of us. Yeah, I'm not looking forward to those days. I literally had a question in my fun section at the bottom of our notes today that was, how are you teaching your kids to thrive in life and the post-intelligence era? And I wrote that as a joke to myself, but I really meant the post-AI era. But I think post-intelligence actually may kind of better encompass.

1:11:56One thing I'm really concerned about is that a lot of people just can't read and can't do math. And I don't think that giving people during their learning years access to tools as powerful as AI is going to encourage them in a lot of cases. And I don't think parents know what the hell they're doing either. And now that I have kids that can reach for things, I see how they react to screens and it's made me rethink my entire relationship with technology. But here's some good news. Technology historically has made things better. And I think it's going to keep doing that, even though there will be some bumps in the road.

1:12:25I'm a long-term optimist, and I think that this is all going to end up being very good. I just hope we don't throw a couple generations of kids into the maw as we get to that point. This is not where I thought the show was going to end. I'm not going to lie. Maybe I should have shaken up how we did things. But guys, an absolute real treat today to have you on to talk about all this stuff. Just before we go, where can people find you online? And is there a category your firm is looking to invest in? And Mike, let's start with too. Yeah, I guess you could find me on X at M2JR. And then our website is www.floodgate.com.

1:13:00And basically, I'm investing in companies that complement the abundance of generative AI. I call it acceptance AI, but it's the companies that ensure the correctness of the work, rather than just generating more stuff. And so that's what I'm looking for. All right. Aileen. I'm Aileen Lee on X and also on LinkedIn and we're cowboy.vc. And we're generalists. I'm like these guys have been doing this for a while. So some of our best investments have been things that we never would have had on kind of like our shopping list. It's really what founders are have insight about. And so open for business.

1:13:42And Ben, take us on. Yeah, as you mentioned earlier, I'm not very active on Twitter, but I'm BenjLirer. And I use LinkedIn a little bit more, but not a ton. And we're at LirerHippo.com. But probably LinkedIn is the best place. And like Aileen, we are generalists. We're looking for great people early. Yep. If you guys are doing the investing, I think the future will be okay. Thanks all for coming on. This has been Twist. My name is Alex. We'll see you all next time.