00:23:38 · Luca Ferrari, Anish Acharya
[00:23:38] S2: Well, I tend to agree. I mean... But not necessarily to the to the same extent. Yeah. Exactly. That's right. [00:23:43] S1: That's right. Okay. Let's quickly talk about durability and moats as a favorite topic in Silicon Valley right now. You know, a year ago was SaaSpocalypse around Opus four five four six. All those names traded down 40% in backup, so I don't know what folks accomplished with the SaaSpocalypse. It seems like it was a little overblown, but it has brought about this discussion of where durability lies in a world where you can trivially create software with AI. How do you think about it?
00:24:08 · Luca Ferrari
[00:24:08] S2: I never fully... So I think the valuations for SaaS historically were for the most part absurdly high, but not because AI was not a risk, because they didn't make any sense from at least from the perspective of someone going back to what we were talking about earlier, someone who values assets based on reasonable expectations of free cash flows in the long run. In most cases, I think you could only justify those valuations based on the speculation that you'd be able to sell them to someone for, say, the same multiple down the line, but in the meantime you've grown 50% or something. Right? Because they were based often on assuming that the current 12% organic growth would last for fifteen years. And it's like, sure, but you know, the world changes so quickly, that's insane. Right? Organic growth almost never goes up, by the way. I mean, very rarely. Again, going back to Mark's. It typically goes down and so those assumptions made no sense. Now what I think AI gave people an excuse to go back to their models and say maybe I don't fully believe in those assumptions, let me be more cautious, and the reset valuations were, I think broadly speaking, reasonable. I mean you could always argue plus minus 20%, but they were more in the range of reasonable in my view. Now I never really bought too much into the risk of now writing software is cheaper, therefore these companies are screwed. I do buy in the idea of it's gonna be much cheaper, therefore they'll suffer on the margin, but now they're screwed. The reason is I think writing software has been reasonably cheap for a long time and most of the companies that matter, you know, companies worth at least, I don't know, a billion dollars, the cost of replicating their product technically Yes. Was negligible relative to the value of being a, you know, a winner or significant gaining significant market share in that market even five or ten years ago. Like, yeah, maybe you needed 20 good engineers, but it's not like replicating Salesforce is not rocket science, you know. We own WeTransfer. If you had 10, you know, five great new graduates who wanna spend three months in a garage, they could probably build a product that's 80% as good, you know, and in fact, guess what? A 100 teams have done that. There's plenty of clones, right?
00:26:22 · Luca Ferrari
So mostly software or digital products do not succeed because it's very expensive to replicate them. They're now SpaceX. They mostly succeed because of other factors. First mover advantage, sometimes you start, you build a brand simply because people know you and nothing else. And then there is a lot of habits people don't, you know, the actual switching costs but just also just psychological switching costs people don't want to bother switching if it means you know our company making a billion in revenue you're saving $3,000 for that such who wants to even bother taking the time and risk and it's rational not to do so. Two, it could be network effects. We own Eventbrite. People don't go to Eventbrite because you know nobody among all homo sapiens has managed to replicate a ticketing platform but it's because on Eventbrite you have 90,000,000 people who go monthly look tickets, and so you sell more basically, you know, right? And so AI can replicate that a million times, but they don't have the audience. And people go there because everybody posts events there, right? Or something like Komoot, you have proprietary data. With Komoot because of the history with all these users, we know that that particular path is muddy and you know it rained and there is a dog barking like and we have photos of those corners and trees and whatnot and turns that are not available, they're not available on satellite, they're nowhere to be found. And so again, it's not... It's actually... Komoot is quite sophisticated actually the routing, but still, know, smart people can build that over time, but they don't have the data or scale advantages, you know, Airtable, our most recent acquisition. Most customers pay Airtable less than $20,000 a year. Mostly there are large enterprises. Even if you get to a point where you can ask your agent by AI agent to press a button and it replicates the code for air table pretty much perfectly, which we're very far from. But we'll get there I think, you know, soon enough probably within the next few years. Supposedly, you'll still want to have someone at the company who's responsible to making sure that that code seems legit, that if something breaks goes to the agent and asks the agent to fix it and whatnot.
00:28:33 · Luca Ferrari, Anish Acharya
I think there's a far future where AIs run our lives and companies nobody works any longer, which is interesting, but you know, it's not... I think most people would agree it's not in the next five That's right. As long as you have someone being responsible for things, even just a 0.2 FTE of a software engineer would still cost you more than paying air table for the same thing. So, sure it's an interesting philosophical point but ultimately you're not gonna do it rationally. It's dumb. So there are all sorts of reasons why I think most not all but most digital products are reasonably well protected. Again they will suffer on the margin new customers you know maybe people search for products through their agents and agents are have a different set of criteria they follow to choose and so before they people tended to choose product A now it's more product B that will certainly hurt product A not to the point that they melt away overnight, you know, because of this. [00:29:29] S1: And those are natural market evolutions, right? Like web to mobile and preferences change, maybe it's agents instead of humans that are searching, but, you know, our data shows that the enterprise spends eight to 10% of total spend on software. So as you think about where you wanna point your AI and your token budget, it's probably not in replicating payroll. It just seems like a high downside, low upside proposition. [00:29:51] S2: I... Sounds so great.
00:29:51 · Luca Ferrari, Anish Acharya
[00:29:51] S2: I... Sounds so great. [00:29:52] S1: Yeah. Excellent. Hey. So you mentioned proprietary data, which I think is actually under discussed these days with everything happening in models. You know, there's a quip that many companies are sort of trading for less than the value of their tokens, which is the value that they would have if they were simply acquired for training data by models. Do you think that's true? Do you think proprietary data has changed?
00:30:13 · Luca Ferrari, Anish Acharya
[00:30:13] S2: So I I I haven't thought enough about this and I don't know enough about it, but I will say that claim strikes me as potentially a little bit misleading. Meaning, the value of data within new pools declines pretty rapidly. It's like if today I write a new blog post on the internet, nobody's gonna give me money for that. It's private and I'll go to open... I'm about to, you know, I may publish it, I may not. Do you want to give me it on top? You know, the first blog post they ever got access to or could ever access, they would probably have paid a thousand dollars for it, but now zero because incrementally it doesn't add anything to the corpus of knowledge that you can train AI models. So yes, maybe there's now a need for plenty of data on how people do project management or whatnot. The first companies selling that maybe make a bunch of money, but I think over time that trends to zero as the, you know, the quality of incremental training you know, returns a decelerate for incremental training. But I'm I'm in no position to determine whether, you know, companies could sell data for more of their enterprise value. I'm not I'm not certain about that. Interesting. Let... [00:31:24] S1: Let's double click on the kind of operating system you use at... Within Bending Spoons to help improve and optimize all of the companies. Can you talk about some of the key components of it? And which of the components do you think would have been most useful to you as a single product founder versus are very, you know, specific to what you do with the portfolio?