--:-- Dwarkesh Patel: Today, I have the pleasure of speaking with Agustin Lebron, who is the author of *The Laws of Trading: A Trader's Guide to Better Decision Making for Everyone*. This is one of those books—Tyler Cowen calls these "quake books" that completely shift the models you have of the world. I really enjoyed reading this book. I'll let you describe your background, Agustin, but before that, let me ask this question. Peter Thiel says the Straussian reading of *Zero to One* is that you shouldn't start a startup. I think the Straussian reading of *The Laws of Trading* is that you shouldn't trade. You probably don't have edge because you're not better than a marginal trader. If you think you have edge, it's probably because you haven't factored in risks and other costs. So don't trade. Is that what I should take away from this book?
--:-- Agustin Lebron: I think you pretty much hit the nail on the head. A lot of the time, when people start thinking about trading seriously, they realize more and more how hard a job it really is to do well. The answer is probably: if you're smart enough, good enough, and hardworking enough to make a go at it and make a living in financial markets, there's probably an easier way to make money and have a satisfying life most of the time.
--:-- Dwarkesh Patel: Okay. Do you want to talk about your background, what you've been working on in the past, and what you're working on now?
--:-- Agustin Lebron: My background is engineering. That's what I did in university, and I did it professionally for about six years as a chip designer. At the time, I was playing a lot of online poker back when that was profitable and arguably legal. Engineering was getting kind of boring, and I wanted to do something else. I thought, what's halfway between engineering and poker? Of course, that's quant trading. So in January 2008, I walked into my boss's office and said, "I want to quit." He said, "Where are you going?" I said, "I'm going into finance." He said, "Are you sure this is a good time to be doing that?" I said, "Yep, I'm dead set on it." A few months later, I managed to get a job at Jane Street and rode out the implosion of Western civilization from the seat of a trading desk. I did that for a few years, then left Jane Street a few years ago and started my own consulting company, helping tech companies with growth, management, and hiring. In the last few months, I started a new company in the crypto space.
--:-- Dwarkesh Patel: How much are you willing to give up your edge by telling us what this is? If you're not willing to talk about it, that's okay.
--:-- Agustin Lebron: Big picture, we're building a crypto protocol that is kind of new and has some pretty cool cryptographic guarantees against things people don't like when they trade in crypto.
--:-- Dwarkesh Patel: Let's get into some of the topics in the book. First, I want to talk about adverse selection, because this was the most interesting part for me. If we think of hiring workers as placing bids on them—multiple employers can place bids—doesn't winner's curse imply that the average worker is probably overpaid? The true value of the employee isn't the highest bid, but the average bid they would get on the market.
--:-- Agustin Lebron: From the employer side, it's definitely adverse selection all around. If you're just posting a job ad, the applicants that apply are selected against that pool. The people who are really good, their employers know they're good and are incentivized to keep them. So the people on the market are probably at the margin not as good. Not only that, but the person who has the final say on whether this happens is the employee. You get adversely selected there because the really good people have lots of job offers and pick from many, while the not-so-good people pick from few. Employers systematically get adverse selected that way. Whether that means they're systematically overpaid is a different question, because companies have a pretty good idea of the marginal value of an additional employee. People by and large give up things to get the security of working at a company, so maybe that counteracts the adverse selection in terms of pay. It's not clear which way it washes out.
--:-- Dwarkesh Patel: Bernd Hobart recently wrote a blog post about this chapter on adverse selection. In a footnote, almost in passing, he said there should be more adverse selection in industries like finance, where the motivation for people to work is money. If a worker wants to work for SpaceX, there's a story you can tell about why they're working for you and nobody else. In finance, there are a lot of people bidding for really talented people. So if they're working for you, there's something suspicious about that.
--:-- Agustin Lebron: I think there's something to that. Doing a lot of the hiring I used to do, one of the biggest red flags is when somebody comes to you and says, "I've been wanting to be a trader my whole life." They don't know what trading is. They haven't known what trading is their whole life. They don't know what the job really involves—it's not tangible like being a doctor. What they're really telling you is, "I've been wanting to make a lot of money my whole life." In some jobs, that's a good motivation, but it's not necessarily the motivation you're 100% looking for out of the gate.
--:-- Dwarkesh Patel: In your chapter on motivation, it seemed like you were implying that making money *is* the motivation you should be looking for. If their motivation is emotional, they're going to lose to people whose motivation is to make money. Can you talk more about what motivation you are looking for?
--:-- Agustin Lebron: The motivation of winning the game—making money is how we determine who wins the game. That part of the making-money motivation makes a lot of sense for a trader. But the "all I want to do is make the most money possible" is correlated to things that aren't so great. A lot of the job is having an inherent curiosity about random things. If your whole motivation is "where can I make the most money today," it's not necessarily optimal over the long haul. You need to balance that against other things, like enjoying the game for its own sake, as an exploratory kind of thing. Maybe that's a little inconsistent with something I wrote in the book, but at the margin, people need to hear the other thing more.
--:-- Dwarkesh Patel: How do you figure out if somebody enjoys the game for its own sake? You said in another interview that a company like Jane Street would hold it against you if you have retail trading experience. If that's not how you judge whether they would intrinsically enjoy the job, how would you judge that?
--:-- Agustin Lebron: One thing I've always said—I would love to talk to the person who is the third-best player in the world at some weird, obscure chess variant. That is probably very correlated with things I care about, such as a willingness to really grind and try to get really good at something, not because there's a huge pot of gold at the end of the rainbow, but because you find inherent enjoyment in getting really good at something. Aside from the mathematical and risk-taking parts, certainly a strong desire to be in a competitive environment and to enjoy being in that environment—that can take many forms, but it's a big part of it.
--:-- Dwarkesh Patel: Why is having domain expertise in trading not important? In other industries, the more experience you have, the better. You guys are often hiring people who are analytically smart but haven't been traders before. How do you manage that, and why is that important?
--:-- Agustin Lebron: The concept of a domain is probably a lot narrower than people understand. If I'm sitting on my Robinhood account punting stocks back and forth, that is not the same domain as what a trader at a market maker or a top trading firm would do. To the extent that you think that's the same domain, that is a thing you have to unlearn when you come work at a real company. It's kind of a problem—it's a thing you have in the back of your mind. You'd rather take a really smart, motivated blank slate and teach them what they need to know, than undo something and then teach them. At a meta level, the person doing trading in their personal account probably isn't doing positive edge trades. They're probably losing money on average. You'd like the person to realize that maybe this is not a winning game for them. If they can't realize they're playing a losing game, that's probably not great.
--:-- Dwarkesh Patel: You said in the book it takes six to 18 months to train a trader to be net positive. What is happening in that time? What skills are you teaching them?
--:-- Agustin Lebron: This varies from company to company and has varied over the history of Jane Street. When I started, it was very much the Socratic method. You sit next to a senior trader, and their job is to teach you everything they know—a continuous stream of questions, answers, conversations. Jane Street has improved on that. There's now a boot camp where you intensively learn the fundamentals of everything the firm feels you need to know as a trader. That accelerates the process. But it's very much putting people in situations to experience the decision-making process and iterating on it. "What are you thinking about here? What do you think about that? Did you think about that? What would you do in this situation? Why, why not?" That just takes time.
--:-- Dwarkesh Patel: You've done a lot of hiring for tech companies. How applicable is this model to the tech industry? Could a company like Google have a very effective boot camp where they get people who studied physics or math at MIT—maybe not computer science—and make them 10x in a short amount of time? Or is that something special about finance and trading?
--:-- Agustin Lebron: I don't think so. The most common failure mode I see in tech company hiring is hiring for skills instead of abilities and potential. It's just because skills are very legible. It's straightforward to spend an hour with somebody and understand whether they can write code in Python. It's like the drunk looking for the keys near the lamppost—you evaluate what's easy to evaluate. My dream, something I can't really work on right now but who knows someday, is doing mass screening for people around the world. I'd love to find the smartest 0.1% of high school graduates around the world—India, Nigeria, all these countries being massively underserved by their educational system and opportunities. Put them in boot camp situations for six months where they learn useful skills, and at the end, there's a six-figure job with a Western company. There's no reason companies like Infosys should be taking the lion's share of that arbitrage opportunity. There's an incredible need in the world for people who are smart and motivated, and an incredible supply we're systematically undertapping. My answer to your question is yes. I strongly believe there is a trillion-dollar business—or maybe a nonprofit—in closing this arbitrage gap.
--:-- Dwarkesh Patel: Your former colleague Sam Bankman-Fried, CEO of FTX, started a big charity called the Future Fund. One of their project ideas is exactly what you're talking about—large gains from enabling talent in the developing world. What would you look for when scouting out this talent?
--:-- Agustin Lebron: One thing that isn't terribly polite to talk about but is critical is just intelligence. It strongly predicts outcomes across jobs and industries. That's certainly some element of it. But also, in an ideal world, you would build the selection process kind of like a game—maybe a mobile game—where people are incentivized to keep trying at stuff, a little bit of a grind. You're selecting for that hardworkingness, stick-to-itiveness. Some combination of those two things is almost definitely predictive of actual value.
--:-- Dwarkesh Patel: Have you heard of Pioneer, started by Daniel Gross?
--:-- Agustin Lebron: Yes, I've heard of it. I don't know much about it.
--:-- Dwarkesh Patel: This sounds a lot like it. I think they're trying to make building a startup like a video game, with associated risks and rewards.
--:-- Agustin Lebron: [Nods in agreement]
--:-- Dwarkesh Patel: How do you deal with adverse selection in cases where it should theoretically work for you, but the counterparty prices in the possibility of getting a lemon? For example, I'm 21 years old and male. Car insurance premiums for me are huge, even if I'm a good driver, because of the adverse selection the insurance company faces. Or if there's a great employee, they might be underpaid because the company hiring them doesn't know how good they are before they're hired. How do you deal with such scenarios when you're on the other side of the adverse selection?
--:-- Agustin Lebron: In the car insurance situation, there are now car insurances that put an accelerometer and GPS on your car and monitor how safely you drive. You can decrease your adverse selection by taking advantage of those things. In the employment case, that's tougher. The most important thing you can do is select your coworkers as a potential employee. Getting really good at evaluating your interviewers is an undervalued skill—not so much to tell if they're good or not, but to tell if they're a good fit for you. The best signal of whether a company is a good fit is who the people are interviewing you and what they ask you to do. If a company is sensible, what they ask you to do in the interview is highly correlated to what you do in the job. Don't adverse select yourself by being "meh, I think this will probably work out," or more importantly, by letting the status of a company override your personal understanding of the experience. That happens very frequently. Once you get past that, you're probably in good shape. It comes down to putting yourself in the right positions, which is a skill you learn over time.
--:-- Dwarkesh Patel: My friends who are programmers complain that when they interview, they get asked questions very unlike their actual jobs—almost brain teasers. There's a Chesterton's Fence argument: if all the tech companies are doing it, there must be an important reason. Have you figured out why brain teasers are so common? Is it just that IQ is so important that this is the best way to measure it?
--:-- Agustin Lebron: The dirty secret is that explicitly testing for IQ is illegal in the United States as an employment practice. However, you can drive a truck through it. Companies like Wonderlic—people might have heard of it because it's the test given to quarterbacks in the NFL—build employment testing that is essentially IQ testing, but with a fig leaf or legitimate justification that as long as you can show it's important for job performance, you can do the testing. I feel like a lot of these brain teaser questions are IQ tests disguised. Oftentimes, they are badly misapplied by interviewers. It takes a lot of really hard training and experience to ask these questions in a way that gets you the signal you want. The extent to which you view your job as vocational is the extent to which you'll hate those brain teasers. If I'm a programmer and I want my job to be just writing code all day, you won't like brain teasers because you don't think of them as part of your job. If you think of your job as more expansive—"I'm here to think about hard problems and I happen to implement them in code"—then you might think of brain teasers as more correlated to what you want to be doing. Select for what you like.
--:-- Dwarkesh Patel: Maybe it makes sense to select for the latter type of person.
--:-- Agustin Lebron: There's a lot of schizophrenia in tech hiring. Everybody says they want to hire A-players, but only a small fraction can hire those high-percentile people. A lot of startups have the failure mode where they try to build incredibly selective processes, but the people they really want are never going to accept their offers—they'll go somewhere higher status or higher paying. So you try to select for an 80th percentile person, but you end up selecting a set of 50th percentile people who look like 80th percentile people, which is really bad. What you should actually do as a startup is be clear-eyed: if you have a team of 10, you probably need one or two 90th percentile people. Evaluate for and pay for that. Then the rest, you should try to hire 40th percentile people and put them in situations where they can be effective. That's a much more cost-effective and stable way to build a company, but nobody wants to hear that or build a company like that.
--:-- Dwarkesh Patel: That's a great example of a barbell strategy. Do you have any ideas of what good arbitrage opportunities in tech hiring might be? I know SpaceX's early engineers were from the gaming industry because they're used to optimization problems, but it's not necessarily a high-status career. Are there good places you would look for talented potential future programmers if you couldn't compete with pay at Google?
--:-- Agustin Lebron: One thing I always tell companies is go more junior. Look at the salary of somebody who just comes out of school—not Stanford, but a reasonable CS program. Look at their salary three years later. It could be almost double sometimes. That jump is kind of unjustified. There's a kink at the two-to-three-year point because every tech company seems to want two years of experience. A lot of it is because companies don't want to or can't see themselves investing in training those first two years. If they do, they tell themselves the employee will just leave after two years for a higher-paying job. I think those are terrible answers. You should be investing in training your people. You get the benefit of training them exactly the way you want. If you put in that work and think carefully about what people come to work to do for you day to day, they're probably not going to leave. Switching jobs is incredibly costly and risky. People don't go out of their way to do so. You have inertia working in your favor anyway.
--:-- Dwarkesh Patel: That sounds very similar to the sheepskin effect of the last semester of college. Bryan Caplan has a good argument about this in the case of education: the last semester of college boosts your earnings many times more than the percentage of college you spend in that last semester. It can't be because you're learning that much more. That sets up an arbitrage opportunity for hiring people right before they finish their last year. Give me a perfect example.
--:-- Agustin Lebron: In San Diego, startups and tech companies love to hire Intuit employees with two to three years of experience. Intuit hires a bunch of people and trains them pretty well, and then they get poached. But nobody actually thinks about the idea that Intuit knows who the good and the bad are after two years. You're not seeing the really good ones—Intuit is keeping those.
--:-- Dwarkesh Patel: You say in the book that over your long career in trading, you've traded all kinds of different financial instruments. What is the reason? Is it just that you had to trade whatever market you had to at the moment? You say in the chapter on edge that one way to get edge is to specialize. Is it a mistake for firms to let their traders trade in multiple categories over their career, or is that necessary to build general aptitude?
--:-- Agustin Lebron: It's a balance. It depends on how big the reference class is. I have never done any trading that looks like looking at a balance sheet and income statement, listening to an earnings call, and making a bet—that's fundamental trading. I think it would be a pretty big mistake to put me in that situation. But within the well-defined realm of quantitative trading, I think a lot of the same skill sets apply in different markets. You're bringing the same skill set to different markets, and having that experience of going around and looking at different kinds of markets and how they work informs how you think about things and gives you that wider vision that makes you a better trader. So yes, it's a balance.
--:-- Dwarkesh Patel: Finance is 9% of GDP. I understand the argument that finance helps allocate scarce resources where they're needed most. But if we're giving up a tenth of our resources to make the allocation of the rest more efficient, is that too high a price to pay for liquidity and price discovery? Is finance too high a fraction of GDP?
--:-- Agustin Lebron: I go back and forth on this question. When you see it from the inside, a lot of it is zero-sum competition. It feels like there's got to be a more efficient way to do this. But from the outside view, we haven't come up with a more efficient way to do this, and it's hard to argue with GDP growth. There are two countervailing forces. You can be inside something and be really familiar with it, and the act of being familiar gives it legitimacy automatically. But if you look at something from afar, you think, "That's ridiculous. That shouldn't exist." The people most well-informed, who could or should be making decisions about whether this is legitimate, are biased towards thinking it's probably a good thing. It's hard to disentangle the experience and the biases that experience gives you.
--:-- Dwarkesh Patel: Would that fraction shrink without harming efficiency? Are there inefficiencies created by garden regulation or restrictions on capital flow? Or is that basically what you should expect in a free market or an optimally regulated market?
--:-- Agustin Lebron: That's also a tough one. I don't have a great answer. If you made me regulator of the world, at the margin, there are some things I would regulate more. This is probably an unpopular opinion among my financial friends, but I think leveraged ETFs should be banned for retail trading. They're kind of a bad instrument, in particular all the volatility products. That should probably be regulated more. At the same time, the qualified investor status thing that people are driving a truck through seems weird. Should we eliminate qualified investor status and let people invest in whatever they want? Or should we make it even more restrictive?
--:-- Agustin Lebron: I'm not sure about that one. And certainly the other thing about it is, a lot of the regulations, especially around capital requirements for banks, are incredibly baroque. They feel like job Ponzi schemes a lot of the time. Like, we need to figure out a way to employ all these people, so we're just going to create Basel III, and that's going to be an extra thousand employees for every large bank in the world. That's probably a deadweight loss. But doing things more simply doesn't seem like it's going to get you the stability outcomes you want. So yeah, it's just poor trade-offs all around.
--:-- Dwarkesh Patel: What does the long-run future of trading firms look like? If economic growth continues to stay low, you would expect other financial instruments to stop growing at high rates as well. But even if economic growth speeds up, if markets get more efficient over time, you would expect the profits that any one trading firm can get to decrease. So is there a future for highly profitable trade firms like Jane Street in the far future?
--:-- Agustin Lebron: To the extent that Jane Street and companies like it provide a service to the world—and I really do think they provide a service to the world—they're going to be around and they're going to be profitable. Are they going to gain excess returns? Even that's not so obvious. The thing about trading firms, especially market makers, is that most of the time the business is pretty good if you're really good at it. But sometimes it's really good, like when there's lots of market volatility. That's precisely because you are the entity that is willing to take the risks that nobody else is willing to take. To the extent that we're going to continue to have volatility—market volatility, economic downturns, whatever—there's always going to be a service that these companies provide.
--:-- Over the long run, there's probably going to be more consolidation. That seems unlikely to stop because the benefits of economies of scale just keep going up. But then again, you have new things that come up like crypto, where it's the Wild West right now, and there's going to be a big consolidation over the next 10 years. I think that's the natural arc of things.
--:-- Dwarkesh Patel: Interesting. Can you describe what these economies of scale look like in finance? And what is the trade-off where if you're too big, it's not even worth your time to look at smaller investments because you can't take as big a stake without moving the market?
--:-- Agustin Lebron: The thing about finance, or market making and trading in general, is it's very labor intensive. You should think of it almost like the value of a seat or the value of a person's time. Are there going to be inefficiencies in the market, like pockets in the pink sheets, where it's just not worth a large company's trader's time to look? Yes. Those will always exist, and they'll get slowly competed away by the mom-and-pop trading operations, or even former Jane Street traders who are now at home doing it on their own for fun. So those will always be there.
--:-- Dwarkesh Patel: Is there a potential that markets can get way more efficient if we develop much stronger AI? At what point will the work that even traders do—the more model generation and abstract thinking—get automated away, not just the rote calculations?
--:-- Agustin Lebron: It's already getting more efficient. When my former boss started, the idea of an options market maker having 10 stocks that they were market makers in was kind of the limit. When I was doing it, we could handle 100 stocks. Technology just made everything more efficient in human time. That will continue. You can set up things where you look at data, run a bunch of different models, select the good ones, and make sure you're not overfitting. This is stuff you can do now that you couldn't do 20 years ago.
--:-- When people talk about AI and trading, I think it's very hard. We have to define terms. If you ask a reasonably aware person what AI means today in 2022, 90% of people are going to say we're talking about large language models. Is the question, "Is GPT-N going to be a significant force in markets?" I'm honestly kind of skeptical about that. I don't know that just making larger transformers is the way we're going to get to AI. That's my personal, parochial opinion.
--:-- But if we think of AI more broadly as slowly increasing the range of things that machines can do that humans can do, then the human part is going to slowly start to get disappeared away. The natural analogy is what happened in the 20th century with manufacturing. It used to be all human power and a little bit of machine power. Factories in the 19th and early 20th century were tall and thin because they had one steam plant and had to use belts to distribute that power. Then electric motors happened, and factories became horizontal. Over time, the trend is for less human power and more machine power. I think the analogy is perfect. AI over time is going to take more and more of that cognitive load from the human. That seems inevitable to me.
--:-- Dwarkesh Patel: I'm curious why you're skeptical that a scaled-up GPT-3 or other large language model has applicability in financial markets. I know there's a toy version where you have GPT-10 and you ask it to complete the sentence, "The best trade I can make today is..." Why is that unlikely to happen?
--:-- Agustin Lebron: There are a couple of things. One is sample efficiency. These things are incredibly sample inefficient in a way that humans are not. There's something fundamental there that we're not getting. The thing we're not getting is the structures our brains have for semantic understanding. To the extent that these large language models have semantic understanding, it's kind of by accident. It's the Clever Hans phenomenon. It's super impressive, and I'm not criticizing the models, but it's still a Clever Hans thing. There must be a better architecture out there, much like our brains have architectures that specialize in certain things that give machines semantic understanding, or at least the potential for it, that I don't think GPT-3 has evidenced.
--:-- Dwarkesh Patel: Jane Street seems like a mysterious place. What's interesting to me is there seems to be a large overlap with the rationality and EA community. Sam Bankman-Fried went into Jane Street with the explicit goal of earning to give. Tyler Cowen announced that $20 million had been donated to his Emerge Adventures grant program from Jane Street traders. Even reading your book, you reference so many thinkers prominent in the rationality sphere. There seems to be a big overlap between this community and at least a part of the trading world. That could just be selection effects, but what is going on here?
--:-- Agustin Lebron: That's a great question. I think at two levels. One is the idea of being very rational and not fooling yourself—to use a Yudkowsky term, "shut up and multiply." I think that is very common in the two circles, or at least it should be. Try to really understand the real world, and do so using rational, mathematical, logical approaches. I think there's a lot of overlap inherently there. But you could say that about any number of finance or Wall Street trading firms.
--:-- The one thing Jane Street has going for it differentially from those other firms is the culture of collegiality. I think that's an important thing that Jane Street has developed over the years and continues to have. If you are an EA person who thinks about things rationally and enjoys the process of collegiality and working with people and thinking interesting thoughts together, Jane Street is going to be a very natural fit.
--:-- Dwarkesh Patel: When I had Berne Hobart on the podcast, we talked about whether debugging or finance was a better application of rationality principles. In each case, you have to update your beliefs. One interesting point he brought up was that in finance, you not only have to model a static system as you would in debugging, but you also have to model other agents and their incentives and motivations. That makes it a much more dynamic system to get a hold of in your brain. It could even mean that the tools of the current rationality movement are not good enough to think about those things as well as you guys who have natively developed in the industry.
--:-- Agustin Lebron: The cross-pollination goes both ways. But the idea of you being an agent in the world you're trying to study is fundamental in trading. It makes it so much more interesting. Getting back to the AI thing, one of the big failure modes is to think, "I'm just going to throw some AI or machine learning at this dataset and get a trading strategy." Let's say you've figured out something that predicts price movement 55% of the time. That thing can still actually lose a lot of money in production because of adverse selection. You're only going to do a small fraction of the good trades, and you're going to do all the bad trades you want. Also, if you are actually making money at it, that's a big, shining signal to the rest of the world: "Hey, there's money over here. Why don't you compete it away?" That's definitely a huge component of it.
--:-- Dwarkesh Patel: You have a very interesting chapter on software and technology in the book. One of the things you argue for is that we should take the concept of technical debt seriously in a financial sense. Is one implication of this interpretation that you should be willing to accept technical debt more if you're a rapidly growing company? If you're a startup that's growing fast, it makes sense to take out a lot of loans because you can pay back the interest plus way more. But if you don't take it financially, maybe you would think that if you're scaling rapidly, that's the worst time to take on technical debt because you're just going to be hampered the entire way. More generally, what kinds of firms should be more willing to take on technical debt?
--:-- Agustin Lebron: Startups are the classic example. It's non-recourse debt. If it goes belly up, you don't have to pay it back. You're done. Startups should definitely do this. You see it all the time with the concept of an MVP: let's just get something out there, get some feedback from users, with the understanding that you're going to have to essentially rewrite it from scratch if it's successful. I think it's a very productive way to do software startups. The implied interest rate you're willing to pay is incredibly high.
--:-- Larger companies, it's interesting. I had a conversation with a good friend who did consulting with me. He worked at Qualcomm for a lot of years and worked very closely with Microsoft. Microsoft employs tens of thousands of software engineers. I asked him, "What do they do all day?" He said, "I don't actually know for a fact, but I'm pretty sure the vast majority of them are doing things like, 'Well, this library is deprecated. We need to upgrade this thing. Let's change all this code in all these different little places.'" There's a sort of archaeology of software that occurs. If you've been building a piece of software for 20-some-odd years, there's all this cruft in there that you're continually trying to maintain so that it's functional as you go from one OS to another to the cloud to whatever. That's accumulated debt that large companies certainly have.
--:-- Dwarkesh Patel: That's so interesting. They're just servicing the debt they accumulated in the 80s and 90s when they were growing rapidly. You can even think of them moving to a new platform or rewriting their code as refinancing their debt.
--:-- Agustin Lebron: Exactly. Probably the best book I have ever read about software development is actually science fiction. Vernor Vinge's *A Deepness in the Sky* is very crucially about this idea: what if we've been building on the same software stack for 6,000 years? What does that look like? I think it teaches us a lot about how to think about large, long-term software projects.
--:-- Dwarkesh Patel: I'm super interested in how you guys think about software in the financial industry. I know Jane Street uses OCaml. From what I understand, there's more safety in a functional programming language. How do you think about engineering decisions and the trade-offs involved when you're working in finance?
--:-- Agustin Lebron: Jane Street uses OCaml. One of the biggest advantages of that language is it is strongly and statically typed. You can use the type system to make impossible states unrepresentable. That's a really good software engineering thing to do, and it makes it a very rich environment to do that in. The "I didn't know I had to handle this explode" problem is minimized.
--:-- Jane Street and companies like it obviously optimize for avoiding hot loops in code that incinerate money really fast. That is not what your average SaaS startup optimizes for, nor should it be. But the thing I keep coming back to when talking to technology leaders is that software development is fundamentally an exercise in sociology: organizing teams, creating processes, culture, and conventions around building software. Software development is fundamentally the management of complexity. All that sociological stuff ends up being some of the most important stuff to think about.
--:-- Dwarkesh Patel: Now that you're working in finance but have a startup, you have to think very carefully about this trade-off. How are you managing this, given that you have to move fast but also need to be safe?
--:-- Agustin Lebron: Hire really, really good people, honestly. Don't skimp on those first few employees. The bar is kind of weird; there's not one total ordering over the quality of an engineer. It's incredibly multivariate. But certainly, one really good, thoughtful engineer who can build correct code is worth four not-so-thoughtful people in a spot like that. That's what we're optimizing for right now.
--:-- Dwarkesh Patel: Do such engineers need to have a lot of knowledge about finance, or can they function knowing just about engineering? Can you just tell them, "We need a program that does this," or do they need to understand how trading and finance works?
--:-- Agustin Lebron: "Need" is probably a strong word, but certainly the culture I want to build is one where it's almost a want. I want to hire somebody for whom understanding the problem domain deeply is a critical part of the job they feel they're doing. Is it possible to build something another way? Probably, but that's not the company I want to build.
--:-- Dwarkesh Patel: In your career, you've done so many different things: engineering, trading, consulting. How much carryover and lessons do you feel you've had between these different domains? Or do you feel they are self-contained pools of knowledge?
--:-- Agustin Lebron: If there's one constant for me, it's that I am surprised by how much my previous careers inform my next careers. When I wanted to move from engineering to trading, it continually surprised me how useful the actual engineering training was, as opposed to just me being generally smart and able to figure things out. Then coming back to consulting with companies, again, really surprising. I expected that when we were doing management and hiring consulting, it would be about the nuts and bolts of what a good hiring process looks like. There's a component of that, but all the other trading stuff—how to think about the market for candidates—it surprised me how non-obvious a lot of that was to the people I was talking to. Now, hopefully, I'm bringing all of those experiences to the table in this new startup. I'm optimistic that'll occur again.
--:-- Dwarkesh Patel: You would think that people like you who have so much experience in so many different industries would be the most common archetype of a startup founder because they have so many general skills. At least in popular culture, it seems like the trope is somebody who has no particular skills and starts a startup out of college. Why are there not more founders who have a broader skill set and lots of experience?
--:-- Agustin Lebron: I think there actually are. If I remember correctly, the average startup founder is actually significantly older than the popular conception. It's just that the young, flashy startup founder gets all the press. Perhaps rightly so. I'm not besmirching the young founders' press, but I think there are a lot of people doing it with similar backgrounds to mine.
--:-- Dwarkesh Patel: I forgot to ask about adverse selection. If a counterparty knows they are trading against Jane Street, and they know Jane Street has a great reputation for making profitable trades, why does anybody even make that trade? Does that mean Jane Street has to pay a higher cost to make the same trade because of its reputation? Is there a negative feedback loop where if you become too successful, the market makes it really hard for you to continue being successful?
--:-- Agustin Lebron: The answer is no. The reason is, getting back to the idea that Jane Street provides a service to the world. Jane Street doesn't want to trade against other market makers, and other market makers don't want to trade against Jane Street because they're in the same business. That's not who they're going to make their money from. They're going to make their money from people who need the service Jane Street provides.
--:-- For example, if I am a pension fund or a large hedge fund and I want to put on a bet in some random country, maybe I should just buy that country's ETF. It's certainly a lot easier and more convenient than going to that country's stock market and buying all the individual stocks. They're not an expert in trading Vietnam stocks; they're an expert in making macro bets. Jane Street provides them the service of being able to sell them that ETF, and then Jane Street takes care of all the little details. There are gains from trade there. It's not zero-sum in that sense.
--:-- Dwarkesh Patel: What is the role of market makers in crypto if you have automated market makers like Uniswap? What is the comparative advantage of smart market makers?
--:-- Agustin Lebron: Perhaps you've seen that paper from last fall that shows at least half of liquidity providers on Uniswap V2 lose money. That's what you would expect on priors. If there was no fee on Uniswap and liquidity providers just tossed their money in, they would be systematically adverse selected against by every trade that happens. The fee you collect as a liquidity provider is compensation for the adverse selection you are undertaking. But that fee is set by fiat—it's either the 5-bip pool or the 30-bip pool. It is not adaptive to market conditions.
--:-- I am personally long-term skeptical about CFMMs as a market mechanism that is going to work. I don't see how it makes sense for somebody to just throw some money in a pool and expect to get outsized returns by doing nothing. Outsized returns come from you knowing how to do something or being able to do something nobody else does. It doesn't strike me as an exciting thing long term.
--:-- Dwarkesh Patel: Does that mean you're also pessimistic about passive investing in the long term? Somebody just putting a certain amount with their money?
--:-- Agustin Lebron: I think the difference is that passive investing, over a long haul, is providing risk capital to companies that are hopefully sources of discounted future profits. There's a reason you might expect that to make money for you. Whereas when you're trading FX or something that doesn't earn yield from actually providing value to the world, you probably shouldn't expect to make money by passively investing in that.
--:-- Dwarkesh Patel: What made you interested in getting into crypto at this time, transitioning to that industry?
--:-- Agustin Lebron: The thing about crypto that I like is, to the extent that you believe in a somewhat stagnationist theory—that whether it's regulation or cultural changes, we're not doing bold, new, exciting, weird things—crypto is a Schelling point around which everybody has decided, "All the crazy, weird stuff that we want to try, we're going to do it here." I like that. I think that is a very good thing for the world.
--:-- I'm not saying this is going to be the case, but if all of crypto goes to zero and all that's happened is we've had a large wealth transfer from the rich olds to the young people who want to build cool stuff, that's still good for the world. I think it's going to be more than that. I think there are a lot of interesting, exciting things that are going to come out of the crypto world. But we get at least that: a coordination around trying new things.
--:-- Dwarkesh Patel: Taking your negative example as a hypothetical, I actually do wonder what the actual wealth transfer has been. Has it actually been from wealthier to poorer? If institutional investors haven't gotten that much into crypto compared to some middle-aged guy, has the wealth transfer been the other way around?
--:-- Agustin Lebron: I think it has to have been. Look at all these VCs raising all these funds. The LPs in those VC funds are old. There's a good story to be told about adverse selection in venture capital as well. It's basically a transfer of wealth from VCs to 21-year-olds.
--:-- The other thing about crypto: people from Jane Street, whenever I meet them and they find out what I'm doing, the first question they ask is, "Are you all laser eyes now? What's your deal?" By crypto standards, I think I'm very non-laser eyes. This is probably an unpopular opinion within the crypto world, but I think success for crypto definitely looks like integration into the financial system. It's not going to replace it. It's not going to be, "Goldman and Chase are going to go to zero and Coinbase is going to crush it." Success for crypto looks like traditional finance integrates, takes the best ideas, and crypto companies are incredibly successful in that process, but we end up with something that's a hybrid of the best of both. I think that's success.
--:-- Dwarkesh Patel: What does the future look like in a world where crypto is very successful? What would something like the stock market look like? Would it be far more efficient if it's over crypto, or would it be less efficient because of gas fees?
--:-- Dwarkesh Patel: Or maybe it's like payments internationally. But yeah, I'm curious what you think 20 years down the line success case for crypto looks like.
--:-- Agustin Lebron: So I'm gonna leave the financial markets to last. I think Western Union being out of business is probably a good outcome. It's probably good all those stupid Thomas Cook money exchange things in the airport are out of business. That's probably good. So if only that happens, I think we're already in good shape.
--:-- Certainly the idea of NFTs as transferable signals of facts about you, or facts about whatever persona or avatar you want to have, I think it's pretty exciting. The idea that I have to call my university and get a transcript from them seems insane to me. The way some of these credentialing systems can work with NFTs strikes me as a fairly natural thing.
--:-- I think to the extent that crypto is breaking the oligopoly of a few large financial participants in the market today... I don't know if you followed the saga of the CFTC review of FTX's proposal to do a different kind of futures margining process on actual real futures. I think this is a good thing. There's a lot of vested interests and entrenched interests that are getting their bell rung. And that's a good thing.
--:-- So that's the direction I would take it. The financial infrastructure or the plumbing of finance is probably going to be crypto-ified. That doesn't mean it's all going on chain. All the blockchain is, is a very slow, weird database. But it is a very slow, weird database that has some useful properties in some situations.
--:-- Dwarkesh Patel: What are three books you would recommend?
--:-- Agustin Lebron: *A Deepness in the Sky* by Vernor Vinge. If you're a software engineer and you like science fiction, read it with the eye towards thinking of it as a software book.
--:-- If you want to think about risk-taking in general, Aaron Brown's *Red-Blooded Risk* is really, really good. In fact, all of his books are really good. The thing that got me interested in finance was reading his book *The Poker Face of Wall Street*, because I was playing poker at the time. That was kind of like, hey, maybe Wall Street is a thing. So *The Poker Face of Wall Street* by Aaron Brown, or *Red-Blooded Risk*.
--:-- And one that's kind of off the wall is *Kolyma Stories* by Varlam Shalamov. It is a collection of stories about people who lived in the gulag in Siberia during Stalinist times. I think it is possibly the most revealing book about human nature that I've ever read. It's depressing. I don't read it in a sunny place, but it's revealing.
--:-- Dwarkesh Patel: Is this a covert way of telling us about the working conditions at Jane Street?
--:-- Agustin Lebron: No, not at all.
--:-- Dwarkesh Patel: A final question. You've been successful in so many different industries and you know the lessons of working in so many of them. Is there advice you would give to somebody who's in their early 20s? Most of my audience are probably going to be programmers, but maybe after interviewing you, I'll have some traders or wannabe traders listening as well.
--:-- Agustin Lebron: Number one thing I tell my kids all the time: life is long. Life is not short, life is long. What that means is you should think of yourself as having many opportunities to learn things and try things and do things.
--:-- This is just my own experience, but I feel like I'm sequentially obsessive. I will block off six years of my life—empirically, this is what has happened—to get really, really good at a thing. And then, okay, next six to seven year period, I'm going to try to get really, really good at another thing that is kind of different.
--:-- That's worked out for me because it's easy to undervalue the importance of deep, deep expertise in a thing and the process required to get really good at something. This idea of sequential excellence is something I think about a lot, because you have the time. Spend five years being a front-end developer and get incredibly good at that. Then go do something else. Maybe it's not programming. Maybe it's something else. And maybe come back to it, and you'll have this other perspective.
--:-- Dwarkesh Patel: That's super interesting. Let's say you had been a trader from the very get-go instead of doing electrical engineering and computer science. Would you have been more successful in the counterfactual where you were a trader the whole time, or one where you have the experience from engineering? Does the career path with a lot of deep specialties, changing what that specialty is over time, lead to a higher peak in the end, or the one where you just focus on one career?
--:-- Agustin Lebron: It's a good question. It probably varies by person. If you are destined to be a world-changing physicist, then probably you don't do any of the sequential weirdness. You just go down that road.
--:-- But I might edit that. One thing I do believe about discoveries, whether in trading or science, is there are two types. There's the evolutionary type, which is take a body of work and push the boundaries out a little bit. And then there's the revolutionary type—the Albert Einstein, the Claude Shannon—where it's like, I'm going to invent a whole new field.
--:-- Claude Shannon is an informative case. He famously just basically played games all day and thought about random things and tried to have as broad an exposure as he could. He rode unicycles and that sort of thing. So maybe you need to be a little self-aware about which of the two you might be.
--:-- Dwarkesh Patel: Okay, so the book is *The Laws of Trading* available on Amazon. Do you want to give your Twitter handle plus any other place where viewers can find you?
--:-- Agustin Lebron: Sure. It's @AgustinLebron3. I don't know why three, but that's what it is. I mostly talk about trading, sometimes talk about software, sometimes talk about random things in the world. I'm not much of a social media guy. If I hadn't been for the book, I would still be a social media non-existent person. But I've gravitated towards Twitter. It's where I ended up having interesting conversations.
--:-- Dwarkesh Patel: I've enjoyed being a follower. Is there anything we didn't touch on that you think might be interesting to close on?
--:-- Agustin Lebron: I'm selfishly going to ask you, Dwarkesh. Tell me about what you're doing. What is this that you're building here?
--:-- Dwarkesh Patel: That's a good question. To give you the backstory, it was my sophomore year of college, or maybe junior year. COVID hit and I was really bored because classes went online. So I just started the podcast. I cold-emailed my first guest. Actually, at the time I was releasing, I didn't even have a name for the podcast because I just had a recorded episode.
--:-- I just kept it up. I graduated like four months ago—technically graduated like two weeks ago, but I was done with classes four months ago. I thought, all right, I have a little bit of money saved up from an internship and another grant. So I thought, let me just do this full-time for a few months and see what happens. Got some traction.
--:-- I don't know where this leads. But actually, the comments you made about going deep on one particular thing and then using the skills you learned there to transition to another—that makes me feel a lot better. Because I don't think my long-term trajectory is being a podcast host or writing a newsletter. I would love to go back to tech and startups in the future. I didn't have a computer science degree, but I'm hoping to learn as much as possible through the podcast and writing, and then use the skills I learned there to do some cool things in other fields.
--:-- Agustin Lebron: Love it. Sounds like a great plan.
--:-- Dwarkesh Patel: Yeah, thanks.
--:-- Agustin Lebron: Awesome.
--:-- Dwarkesh Patel: Yeah, thanks for coming on, Agustin. This was one of my favorite podcasts I've done. So many insights.
--:-- Agustin Lebron: Awesome. Thanks, Dwarkesh. Really enjoyed it.
--:-- Dwarkesh Patel: Bye.
