The a16z Show - Where Value Will Accrue in AI: Martin Casado & Sarah Wang

Starting point is 00:00:00 Zero-sum thinking has been wrong. That doesn't mean that you can't get in trouble. Every SaaS company under the sun has launched an AI product. They're not just sitting on their hands. And you'd think that they'd have a huge advantage given distribution. But we're just seeing classic innovators dilemma. GPG wrapper was this like derogatory term. I think we'd come as a conclusion.

Starting point is 00:00:20 Like that's not even a thing. When someone writes software on the cloud, you don't call it a cloud wrapper. The success of these companies actually also reflects, obviously the customer love, but I would also add on top of that tangible value that they're bringing their customers. Conflicts really matter in this space. And so if you're too aggressive early

Starting point is 00:00:40 and you don't really think through things, it can really keep you from investing in the one that's winning. Recorded live at our annual LP Summit in Las Vegas, I sat down with general partners Martin Casado and Sarah Wang for a deep dive on the current state of playing AI. We covered where value is occurring across the stack, how this wave compares to past platform shifts, and what it takes to build and invest in enduring companies in an era of exponential acceleration. From the myth of the GPT wrapper to the

Starting point is 00:01:10 rise of AI native apps and the unexpected lessons behind cursor's breakout growth, all with an eye toward where AI goes next. Let's get into it. As a reminder, the content here is for informational purposes only. Should not be taken as legal business, tax, or investment advice, or be used to evaluate any investment or security and is not directed at any investors or potential investors in any A16Z fund. Please note that A16Z and its affiliates may also maintain investments in the companies discussed in this podcast. For more details, including a link to our investments, please see A16Z.com forward slash disclosures.

Starting point is 00:01:52 Martine, Sarah, we just went through the state of the firm. What's the state of play right now in AI? The last two and a half years have really felt like a blur. Maybe just to set the table, given that the AI landscape is changing so quickly, Martina and I thought it would be valuable for our internal team, actually, to reflect and take stock of where values accruing in the AI ecosystem. You know, I think it really distills into a couple of key takeaways, and that's one.

Starting point is 00:02:18 AI companies are growing faster and are larger than even we expected. There's value accruing across every layer of the stack, models, infra apps, All that being said, there's this paradox that we're seeing where more value creation is occurring in a shorter amount of time paired with more wipeout potential happening over a shorter period of time. We'll definitely dive into that dynamic more. And then finally, our conclusion is that you've got to be on the field, but you have to be smarter about where you're taking those bets than ever before

Starting point is 00:02:48 because the stakes are higher. Yeah. And I think as you're listening to this, it's probably worth pointing out that we've come to an opinion that there is no AI. There's like a bunch of subspaces that are totally different that all require their own strategy. So, for example, the language models are very different than the diffusion models.

Starting point is 00:03:06 The apps are very different than the models themselves. The tooling is different than that. And all these subspaces are very different. And so we're starting to learn that this is as big as software and the strategies need to vary as much. Yeah. Let's put that first point you made more into perspective. What's the scale that we're talking about

Starting point is 00:03:22 when we say foundation models are growing faster than expected? Yeah. So two years ago, I would have actually said we were probably the firm, maybe the most bullish, on where the market could go. And I've got to say even we are surprised by how large and fast-growing this market is. And if you look at the revenue of just two of the top-tier frontier labs, not only have they surpassed the early revenue ramps of some of the best SaaS companies in history, they're actually starting to pass the early ramps of some of the hyperscalers. And I think these stats are even more breathtaking when you think about just the time. of when their products launched, I think what we find even more exciting in the space is that

Starting point is 00:04:00 it's not the case that just two companies are growing very quickly in AI. And that shows markets are not only growing faster and much larger than expected, they're also fragmenting. Let's get deeper into some examples. It's obvious Open AI and Anthropic have tremendous growth opportunities ahead of them. Why are we excited for leadership across the stack? There was this view that Open Iowa would win everything or these large models would win everything early on. But if you actually look at the history to now the last three years, it's been the opposite. So if you remember, like, what was the first use case at opening idea? It was code. It was co-pilot, right? But they lost that. And then they were actually the first to image, really,

Starting point is 00:04:34 with Dali. They lost that, right? Mid-Journey came up. They were the first to, like, real video with Sora, and they lost that. And yet they've gotten tremendous amount of value out of text. And so, like Sarah said, I think this is right. The primary takeaways, these markets are larger, and they're going faster than than we expected. And so you result in fragmentation. So things before that we would have said, oh, oh, this is like some sub thing, an opening out will get it, or this is a minor market, or whatever, ends up turning to be large enough to multiple companies

Starting point is 00:04:59 with tremendous growth and tremendous value, right? And so we think the only crime, this is going to be caveated later on, is zero-sum thinking. Like, anybody that decried out of defensibility isn't going to work has been wrong. Anybody that's decried like it's all going to aggregate has been wrong. So zero-sum thinking has been wrong. That doesn't mean that you can't get in trouble.

Starting point is 00:05:17 And so we'll talk about that. It's funny because there was all this talk a year ago about GPT rappers. Every app was a GPZ rapper. where, you know, foundation models going to win everything. What changed? And why aren't the foundation models winning everything? I think this is a fun one for a lot of the investors in the room because we don't all invest in just foundation models. And the fact that we're seeing this gangbusters growth in AI apps is really exciting to see. So I think two parts to your question. One, it makes sense that a lot of the focus in investment early on in this cycle was

Starting point is 00:05:44 on the infrastructure side. And now what we're seeing is apps are benefiting from that massive investment as intelligence effectively has become free. And you have this aligning of the stars where the fierce competition on the state of the art model market is driving both nonstop continued capability improvement paired with continuous price decreases. And in fact, I think model inference costs have gone down 10x year every year. So the stars, as we say, are aligning on that front. And then on your question of why don't the foundation models just take over everything? I think this is a question. I mean, it's a very reasonable question. It's one that we sort of ask. ourselves for every new investment that we make. And we certainly saw this with the early

Starting point is 00:06:24 marketing copy AI apps. But I think in talking to the founders of these app companies themselves and then also the model providers, the answer to this question has become increasingly consistent. And that is where you have complex workflows and a ton of customer data where deep integrations actually are necessary to get that last mile value for the customer. This is where the specialized AI apps are sort of crushing any either foundation model layer or otherwise company in the market. We have actually a couple of examples that we'll cover on this. But to see that has been very interesting given the landscape initially was going in a different direction. I got to say, like, GPG wrapper was this like derogatory term. I think we'd come as a

Starting point is 00:07:07 conclusion like that's not even a thing. When someone writes software on the cloud, you don't call it a cloud wrapper. Like you have all of the complexity in the software that's on top of these models, there's tremendous amount of opportunity to add value with traditional software, but also by building your own models. And so, like, we kind of view that like, you know, listen, this is an evolution in software. These models are an evolution in infrastructure and there's tremendous opportunity to add value above the stack. Someone tweeted that venture capital is just a wrapper around LP Capital. Yeah, to speak to the point. How should we think about AI native companies relative traditional SaaS companies? This is an interesting one. And the first thing I call out

Starting point is 00:07:44 that sort of just jumps off the page is that the AI-native companies are far outpacing their SaaS counterparts. And you can see it in terms of new companies blowing past this golden metric of time to 100 million of error, which is truly an incredible feat to accomplish on the left. But what's amazing is that it's not just the best companies that are doing really well. It's actually on average. You can see this from the aggregated stripe data where AI-native companies are growing faster and then sort of the SaaS 2.0 generation, if you call it. And we'd sort of attribute this to a number of factors. One is just looking at the compelling ROI out of the box.

Starting point is 00:08:21 And with AI improvements and capabilities, you're seeing this 10x plus improvement in the customer experience as well. Whereas SaaS 2.0, you generally saw a little bit more of an incremental improvement, you know, call it 25, 50%. And then the second piece is it's early days for this, and it's related, but you're starting to see replacement of some of the services budgets versus just soft. And the second thing that I'd say on top of just the fact that the growth itself is happening is that the relative growth is particularly interesting when you take into account that every SaaS company under the sun has launched an AI product. They're not just sitting on their hands. And you'd think that they'd have a huge advantage given distribution. But we're just seeing classic innovators dilemma starting to play out already. And the fact that they have revenue generating products where they have to devote time and resources changes the game for them. A.I companies aren't building an AI product. They're just building a product. A lot of them are newer, and so they don't

Starting point is 00:09:17 have this 2021 imposed remote culture. Most of the founders that we work with are in the office, six to seven days a week. And then finally, a lot of them tend to be these essentially applied AI engineers where they're just incredible at ringing out every last drop of value from the LLM in a way that actually translates to customer value. It sort of circles back to the compelling ROI piece. And I think the results speak for themselves. Something else we debated which inspired this conversation was defensibility. How should we think about defenseability for this company? Are they defensible? Where does the defensibility come from?

Starting point is 00:09:48 Does it come from state? He's come from contacts. It's come from brands. Some hybrid. Our team, when you take that? The actual data on this stuff is really noisy because everything's doing well. So it's kind of hard to have a theory. But if you actually kind of dig into it and you watch this thing for three years, something seems to be pretty clear. And that is a really hard thing about building any startup or software is the bootstrap problem. Like, how do you get like the first 100, 200 customers?

Starting point is 00:10:08 And like AI actually solves that problem. It just solves the bootstrap problem. It's like these models are so magical. You wrap one of these models, you know, you make it available, and people think it's amazing they show up. But what's also clear is that doesn't solve your retention problem if you're a software company. It solves a very hard problem, but doesn't solve another problem.

Starting point is 00:10:23 You know, and arguably, there's actually a lot of perverse economies of scale that are actually in play with these AI companies because, like, the models that commoditize very quickly, anybody can kind of use them, et cetera. And so what we found out is the pattern that seems to work is, you know, a startup will come and it'll do a model and we'll get a bunch of users on that metal, that'll be great,

Starting point is 00:10:41 but then they have to kind of revert to traditional software to build traditional modes, right? And so these modes can be anything. You have two-sided marketplace, it can be long-tail integration mode. It could be a workflow mode. Whatever it is that we've figured out how to build in the past,

Starting point is 00:10:52 they end up having to do. But again, I mean, one last thing to notice is like some of these companies are growing so fast and the space is so new. We're even seeing effects that we haven't seen in a long time, like brand effects, right?

Starting point is 00:11:01 Like, these companies are entering these massive vacuums, and then we all know them. And even though you're like, the competition is just as good. right? You know, like how much better is Open AI than Anthropic? I don't know, but everybody seems to use it because of the brand. Like, how much better is cursor than the competition? It's a lot better. But like, everybody knows the name, right? And so, like, when you know the name, you do this. And the early internet was like this. Everybody knew Google and everybody knew Amazon and you had these big brand modes. We're starting to see that come back again in this space. But as far as I can tell, as far as we can tell, there is no inherent endemic mode in the technology stack to AI other than just overcoming the bootstrap problem. Let's double click on cursor a bit. What explains it's. astronomical success. This is just crazy, right? And so you could ask this question, like,

Starting point is 00:11:42 you know, is it they've figured out cold fusion. And like actually, some of the answers are actually pretty banal. It's actually the first kind of monetized AI app was code, right? It was copilot. And so Microsoft had invested a ton of money and matured the market with copilot, with DS code. And so everybody knew it. It's just like the models weren't quite ready then. And so you had, you know, probably say 400 to 600 million in ARR of, you know, users out there in user behavior. So when cursor came out, two things happened. A, they basically followed the same behavior that you saw in Viscode, which is like, you know, this code editor is one.

Starting point is 00:12:14 And the second thing is you had the RL wave where you have these models with using RL. So coach just got way, way, way better, right? And so it's a phenomenal team, very product focused. They caught the model wave and there's existing user behavior. And like the rest is kind of history. And like I mentioned just previously this notion of brand. I mean, they just entered the Zite guys. I can't tell you how often we'll have a founder show up.

Starting point is 00:12:36 and they're like, we're the cursor for X. It's hard to articulate the actual user love for these products, which goes a long way. I mean, we really haven't seen it, in my opinion, since the Internet. Absolutely. The success of these companies actually also reflects, obviously, the customer love, but I would also add on top of that tangible value

Starting point is 00:12:53 that they're bringing their customers. And I think one thing that's changed over the last 12 months is this shift from just, I need AI, like, you know, experimental vibes buying, I'll call it. You have vibe coding, you have vibe enterprise AI purchase. but to tangible ROI focus. And so two examples of that. On the cursor side, we host these annual portfolio CTO dinners and some of the questions that we'll throw out, of course, in the last few years, AI has come up for a significant portion of the dinner. And last year,

Starting point is 00:13:23 it was notable that when we asked, hey, CTOs across 24 portfolio companies, how much is AI actually impacting your productivity? And the answer across the board was pretty much 10 to 15 percent. We're all using GitHub co-pilot. And the implication was that there's a lot of hype, not a lot of results. This year, I was pretty blown away by the answers that we got. They spanned from, call it, 30 to 50% on the low end in terms of productivity gains to, I kid you not. One CTO told us that he had seen a 10x productivity lift from himself and his team. They were all using Cursor.

Starting point is 00:13:57 I think 24 out of 24 portfolio companies were using Cursor. And that 90% of the code in their company was AI generated. this is in a short 12 months, maybe not even 12 months. And so you really are seeing this bump in hardcore ROI. I think customer support is another use case that has been hyped up, if you will, in terms of, hey, this could really have an impact on the industry. But the early results were, let's call it mixed. If you talk to a Decagon customer, they're actually slashing their customer support costs by up to 80%.

Starting point is 00:14:26 And not only that, they're seeing deflection rates go up from 30% to anywhere from 60 to 80%. and their CSATs, their customer satisfaction scores, are doubling. So this is, like I said, tangible ROI, and that's what's really driving a lot of this growth. And honestly, it's the underlying productivity and impact gain that gets us really fired up. Let's go deeper on the customer segmentation part. What are the ramifications of the fact that a lot of the growth

Starting point is 00:14:53 is being driven by the presuming level? So like I mentioned before, AI really helps overcome the bootstrap problem. It doesn't have the retention problem. So it's kind of a separate thing that we look. at. And it just turns out these companies, they look like these prosumer companies that we've been looking at for quite a long time. And they all have different profiles. So I just think we should remember that every time we have a super cycle, it tends to start in these prosumer arrays, right? The internet did this, right? I remember when Sun outlawed the browser, right? This is

Starting point is 00:15:19 Sun Microsystems, right? But they didn't really know how to consume it. So the enterprise doesn't know how to consume these new technologies, but there's clearly a lot of value. And so, you know, the individuals pick them up and they use it. And we're seeing a lot of the new behavior. And what's been very interesting is that has already led into enterprise pipeline like we've never seen. So the fact that these are prosumer businesses, like very specifically to your point, the fact that these are prosumer businesses is not in some way because that's where they always sell to. It's just a natural maturation of the cycle. And if anything, it looks far more promising than it did the end of that time. To build on Martin's point a little bit more, I think the high

Starting point is 00:15:51 amount of pro-sumer revenue does mean that we're paying attention. I mean, we always pay attention to retention, but we're paying attention to it more than ever before. And a lot of these high-growth apps are not your typical system of record, 95% gross dollar retention companies that we sort of saw in the 2010s. But importantly, that doesn't mean you should throw the baby out with the bathwater. It's not like, hey, these have terrible retention. And so these are terrible companies, right? And then I think the other piece is for the companies with questionable retention, that's something that we're being cautious about because the valuations that are being demanded in this market

Starting point is 00:16:26 do require some sort of customer stickiness and base. to build upon, or at least the ability to show that the top of funnel is actually converting into enterprise revenue, as Martine mentioned. And so this is an area that I think requires a lot of nuance and is one, frankly, that the team is spending a lot of time on. So we've talked about the big winners. There are also some big wipeouts, as we mentioned. What have we learned about the commonalities between ones that win and ones that don't? So I think what's interesting about this funding cycle in particular. I know history repeats itself sometimes, but in this case, and the foundation model layer side,

Starting point is 00:17:01 I think what's remarkable is just the massive size of the rounds that folks are raising before any traction. And we're participating in some of these rounds. We'll talk more about what our thesis is when we do. But as we all know, the more money you raise early on, the more pressure you have to really show performance. DG likes to call this transition going from a tell-the-story company to a show-not-tell company.

Starting point is 00:17:24 So you really need to show-not-tell when you've raised hundreds of millions of dollars. So there's a couple of themes that we would flag. The key ones are passing on good, but not exceptional teams, has generally paid off. And then the second one I'd highlight is that researcheritis, as Martin and I call it, is a real thing. We've seen this up close and personal. It's particularly important given a lot of incumbents and Chinese companies are pouring money into these spaces, so it's really not for the faint of heart. And then finally, I think the broader point that we've made,

Starting point is 00:17:57 make here is that this is not a market where a rising tide lifts all boats. Picking actually matters more than ever. Also just important to realize, unlike even crypto in the early days, conflicts really matter in this space. And so if you're too aggressive early and you don't really think through things, it can really keep you from investing in the one that's winning. Martine, why don't you talk about what's happening in China affects the market? Yeah, so let me just be very quick about that. It's kind of a mixed blessing, right? So on one hand, they build these great open source models. They're not hindered by copyright. They get very cheap access to data.

Starting point is 00:18:27 But on the other side, like historically, China's just not been able to build software, at least for like the prosumer enterprise market, which is my world. They've just never really been able to do that. And so I think that their ability to compete at like a software level is pretty limited. At a model level, they're actually quite good,

Starting point is 00:18:43 but, you know, we benefit a lot from that. And of course, in the consumer space, like with TikTok, they've historically been very good. And so I think that's TBD. But for us, you know, for me, I think it's a mixed blessing, but more of a blessing than not. I mean, I think it's actually great to have the competition. and it's great to have these models out there.

Starting point is 00:18:57 Let's transition to our thesis. We talked about what to avoid in terms of pitfalls. Let's talk about what we're looking for, starting on the foundation model side. Yeah, so we split the foundation models into kind of two, and I'll just talk about the sort of models because everybody does. So the state-of-the-art model market,

Starting point is 00:19:09 this is like the Anthropics and the Open AIs, it's incredibly competitive, and it's very heavily subsidized that being like with meta and like Google, et cetera. And so kind of our view is, you have to be very, very careful before you go into it. And there's a lot of companies you've never heard about that are in this space that we avoid.

Starting point is 00:19:24 So our view is, you really want to back the primary names that have done it before that are able to raise capital and put together in the best teams, right? So you know that we invested in I mean, the guy's Oppenheimer. He's been close to every major advancement in the last 15 years in AI. Our view when it comes to these stated-art models is really just like the premium teams that can get the capital. Yeah, exactly.

Starting point is 00:19:47 And I won't go into our thesis on the other categories one by one, but I think the common theme here, and you'll hear it over and over again during this, is that the goal, the thesis, is to bet on market leaders with demonstrated momentum and are led by founders that are visionary and how they're applying AI to their verticals. Let's close with one or two spicy takes.

Starting point is 00:20:06 What's something that other firms think is real that we don't, or vice versa? I think just at a high level, I'd say, it seems like a lot of firms are failing on either side of this. Like some of them are like, it's not real. We're not going to invest. And it's amazing, like, some firms that were very, very relevant, we just never see anymore.

Starting point is 00:20:22 I mean, like, the founders don't talk about that. they're not there, they're not in the deal. For the 10 years, I've been doing this, the most remarkable transition. And then there's others that got so excited, so early, and did all the deals. And like I mentioned, the guy got conflicted out,

Starting point is 00:20:34 like they're dealing with a lot of companies that aren't working. And so, I mean, we think you have to be very thoughtful and realize that this is its own space that need to have the same level of sophistication for any software. In closing, what are the key messages we want to leave this audience with?

Starting point is 00:20:48 You've already heard these messages over and over again, but they really are the key ones that we want to impart from our work and the way that we think about investing. And that is the market is growing faster and it's larger than anyone anticipated. You've got to be really thoughtful about where you're placing the bets.

Starting point is 00:21:02 The stakes are higher than ever. And this is a market where heat cannot be confused with momentum. And then finally, as Martine said, you're on the field or you're irrelevant. We're incredibly bullish on the opportunity ahead and we think this is just the beginning. Thank you very much. Thank you, Martin.

Starting point is 00:21:20 Thank you, thank you, thank you. Thanks for listening to the ACE. 16Z podcast. If you enjoy the episode, let us know by leaving a review at rate thispodcast.com slash a16Z. We've got more great conversations coming your way. See you next time.

The a16z Show - Where Value Will Accrue in AI: Martin Casado & Sarah Wang

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.