The Exchanges, every show

Q How does the UI paradigm change in the world of AI? Like, everyone's like, now we're just all going to be voice. Do you agree with like, it's all voice.

A I think, so voice is amazing for enterprise. I think that, uh, one dynamic UIs and two chat UIs are overstated in consumer. And the best thinker on this is actually Eugenia who founded Replica and now Wabi. She's, she's great on this. And what she would tell you if she was here is that, look, most people don't want to save time. They want to spend time. Okay. And the products are designed by the most high agency people in the world. Like Sam and Elon are the most high agency people in the world. For them, the optimal UI is a chat box where you say exactly what you want and like, voila, there it is. But for many people, they're again, looking to waste time, spend time. They want a browse based interface. They're not quite sure what they want. Can't always articulate it. So I think that in a world where we have intent-based and browse-based, browse-based largely stays the same. And perhaps the future of intent-based is chat. I'm still a little skeptical.

AI assessment note: “browse-based largely stays the same. And perhaps the future of intent-based is chat.”

Q How does the UI paradigm change in the world of AI? Like, everyone's like, now we're just all going to be voice. Do you agree with like, it's all voice.

A I think, so voice is amazing for enterprise. I think that, uh, one dynamic UIs and two chat UIs are overstated in consumer. And the best thinker on this is actually Eugenia who founded Replica and now Wabi. She's, she's great on this. And what she would tell you if she was here is that, look, most people don't want to save time. They want to spend time. Okay. And the products are designed by the most high agency people in the world. Like Sam and Elon are the most high agency people in the world. For them, the optimal UI is a chat box where you say exactly what you want and like, voila, there it is. But for many people, they're again, looking to waste time, spend time. They want a browse based interface. They're not quite sure what they want. Can't always articulate it. So I think that in a world where we have intent-based and browse-based, browse-based largely stays the same. And perhaps the future of intent-based is chat. I'm still a little skeptical.

AI assessment note: “voice is amazing for enterprise. I think that... chat UIs are overstated in consumer”

Q terms of like having a command of technology, and you said earlier about the challenge of shipping products and getting from zero to one, showing that you've had a success building in the past is a great way to prove that you can do it moving forwards. Do you have an unreasonable or an unwavering leaning towards serial founders who've proven that they can do it because of their track?

A Yeah, I'll give you a nuance take on this. So I think that repeat founders working in their domain of expertise Are formidable. Like the Clutch guys sold a company to Carvana. Uh, you know, they weren't super happy with the way that the whole thing, you know, ended up in terms of their startup achieving their ambitions. They went and then started another company out of that also in the auto space called Clutch. It's going extraordinarily well. And they know, you know, they're taking all the shortcuts because they know the market. So I do think particularly in enterprise, um, working in the same domain and, you know, being a repeat entrepreneur is a huge, uh, source of alpha. I actually think conversely in consumer, having a beginner's mind and a high willingness to be embarrassed is a competitive advantage because so many consumer products feel embarrassing and, you know, they're immediately dismissed as embarrassing or impossible or a silly, non-serious thing to be working on. When you're 25 and like the stakes are low, you just want to make something happen in the world. That is a perfect setup. Once you've sold a company, all of a sudden it's like, your venture friends are like, what are you working on? You know, like your girlfriend or your boyfriend's like, what are you working on? You want to sound cool at dinner parties or at the bar, and that slight hesitation to be emb…

AI assessment note: “repeat founders working in their domain of expertise Are formidable”

Q So when you think about market composition for that market and the kind of developer tooling, Uh, space. Does that look more like cloud, or does that look more like Uber and Lyft?

A I don't think it looks like Uber and Lyft, right? I think Uber and Lyft are the, to my mind, the most extreme examples of pure substitutes, and a lot of the sort of price has been computed away. You look at cloud, you sort of have this oligopoly where they all actually have pretty reasonable margins, right? And, you know, you can squint and say, of course they have their specializations, but they're roughly substitutes, and yet they've all done well. I think the, uh, the foundation model companies look a little bit like that. And I think in the apps layer, you're just going to have people that want to Consume the code they generate through a rich IDE and those that want to be closer to the metal, and that's probably closer to AWS Google Cloud than it is Uber Lyft.

AI assessment note: “I don't think it looks like Uber and Lyft, right?”

Q many companies and so many startups, so many products, everyone's launching the speed at which companies and products are shipped is like thousand xing. How do you think about durability and moats when you look at a startup? Because a lot of founders get that question. Everyone's getting that question. How am I not a rapper? What are, what are signs that tell you this might be a durable thing?

A Well, I think there's two important ideas. One is something that, um, Jesse from Decagon said, which I love, and that is that moats are most often discovered, not designed. I think it's really easy. I've done this as a founder to get in your own head about like, Hey, I need a business plan that survives scrutiny from MBAs and VCs. I've got to have some really sophisticated, you know, idea of what my moat will be. And for that team, they just started shipping and it developed over time. Another great example of this is cursor. You know, they were criticized a lot for not having a moat, but it turned out that initially being a high NPS DAU product was really good. And over time they captured all the reasoning traces. They train their own models, the composer one, two models, and you know, so on and so forth. We know how that story plays out. So moats can be discovered. They don't have to be designed is one. And then I think the other is that we seem to have forgotten that the classic moats, none of the classic moats are based on how hard it is to make the software. You know, like we're not building self-driving cars. Most of us aren't. So it's network effects, it's scale advantages, it's brand effects, proprietary sort of data or what was historically called a cornered resource. Every moat from five years ago generally is still a good moat. We just need founders that have ambitio…

AI assessment note: “moats are most often discovered, not designed”

Q To what extent is models invading the apps layer a credible threat to the verticalization of apps?

A Yeah, so this is such an interesting topic. So Granola, which we're not investors in, but I admire them a great deal. It's a great company. Um, they've built a really interesting thing, and they're first, of course, to live meeting recording and transcription, which is awesome. They have been copied to the moon, right? Now, everybody has a meeting transcription feature. OpenAI released one within ChatGPT. Very cool. The thing about Granola, and I assume this is true, is that their vision is not to be a meeting transcription product. I assume it's to be a productivity suite, right? They're going to build Word and Docs and Spreadsheet and all of these other products around that core primitive, does OpenAI have the sort of prioritization, the resources, and the ambition in that direction to build all the feature surface around the primitive? So I think the models will often actually recreate the primitive and even do product marketing, which I think the Claude legal stuff was. But if you have a market that demands a lot of feature surface, I just think the model companies are less set up to prioritize it.

AI assessment note: “if you have a market that demands a lot of feature surface, I just think”

Q Are there any other locations where you think there is actually positivity associated with that being located there?

A Tel Aviv. I think Tel Aviv, you can be incredibly ambitious and uncompromising on that ambition and have a really, really good reason to be there. I think the other nice thing about the Tel Aviv ecosystem is that the country's so small, it's ten million people, that you can't possibly fool yourself into thinking that the domestic market is going to be big enough for whatever you're doing, so you immediately go outside. Whereas if you're in London, there's sixty million people here. Okay. And you might say, well, that's actually a lot of people. And you know what? There are parts of the market like fintech where the LTVs are so high that perhaps sixty million is sufficient. But for most mass market products, it's just not sufficient. And if you end up starting focused on the domestic market, it's often hard to actually move on to a bigger market. So there are all of these reasons that it's just, it's not that it can't be done. And there are incredible counterexamples like 11. Um, but I do think that is just that much easier in SF. And that's why that's where I focus.

AI assessment note: “Tel Aviv. I think Tel Aviv, you can be incredibly ambitious”

Q To what extent is models invading the apps layer a credible threat to the verticalization of apps?

A Yeah, so this is such an interesting topic. So Granola, which we're not investors in, but I admire them a great deal. It's a great company. Um, they've built a really interesting thing, and they're first, of course, to live meeting recording and transcription, which is awesome. They have been copied to the moon, right? Now, everybody has a meeting transcription feature. OpenAI released one within ChatGPT. Very cool. The thing about Granola, and I assume this is true, is that their vision is not to be a meeting transcription product. I assume it's to be a productivity suite, right? They're going to build Word and Docs and Spreadsheet and all of these other products around that core primitive, does OpenAI have the sort of prioritization, the resources, and the ambition in that direction to build all the feature surface around the primitive? So I think the models will often actually recreate the primitive and even do product marketing, which I think the Claude legal stuff was. But if you have a market that demands a lot of feature surface, I just think the model companies are less set up to prioritize it.

AI assessment note: “if you have a market that demands a lot of feature surface, I just think”

Q What have you changed your mind on most in the last 12 months?

A I think the thing that surprised me about this product cycle, you know, I was building my, my first company in the mobile product cycle and, um, in the mobile product cycle, the sort of like the anointed winners in 2008, 2009 were not the eventual winners. So we had the cycle where you sort of had the friendsters and then two or three years later you had the Facebooks, right? In this product cycle, what's actually interesting is a bunch of the early leaders from 23 and 24 have maintained their lead. You know, we talked about Harvey. That's a really impressive company. You know, Gamma is a really impressive company. Cursor is a really impressive company. Like the companies that were early have so far continued to actually be dominant, and that's something that I've sort of changed my mind on. I think in 26, we're going to see a whole new set of categories. Can I share my view on where we are on the market? I think that late 22, November 22 is ChatGPT. 23, a lot of the kind of obviously good ideas, and that's not to denigrate them, they were obviously good were started in some of 24. In the end of 24, reasoning models started working. So even the ideas that were obviously good but not working, suddenly, many of them suddenly started working with the advent of O-one and Deep Seek. 25 of those companies scaled, so now we are starting to see for existing markets, which is like custo…

AI assessment note: “companies that were early have so far continued to actually be dominant”

Q we see the true substitution of AI products based on price. He's like, 11 Labs. I love it. It's amazing. It's too expensive. It's too expensive. This is the year where we've moved from trying things to shit. It works, but it's too expensive. No comment on 11 Labs. But do you agree that we're going to see this transition in mindset from shit, it works, to shit, it's expensive?

A I don't think so. Because I think what we keep seeing is as the models get better, downstream players' ability to take those capabilities, productize them, and raise prices, Has outstripped the raising costs, right? So the incremental cost increase potentially, and in many cases not a cost increase, but it's, you know, not, not a cost decrease, uh, is so far outweighed by what the new capability unlocks. Like coding agents, right? What you can do with coding agents, Cloud Code came out last February, is dramatically better. Is anybody here saying, well, I should go back and use Sonnet three seven because it's cheaper, you know, or I should use something other than Opus four five or Codex It's five two. Like nobody is saying that because the capability is so much more powerful. It really like sparks your imagination in the other direction. What more can we do rather than how do we make the existing thing cheaper?

AI assessment note: “I don't think so. Because I think what we keep seeing”

Q Do you agree that the best founders you work with don't need their VCs?

A I think the best founders that I work with know how to maximally leverage their VCs. And look, I think there is a set of founders who perhaps would never need their investors, but I do think that the best founders how to, you know, know how to sort of, uh, extend their success and increase their momentum by leveraging the right investors. Like Alex does, right? Dude. I mean, I basically have a sales quota with Alex, you know, and DG would say the same thing. And Ben, he's even calling Ben saying, Hey Ben, can you help make this introduction XYZ? Like, He knows how to get the best out of Andreessen Horowitz, and all of the, it's not just the investors, the entire team shows up for him that way. Could he do it without us? Of course he could.

AI assessment note: “Could he do it without us? Of course he could.”

Q I think people are consistently concerned by, we mentioned it earlier, but defensibility, switching costs, durability. When we think about moats and Alex's statement of hostages, not customers, in a new world of AI, do we just accept that there's no defensibility or are new moats created?

A I think defensibility still exists and still matters. Like, networks are the gold standard, and they still are. You know, a network effect product is incredibly powerful. Now look, you might argue that something like Moldbook is a new type of synthetic network that perhaps means there are certain types of networks that are less defensible than they were, you know, once were. But something like an Airbnb, you know, you can have all the vibe coding in the world. Like, their network effect is incredibly powerful. So one, I think defensibility matters. Traditional modes still do matter. I do think that within modes like systems of record, there will be some who are more or less prone to disruption. So if you're an on-prem database, and there's no engagement layer, and there's not a lot of human workflows built around the so-called system of record, I think that you actually are at some risk. If you're the core system for a bank, you've got thousands of transactions per second. You've got hundreds of humans that interact with you. You have this incredible demand for accuracy. I still think that you're sort of as good as gold in terms of defensibility.

AI assessment note: “I think defensibility still exists and still matters. Like, networks are the gold standard”

Q Mine you should do. Yeah, a hundred percent, yes. Uh, so you said about market underestimation. I underestimated the, the payroll market, um, specifically. I thought Alex, I didn't underestimate him, but the market I did. What market did you underestimate, which you later realized you were wrong on, and what did you learn?

A It's such a good question about which market did we underestimate. I mean, I've made this mistake a couple of times, you know, like, for example, I remember when we were acquired by Google, and I remember looking at the sort of the stock price and telling my co-founder, like, well, maybe this can go up 10 or 15 or 20 or 30%. How much bigger can it possibly get? So if you look at a company like that, which was so capable but seemed dominant in their core market, it was very hard to squint and see what they would become, and they're so much more valuable than they once were. Right. I think another interesting example of this is credit karma. You know, free credit scores for Americans, and I know the credit score is a much bigger concept in America than it is where I grew up in Canada or even here. I think you'd sort of ask yourself, if you did the back of the envelope, you would say, well, most people tend to use their credit score once or twice a year. Right. And most people don't even actually need it that often. It's only when you're applying for a new financial product. And then for most people, you either have Exceptional credit and you don't really need to look at it because you already know that or you have terrible credit and you just don't want to look at it and you kind of already know that. So now you've got this torso of people who infrequently need access to their cr…

AI assessment note: “I think another interesting example of this is credit karma.”

Q Andres Guido, uh, we want to talk about sovereign AI, uh, AI and geopolitics, and let's start with the news. Uh, our partner Ben is in the Middle East right now. Um, what happened, uh, and, and why is it so important?

A What happened is, um, the kingdom announced that they're going to build their own local hyperscaler or AI platform called Humane. And I think why it's notable is that they are As opposed to the status quo of the cloud era, um, they're viewing the AI era as one where they'd like the vast majority Of AI workloads to run locally. If you can, if you think about the last 20 years, the way the cloud evolved was that the vast majority of cloud infrastructure basically existed in two places, right? China and the US. And the US ended up being the home for the vast majority of cloud providers to the rest of the world. That doesn't seem to be the way AI is playing out. Because we have a number of frontier nations who, who are basically raising their hands and saying, We'd like infrastructure independence. The idea being that we'd like our own infrastructure that runs our own models that can decide where we have the autonomy to build the future of AI independent of any other nation, which is quite a big shift. Um, and I think the headline numbers are somewhere in the range of, um, a hundred to two hundred and fifty billion worth of Cluster build out that they've announced, of which about 500 megawatt seems to be the atomic unit of these clusters that they're building. So what's going on? A number of countries, the, with the kingdom being the one that's most recent, are, have been announcin…

AI assessment note: “the kingdom announced that they're going to build their own local hyperscaler”

Q down their AI development. They paused Uh, their RL kind of phase on their latest model because of what they're seeing. So that's obviously a big concern for people, just how fast and smart these models get. Any thoughts on just that? That's like a big shift now instead of race ahead to the fastest, best model ever. Okay, we actually have to slow these things down. That feels crazy.

A I mean, without commenting on OpenAI specifically, I think that, um, maybe I'm a little skeptical on some of these things where I think the kind of the aura That Anthropic got from having a model that was too dangerous to release was extraordinary. And maybe they had a GPU shortage, you know, maybe it was, you know, the capabilities were more advanced than they actually wanted, you know, maybe they actually wanted to keep that proprietary model internal to extend their own lead. So I think there's a lot of sort of confounding factors that would cause you to pull back a little bit. Look, I do take the points about offensive cyber seriously. Which is we should harden all of our systems before we make them trivial to penetrate. Um, but I think the sort of concept of the model that's too dangerous to release it, it kind of conflates marketing, um, inference capacity, and then also economic considerations, like, do you want to externalize your competitive advantage or use it to make yourself better?

AI assessment note: “maybe I'm a little skeptical on some of these things”

Q Yeah, it's like you have to click three different matching shapes, which, like, come on, you can't do that? Hopefully by the time this comes out, I haven't seen it. I think, uh, it's a few people I guess in a row, I'm like, I haven't seen it yet. Oh, man. Okay, uh, favorite new AI product? I don't know, favorite AI product recently that you've, they've given me joy.

A Oh man, I've, I've spent a lot of time with the personal agents. I think Grok's okay. I'm going to give it to GrokBots only because it's just so unhinged for it to be so ambitious about sort of caching credentials and getting work done on your behalf. Like I love it and I expect it from a startup. Um, but they actually are doing things that I think no other sort of big company would do. It's really, really well done. It's a really thoughtful UI. It's got a really powerful foundation model. I think it also is like, okay, wait, maybe this is not a two horse race on the sort of model side. So I, I just think the product is like fun and ambitious and, uh, and is sort of taking risks that other products like that wouldn't take.

AI assessment note: “I'm going to give it to GrokBots only because it's just so unhinged”

Q So when you think about market composition for that market and the kind of developer tooling, Uh, space. Does that look more like cloud, or does that look more like Uber and Lyft?

A I don't think it looks like Uber and Lyft, right? I think Uber and Lyft are the, to my mind, the most extreme examples of pure substitutes, and a lot of the sort of price has been computed away. You look at cloud, you sort of have this oligopoly where they all actually have pretty reasonable margins, right? And, you know, you can squint and say, of course they have their specializations, but they're roughly substitutes, and yet they've all done well. I think the, uh, the foundation model companies look a little bit like that. And I think in the apps layer, you're just going to have people that want to Consume the code they generate through a rich IDE and those that want to be closer to the metal, and that's probably closer to AWS Google Cloud than it is Uber Lyft.

AI assessment note: “that's probably closer to AWS Google Cloud than it is Uber Lyft”

Q What does that look like? That means like AI native law firms,

A Possibly. I think it means dramatic productivity increases for lawyers, dramatic productivity increases for programmers and engineers. Like I think that the, the difficulty of doing a hundred percent of a job is really, really high. It's this thing of like pretty easy to get to 60, 70, 80%. So I do think that's why a 20% productivity increase so far we're seeing it show up more as, you know, a four day work week than 20% less jobs because jobs as bundles of tasks don't set themselves up to be 100% automated so far. Right. You can do all the customer support you want, but sometimes you got to take the customer out for a steak dinner. And so far the models are not doing that.

AI assessment note: “I think it means dramatic productivity increases for lawyers, dramatic productivity increases for programmers”

Q terms of like having a command of technology, and you said earlier about the challenge of shipping products and getting from zero to one, showing that you've had a success building in the past is a great way to prove that you can do it moving forwards. Do you have an unreasonable or an unwavering leaning towards serial founders who've proven that they can do it because of their track?

A Yeah, I'll give you a nuance take on this. So I think that repeat founders working in their domain of expertise Are formidable. Like the Clutch guys sold a company to Carvana. Uh, you know, they weren't super happy with the way that the whole thing, you know, ended up in terms of their startup achieving their ambitions. They went and then started another company out of that also in the auto space called Clutch. It's going extraordinarily well. And they know, you know, they're taking all the shortcuts because they know the market. So I do think particularly in enterprise, um, working in the same domain and, you know, being a repeat entrepreneur is a huge, uh, source of alpha. I actually think conversely in consumer, having a beginner's mind and a high willingness to be embarrassed is a competitive advantage because so many consumer products feel embarrassing and, you know, they're immediately dismissed as embarrassing or impossible or a silly, non-serious thing to be working on. When you're 25 and like the stakes are low, you just want to make something happen in the world. That is a perfect setup. Once you've sold a company, all of a sudden it's like, your venture friends are like, what are you working on? You know, like your girlfriend or your boyfriend's like, what are you working on? You want to sound cool at dinner parties or at the bar, and that slight hesitation to be emb…

AI assessment note: “Yeah, I'll give you a nuance take on this. So I think that repeat founders”

Q We say about kind of trying product. I mean, 90% of the companies that I get pitched say, especially on the application side, they're agent-led or agent-first. You said before there might be agent overhype. Can you talk to me about this? Why do you feel there's agent overhype today, and what does that mean?

A Here's what I think. I think that the extremist view that we are going to have autonomous agents that simply do everything over incredibly long time horizons, like maybe we'll get there someday, but I do think that at a minimum, you need humans in the loop for exception handling. And then these models are only as good as the instructions that we give them. And our instructions, I mean, think about the way you manage your team. Your instructions are often frustratingly vague. So I do think that we need people in a tight loop with the models to actually achieve our objectives. And I think that the, the sort of agent maximalist view, which is, you know, you just like chill out for the day and your AI does everything you need to do is probably a little bit ahead of where we actually are.

AI assessment note: “the sort of agent maximalist view... is probably a little bit ahead of where we actually are”

Q we see the true substitution of AI products based on price. He's like, 11 Labs. I love it. It's amazing. It's too expensive. It's too expensive. This is the year where we've moved from trying things to shit. It works, but it's too expensive. No comment on 11 Labs. But do you agree that we're going to see this transition in mindset from shit, it works, to shit, it's expensive?

A I don't think so. Because I think what we keep seeing is as the models get better, downstream players' ability to take those capabilities, productize them, and raise prices, Has outstripped the raising costs, right? So the incremental cost increase potentially, and in many cases not a cost increase, but it's, you know, not, not a cost decrease, uh, is so far outweighed by what the new capability unlocks. Like coding agents, right? What you can do with coding agents, Cloud Code came out last February, is dramatically better. Is anybody here saying, well, I should go back and use Sonnet three seven because it's cheaper, you know, or I should use something other than Opus four five or Codex It's five two. Like nobody is saying that because the capability is so much more powerful. It really like sparks your imagination in the other direction. What more can we do rather than how do we make the existing thing cheaper?

AI assessment note: “I don't think so. Because I think what we keep seeing is”

Q Do you agree that the best founders you work with don't need their VCs?

A I think the best founders that I work with know how to maximally leverage their VCs. And look, I think there is a set of founders who perhaps would never need their investors, but I do think that the best founders how to, you know, know how to sort of, uh, extend their success and increase their momentum by leveraging the right investors. Like Alex does, right? Dude. I mean, I basically have a sales quota with Alex, you know, and DG would say the same thing. And Ben, he's even calling Ben saying, Hey Ben, can you help make this introduction XYZ? Like, He knows how to get the best out of Andreessen Horowitz, and all of the, it's not just the investors, the entire team shows up for him that way. Could he do it without us? Of course he could.

AI assessment note: “Could he do it without us? Of course he could.”

Q When you reflect on companies or investments that you've made that were not good, What did you not see?

A I mean, I, I think, you know, again, if there's a mistake that I've made, it's been, um, being a bit too casual about product market fit, and this was more of a 2021 mistake, which is assuming something had product market fit, and perhaps it didn't, and perhaps the founder had a super credible theory, which by the way, matched my theory for why it would get to product market fit. But as I said, it's easy to overestimate the sort of path from zero to one. And, uh, and I'd say if there was a sort of mistake I made, it was, you know, not being intellectually honest about is this actually working or do I think it will work in the near future? Now, look, I've done a bunch of seed investing and I've made the bet. And I think if you're intellectually honest and sort of clear sighted about a belief that it will work, then that's a fine way to invest. But investing with the sort of self-deception of like, well, let's just assume it's working when it's not quite working is a mistake.

AI assessment note: “if there's a mistake that I've made, it's been, um, being a bit too casual”

Q What does that look like? That means like AI native law firms,

A Possibly. I think it means dramatic productivity increases for lawyers, dramatic productivity increases for programmers and engineers. Like I think that the, the difficulty of doing a hundred percent of a job is really, really high. It's this thing of like pretty easy to get to 60, 70, 80%. So I do think that's why a 20% productivity increase so far we're seeing it show up more as, you know, a four day work week than 20% less jobs because jobs as bundles of tasks don't set themselves up to be 100% automated so far. Right. You can do all the customer support you want, but sometimes you got to take the customer out for a steak dinner. And so far the models are not doing that.

AI assessment note: “I think it means dramatic productivity increases for lawyers, dramatic productivity increases for programmers”

Q You said the word sufficient there. Yes, you can build a sufficient size business, say, in the UK, a three to five billion dollar business, say, as an example. Respectfully, when we look at companies being created today, three to five billion dollars just doesn't seem like it's interesting enough. Has the world of venture changed so significantly on what is sufficient for a venture outcome?

A I mean, three to five billion is an extraordinary outcome. Don't get me wrong. So in no way, no way am I like minimizing that. And look, I do think that those types of venture outcomes stack to create really meaningful funds. So this is not about working backwards from venture economics, but the biggest companies in the world today are trillion dollar companies. If you want to build a trillion dollar company, if that's your intention, you've sort of got to start with a set of assumptions that can lead to that. If your intention is to build an extraordinary enterprise, and you build a three to five billion dollar enterprise, like, you are one of the few people in the world.

AI assessment note: “three to five billion is an extraordinary outcome... those types of venture outcomes stack”

Q Um, no wonder she wants to finish the walk. Um, my, my question to you is, do margins matter as much in a world of AI, and are we entering a new way that we should be thinking about margins?

A Yeah, so here is actually where I think there's nuance in the margin conversation that's important, okay? So if you look back at any time you've gotten, we should talk about, you know, the bubble that doesn't exist, or perhaps there is some sort of subsidization and distortions happening in the market. For the record, I don't believe we're in that period, but I do think that any time you have this sort of superheated markets, you have some distortion, okay? If you look at the distortion from 2021, you essentially had this indirect subsidy of Google and Facebook. So you would invest in a fintech company. You would give them ten million dollars. They would go spend eight million dollars on Google ads and Facebook ads. So there's a subsidy that was happening that were sort of these empty calories for the startup. If instead you look at the form of subsidy that happens today, what it typically means is zero margin or negative gross margin credits for the user to try the product. So these things tend to be a drag, but these are actually very healthy calories for the companies because out of that you get conversion into High paying users and many of whom are actual power users. So I do think that, um, the, the blended margin story for AI native companies tends to be worse. But if you look at the overall sort of form of distortion that's happening, it's a much better one than we had f…

AI assessment note: “the blended margin story for AI native companies tends to be worse”

Q I, I do a show with Jason Lemkin and Rory O'Driscoll, um, and Rory said something brilliant, I think, which is like, this will all work out if we see the transition of spend from the 12% SAS budgets that we operate in today, transition from that budget to the human labor budget. Do you think we will see that transition?

A I mean, we're already seeing it. I think DG was on the show talking about CH Robinson, right? Like we're seeing a lot of companies start to see the productivity improvement from this new technology. I mean, how can they not, right? It's not just coding agents in which this is showing up. You talked about voice. Voice is the wedge into the enterprise. Voice agents are so powerful. And by the way, I think that the, the, uh, the near term story of a lot of voice, I know we talked about support. You talked about sort of customer support. That's interesting. But the more interesting thing is why is support an isolated function? Why is it? And let's go through it. It's you typically have had sales, support, operations, and collections. You know, who is the person that's really good at customer support? Empathetic. They're a listener. They really understand the product well. Who is really good at sales? They're more of a yapper. They're a talker. They're high energy. They're very charismatic. You know, they're sort of good at the upsell. They're always in a good mood. You've got these two different human archetypes for these two different roles. We've typically organized the enterprise around these two archetypes, right? But now the models can be either of those people at any time. So the most sophisticated companies are starting to take support, sales, collections, operations, bundle…

AI assessment note: “I mean, we're already seeing it.”

Q I have a lot of enterprise SaaS companies that do double-double or triple-triple-double-double. Is the world of triple-triple-double-double dead, and do we all have to be lovable, wrapped in 11 labs to get funded?

A I don't think so. I mean, I think that a lot of it is dependent. It's calibrated to your part of the market. So product velocity plus business velocity. I do think that you have to be top quartile compared to your peer set, right? I think there are some markets that are consumer led or bottoms up where you can just see this explosive growth, and that is awesome. It's extraordinary to see. But look, if you're selling an ERP, you've got a much more cautious customer. It's a much more high stakes sale. If you're selling payroll now, granted in the case of payroll, Alex and team have done a tremendous job of what should be like a slow boil sale and turning it into a fast boil sale. And they've got some very specific ways that they do that. But typically that is an industry that moves on slower cycles. So I think that there is just physics to some of these markets that mean triple, triple, double, double is phenomenal. But there are other markets in which, especially with these new primitives, you can go like 10 to a hundred or 10 to 200.

AI assessment note: “I don't think so. I mean, I think that a lot of it is dependent.”

Q We say about kind of trying product. I mean, 90% of the companies that I get pitched say, especially on the application side, they're agent-led or agent-first. You said before there might be agent overhype. Can you talk to me about this? Why do you feel there's agent overhype today, and what does that mean?

A Here's what I think. I think that the extremist view that we are going to have autonomous agents that simply do everything over incredibly long time horizons, like maybe we'll get there someday, but I do think that at a minimum, you need humans in the loop for exception handling. And then these models are only as good as the instructions that we give them. And our instructions, I mean, think about the way you manage your team. Your instructions are often frustratingly vague. So I do think that we need people in a tight loop with the models to actually achieve our objectives. And I think that the, the sort of agent maximalist view, which is, you know, you just like chill out for the day and your AI does everything you need to do is probably a little bit ahead of where we actually are.

AI assessment note: “I think that the extremist view that we are going to have autonomous agents”

Q When you reflect on companies or investments that you've made that were not good, What did you not see?

A I mean, I, I think, you know, again, if there's a mistake that I've made, it's been, um, being a bit too casual about product market fit, and this was more of a 2021 mistake, which is assuming something had product market fit, and perhaps it didn't, and perhaps the founder had a super credible theory, which by the way, matched my theory for why it would get to product market fit. But as I said, it's easy to overestimate the sort of path from zero to one. And, uh, and I'd say if there was a sort of mistake I made, it was, you know, not being intellectually honest about is this actually working or do I think it will work in the near future? Now, look, I've done a bunch of seed investing and I've made the bet. And I think if you're intellectually honest and sort of clear sighted about a belief that it will work, then that's a fine way to invest. But investing with the sort of self-deception of like, well, let's just assume it's working when it's not quite working is a mistake.

AI assessment note: “being a bit too casual about product market fit, and this was more of a 2021 mistake”