The Exchanges

Q Is there an ideal growth rate, do you think? Like, can you grow too fast?

A The, the answer depends. I, I, I think the easy, if you're growing so fast and the customer experience starts to degrade, you're hurting yourself. And so, yes, you can grow so fast that it's just a complete mess. You're losing customers because you're not servicing them correctly. And at Zappos, we were very mindful that whatever customer experience that we delivered, it was the best customer experience that we can deliver at that growth rate. And I think if you ask, uh, Tony Hsu at DoorDash or Brian Chesky at Airbnb, they would probably say the same thing. It's like, we want to grow as fast as possible, uh, With the constraint that we don't, we want to make sure that there's a great customer experience.

AI assessment note: “yes, you can grow so fast that it's just a complete mess.”

Q Well, I want to go back to something you said earlier about crucible moments. What are they, and how do you identify them?

A Crucible moments, um, is, is the term we use at step, at Sequoia for things that are basically very important type one decisions, and I've had to deal, as an operator before Sequoia, we didn't call them Crucible moments, but I've gone through a fair share of crucible moments. You know, link exchange, do we want to, do we want to sell the company to Microsoft? At, um, tell me networks, when we're burning sixty million dollars a quarter, how do we get through that, and how do we pivot the company from a consumer business that wasn't working into an enterprise business? These are decisions that change the trajectory of a company. Um, at Zappos, it was, what do we do, um, after nine-eleven, um, when sales went to zero, what do we do during the financial crisis when credit became very, very difficult, and we were borrowing, uh, sixty million dollars, and we had sixty million dollars that we did nothing wrong, but we needed to, but the banks didn't have the liquidity anymore, they were calling the loan. And how do you, how do you go through these things? At Airbnb, there have been a number of crucible moments as well, one of which started very early on that Brian Chesky talks about a lot, which was the PR crisis with, um, uh, a guest trashing a host, um, and that led to host guarantees, but there were many of those. There was a fight to win Europe, and a competitor called Windu that …

AI assessment note: “term we use at Sequoia for things that are basically very important type one decisions”

Q first order. Uh, I want to get into more of your experience, not only at Zappos, but, um, being on the board of some of the companies that everybody has heard of, um, today are being involved. But before we get there, I want to come back to the school for a second. Are there any other experiences that you had during school with teachers that might have impacted you?

A You know, earlier on in junior high school, I was, um, I was suspended from the computer lab because I built, this was very old, so they're a Radio Shack TRS-ADs, so we call them Trash-ADs today. Um, and there was the computer lab, you know, you had to run time in our computer lab to, to use the computer, and I built this game, and the central server, um, At the time was literally a floppy drive, where all of us saved our programs there, and one day, uh, a bunch of the students, uh, in the computer lab all found the game that I programmed and then started playing with it. The teacher and the principal just happened to walk in and saw that we're all playing this game, and, um, I was told that computer lab is valuable time, and you should be doing something much more productive than Then producing a game, so I, I was no longer allowed to work in the computer lab, and the computer lab teacher was Mrs. Pitosa, and she's, she also ran the math team, and she said, well, you, I'm sorry, the principal wants you out of the computer lab, but you should join the math team, and I joined the math team there, and, um, one of the things I was very good at was math, and she told me that if you want to be a leader and you want the team to win, It's not good enough for me to just solve the problems. I mean, figure out how to get the rest of the team to perform, and so I started teaching the rest…

AI assessment note: “the computer lab teacher was Mrs. Pitosa, and she's, she also ran the math team”

Q Is that what you mean by first order issue? I think you mean something a little more nuanced.

A Some people talk about the most important thing. Um, and I think, I think about first order. If I get this problem, I have a problem on my hands. If I get to the first order issue and get to the root cause of that, usually that helps solve that problem. And there are other issues that are not first order. Um, and that concept is quite important. We also sort of navigate that into other, uh, situations where, um, where they are crucible. So as an example, like, you have, you have situations where The website's not working fast enough. At Atlas, we had a situation where the website's not fast enough. Is the first order issue that we have too many pictures? Well, we want the pictures. We have lots of photos. We want to show those photos. Is the first order issue that we need to, um, trim the number of search results? Well, customers want longer search results. It's like, no, it's none of that. We need to figure out how to make the website go faster. And so we start caching, um, The, the search results. We start caching things, and so you start developing the technologies that solve the speed issue, but the first order issue is that we need to solve this with technology, not with a bunch of either-or solutions. Um, that's an example that I learned a long time ago. Another situation is the distribution at Zappos was not flowing well, um, and we couldn't figure out which Process was …

AI assessment note: “If I get to the first order issue and get to the root cause”

Q How do you want somebody to call you out in those situations, or how do you call out other people?

A The way I do it is try to be fact-based, and, um, just give examples specifically of what What has been working and what has not been working, and I, that tends to work pretty well, and there, there are times when Rebecca, my wife, will say, hey, this is not working for us, so what do you want to do about it? Uh, you can try to keep going the same way, same way down this certain path, or we can try to change it. If we don't like something, we just change it, and it's just much easier to have those kinds of conversations. Um, my son a few days ago was being, um, quite negative about his school, and You know, he's starting, he's 13, he's gonna be 14 this year, he's starting to be like me when I was younger, and think that he's smarter than the whole rest of the world. And he said, hey Atticus, do you wanna be, do you wanna have a negative attitude because you'll spin into this negative spiral? Um, and by the way, if you have a negative attitude, the world just looks worse. When you have a positive attitude, The world just looks better. And we played this, let's use yes and for a bunch of things, and let's create options, and then you're going through the no but situation, and it doesn't work. So let's try to just change your attitude about some of these things. And I, I thought that was quite an interesting conversation where just by simply changing the framework, he became much …

AI assessment note: “The way I do it is try to be fact-based, and, um, just give examples”

Q Let's go back to the push-pull thing for, you know, you said companies don't generally increase velocity without force. How do you think about the difference between adding force or removing obstacles?

A That's a very, that's very, very insightful of you. I think many times applying force, um, is what What leaders try to do, which is just to apply a forcing function and make you make a choice. And often I think removing obstacles are just as important as, as applying the pressure to keep going on and on and on. But often, you know, when you apply force, you realize where the obstacles are. Because you, you butt against that obstacle. And then you realize, okay, I keep butting against this. Maybe you have to figure out a way around it. Um, and I, you know, my job as a board member is to help you identify obstacles and help you remove it, but sometimes you just need to be pushed against it.

AI assessment note: “when you apply force, you realize where the obstacles are”

Q Where do people tend to go astray, right? So you, you have a great core product, and I'm just imagining you start building up a team, you start getting out of your focus, like where do companies go wrong?

A A number of places, if you just look at great companies, they compound at a very high rate for a long period of time in their core business. Google for a long period of time was search. Amazon for a long period of time was e-commerce. And there's a difference, I think there's a difference between act two and category expansion. So in Amazon, they went from books to music to electronics to a whole bunch of, they just went category by category. But it was the same real, like, core business, which is getting product to the distribution center, pick, pack, and ship the products that you want, and shipping it to you. Um, and it could be books, it could be music, it could be electronics, it could be shoes, it could be, um, tools, it could be a hammer, but, but that's one core business. And I think founders start their company because they have novel and compelling insights into the world, and specifically for their company. And they like the new shiny penny. I'm not saying that you shouldn't focus on the new shiny penny that is adjacent to your business. I'm just saying that you shouldn't focus on the new shiny penny that is like way out, away from your business. Eventually, all companies need an act two. AWS was an act two. That was very different than the, than the e-commerce business. But even the way that Amazon thought about the, their web services business is like, well, we're …

AI assessment note: “you shouldn't focus on the new shiny penny that is like way out”

Q What's the difference between working backwards and working forwards?

A So, working backwards is this, um, idea that Amazon, um, popularized, which was, you start with the vision of the world that you want to imagine, and you, you think about if everything goes right, what is this, what is the world that we live in, and what does this product become that you're trying to build? And at Sequoia, we call that the pre-parade. And, um, You know, if everything goes right with this company, what does it become? And then you work backwards from there. What, what is, you know, one, one year before that, what does it look like? Two years before that, what does it work, look like? And Amazon popularized by working backwards, meaning that go from the future all the way to today. Um, and I love that concept in a sense, um, but I've also seen lots of founders struggle to take it from the future, which they have this great vision of the world, and really map it all the way back to today. They map it somewhere five years out, 10 years out, but they can't map it today. And I also tell founders, okay, that's the future that you're aiming for. Here are the realities of today. What does it look like one year ahead, two years ahead, three years ahead? That's working forwards. And if the two don't match, It probably, you're not on the same trajectory, and so I believe in having founders work both backwards and forwards. Often when you don't, when you work backwards and …

AI assessment note: “go from the future all the way to today... That's working forwards.”

Q Let's go back to hiring and firing a little bit. Was there any moments sort of at Zappos or earlier with LinkExchange where you thought you had the wrong fit, but it turned around?

A That's a great way of asking that question. I, I would say probably not. In the early days of Link Exchange, we hired, and back to, like, conventional, and, you know, we, we hired the conventional right person. We were looking for a marketing person, we were looking for a marketing person that marketed to small and medium-sized businesses. They understood the internet. They, they fit all the specs that you would have in a job description. Um, But they didn't fit the way we operated. They didn't fit the way we, what we valued. And, you know, it's easy to sort of limit that down to, like, they're just a, a professional marketer, and they're a mercenary, not a missionary. That's how Tony would tell the story. But at the end of the day, they didn't value what we were trying to build. They didn't have the same values that we had. And which is why when Zappos started, Tony spent a lot of time defining the values. This is Tony Hsieh of Lake Exchange and Zappos, the values of the company. And it's something that I talked to founders about. If, if you don't have the same mission, you don't have the same values, you don't have the same operating, um, understanding of the company, it's very, very hard to work together. And then, you know, you're asking someone to change when they fundamentally don't fit to, um, to the company and how the company works. That is generally the, the, the issu…

AI assessment note: “I would say probably not. In the early days of Link Exchange”

Q of like the next revolution. Where are we going with AI? Where do you see that? And obviously the longer we go out on the horizon, the harder it is to predict. So maybe let's start with like, what do you see as the next, 12 months in AI, and where do you think we, we sort of, like, how do you envision sort of 12 months to five years?

A It's funny, because I think in the history of technology changes, it's, and then Bill, I think Bill Gates said this, it's easy. We, we almost always, uh, overestimate what things are going to happen in a year, and we underestimate what's happened in 10 years. Yes, it's harder to imagine 10 years out, but, you know, in, in the business I'm in, at Sequoia, we are paid to make investments and to hold them long term, and we do have to sort of imagine 10 years out. Um, I mean, the next year, you know, these foundation models are getting so Good. That you can see certain things being automated that we used to do that we're not going to do as much. But I think, you know, the next step after that is if you think about, lots of people talk about customer service and how we can automate some of the customer service emails or calls or, and things like that. Okay, let's say we do that. What's next? Well, then you should reimagine the customer experience. And so, AI will help us automate some of these things, and there'll be massive cost savings, but AI should also allow us to reinvent the customer experience, and that's why we're very excited about a company like Sierra. They're not just going to help you automate customer service tickets, they're going to help you reimagine what customer service looks like. Just similarly to how, what happened in the internet, you could, in the internet h…

AI assessment note: “the next year, you know, these foundation models are getting so Good.”

Q Sometimes you've said in the past that sometimes the path to greater efficiency is doing things sequentially and sometimes it's doing them in parallel. Can you explain that?

A I think most people are trained to do things in serial. They, they take a problem, they break it down into pieces, and they do one piece at a time. Step number one, solve the problem number one, solve problem number two, because we need the solution to problem number one to feed into problem number two. And they eventually, uh, go down the path. And a lot of, like, if you're an individual contributor, you just, you write code, you just keep writing the code, and you finish the code, and you submit it. Um, but if you're managing a team, you need to break down this piece of software into components so that you can give it to your team members. You need to have all 10 of them working on it. Um, as opposed to you write it, write it yourself, and even if you're a really good software writer, you literally have to be a ten-ex engineer on top of your ten-ex engineer, so you have to be a hundred-ex engineer to do it in the same time frame, but 10, 10 ex-engineers can do it, and sometimes it's just more efficient to just break the problem down so that you can put them in parallel and then stick them all together.

AI assessment note: “sometimes it's just more efficient to just break the problem down so that you can put them in parallel”

Q How do you gauge somebody's potential? It's like, you might be able to do this job, and you might be able to grow into it. But like, coming into it, it's like, they're not going to check the boxes. They're going to look different.

A The future does not look like the past, but you have nothing else to go on but the past, and you can see whether someone moved really, really quickly in the previous situation, where they promoted three times in the same year at a different company that was moving very, very quickly. Do they have a sense of urgency? You can test for some of these things. You can ask about these things. But yeah, you don't know whether they have the potential to grow into the next stage of the company until you You, you're, you put them in that position. Often I think the, the challenge is the company is moving so fast, and you're hiring someone that has done the job before, and you know that they can do the job for the next year or two, and that's maybe good enough. If you hire someone, you have to let them go in three months, that was, that was a messire. There's almost no reason why that person was not going to be able to do the job for the, for three months. They either were not a cultural fit, they didn't have the right skills. If they stick around for a year or two, it's usually because they're They've done the job before, and they're picking up a bunch of low-hanging fruit. Um, but I look at, you know, Tony at DoorDash. Many of the people who work for him have worked for him for a long period of time. He likes to grow people from within. He obviously hires from outside as well. But many o…

AI assessment note: “you can see whether someone moved really, really quickly in the previous situation”

Q think Google and Microsoft, like how, how do you think you compete with them on a foundational model basis? Like they, A, they can scale into enterprise overnight. B, they can spend a hundred billion dollars on, on GPs. How do you think about that? Are we going to end up in a world where there's like only a few foundational models and we're all going to plug into them?

A To be honest, I don't know, but I'll tell you this. In the history of technology investing that I've been involved with, if you're afraid of the incumbents, you should just punch out. But in 1999, you could have been afraid for Microsoft because they have lots of money, lots of servers, and yet they didn't win search. Someone else won search. Microsoft won the browser war for a hot socket. It was Netscape against, ah, Internet Explorer. I don't know many people use Internet Explorer today. Most people I know use Chrome. So, this is not a static situation. I think you just have to sort of have an open mind on why you will win. Um, and often the reason why large companies don't win is because they have their own internal issues. That slowed them down. So Microsoft during the internet, it was, it was the, it was the antitrust situation. And that slowed them down. They could no longer bundle IE into, into the operating system. And so they had to break all those things apart. Um, Google has its own regulatory challenges, and so does Microsoft. So, I don't presume that they're all just gonna win, because they have a lot more money, and they have more, a lot more capital, and they have a lot more people. Um, And besides, if you thought that, then I would be out of a job.

AI assessment note: “In the history of technology investing... if you're afraid of the incumbents, you should just punch out.”

Q more about that, because you didn't have a huge budget for marketing. You had to get, you know, within the first month, you basically had to pay back yourself for marketing. How, how do you think about that, not only back then, and then what's different now in a world where there's a lot more money, and maybe all the players aren't rational about how they're spending money to grow?

A Back then, we thought that that was a, You know, we probably thought it was a curse that we couldn't raise more money because you probably would rather run the business a little bit looser so we can grow a little faster, etc. But being tight led us to find solutions that was, again, non-obvious or divergent, to use your term. Um, and the divergent idea we had was if we can only spend money that was profitable on the first order, we needed other ways to grow. And the focus was to become, um, Um, much better at customer service. We noticed that it was much easier to keep a customer and ordering, getting that customer to order more than it was to acquire a new customer. And then we just went down the line of what we can do for the customer. We, we, we tried to make the website load as quickly as possible. We tried to pick back and ship within four hours. The solution to pick back and ship in four hours was to solve a operations flow issue and not batch the, the operations, which I don't think anybody else figured out in e-commerce back then besides Amazon.

AI assessment note: “if we can only spend money that was profitable on the first order”

Q So how important do you see what Facebook's doing then, where they're spending, you know, 50 to a hundred billion, but they're making it open source?

A I think open source is a very, very important aspect of, um, what has happened in technology, and, you know, you have these religious fights between closed and open, and you have religious fights between Apple and Microsoft, you know, or, or Apple and IBM, uh, It's, let's get, let's go back to and. The world is built on and, um, and, you know, open source, in the internet days, there were lots of things that were built on open source, and some of it was closed source, and some of it was open source, and I think the same will be true with, with, uh, with, um, AI. There are going to be things where you just want to host it on, um, a closed source model, because you know it has The breadth of functionality, and there are things where you can only make it work if you use your specific data, train it very specifically for you, and you're going to use open source models for that. Um, and just like, you know, Google built a great business that is pretty close source on open, on open source software, like Linux. I think the world of AI is going to be canned as well. They'll be close source, and they'll be open source.

AI assessment note: “I think open source is a very, very important aspect of, um, what has happened”