Invest Like the Best with Patrick O'Shaughnessy
· Patrick O'Shaughnessy, Henry Ellenbogen
[00:00:00] S0: Doesn't matter because you're too far ahead. [00:00:00] S1: Doesn't matter. And as we all have come to understand with Amazon, they took that 3-5% cost advantage of getting that box to you. and their ability to put more than one item in the box, and they used that economic advantage to then go build fulfillment centers that are physical to reinvest into capital and infrastructure that allowed them to go down that 3% to 5% cost curve for 20 years. As I said earlier in the show, they woke up and the only people who could play their game, when eventually everyone realized what they were doing, were the people who still had the scale and the customer relationships and the trust of Walmart and Costco. And then eventually when they figured out, all three of them were great. Now the problem is the rest of retail was not so good. That's what we think about here. [00:00:00] S0: So if I say it back to you, we've seen through Mitch and others like him, this 40 year benefit of Kaizen brought to physical product world, and that AI represents a sort of kick-off of Kaizen to human work world and that that is going to have lots of stories like the Amazon story you just said, where someone gets on one of these curves early and they can't be caught. And so you're trying, I'm sure, to find who those people might be. [00:00:00] S1: We're trying to do two things. We're trying to find the people who that might be, and then we're trying to make sure we don't… [00:00:00] S0: Get killed by them.
· Patrick O'Shaughnessy, Henry Ellenbogen
[00:00:00] S0: Get killed by them. [00:00:00] S1: Mistakenly own a company, that based on last generation competitive risk was a great company and correctly, we thought was competitively advantaged, and had a good operating culture, was led by high quality people who thought like owners. But because discontinuous change changed the world, they weren't able to adapt. Now ideally what we go do is we go find those already advantaged companies and then they leverage this to go from good to great. [00:00:00] S0: If you think about all the people that have navigated change as CEOs the best that you've worked with. I'm an investor with Dave Duffield in his latest company and so he comes to mind 'cause now he's tackling AI and the guy is incredible. If you think about whether it's Mitch or Jeff or Dave or people like this that you've seen operate, what methods have impressed you the most, of how they themselves adapt first, their own mentality and then their teams, to these fast-changing circumstances. And this is a question for everyone out there that's running businesses, that is facing this same change that's a risk and an opportunity at the same time. [00:00:00] S1: Let's go pick on Luis. We talked about Luis earlier. Duolingo has a lot of opportunity with AI and a lot of risk. And the stock, depending on the day, reflects it. When OpenAI demos, how you can use OpenAI to basically do translation. A lot of times the stock goes down, or when Apple shows you how AirPods can be used to, in the physical world… [00:00:00] S0: Live translation.
· Henry Ellenbogen
[00:00:00] S1: Yeah, live translation. Stock goes down. And actually I don't think the market's wrong. Now, it's probably wrong on the magnitude, but I think what the market is saying is there's a risk, there's a risk here. So that's probably in our portfolio, one of the higher risk names. But what do we look for when Luis or Max or Dave Duffield, I think the first thing we look about is a business that already is operating well. Because if you're not operating well when you have to deal with change, you're not gonna be able to do two things at once. You can't do a turnaround and do well. I think the second thing we would say is a business that already was winning in its first end market. All the definitions we would have about winning. Substantially gaining market share, driving real economic profit that allows it to reinvest in this next S-curve. And also the other thing we really care about is people, we really stress resiliency at Durable. It's one of the reasons we like so much investing in Act 2 managers, we can go study their resiliency. When you're an investor in Max and you saw how resilient he was in Slide, and now you understand how he's gotta be resilient to implement AI, to lean out his cost and drive his revenue before his competition does. You're like, we've seen him under stress before. I think you're looking for people who have a perspective, can execute but also are humble. Because we at Durable write down our views on AI every six months and we update them. And as we've gone more into this period of change, probably we're less wrong. Our perspective is more informed you would say, than it was, as we change it every six months. But we're learning, we have the benefit of having a job that allows us to spend our time reading and thinking and talking to smart people. So even if these are really talented people, most likely they don't have as much time to go do it as we do. And we have to be very humble in our approach. So they have to have a perspective, figure out how to go do this, not be paralyzed, but they also have to be humble and constantly learning. And you know, the last thing I think practically we have to think about as investors goes back to the memo.
· Henry Ellenbogen
If you have a high risk situation, which we would all agree, Duolingo is and Affirm is, and you're really close to that part of the change, then you need to be compensated for the risk. And so what we would tell people on Duolingo is yeah, like because of this risk, the discount rate has gone up but probably commensurate, the opportunity has also gone up. We can articulate it. Luis has talked about being 20x faster in generating content. We've already seen it. He's publicly said—he developed chess, which is doing incredibly well and my kids really love that product, I mean they're addicted to it—first it was two people for six months, and then he added another four people and he developed a product in nine months. That's the best product he's ever done. And if you ever talked to him, how long would it have taken in the past? He's like, I don't know, I probably need 4-6x as many people and it would've taken them four times as long. And so you could say a startup could do that, but he's doing it. He's doing it out of his platform. That product is well over a million DAUs and is growing at astronomical rates and it's a great underserved end market. It could be a huge business. And so yes, the discount rate has gone up, but this company was purpose built for AI. You actually have a person who studied AI and taught it at Carnegie Mellon and has an organization of A players who are agile in his company, and is humble and constantly learning. A proof point on how it's making him faster than other people. It's driving real value, and obviously that means the probability of it going from a point solution or a couple products to a suite is higher. And so that's what we look for.
· Patrick O'Shaughnessy, Henry Ellenbogen
[00:00:00] S0: Can I stack a couple of the things that you've said that interest me the most into a bigger question. If I take physical Kaizen and digital Kaizen, just to shrink the concepts down. If those were to have a kid, it might be robotics. And so I'm really curious how you approach the potential, with an unknown timeline, that we might get a second wave of—the physical labor economy is way bigger than the digital labor economy—that we may get a second application of Kaizen over the next 40 years. The first one we saw from Mitch and company. How do you think about that opportunity and potential change? [00:00:00] S1: It's something we have really thought hard about. Here's what we have tried to do. First of all, we have tried to get smart by meeting with the entrepreneurs and talking to the companies that we're involved in that we know that are leading this area, just to try to learn. We only recently did this with robotics. So if we started writing down our conclusions and then being humble that it could change every six months in AI, in what you would say is more data businesses or digital businesses. We only started doing this literally in the last month where we documented it for the first time. I'm gonna do something that I don't love to do, which I know our views here are very early and probably deeply wrong, but I'll give you our initial conclusions. Which is in certain use cases, it's pretty clear that already the cost is lower than the equivalent analog process or… [00:00:00] S0: Physical labor process.
· Henry Ellenbogen
[00:00:00] S1: Physical labor process. And yet as we all know, this is the earliest and worst robotics is gonna be. And because machines are iterating with machines and it's being powered by general purpose models, not by specific purpose models, this is riding a curve that is definitely geometric. And so, back to this mental model of Amazon that drives so much of what we think about durability. What Amazon was able to do is ride a cost curve where they were deflating the cost of sending a box out at 3-5% a year for 20 straight years. And the people who did not leverage the right distribution infrastructure, the right investment in robotics at that time—they bought Kiva. The right ML models when they came into play, about how to basically plan inventory and deal with suppliers, actually were probably at a curve that probably inflated at three to 5%, but at best were flat. And that differential in a low margin business, you compound it over five years. Honestly, that's all you need to know. And that's I think what we're starting to get our heads around at Durable, which is in many areas robotics is at parity. Now there's a lot of data that says it's lower cost in certain cases. And the use cases are about to go up, and the cost on the existing use cases probably don't go down at 3-5% at this point in the curve because of the scale brought in, the human capital brought in, the IP bought on. And the fact that you're gonna be able to use these general purpose LLMs to power it, it probably goes down at more like 15 to 20%, but maybe it's Moore’s law and it goes down even faster. Well then we wake up in five years and the people who put themselves on one cost curve, if they compete against the people who put themselves on the other cost curve, those could be power law businesses. What we have thought really hard about is who's gonna go benefit from this curve, where their competition—even if they woke up tomorrow, even if they put the same amount of money into this problem, even if they could hire the same quality of people, which is unlikely. Because they have not invested in the distribution infrastructure or the technological infrastructure to compete on this curve, at minimum, is probably two to three years behind and every day that they wait, they're probably getting further behind.