The Exchanges

Q content's a really good example. It takes time To know if content works. And so you have to do things with no obvious gain in the short term for compounding long-term advantage. Does that make sense? It's almost a bit like the gym. You don't gain muscle day one, day two. Week one, you'd be like, Harry, quit this. This A-B test is failing. But three months in, it shows.

A Yeah, I, I, I would agree with that. I think there's like, I think the, the, the, the trap you can run into is you only run tiny experiments, right? Like on, on your purchase page or you, you run these tiny experiments on, on copy or something. And I think it's very important that you have your portfolio of changes. So you have lots of, uh, low risk, small changes, but you also have some big changes and a big change could be Adding chess or adding math. And, and that, that's how you stay innovative and relevant. Uh, you can't just, otherwise you end up in a, in a local maxima where, you know, you can't get out because you, uh, optimize yourself into that.

AI assessment note: “Yeah, I, I, I would agree with that. I think there's like, I think the, the”

Q it. How do you think about personalization balanced with interactivity in the future of content and the future of education? Like, to what extent is it multimodal? You know, the thing I love about Duolingo that makes me sound Weird as anything is I'm walking around London and then I'm going Bruno S Americano and it's, it's multimodal. It's my voice playing back. It's me typing. There's many different modes.

A Yeah. Yeah. I think it's going to be multimodal. Uh, a hundred percent. There's, we have this feature inside the app called video call with Lily. So you can call Lily. Lily's one of our characters. She's the one with the purple hair and, and it is. She, she's not a, she's not a one-on-one tutor. She, she, she's more like a friend and, but she remembers stuff. So she remembers, for example, that, you know, that I like cooking. She knows where I live, et cetera, et cetera. So she, she personalizes in that way, but it's full conversation, right? So you learn by conversing with an AI friend. And I think that's going to be super important. Um, now there's always like these, these modalities, like sometimes, you know, there's, Not in every moment, you know, you can't talk in every moment, right? So sometimes you're in the subway, you don't want to use audio.

AI assessment note: “I think it's going to be multimodal. Uh, a hundred percent.”

Q What was it that you saw that gave you such optimism?

A Uh, so we had, we had early access and we saw you could really Used, I mean, I don't know, when you first had access to it, it was impressive. It was like a, uh, an iPhone moment. It's like, you just realize this is the future. We're going to use this everywhere. That was my first personal impression. And then it's like the second question is like, how do you use this at Duolingo? And the first thing we realized is this can really help us accelerate content production. Okay. So in, in the past, it would always take us a long time to produce New courses, right? So, uh, it requires a lot of, you know, effort to, to create these courses. But then with AI, there's, there's a potential that you can just pump out a lot of these courses all at once. In fact, we just did that. Uh, you know, it took us, I think, 12 years to, to, to build the first hundred courses or so. And now, within one year, we built another 148 courses. That was just not Possible. It would have taken us decades to build this.

AI assessment note: “within one year, we built another 148 courses.”

Q You shared an insight with me before that I'd never thought about, about having offices in San Francisco as a non SF HQ company. What was that insight and how did it lead to your not wanting to have an office in SF?

A Yeah, so Duolingo does not have an office in Silicon Valley or anywhere in California, and the reason is, when we thought about opening one, and of course all of our investors, they said, you gotta open an SF office, and we in fact did go to SF, and we looked at office space, and we were this close to opening an office in SF, but before we pulled the trigger, we asked a bunch of founders Who had companies that were headquartered outside of this, outside of Silicon Valley, and then opened an office in, in Silicon Valley. And we asked him, what do you think? It was a good decision, bad decision. And all of them said, worst single decision ever. And I said, why? And they said, well, you know, what happens is, It creates this internal, uh, funnel of your best employees moving from your HQ to Silicon Valley. Which by itself is not a problem. The problem is then they get recruited away by the whatever the hot Silicon Valley company is at the time. You know, like it could be Airbnb or Uber at the time. Maybe now it's OpenAI or Anthropic. And you basically create a hiring funnel for Silicon Valley and you, you lose your, your talent. And that was also like new to me. And then it was like, no, no, let's not do that.

AI assessment note: “creates this internal, uh, funnel of your best employees moving from your HQ to Silicon Valley”

Q What do you think they saw that others didn't?

A Okay, so one thing that was true back then is we did the Silicon Valley Roadshow, and we actually did get a lot of interest from Silicon Valley VCs at a time, but they all made it a requirement To move to Silicon Valley. So we're like, you've got to move here, right? Like, when are you going to move here? That was the, always the question. Okay. We're super interested. We want to make a deal here and all, uh, when are you moving to, to SF or Bay area? And we're like, we're not moving. And, and then they immediately lost interest, by the way, some of them later invested in later rounds. So they, but they didn't want to be the first one in, in a non, Uh, Silicon Valley company, yeah. And so Unisquare, they didn't care. They said, in fact, they said, we're the only investor that doesn't care that you're in Pittsburgh. That's true. Things, you know, have changed completely. Like, I mean.

AI assessment note: “we're the only investor that doesn't care that you're in Pittsburgh”

Q What was it that you saw that gave you such optimism?

A Uh, so we had, we had early access and we saw you could really Used, I mean, I don't know, when you first had access to it, it was impressive. It was like a, uh, an iPhone moment. It's like, you just realize this is the future. We're going to use this everywhere. That was my first personal impression. And then it's like the second question is like, how do you use this at Duolingo? And the first thing we realized is this can really help us accelerate content production. Okay. So in, in the past, it would always take us a long time to produce New courses, right? So, uh, it requires a lot of, you know, effort to, to create these courses. But then with AI, there's, there's a potential that you can just pump out a lot of these courses all at once. In fact, we just did that. Uh, you know, it took us, I think, 12 years to, to, to build the first hundred courses or so. And now, within one year, we built another 148 courses. That was just not Possible. It would have taken us decades to build this.

AI assessment note: “the first thing we realized is this can really help us accelerate content production.”

Q I mean, that is so cool. Um, listen, I do want to talk a little bit about how you work, because you're Fascinating in how you work. I, I saw in, in some of your posts, you said you're not good at finishing the final 20% of a project or a bit of work. Is that, that seems problematic, is it? And how do you reflect on that?

A It would be problematic if it was just me and ever, or everybody was like me, right? So I think ultimately you need to have complimentary skills, uh, within the founders, but then also within the whole exec team, within the whole company, right? So, so if everybody was really bad at finishing or like the last, by the way, the last are basically the details, right? So you need to have someone who's obsessed with details in, in Duolingo that it, that is Luis, uh, to this day. Um, but also, you know, from day one, he was just obsessed with details, and he really cares, and it's not, by the way, I'm like, I also care about details, but he is an order of magnitude better at that.

AI assessment note: “It would be problematic if it was just me and ever, or everybody was like”

Q You shared an insight with me before that I'd never thought about, about having offices in San Francisco as a non SF HQ company. What was that insight and how did it lead to your not wanting to have an office in SF?

A Yeah, so Duolingo does not have an office in Silicon Valley or anywhere in California, and the reason is, when we thought about opening one, and of course all of our investors, they said, you gotta open an SF office, and we in fact did go to SF, and we looked at office space, and we were this close to opening an office in SF, but before we pulled the trigger, we asked a bunch of founders Who had companies that were headquartered outside of this, outside of Silicon Valley, and then opened an office in, in Silicon Valley. And we asked him, what do you think? It was a good decision, bad decision. And all of them said, worst single decision ever. And I said, why? And they said, well, you know, what happens is, It creates this internal, uh, funnel of your best employees moving from your HQ to Silicon Valley. Which by itself is not a problem. The problem is then they get recruited away by the whatever the hot Silicon Valley company is at the time. You know, like it could be Airbnb or Uber at the time. Maybe now it's OpenAI or Anthropic. And you basically create a hiring funnel for Silicon Valley and you, you lose your, your talent. And that was also like new to me. And then it was like, no, no, let's not do that.

AI assessment note: “creates this internal, uh, funnel of your best employees moving from your HQ”

Q one. Everyone has said, including Warren Buffett, the most important thing in life is partner selection. Yeah, like who you choose to spend your life with. I spoke to everyone on your cap table, and they all said about the special relationship you and Luis have. I'm intrigued. We mentioned marriage there. Whether it's Luis or marriage, I don't mind really. What would be your biggest advice on partner selection?

A So for, for founders, I think it's, it's a little easier. Um, Well, actually, okay, let me see. Let me see if I can answer this for, for founders. I think the one thing is like, have you worked previously with this person? And the key is worked, not, you know, hung out at the bar. I was like, have you worked with this person prior to starting this company? And Luis and I, we had worked together for two years prior. So we kind of already knew how we work. And, you know, what we're good at, what are we bad at? Priorities. And we actually had this very early on. We had this because I went, I went from, you know, his, uh, he was my superior. He was my PhD advisor and I was a student into an equal relationship, but we're equal co-founders. And I was like, can I trust this guy? What if he just fires me after a year? So, you know, we, we, I wrote down a little contract between the two of us. I still have a copy at home. Where we outline exactly like, you know, this is how we're going to make decisions. It's like this short. This is how we're going to make decisions. This is, you know, Luis's responsibility. This is Severance's responsibility. And, uh, and we both signed it. And, and I think that avoided a lot of conflict. So, uh, and the other thing that was interesting, I think a lot of the conflicts happened in the first two years.

AI assessment note: “the one thing is like, have you worked previously with this person?”

Q Yeah. How has the way that you build internally in Duolingo changed with AI? What tools do you use? How has it changed cadence? How has that changed?

A We use AI in three ways at Duolingo. Number one is content generation, and that's where we've seen, you know, the biggest success. It's, it's just completely changed it, like how we produce content. Um, this is learning content. Number two is features, AI features that you previously couldn't build. So this is, again, this video called with Lily. It's a interactive conversation with an AI bot, right? That two years ago or three years ago, you couldn't build. Now you can do it. And it's actually the thing that, you know, was, was lacking in the product. If you ask our users, like, what's missing in Duoling? It's like this conversational piece, like speaking. I want to get better at speaking. And, and now we, we can do this, and it's a really, really, really powerful feature, and it's, you know, great adoption with this, uh, video call with Lily feature. So that's number two. And I think there's more to, to be done there. So both making this feature better, but then also maybe there's, there's new AI features that, you know, we cannot build. And the third is, is kind of overall productivity improvements across the company. So that's. Uh, software engineers using Cursor, there's using AI for customer support.

AI assessment note: “We use AI in three ways at Duolingo. Number one is content generation”

Q Yeah. How has the way that you build internally in Duolingo changed with AI? What tools do you use? How has it changed cadence? How has that changed?

A We use AI in three ways at Duolingo. Number one is content generation, and that's where we've seen, you know, the biggest success. It's, it's just completely changed it, like how we produce content. Um, this is learning content. Number two is features, AI features that you previously couldn't build. So this is, again, this video called with Lily. It's a interactive conversation with an AI bot, right? That two years ago or three years ago, you couldn't build. Now you can do it. And it's actually the thing that, you know, was, was lacking in the product. If you ask our users, like, what's missing in Duoling? It's like this conversational piece, like speaking. I want to get better at speaking. And, and now we, we can do this, and it's a really, really, really powerful feature, and it's, you know, great adoption with this, uh, video call with Lily feature. So that's number two. And I think there's more to, to be done there. So both making this feature better, but then also maybe there's, there's new AI features that, you know, we cannot build. And the third is, is kind of overall productivity improvements across the company. So that's. Uh, software engineers using Cursor, there's using AI for customer support.

AI assessment note: “software engineers using Cursor, there's using AI for customer support.”

Q I mean, that is so cool. Um, listen, I do want to talk a little bit about how you work, because you're Fascinating in how you work. I, I saw in, in some of your posts, you said you're not good at finishing the final 20% of a project or a bit of work. Is that, that seems problematic, is it? And how do you reflect on that?

A It would be problematic if it was just me and ever, or everybody was like me, right? So I think ultimately you need to have complimentary skills, uh, within the founders, but then also within the whole exec team, within the whole company, right? So, so if everybody was really bad at finishing or like the last, by the way, the last are basically the details, right? So you need to have someone who's obsessed with details in, in Duolingo that it, that is Luis, uh, to this day. Um, but also, you know, from day one, he was just obsessed with details, and he really cares, and it's not, by the way, I'm like, I also care about details, but he is an order of magnitude better at that.

AI assessment note: “It would be problematic if it was just me and ever, or everybody was like me”

Q Can I be really kind of weirdly granular? How are you using AI to then 12 X content creation within Duolingo?

A Yeah. I mean, there's still, by the way, there's still a lot of, uh, human in the loop. Um, so for example, the curriculum design, that's all human made. It's like, you know, how, how, how do you structure the course? That's all human made. But then, you know, inside Duolingo, there's a lot of these, uh, short sentences, right? It's like, um, You know, the ones you see in, in, in the lessons. And those are now produced with, with AI. And, and there's also this, this, this concept of, uh, viability. You, you want to introduce new words one at a time, and you can, you know, give these constraints to the AI saying, create a sentence that uses these words and doesn't use any of the other words and only uses these grammar concepts, and it can do that. So that, that's kind of the magic. And, and that's how we've been able to generate The, kind of, the sentence content within these courses.

AI assessment note: “those are now produced with, with AI”

Q Any lessons from doing that? It's hard. I remember having Dash to Silver on from Lightspeed, the payments provider, um, in restaurants, and I, they've done the 18, and there was a lot of lessons there.

A You know, it's actually very interesting. So one insight, and in hindsight, it's, it's kind of obvious, but you know, they say like, what, what is the best predictor of, uh, two founders working out? Like, you know, they're actually, uh, they're not gonna get into a fight and, uh, founder conflict. And the number one, uh, Thing to look for is have the two founders work before they started this, this company, like have they work together. Um, for, for example, Luis and I, we had been working, uh, together on, you know, research for two years prior to starting Duolingo. So, and it, by the way, it's not the same as friendship. So it's like, you can know someone for a very long time, but they might Be very different at work. So prior work experience, um, of working together. That is number one. And with M&A, it's the same thing. I actually think it's, do you have prior experience working with this company? Yes or no? So I would say that's the number one advice is like, Have you worked with them before?

AI assessment note: “With M&A, it's the same thing. I actually think it's, do you have prior experience”

Q content's a really good example. It takes time To know if content works. And so you have to do things with no obvious gain in the short term for compounding long-term advantage. Does that make sense? It's almost a bit like the gym. You don't gain muscle day one, day two. Week one, you'd be like, Harry, quit this. This A-B test is failing. But three months in, it shows.

A Yeah, I, I, I would agree with that. I think there's like, I think the, the, the, the trap you can run into is you only run tiny experiments, right? Like on, on your purchase page or you, you run these tiny experiments on, on copy or something. And I think it's very important that you have your portfolio of changes. So you have lots of, uh, low risk, small changes, but you also have some big changes and a big change could be Adding chess or adding math. And, and that, that's how you stay innovative and relevant. Uh, you can't just, otherwise you end up in a, in a local maxima where, you know, you can't get out because you, uh, optimize yourself into that.

AI assessment note: “Yeah, I, I, I would agree with that.”

Q university state or contemplating university courses today, and who are told, ah, CS in five years time, it's not going to be worth it. It's not, Do you still see immense value in CS as a principle and a mind, like, mindset? If I was your little brother, and I'm 18, ok, and I'm like, you know, help me, Severin, should I do CS at university? What would you say?

A So I did my CS degree, it's now quite a while ago, and we, they don't actually teach you coding. You know, people think like, oh, CS means coding, um, but CS is actually, What, what they teach is kind of the fundamentals of, of not even software engineering, but like how computers work. And then, you know, like computer science is kind of applied mathematics. Okay. So I do believe there's still a lot of value in, uh, thinking logically, which is something that the best CS, um, Courses teach you, right? So, so if you, if you go, you know, studies at university, they, they don't teach you, you know, Java or Python, and they shouldn't. I think that that's actually, that, that is, that part is gonna go away. Like all the details of, of, you know, actual coding, but the, the fundamentals of, um, thinking about problems, it's problem solving. I think that will still be valuable.

AI assessment note: “I do believe there's still a lot of value in, uh, thinking logically”

Q What do you hear a lot from investors that you're like, pfft, really?

A So one thing they, they, um, it's not just investors. It's like, they ask this question, like, what is the secret sauce? Why is tooling so successful? Okay. And they look for an answer that is like a one word explanation. And, and you can tell when you give them the actual answer, they're disappointed. Okay. So they wish the answer was, you know, it's the, it's a streak. That's it. That's, that's the secret sauce. You know, it's this one mechanic. That's why we're so successful, and we, we invented it, or whatever. We, uh, maximize it, or whatever. Or it's, it's the leaderboards. Or it's like, oh, we're, we're so good at, you know, like, uh, uh, this particular market. It's like, it's, it's like, uh, Brazil. That's, that's a secret sauce. We figured out how to crack Brazil. And, you know, like, whatever. Like, they, they, they want an answer like that. And then, then Luis keeps telling them the true answer. Which is, it's running thousands and thousands of AP experiments. That's, and, you know, like the streak mechanic, for example, is like, you know, we added it, that was one experiment, but then we've probably run 300 experiments fine-tuning the streak mechanic, and that's where you get the gains in retention, and that's what ultimately drives the growth of Duolingo, which, you know, is still, it's super fast. We're still growing so fast, um, and, and, and, and, you know, it'…

AI assessment note: “they ask this question, like, what is the secret sauce?”

Q Can I be really kind of weirdly granular? How are you using AI to then 12 X content creation within Duolingo?

A Yeah. I mean, there's still, by the way, there's still a lot of, uh, human in the loop. Um, so for example, the curriculum design, that's all human made. It's like, you know, how, how, how do you structure the course? That's all human made. But then, you know, inside Duolingo, there's a lot of these, uh, short sentences, right? It's like, um, You know, the ones you see in, in, in the lessons. And those are now produced with, with AI. And, and there's also this, this, this concept of, uh, viability. You, you want to introduce new words one at a time, and you can, you know, give these constraints to the AI saying, create a sentence that uses these words and doesn't use any of the other words and only uses these grammar concepts, and it can do that. So that, that's kind of the magic. And, and that's how we've been able to generate The, kind of, the sentence content within these courses.

AI assessment note: “give these constraints to the AI saying, create a sentence that uses these words”

Q With respect on the curriculum design, when you look at completion rates of students, and you have the data that you have, in six months, would it not be better to let AI do curriculum design, knowing all of the completion rates, success rates, satisfaction rates, NPS scores, to actually build that curriculum itself?

A That's, that's a great question. I think the, the industry is definitely moving that way. Um, there's, you know, one of the most amazing things about Duolingo is we have this massive user base. Like, we're by far the largest learning platform out there. By far. Uh, order of, orders of magnitude. And we see how people learn. It's like the largest school, you know, like we see it usually in, in education, the, the studies were done with, you know, 20 to 30 students. We have hundreds of millions, and, and that's just, that's allows us to look at, okay, does this curriculum design work versus a, you know, this versus that? And, and we can use that data to improve the curriculum design itself. But not only that, so, you know, right now we still have these, these, these courses and, you know, your curriculum is the same as mine. But really, I think the future is going to be quite different where you say like, hey, you know, you might be interested in, in rugby, and, uh, you have a trip planned to, I don't know, France. I think France also is, is big into rugby. And, you know, you, you could, you could then custom design a course just for you that uses exactly the vocabulary you want to use when you go to France. And, you know, it could be completely different from mine. It's like, why do we even create this? Batch on the, on the back end. Like, why don't we create it on the fly when …

AI assessment note: “I think the, the industry is definitely moving that way.”

Q Yeah, you're not Gen Z anymore, Severin. I'm sorry, dude. You don't fit in that demographic. Um, we, we mentioned that how it changes like engineering, product design, Another big area is customer support, and we chatted about it a little bit before. How does AI change customer support processes workloads within Duolingo?

A Yeah, so we started using an AI tool for, for customer support. I think it's another one of these early applications of AI where AI can really do a, uh, you know, a great job and transform an industry. And, um, we have found for a lot of the, you know, AI can do 70 to 80% of the, of the tickets that we get. Um, and then people always say, oh, well, okay, you guys, you know, you, you don't need, uh, human customer support, um, agents anymore. That's not true. Uh, we actually, again, we, there's, there's, they're still there because you still have the, the other, uh, 30% or so. But then not only that, now that, um, you know, you have this AI, kind of like an unlimited AI Customer support agent, you can actually give customer support to a much broader base. So, so far, we only give it to our subscribers, the ones that pay for Duolingo. But if we can, you know, reduce the cost of customer support by 10 X or a hundred X, then we can actually give it to everyone. So everyone benefits from this.

AI assessment note: “AI can do 70 to 80% of the, of the tickets that we get.”

Q I've, I've seen many competitors. I remember I think you came pretty close to doing a random Practica. Um, how do you view the competitive landscape today and the distribution of value across it? Is it like Uber and Lyft where Uber takes 90% and everyone else takes 10% and that's duo and everyone else? Is it a much more distributed value kind of dispersion? What does that look like?

A In consumer, it's much more, uh, winner-take-all, kind of, like, Um, mechanics. So we've been now running this for 12 years. I would say every single year there was One or two companies that grow really fast in, in language learning, one or two, and inevitably they go down again. And what happened is they, they use venture money or some other funding to grow through paid acquisition. So they spend it on, you know, Facebook ads, Google ads, and, and they grow really fast. But ultimately, the only thing that matters in our space is retention. So if you, you can spend as much money as you want. If you don't have retention, it's a leaky bucket. All these people are going to leave. And so we're not, now we're seeing one of those. We're not scared because we've seen this play out a million times. So the one company I'm Scared of is one that has higher retention than Duolingo.

AI assessment note: “In consumer, it's much more, uh, winner-take-all, kind of, like, Um, mechanics.”

Q If you could go back to the early days of Duolingo and do something differently, what would you do?

A Oh, yeah. So I would say an, We're pretty open up this. I think that the two mistakes we've made at Duolingo, they were not fatal, but obviously the two big, big ones were one, we We waited too long with monetization. So we just didn't take it seriously enough early on. And we were like operating under this business model of let's just raise more money. Like our business, people ask, what's your business model? Like, oh, it's like venture capital. So, uh, and eventually we did take it seriously. And, you know, we're very good at, at this now. Um, you know, we have this internal, uh, concept of the green machine. It's like, You, you run a lot of experiments, see what works, and then double down, and that's what we did with monetization. It's, it's, you know, super good now, but we waited, we just waited for too long.

AI assessment note: “one, we We waited too long with monetization.”

Q Neil Mehta asks a brilliant question from Green Oaks. He likes to go into a company and ask the employees, are your best days ahead or behind you? When I ask you that for Duo, Why would you say they're ahead of you?

A Well, thanks for assuming that they're ahead of us. No, I think, I think, so, as I said, going back to the very beginning of the conversation, it's like, the mission is, is this, to provide the best education, making university successful, but now I think we can really do it. You know, we, we can really build an AI tutor that is as good as the best human AI tutors, and we can do it for everyone, not just the rich, like we can, and, and, and that will change the world. I, I really think this is the moment. Uh, it is, it is, uh, you know, we had to wait for 10 years to get to this stage. Um, but I think this AI wave is gonna, is gonna really enable us to do this.

AI assessment note: “I think this AI wave is gonna, is gonna really enable us to do this.”

Q What do you hear a lot from investors that you're like, pfft, really?

A So one thing they, they, um, it's not just investors. It's like, they ask this question, like, what is the secret sauce? Why is tooling so successful? Okay. And they look for an answer that is like a one word explanation. And, and you can tell when you give them the actual answer, they're disappointed. Okay. So they wish the answer was, you know, it's the, it's a streak. That's it. That's, that's the secret sauce. You know, it's this one mechanic. That's why we're so successful, and we, we invented it, or whatever. We, uh, maximize it, or whatever. Or it's, it's the leaderboards. Or it's like, oh, we're, we're so good at, you know, like, uh, uh, this particular market. It's like, it's, it's like, uh, Brazil. That's, that's a secret sauce. We figured out how to crack Brazil. And, you know, like, whatever. Like, they, they, they want an answer like that. And then, then Luis keeps telling them the true answer. Which is, it's running thousands and thousands of AP experiments. That's, and, you know, like the streak mechanic, for example, is like, you know, we added it, that was one experiment, but then we've probably run 300 experiments fine-tuning the streak mechanic, and that's where you get the gains in retention, and that's what ultimately drives the growth of Duolingo, which, you know, is still, it's super fast. We're still growing so fast, um, and, and, and, and, you know, it'…

AI assessment note: “they ask this question, like, what is the secret sauce?”

Q What do you think they saw that others didn't?

A Okay, so one thing that was true back then is we did the Silicon Valley Roadshow, and we actually did get a lot of interest from Silicon Valley VCs at a time, but they all made it a requirement To move to Silicon Valley. So we're like, you've got to move here, right? Like, when are you going to move here? That was the, always the question. Okay. We're super interested. We want to make a deal here and all, uh, when are you moving to, to SF or Bay area? And we're like, we're not moving. And, and then they immediately lost interest, by the way, some of them later invested in later rounds. So they, but they didn't want to be the first one in, in a non, Uh, Silicon Valley company, yeah. And so Unisquare, they didn't care. They said, in fact, they said, we're the only investor that doesn't care that you're in Pittsburgh. That's true. Things, you know, have changed completely. Like, I mean.

AI assessment note: “Unisquare, they didn't care. They said... we're the only investor that doesn't care”

Q one. Everyone has said, including Warren Buffett, the most important thing in life is partner selection. Yeah, like who you choose to spend your life with. I spoke to everyone on your cap table, and they all said about the special relationship you and Luis have. I'm intrigued. We mentioned marriage there. Whether it's Luis or marriage, I don't mind really. What would be your biggest advice on partner selection?

A So for, for founders, I think it's, it's a little easier. Um, Well, actually, okay, let me see. Let me see if I can answer this for, for founders. I think the one thing is like, have you worked previously with this person? And the key is worked, not, you know, hung out at the bar. I was like, have you worked with this person prior to starting this company? And Luis and I, we had worked together for two years prior. So we kind of already knew how we work. And, you know, what we're good at, what are we bad at? Priorities. And we actually had this very early on. We had this because I went, I went from, you know, his, uh, he was my superior. He was my PhD advisor and I was a student into an equal relationship, but we're equal co-founders. And I was like, can I trust this guy? What if he just fires me after a year? So, you know, we, we, I wrote down a little contract between the two of us. I still have a copy at home. Where we outline exactly like, you know, this is how we're going to make decisions. It's like this short. This is how we're going to make decisions. This is, you know, Luis's responsibility. This is Severance's responsibility. And, uh, and we both signed it. And, and I think that avoided a lot of conflict. So, uh, and the other thing that was interesting, I think a lot of the conflicts happened in the first two years.

AI assessment note: “have you worked previously with this person? And the key is worked”

Q fundamentally, they don't want to work, and AI is as good as they are at sales, at marketing, whether you're an SDR, whether you're a social media content creator for memes or GIFs or whatever it is. Like, actually, and so in 12 months, We're going to see huge, huge unemployment in this kind of lower level tech employee, and it's, it's going to happen. Do you think that's true?

A I'm not sure. So, so here's another thing. Um, so the, the general consensus is that these AI tools are really good for senior software engineers or staff level software engineers, right? So because they, they, um, you know, they have been trained and now they can use AI and they become much more productive. They're now a 10 X engineer. And, and then, uh, they can, you know, AI, the, the, the general consensus is that AI can do the, the job of an entry level engineer or an intern or something like that. Therefore, let's stop hiring entry level engineers and only hire senior engineers. But I think that's a mistake. And, and we're actually not doing that at Duolingo. We still hire from universities. And one of the reasons is That we believe that people that grow up with these tools and, you know, start using these tools early will become much better at using them. Um, so in a way, yes, they're, they're, uh.

AI assessment note: “I think that's a mistake. And, and we're actually not doing that at Duolingo.”

Q Neil Mehta asks a brilliant question from Green Oaks. He likes to go into a company and ask the employees, are your best days ahead or behind you? When I ask you that for Duo, Why would you say they're ahead of you?

A Well, thanks for assuming that they're ahead of us. No, I think, I think, so, as I said, going back to the very beginning of the conversation, it's like, the mission is, is this, to provide the best education, making university successful, but now I think we can really do it. You know, we, we can really build an AI tutor that is as good as the best human AI tutors, and we can do it for everyone, not just the rich, like we can, and, and, and that will change the world. I, I really think this is the moment. Uh, it is, it is, uh, you know, we had to wait for 10 years to get to this stage. Um, but I think this AI wave is gonna, is gonna really enable us to do this.

AI assessment note: “this AI wave is gonna really enable us to do this”

Q But by achieving your mission, are you not disrupting Harvard?

A Yeah. We would be. Um, but one thing on the second thing, so we have not solved the social thing, right? Okay. This is what higher ed does. We have not solved the social thing, but on the number two of the credentials, One thing we're also very passionate about is, is the Duolingo score. I'm not sure if you, if you notice it within Duolingo, but we now give you a score, like how good you are at, at this language. And, and we want this to become the way people talk about language proficiency. So it's like, Hey, my Duolingo score is 61 in French. And then people know, okay, I guess you can have a, this kind of conversation, et cetera, et cetera. So, so we're getting into the credentials or the degree aspect of it with this score. So.

AI assessment note: “Yeah. We would be. Um, but one thing on the second thing”

Q about the not having PMs, and actually, if all the biggest companies have them, maybe there's something to it. And it makes me think of this, like, conventional rules are conventional for a reason sometimes. What conventional rules do you find are conventional for a reason? One. And then what one conventional rule do you find, what the fuck? That, that's just the way it's been, but it's not right.

A Well, first of all, I would say you should always, not always, but every now and then, especially when there's a big technology shift like now, you should question all of these assumptions, right? Like, uh, you know, how you build companies. I think it, it just, we're in the middle of this massive transformation. You can build companies with probably 10 times fewer, uh, employees. Um, so, so that, that, that convention is, is, is, uh, you know, uh, completely You know, that you need to hire a lot of people. Um, that's one thing. Uh, I would say, you know, early on there was like uh,, ah, this is, this is going back. It's not a, not a great answer, but, um, there was like this, this belief that you can run a completely flat organization, right? No, no managers. Um, no hierarchies, no levels. And. I think that's not really possible. Uh, at a certain scale, you need to have managers or layers, or you need to have some sort of hierarchy. You need to have some sort of, um, career progression and stuff like that. I think that You know, the, the oldest organization in the world, the Catholic church that is still around, it's very hierarchical. And, and I think there's probably some, some reason for it. Yeah.

AI assessment note: “that you need to hire a lot of people... at a certain scale, you need to have managers”