E9: Sam Lessin and Seth Rosenberg Debate: Will AI Favor Incumbents or Start-Ups?

0:09Welcome back to Turpentine VC, a podcast where we discuss the art and science of building successful venture firms, VC to VC. This week's episode is a special one. Sam Lesson, who we had on the podcast before on episode four, and Seth Rosenberg, general partner at Greylock, debate what the effects of AI on incumbents and the next wave of AI first products will look like. If you like what you hear, please do subscribe and leave us a review. We also started a companion newsletter, which we'll link in the show notes below. Now onto the interview. Seth, Sam, welcome to the podcast excited to excited to have a conversation and debate seth you just wrote this great blog post on opportunities in ai why don't you first talk about the inspiration for what inspired you to write this in the first place it was mostly to piss off sam i love it no i mean i sit on the board of a company called tome and you know when they talk to investors there's this constant narrative of you can't build value at the application layer of AI, right?

1:13These are thin layers on top of open AI and incumbents are going to capture all the value. And I just fundamentally disagree. And so I just wrote this post as kind of a call to action of, hey, I think creative product-driven entrepreneurs are going to take this technology and find ways to build totally new businesses that aren't going head to head with incumbents, but are actually finding totally new experiences. And I basically just tried to lay out what those opportunities might look like as a call to action for entrepreneurs who are more creative than me to figure it out. And then hopefully give me a call.

1:51Cool. So why don't we set the table here before getting into the debate? Seth, the post is called Product-Led AI. Why don't you walk through the different opportunities that you that you outline by the way i i had sam comment on it before i published it um which which actually helped make it much better such a nice guy it's like i already know i'm the asshole on the podcast this is great keep going i basically laid out kind of three three categories where i think massive companies can be built that um can either sidestep or just not compete at all with incumbents. And those three were AI first networks and marketplaces, right?

2:33So finding new ways, the marketplace of the future is likely not going to look like a website with a bunch of search and filters that match two sides directly, but it might look more like, you know, two seemingly disconnected personal assistants, right? The example I gave was, you know, a career coach and a sourcer for a recruiter, right, that are like two AI based personal assistants that actually match people intelligently on the back end. You know, you could also imagine experiences like Roblox, right, like a V2 of Roblox, that where you create more of a network where you have creators on one side and consumers or players on the other.

3:12The second category was just enterprise software categories, right? So So one just simple example in our portfolio is called Adminable Security, right? There's$20 billion of market cap for email security companies. The previous architecture was like, was analyzing links and attachments for viruses. The next attack factor is actually understanding the content of the email and seeing if someone's trying to impersonate your CFO. And so, you know, we backed a company called Adminable Security that basically takes an AI architecture to email security,$20 billion category. They're now after, I think, four or five years since founding over 100 million of ARR.

3:52Another example is productivity, right? And we can debate that. And then the last category I called out was kind of this co-pilot for services, which I think is going to be just a different product. It's not a question of adding AI to an existing SaaS product, but it's really a question of for every niche knowledge worker, just like we We had specialized robotics to automate manual labor. Now we're going to have specialized AI tools to automate knowledge work. And this isn't like trying to create extreme intelligence. It's really just, you know, understanding that most of the work we do is just dumb work transforming data.

4:28And now we have a tool that can do that for us. What I didn't include are kind of AI native companies like autonomous vehicles or autonomous drones like Andruil. But those are kind of the three categories I dug into. And when you and Sam were discussing, what were the debates you were having as you were putting this post together? Well, I just think he's completely wrong. Please. Film seems to be working really, really nicely. Here would be my take on it. And like why, again, from a seed perspective, I was like, Keith, go spend your time on something more valuable and why it might work and I might be wrong, right?

5:04One is, you know, Keith started out being like, we're going to do lightweight, mobile-first presentation software type stuff. And look, it's fine. Like if you're certainly some product you can find there, that will be better for some set of people. But like, you know, the big platforms around this stuff, Google Slides, whatever, are just enormous and distributed everywhere and baked into every piece of enterprise you're ever going to get. You're never going to build a big business doing this. You might build a small business doing this. And this is with love and respect for Keith, is that they really snatched potential victory out of the jaws of defeat in the last year with AI.

5:38They had built so much infrastructure for years to support this, that they were able to sprinkle some generative AI on top, right, in a smart way with credit to them, and turn it around and create a phenomenon that's going really quickly now. Now, is it actually the next important platform? You know, how long is it going to be before Google Slides or whatever integrates the same functionality and distributes it dramatically more broadly? Like, these are all TBDs, right? But there's no question from a product perspective, you know, Tome is an example of something where AI made a huge, large language models being plugged been made a huge difference at a key moment based on a lot of the work they had already put in and the infrastructure they had built.

6:15Look at this. He's a good debater. He's taking notes. He's ready to go. I mean, listen, Bergen record champion 2001 here, you know, Northern New Jersey debate for the win. And look, New Jersey has more people than all of Canada. So I think we're like on par, But look, I think let's go topic by topic. Marketplace AIs, right? Super interesting vision. And actually, interestingly, this idea that you'd have smart agents that would be your personal agent negotiating with other agents through some sort of ether web of some sort and creating outcomes. Interestingly, that was actually a very popular 1980s vision pre-internet of how technology would work.

6:59Like that was an original framework that came from science fiction, right? It has not played out that way. Like we've built very different, in a very different direction from that. It is possible that that comes back. I actually love that vision. Like you tell an AI a bunch of shit, it does a bunch of the chatter for you, feeds you the best. It's a great product vision. The question is, is that new marketplaces or is that a layer on top of Zillow, right? Is that a layer on top of eBay? And my sense is it's pretty obviously by default going to be layers. It's not going to be new things. It's extremely easy for Zillow to take its brand, its distribution, its relationships, everything it has, even on top of MLS and open data sources, and say, hey, guess what?

7:38We now have this Zillow smart agent. You tell us, what are the types of homes you're interested in? We'll negotiate what to look at. We'll negotiate with brokers to find the right houses for you. That's a much easier vision of the future for me because I think AI, in that case, is a great extender of existing platforms. it's not necessarily a fertile ground or an angle to start something brand new. Enterprise software, and specifically security. Again, I go back to the fact, are there super cool things you can do, large language models and the direction of AI? Forget AGI and all this bullshit, but just like with what exists now that increases everything from productivity to security?

8:13Of course, right? But it doesn't shift the table. The big security companies slot this stuff in. My line on this is a thousand people have pitched me, I'm sure you too, some version of Adobe or Canva for AI, it's bullshit. Because the reality is this is not a platform ship. This is just an extension. You don't want to lose magic lasso. You just want an extra feature set in Adobe. And unlike the shift from packaged software to the internet, this is not a pivot that's hard for big companies to make. In fact, it's quite easy for them to make. Most of them are stockpiling talent. They're ready to go the whole nine yards.

8:47And so I think if you go down the lines, we can talk about co-pilots. I just think what you're going to find is not that AI isn't an important shift. It's just that it's kind of like mobile shift on steroids where like mobile is like, it's a new platform. Like the table, it turns out 95 % of the value goes to the incumbents, right? Like sure, there's going to be one or two sneaky wins here or there, but it is not a paradigm shift that creates new winners. If you want to bet on edge technology, it's like, you know, metaverse not going to create new winners too expensive to play. You know, the existing company might be, might be the big deal might not be, but it's a big company play.

9:21AI is so obviously a big company play. And I think it will also empower tiny businesses. So like solo entrepreneurs, people that you can do way more just as cloud does that for people. The information, my wife's company spending thousands of dollars a month on open AI, it has superpowers for them specifically, but it's not like a new startup opportunity. And like I did the other one, like what you and I have both been interested in over years is crypto, whether or not it happens is structurally so disruptive. If it does happen, that it generates startups. I think AI just extends winners. Starting with Tome.

9:56First of all, you mentioned Tome might be the exception that doesn't prove the rule, right? Or the exception that proves the rule. But I think you and I are both looking for outliers. Definitionally, what we're looking for here is companies that are going to be worth over 10, 20,$50 billion. And those are always the exception. What is our job as VCs? And I think the biggest thing to say is there's a huge amount of random walk to venture capital. I've made tons of money on things I thought were dead. And I've had things that I thought were amazing go to zero. There's a ton of random walk. And if you just shoot the shotgun at everything and just say, as an asset class, VC works, and therefore the winners outweigh the lures.

10:42I'm not even sure that actually works in this era, but there is times in history where that has worked. And so I do think it's like kind of like, where do you point your shotgun? And my basic point would be, yes, there will be someone will make a ton of money on AI. It will be random. It's cool. Yeah. So I guess we both agree that there's going to be outliers. The question is like, how deterministic can you actually pick them at an early stage? And I agree. Like, you know, you have to be interested in a market and a problem statement and back the absolute best people in the world that are able to solve that problem.

11:12And so with Tome, obviously, they achieved the impossible by building a product that people really love. In nine months since launching, they're at over 10 million users. And they had a clear problem statement, which is, especially in 2020 when it was started and everyone was doing remote work, that there was actually no tool to communicate ideas in a really lightweight way in the workplace, especially when you're not in the same place, right? PowerPoint was a bloated tool that was built really for quarterly alt-hands, right? And then you had Slack that was for real-time messages. But if you wanted to do product updates or any substantive kind of artifacts that was linked into all of your different SaaS tools, there wasn't a good way to do that.

11:59And AI definitely accelerated and transformed the company. And now that's their primary focus. But that's the winner. That was the winner. And that was not the original plan. Again, that's not to knock the company. Great entrepreneurs pivot, right? And I do think because they built a bunch of infrastructure and cool stuff, they were able to quickly integrate AI and get a few months ahead of everyone else, at least. Not a company that was started as an AI company, right? And that had nothing to do with AI. And AI is the reason it's taking off, right? Yeah. Yeah. And that gets to this question of like productivity more broadly, right?

12:32Productivity software is like multiple hundred billion dollar categories, mostly owned by Microsoft and Google, like you mentioned, right? There's been these three technology waves that have happened over the last 15 years, right? One is the cloud, one is mobile, and then this latest one is AI. And the thing with incumbent software is their primary advantage is that people know how to use it, right? So Microsoft doesn't want to blow up how to use PowerPoint. And so we saw this in the shift of the cloud. Even though Microsoft owns the cloud, PowerPoint is not really, and Microsoft Office in general, doesn't feel like cloud-native software.

13:10But does it matter? The reality is, I think that may be true to an extent. Adobe took a while to bring stuff online. There are opportunities. Like I think Canva, for instance, in the creative suite found a really interesting opportunity. It was really specifically originally about like creating Facebook ads, right? Like we're like the tools weren't covered. So there are things that can open up, but that was more like a market segment than opened up. It wasn't like customers like, oh shit, let me try something new. In fact, I'd say especially in enterprise or anything that touches big businesses in any way, shape or form, the defaults are God, right?

13:43The fact that like, you know, Google's just like, here's Sheets, it's fine. Here's what is it? Duh. slides. It's like, you don't even, it doesn't even need a name. It's so like embedded that it doesn't even need a name. Right. And like, I just think it's like, again, can it be done? Yes. But I don't think those things shift because of platform. That's why actually through cloud shift and, you know, through mobile shift, it's still basically Microsoft and Google, right? Like that's why, what it is, is business shifts of a new customer segment for a new reason. There's like that creates opportunity.

14:16If the distribution isn't there yet, you know, The people are stick with what they know. You know, again, like it sounds like Tome, for instance, my understanding, you can correct me. But just since it's an obvious one, that's a potential winner. And like, you know, we both know the people and like it. Like the real win there is not like companies, right? As much as college students doing super quick presentations for things, right? It's like, and that's cool. That's not a bad thing. That's a new segment they're trying to open up. But they're going to win not because of a platform shift. they're going to win because of like effectively finding some vein of presentation making that didn't exist before.

14:50Hey, we'll continue our interview in a moment after a word from our sponsors. Yeah, starting with prosumer creators, and I think eventually they'll get into more work use cases. But that's going to be the question, right? That is the multi billion dollar question is like, is that a beachhead to the money and to all the users, which is the presentation at enterprise or not. And my bet is like if Google was structurally unable to add AI tools or Microsoft was structurally unable to, I'd be with you. Like there'll be some, but they're not. I mean, if anything, they're incredibly well positioned to do exactly the same thing.

15:26It might not be as good, right? It might be actually slightly worse, but it won't matter. Well, I think what they're structurally unable to do is totally blow up how the product works. They haven't actually done it. Why do you think they need to do that? I mean, I think, again, I go back to the Adobe example. It's like, it's so easy for them to add generative fill. And the reality is, is like, just because you have generative fill doesn't mean you don't want the rest of the tools that are in Photoshop, right? Like, it's just a tack on, right? And to me, it's like, sure, you might change the creation route where you start with Word prompts or import a document.

16:01Like, the onboarding might be different, but you still are going to want all the same tools and collaboration features and whatever that exist in the existing platforms. Yeah, well, I think two things need to be true. Like one is like, AI needs to be the dominant feature that matters. And, and like a ground up tool that is AI native needs to be much better. And I think the example is like, when collaboration is the only thing that matters, and the ground up tool is better at collaboration than that, that ground up tool one, right? So Figma$20 billion company, because it was better at collaboration.

16:37Even we're talking about G Suite as equal to Microsoft, but that product didn't exist. And it succeeded because Microsoft left an opening with cloud-native collaborative software. Yes, or because we honestly, Google had such an unbelievable distribution in the enterprise, they were able to easily jam it down to people's throats. You and I know from Meta that just because you have dominant distribution doesn't mean every new product launch works, right? True. Although you could argue that some people are better or worse at doing that strategy. But the I don't know if Google's like the king of I don't know.

17:11I think I think it goes back to what you said. I agree with and I think is the many billion dollar question is like, is AI a paradigm feature dominant feature or is it a piece of the equation? I think it's like obviously an accelerant to lots of things. It will make Zillow's marketplace better. It will make eBay way better. Like it's going to make everything better. But it's not technically it's not only is it technically easy on a relative basis to make the shift to doing a lot of this stuff for these marketplace players but i go a step further which is which is not only is it technically easy it's like it's actually like not even mentally that hard right to think about all these same marketplaces that have the distribution that have the data as a tack on on top of it um and look maybe the front end does change but like i don't think it's gonna matter that much like i think the reality is again like you know everyone gets smarter and richer if you're huge already with data and distribution and again i want to be really clear if you were a micro business this stuff is fucking magic but like you're not a pure ai company you're just like building something that is good or has a clear market with clear users and ai makes a thing that used to be super expensive much cheaper you know yeah and like moving on to marketplaces where i agree with you is if the demand and supply is the exact same and you're really just talking about a ui change then you know the incumbents are gonna are gonna win and AI is irrelevant for startups.

18:33I think what I'm curious about is can AI unlock different data sets and different networks, right? And I think like the example here is with mobile, right? Uber, because everyone had a GPS in their phone, right? That was a technology that unlocked a new network, right? Which is, you know, supply and demand drivers. Yes. Although here's the thing, I think Uber is a great thing to bring up, right? Because if you think about the history of technology and platform shifts, right? You had the PC era generates Apple and Microsoft,$3 trillion plus ridiculously huge companies. All right. And then you have the internet comes along.

19:09Who benefits the internet? Well, you get two new guys, Google, you get Google and Meta, basically, trillion dollar companies. And Amazon, right? Trillion dollar companies, not$3 trillion, trillion dollar companies. And then those two big guys benefit a shitload along with Oracle, right? from the internet. Then you get mobile. And the story of mobile is supposed to be super disruption, all these amazing companies, everything's turned over. That didn't happen at all. Actually, the internet companies won mobile, right? Because it turned out that taking the internet from your desktop computer and putting the internet always on in your pocket, the really big shift, right, was, well, obviously, the real winners, again, are Apple, Microsoft, etc., right?

19:48But then the internet companies got the echo boom of basically all the things they already built applied perfectly to the screen boom crush it uber is the exception where i agree with you like uber doesn't exist pre-mobile but here's the thing uber is not that big a company 90 billion but that's that's my point that's actually think about it three trillion one trillion hundred million by the way and it took a shitload of capital to get there like i'm not it's not like if that's our paradigm of like paradigm shifts the biggest winners you get a uber and by the way lyft is worth nothing, right? And there's not a lot of other good examples of mobile companies that are worth anything.

20:24Instagram is obviously very valuable, but it's valuable because it rode Facebook's distribution and was very quickly acquired by Facebook, right? So will there be some acquires? Sure. But I think what you're going to see with AI is mobile, but on steroids, right? Whereas the big company is going to make so much money, right? It slots perfectly into what they already do. Will someone find some new marketplace opportunity? Sure. In fact, the irony of all this is I'm actually working on kind of a new marketized opportunity that only AI. As I like to say, I'm a fan of hypocrisy. So like, it's not that I don't think there are some opportunities that fit this, but I think as like a cultural thing or where to spend VC dollars or VC energy, right?

21:02I just think we're going to see mobile all over again. Great technology does change the world, but not a good place to buy to build startups. Yeah. I mean, in some sense, you and I, or you were like, I'm actually, I'm just trying to look for the Uber of the AI era, right? What networks can AI unlock that are not the same networks as incumbents, where it's not a UI change? That's fine. And like, someone should do that work. But like, I think you got to keep your hat on that, like, you can't play trillion dollar VC games, multi stage games, believing you're going to get something as big as a Facebook or a Microsoft out of this.

21:36I think people who believe that are generally wrong, might happen, but very unlikely. I think what you kind of get is an Echo Uber, right? Where like, and then maybe it's, you know, a$50 billion company. There's one or two of them out of this wave, right? And then everyone else gets super rich, right? And then the question as an investor is like, aren't you just better off buying more meta stock, right? Like why even bother dicking around with the private markets, right? Like when you can just do that and then go use your VC dollars in places that are more capital efficient and more clearly going to have more winners.

22:05Yeah, but Uber's capital inefficiency is not necessarily true of the next AI marketplace, right? That was a specific market. H100s are pretty expensive. I mean, I agree that Uber was particularly egregious, right? In terms of for a whole bunch of reasons we can get into. But ironically, one of the reasons they were so capital inefficient, Seth, was because everyone thought that this was the thing. so everyone shoveled all of their VC money into fighting with each other over giving annoying San Francisco people free rides. And so the other reason that I'm pretty short AI investing, which has really little to do with the technology, is not only do I don't think the technology stacks up well for VC investment, but then ironically, since all the big boy VCs with big checkbooks are excited about it, not only is it all super overpriced, but everyone's going to shovel money into fighting each other and it's going to drive down returns.

23:00I agree with you on that. Likely at Greylock, most of our best AI investments, like AdNormal and Tome and others, were pre-this most recent. And look, I think we too have companies in our portfolio that I would say dramatically benefit from AI, right? And I'm fine. I think that's great. I'm like, well, we're smarter at AI and therefore. It's like, no, everyone's going to be just about as smart at AI, right? Yeah. Yeah, the example I give is, you know, we backed this seed stage company, making mortgages much more efficient for people, right? It's called Pine. And if someone came to me today and said like, hey, I want to back like an, or I want to start an AI mortgage company, they still need to spend the next two years lining up their capital markets, figuring out their distribution, you know, understanding their underwriting and building a fintech company.

23:47And then they just layer on AI for the last 20 % of efficiency. And so I think that's a good example where like you actually back this series A or B startup in that case, rather than like the seed stage company that's two years behind. Or as a seed is actually where the ingress is. The reality is like, again, I think this goes back. This is less about the company, more about the markets, which is like once everyone knows that you just pop in AI, it's like going to be priced that way. Right. The real risk and the real reward is in the operator and the space and the details of the business. You know, it's not dissimilar to like the whole trend around like scale AI and remote labeling and things like that.

24:23Right. Which is basically just like steampunk AI. Right. Like the same thing. It's just like it was more expensive and kind of the steampunk human version of it is like there's a bunch of businesses that like work better because their cost structure drops because they could use that stuff. But like that wasn't the differentiator. Right. Like that is not the key input. But so again, it's fine to think about AI and think about it as like an input to the overall model. And again, I do think it's cool. Like I actually, ironically, I have to imagine one of the only VCs in Silicon Valley that like is building AI tools and spending thousands of dollars a month with open AI.

24:57And so I have intimate relationship with it. And I think it's super cool what you can do with it. I just don't think it's where are you putting seed dollars? And maybe we agree on that. Well, I actually disagree on that. Yeah. I would play both strategies, right? I would play both the strategy of existing company that's getting better with AI and just totally new markets that can be unlocked with AI. So what are those? Give me some examples of what are the new market that truly everyone's begging for that could be unlocked with AI that wasn't doable before? Yeah, I mean, look, you have character AI, right?

25:34And I think the knock on that is the question on defensibility, right? But you have tens of millions of people using this new interaction model. I would say that's a new consumer behavior. And then the question they need to figure out is defensibility. And that's where I lean on this AI marketplace or AI networks is, can you actually get the data that you're extracting from these conversations and do something with that, either by making the product exponentially better than the next startup that doesn't have that data, or finding some matching algorithm where you actually can benefit from the user base.