Shishir Mehrotra on Building Tools Creators Love

24:41He lays out three charts and the first chart says, this is how much money YouTube is losing per year. And it was like about to be about a billion dollars a year. We paid 1.6 for the company. We're about to spend another billion every year for the same company. That's not what we had in mind. Second chart said, here's how much money YouTube loses per view. And it wasn't quite a penny per view, but it's very close to a penny per view. And the third chart said, here's what views are doing. And it wasn't like a little bit up. it was like a fricking rocket ship. Like the only chart that was a rocket ship was the viewership.

25:13He's like, this is the worst business on the planet. I mean, this is on path to lose tens of billions of dollars. Like this is never gonna go anywhere. And his question was, who else were the buyers? Can we sell it to them? Because this thing is never gonna work. And that was like, so our first two years was under deep scrutiny. We were broadly known as Google's first mistake. How was YouTube spending a billion dollars? So what was it going to? Networking and music. and the licensing. The licensing. I mean, YouTube is still, besides Spotify, depending on the year, the number one or number two biggest revenue source for the music labels.

25:49That's, and then networking wise, I mean, we were, by the time I left, we were 20 % of the bits on the internet. And so, you know, that was like a lot of networking. And so in between those two things, it's just a lot of money. Now, it turned out like all of that was very correctable. And thankfully, like I'm sitting in this meeting with Patrick and Eric Schmidt Eric was the CEO at the time. And Eric said, don't worry, Patrick, we're sending in the right team. They haven't really tried to turn into a business yet. They've been focused on user growth. Give them a chance. And thankfully we turned it around and within two years we were profitable and reasonable business and so on.

26:21But I spent almost six years there and the first two, three years was like on knife's edge of like every quarterly meeting. We basically, we had board meetings and every board meeting was like, not only like we might get fired, but like the whole thing was gonna get shut down. Like it wasn't a, it was like not a, there was a contingent of Google executives that thought this was a total lost cause. It's never gonna work. It's funny hearing you say that, it reminds me of another business today. Like that story is kind of ChatGPT. Yeah, I mean, I don't - The part of usage going bananas. Yeah. And spend going bananas.

27:02Yes, but I don't think it's, I think ChatGPT is a much better business because it wasn't, in our case, it was marginal spend per view was negative. That's not really the situation ChatGPT in. If ChatGPT stopped training new models, it'd be very profitable. They've got Ford R &D, thankfully for the rest of us who are deep, deep users of their product, where they're investing on our behalf. That's where billions of dollars is going. The actual ChatGPT product is quite profitable. Yeah, I don't think that's a I think it's a pretty good business. Yeah, that's fair. But like, if you look at, I don't know, let's just say Gemini versus ChatGPT today in the Google world.

27:40Yeah. You know, like Google can just spend unlimited. Basically, they can just spend. It's interesting you say that because I think there's this perception that at these bigger companies, they have this big war chest and you can just spend it. And I don't actually think that the dynamics of the companies allow that to happen as much as you would think, because everything gets managed through a P &L. And like you're publicly reporting and your incentives inside that company are not really to go take that money and spend it that way. I think it's one of the reasons why what Zuck is doing right now and bringing in talent is like a real founder move.

28:12Like that's a thing a founder can do and say, you know, I'm going to kind of ignore a little bit what the street is going to say. I'm going to think in like startup terms. I mean, he's not thinking in dollar terms. And she's thinking like, you know, I'm going to spend 1 % of meta on building the AI future. Like, do I increase the chances of meta succeeding by more than 1 %? I should do it. That's like, that's a founder way of thinking. Yeah. But if you're inside a big company, you own a budget. Like, you don't just get to go deploy$100 billion from somewhere else on a whim. Like, it's not like slush money that goes around.

28:45Like, that's all committed. It's showing up in an earnings report. The CFO has got a hold on it. It's fun. Like my whole time at YouTube, every first two years or so we were unprofitable and it was get profitable as fast as possible. After that, every dollar we made, 25 cents went to corporate and 75 cents we could spend. Like I effectively had a dividend at Google. And so like the idea of like, oh, YouTube has infinite money. They can spend whatever they like. Yeah, Google had a cash pile over there, but I had to go ask for investment from it. It's just like raising money from a venture capitalist, except like the irony of it is, the analogy I would give people is if you're an entrepreneur inside a Silicon Valley tech company versus you're a business owner inside of Google, inside of Google, like everybody around you can say no and only one person can say yes.

29:32Like Sundar can say yes, everybody else can say no. If you're Sam Altman, everybody can say yes and nobody can say no. Like he can just keep hunting for whoever can give him money. So I actually think it's easier to raise capital when you're not in that environment. Interesting. Than when you are. So let me ask you this. If Sundar, Larry, and Sergey all decide that the future of our business hinges on AI. Then they can do it. Absolutely. Then they can just reappropriate the money printing machine from search and ads. They can. Into Gemini. Yeah. I mean, practically speaking, and there are examples, I mean, Satya is, I think, a good example of, I think he made some bets.

30:12He's not the founder, but he made bets like a founder did. But my point is that, yes, the CEO can do that. The founder can often do it in a way that is like a little bit more forward-looking than sometimes even the CEO can. But the CEO can almost always do it. But if you're the business owner, you're asking me earlier, as a YouTube owner, could I make product decisions on YouTube? Absolutely. Could I spend the YouTube P &L however we wanted? Absolutely. I could decide, this is going on networking, this is going on music, all the music deals. We didn't have to get any approvals. We did it all ourselves and any content deals we want to do, we do ourselves.

30:47But if I wanted money out of the Google cash balance, that's like going and getting an investment decision. And, you know, what Mamoun is here is what Sundar is there. Right. Except there's only one of them. You don't get to go down the street to, you know, to read Hoffman or Hamon. You don't get to do that. There's only one of them. And they say, no, that's it. You go back with the tail between your legs. It's like there's no, you come back next quarter and you try again. And so that's the dynamic that prevents actually using that cash pile. And so it's not. And I think it's one of the reasons why some of the big innovations, even big spending innovations happen outside those companies.

31:24That's a very interesting perspective. And then maybe can I press on that a little bit more on you had mentioned on the reels and shorts. Yeah. There is a data deficiency, like meaning you need enough data to continue to power more of these things. And so let me pull that like analogy forward into the current, the current world where again, we're like kind of living in a world where data and money seem to be the fuel that powers these models. Like, how do you think about the data that Google has that YouTube is producing versus like Sam can't, turns out like Sam probably can't just go buy a bunch of data.

31:58Like we've already scraped all the data in many ways. Like, how do you think about that? By the way, when we say data, I think there's, there is like data at rest, like scrape all the data. Yeah. And then there's activity data. But like when you think about things like recommendations, things like improving these models, like it's the activity data that is actually the hardest to get. It's the, like what makes TikTok's algorithm good is the fact that they have, you know, hundreds of millions of people doing it every single day. And it's not really static analysis of the videos that they do a good job of.

32:27It's that they just have a lot of people that are constantly swiping and unswiping. And that's great signals. YouTube, same thing. And it wasn't really that we had great intrinsic knowledge of every video. We actually had very little knowledge of what was in the videos, but we had a lot of data of watching people swap back and forth between them. And so I think to go to other markets, I'd say, you know, if you look at ChatGPT, like what's its biggest advantage is that it has a lot of usage. And yeah, it's got the same index of the web that everybody else does now. And maybe they can like on the margins, like one of them is going to scrape books better or scrape magazines better or whatever.

33:03but the biggest advantage is like they have this usage data. I think like self-driving cars, same thing. Or what's the difference? Why is there so much question or enthusiasm of like, can Tesla do better than Waymo? And so on, it's like, Tesla has a lot of cars on the road. Yeah, they get that data for free. They get that data. Like, yeah, can you map every street? Yeah, lots of people can map every street. But can, like, how do you get people actually like jerking to the right on the pothole on that street? Like, how do you get that? Like, you just have to watch people do it. And so that kind of data requires scale.

33:33And that's an economy of scale business that I think many of these businesses need. You just need, I mean, if I go to my current business, grammarly, like we, like the rules of grammar are the same. Everybody has that. What do we have? We have 40 million people a day accepting and rejecting our suggestions. And so we - I'm one of them. Yeah. So we're reading what they're doing. We're able to make a suggestion. We watch really closely when we make good ones and bad ones. We now do way more than grammar and we're about to do a lot more than that. But like our biggest advantage is we're just in front of people in a lot of places.

34:02And so our data moat isn't a static data moat, it's a living data moat. And I think that's really what powers the long-term gap between these businesses. Okay, it's interesting you say that because on the ChatGPT point that you're making, it's basically like, hey, there's a bunch of activity that's continuing to power these models. And as ChatGPT grows, the activity grows. But in the Google world of this, they're like, great, activity is great, but context also matters. And so what Google would say is like, actually what people really want from these models now is actually to know more about like, go to my calendar, go to my email, like go read all these things instead of me training my chat GPT, just Google, go pull it for me.

34:42And the context plus whatever, if the world stopped and there was no more new models, the context plus that model would probably be their, what their argument would be of like maybe their future. I mean, I think, yes. I don't know, I have no idea. I mean, and this is a part of the Grammarly strategy as well. I mean, a big piece of what we're doing is we're turning Grammarly from being a single agent to a grammar agent to being a platform for agents. Or one of our superpowers is we have a lot of context. We're with you in 500 ,000 different applications, whether you're on a website, a desktop app, a mobile app.

35:13We're watching what you're doing. We're able to make suggestions on that behalf. We don't make good usage of that today. Like today, we only run this. The analogy we use is like we built an AI superhighway to bring AI right to where you work. And we only run one car on that highway today. That's your high school grammar teacher. And so we're going to broaden that. And a lot of that, what we're bringing is context. And so now when you're sitting and writing an email in whatever tool you're using to write an email, we'll recognize that, oh, you're writing an email to a customer. We have one agent that can help you with your grammar and another one that has read everything in your Salesforce and can tell you this customer has an active opportunity.

35:55It's read everything in your support system and says, this customer had an outage last week. You should probably respond to that or has read all your product marketing materials and says, like, the thing this person is asking for, that feature is actually shipping next week or whatever it is. And so I think that context really matters. So we're a big believer in that, too. The context of where you are, what you know, what you're trying to do, all that really matters. I'd say in terms of your question of, like, does ChatGPT or Google or Microsoft have a particularly good moat that way? I don't think so because I think the number of applications people use is much wider than people think.

36:28I mean, I'll give you a fun stat. I took over as CEO of Grammarly six months ago. One of the early reports I got was our SaaS list. How many applications do we pay for? Take a guess. How many applications do you think we pay for at Grammarly? I'm going to say like, I don't know, 150? Okay, 972. So that's my list. And how many of them come from Google? Five? How many come from Microsoft? Maybe five. And so the idea that that's a massive moat, it's like, yeah, there's a moat there. Those data sources are mostly available. We can talk separately about what Slack's doing right now, but I think for the most part, you can get to those things.

37:12But people just work on a lot of tools. And so that context moat, I do agree it exists, but I'm not sure that Google or Microsoft or someone have any particular advantage on it. very interesting. And you got, you started Coda. You spent eight or nine years building that business. How would you like in 20 seconds or less, how would you have described, how do you describe Coda? Oh, Coda's all in one doc. Doc. Okay. Yeah. Like a very rich, extensible, robust way to build documents basically. I mean, I think the core observation of Coda was that the world has fragmented productivity into documents, spreadsheets, presentations, applications, so on.

37:57And we thought that was silly. People don't think that way. People naturally want to solve a problem. I'm trying to run a podcast. I'm trying to do an investment. Like I don't want to think about my, my list of podcast guests goes over here and my scripts go over here. And so that's not, I just don't think that's how people think. And so we built an all-in-one doc. And I'm curious, like in hindsight, using this safari versus zoo analogy, did you think, because when I said that earlier, you maybe shrugged. I'm curious. Did you think that Coda ran into the safari versus zoo thing? Yeah, for sure.

38:29I mean, Coda is a safari. That is, I mean, that's the value proposition. It's like you're, you feel like if you use the office suite and you're sort of in that dated world, like I watch people do it. They'll have this moment in the meeting where they're like, okay, somebody should write this down. What are we going to do? Are we going to do it as a document? Is it a presentation? Is it a spreadsheet? And you go through this like complicated algorithm in your head of which one is it going to be? And if you're a Coda user, you just don't do that. Just create the doc and you can do all of it in one place.

38:58And yeah, it's a Safari. Like sometimes the needs change as you go. It's absolutely the right analogy. It's like the media analogy is a little bit different, but I think it's how I would think about Coda. And then was there an equivalent tension of zoos that were popping up around Coda of the Spotify's? Yeah, I mean, like Airtable is a good example. I think Coda, like one of our biggest strengths at Coda is like we have a very robust, we call them tables, but some people would call them databases. And it's very robust and it gets compared all the time with like a company that's only focused on that is Airtable.

39:31Or if you look at our document surface and sometimes gets described a little bit like a wiki, that gets compared a lot with products like Notion. And so like we ended up in lots of comparisons. It's kind of like the best parts of that product and this product put together in one Safari. And yeah, I think that's like a reasonable tension as you build up that product. Yeah. And then obviously maybe just by revenue, Grammarly was a much bigger business. Yeah, for sure. Right? Yeah. They acquired Coda. And you. And then you became the CEO. Yeah. Like it was eight years of Coda? Ten. Ten years of Coda.

40:11And like, I don't know. I'm really curious what you say. Like if you were evaluating, like when you were evaluating, like, all right, now what? Like I've been doing this for 10 years. Um, like why do I do it this way? What's the, what's the future of this business? Yeah. Like, did you want to keep chipping away at it? Was 10 years, like 10 years is enough of anything. Like, I mean, so this was definitely not that because still running it. So there was, this was like, I think it was an independent thing. Yeah. But it's like, I think there are people that they asked me, it's like, was it an exit?

40:43And it's like, actually, I don't think it doesn't feel like an exit. Yeah, it's not not not the way I would describe it. The story is I was in the middle of raising our next financing around for Coda. I had written a memo for that. The title of the memo was the future of the AI native productivity suite. And that was my investment memo. This is what this is what we're building. I had a term sheet for that. I brought it to some of my other investors. And one of them happened to be Hemant Taneja, who sits on the board of Grammarly. And so I said, Hey month, would you, would you like to participate and invest?

41:15And he said, before you do anything, why don't you go talk to the Grammarly co-founders? I think you might find it interesting. And my first reaction was why? And I didn't like, I think like most people I had put Grammarly in this like different bucket in my head. And he said, just trust me, it'll be more interesting than you think. And so I flew up to Vancouver. That's where Max and Alex, the co-founders of Grammarly live. Some of the Grammarly executive team joined. And we started by going through both of our vision memos and they pull out theirs. And it has the exact same title. And it says, the AI Native Productivity Suite.

41:45Now, they had a completely different view of how to think about it. And they sort of walked me through, you know, hey, Grammarly is a much bigger business than people think. You know, about$700 million in revenue. And it's like 40 million daily active users. And it's a big business, right? It's a very, very underestimated business. And they said, but the thing that people most misunderstand about the business is what the core technology is. And they walked me through this view of how they thought about the AI Native Productivity Suite and how it basically, by the end of the day, it felt like this obvious mesh.

42:16And there are basically two parts to what we talk about here. One is the future of agents. And I kind of give a little bit of a preview of this, but the idea is most people think the technology of Grammarly has to do with grammar. And there is a big technology base that's about grammar. But actually, the hard tech of Grammarly is about running AI right where users work. And so this is, sometimes we refer to it as a ubiquity layer or the AI superhighway. And the idea is that we can take about 500 ,000 or so websites, applications, desktop and mobile, and we can read whatever you're doing. We can annotate it in a way that's unobtrusive and we can make changes on your behalf anywhere.

42:52Whether it's Gmail, Slack, Salesforce, you know, you could be in Apple Notes, you can be on Twitter, you can wherever. we can help you right where you are in the applications you're already working in. And the analogy was that we did all that work to do all those integrations. And we're only, you know, we've got this highway with AI running right to the user. We're only running one car on that highway, this grammar car. And we started looking at Coda and Coda was building, we were building our new product is called Coda Brain, which was basically taking the Coda integration layer. One of Coda's hallmarks is these deep integrations into many different tools.

43:26about eight or 9 ,000 of these integrations, we call them PAX. And we were turning them into agents. And they said, what if you could immediately fill this highway with agents that could read all your context from whether that's Salesforce or Gmail or wherever else, could take action in all these different tools and so on. So that became one part of it. And the second part of it was, Grammarly is kind of like, one description is it's kind of like a moat without a castle. It works everywhere, but it lacks a great destination in the center of it. And the best analogy from my experience with this was, it's kind of like imagine if YouTube had YouTube embeds, but no YouTube.com.

44:06Kind of what Grammarly feels like. And it turns out it's not quite true. Grammarly has an editor that it's a little hard to find, but it drives a massive amount of our business. It's actually where people get the best experience with agents. So, you know, Grammarly works everywhere you work, but it turns out that when you come to Grammarly's own editor, you have an even better experience and you're much, much more likely to upgrade, much more likely to retain and so on. And we had zero engineers, the Grammarly team had zero engineers working on it. So they were going through this observation of like, maybe we should go build a document surface and we should really rethink.

44:40And if we're going to build a document surface, we should probably think about the rest of the suite. and what if we were really not just a place where people came in and out of it, but really built their knowledge base and they're out looking at all the different products in that space. And there's only a couple of products that have meaningfully innovated in the document category. And I said, well, we're going to go build this. And so by the end of the day, I was looking at it. It's like, well, the two of us are going to end up building very similar things. And now it's a question of would I rather do it on my own or do it as part of a bigger company with a much bigger distribution footprint, um great capital base and you know it just seemed kind of obvious and at some point in the conversation max and alex said and by the way we think it'd be good to have a founder as a leader and neither of them wanted to be the ceo and said would you be up for it i thought that would be i thought that would be a lot of fun so i kind of viewed it as we're going to keep running we're building roughly what we were planning to build with some with some twists and changes but with just a much bigger set of users and distribution and infrastructure and capital and so on.

45:42So I view it as, you know, the next wave of what we're doing with Coda. And then boom, you raise a billion dollars. And then the first thing that you go do is buy another car for the highway, right? Well, in that analogy, you bought another castle, actually. You bought another castle. Yeah, yeah. So if Coda was a castle in that moat, then Superhuman is... Yeah, so the email client. The email client, yeah. And by the way, the raise a billion dollars. And I think it's very important to have capital to be able to, you know, we talked about earlier this idea of there's many moat, but capital is a moat.

46:14And it's an important one to be able to make real moves. and then we went looking for, okay, we're ready to finish, you know, start building out the rest of the suite and one of the most interesting pieces of data we got was if you think of Grammarly, the number one use case of Grammarly is actually email. 17 % of words written in Grammarly That's probably where I use it most. Yeah. And it's three of our top 10 applications are Gmail, Outlook, Web, and Outlook Desktop. And so we started looking at that and saying, what would you do if you built an agentic email experience. And kind of like we're looking at the world of Coda as if you think about the world of productivity, I kind of think of it as there's one half of it that's work artifacts.

46:59You build a thing that we can talk about, we can collaborate on and so on. And there's a world of communication because this is how we talk to each other. Some of that's messaging, email, video, so on. And I think of that as slightly different kind of software that people build. And so we started looking over there and saying, well, email is a dominant use case for us. So that seemed like an obvious place to go look. And then I sat down with Rahul Vora, the CEO of Superhuman, and we just did a big design sprint. And we said, what would we do if we were doing this together? And it was just crazy exciting.

47:31We just walked out, here's the next, I think we have over years of roadmap of what we know we want to build. And it was just really obvious. If we did this together, we could build an email experience that, I mean, the existing Superhuman email experience is amazing. I've been a customer user for eight years now. I have a inbox zero streak that they told me is at 144 weeks. So it's two and a half years of getting inbox zero with superhuman. But I think what we're about to build with it is much more exciting even than what's there today. And it just became an obvious one to build as the next part of the suite.

48:06And this word agents, you've used it a few times. I probably hear it like more than any other word right now. It takes different meanings depending on who's saying it. How do you define it? Yeah, so we sort of, it's a big debate and we've narrowed it down to saying agents have four characteristics and we use that as our frame of where do we put agents? So agents have knowledge, agents have skills, agents have assignments and agents have, sometimes we call it a soul. um and so maybe i can break those apart so and you can sort of map it to humans if you think about i'm hiring new teammates what what what do they have so they know a certain set of things that's like the facts they know sometimes those are global facts like i know all the rules of the english grammar language that's grammarly sometimes that they're team level facts i've read our support knowledge base i've read um uh you know our wiki uh and sometimes they're they're personal facts They're like, I've read my email and so on.

49:11So they have knowledge. They have skills. There's things they can do. Skills, there's a bunch of different things. Sometimes skills are as simple as answer questions. That's how I think about ChatGPT is like it's number one skills. It can answer questions for me. Sometimes it can assist me. And it's the thing that Grammarly is really good at is it can come along and it can sort of nudge me and say, hey, I'm going to underline this thing and mark it up for you and nudge you in a certain ways. Sometimes I can do action on my behalf. I can actually send an email. I can actually go update a record or whatever it might be.

49:39This is the thing that Coda Packs have been really good at. So you have knowledge, you have skills, you have assignments. I think those are the jobs where you say, every time this happens, I want you to do this. And like, if you think about Grammarly and that vantage point, you say Grammarly is an agent that says, every time I type a character, I'd like you to read what I did, look at the rules of English grammar, and go and underline with a set of rules. In red for anything that I absolutely must change, and blue for anything that looks like suggestions. And then finally, you have, sometimes we call the soul, which I think, you know, a technical term for it would be the system prompt.

50:12And it says, this is how I want you to behave. You are the grammar agent. Your job is to help this person. I want you to be super factual. You are, you are this person's personal assistant. I want you to be collaborative. I want you to be funny, whatever, whatever it might be. If you take those four things, I think you could use that to describe a person and you can say, what is, what is this person? Well, they have kind of a personality. They know a set of things. They have a set of skills and I give them a set of jobs. And those jobs can be short running, long running, you know, short running isn't like, can you, can you get me a coffee?

50:39Long running isn't like, can you help me build this product? But you take those four things, I think it defines that space of agents pretty broadly. And Grammarly fits very neatly in that. Like it's a particular type of knowledge and skills. Everything we're building with Coda and PAX and so on is a different sort of different type of agents. But I'm really excited about ones that really broaden that definition of agents. Maybe I'll give you an example that was really fun for me. I was, and I don't know yet if they'll end up building this agent, but I'd be really excited if they did. I was talking with the Duolingo team and the, Luis and his team, and we were just brainstorming on what a Duolingo agent would feel like.

51:19And I just thought it was a really interesting example. And, you know, he sort of heard me out on what's the AI superhighway and agents running on it and so on. But I still don't quite get it. Like, tell me what a Duolingo agent is. And so we started brainstorming and said, well, from a user of Duolingo, I'd say, well, one thing I want is I want an agent that if it could be with me everywhere, what's the first thing I want it to do? First thing I think I want it to do is maintain my streak even when I'm not using the app. I want it to say, hey, I'm trying to learn Spanish. Like, if I read three articles in Spanish, maintain my streak.

51:50That seems like a positive thing. If I could have my coach with me all the time, they wouldn't require me to go to tutorial today if I clearly have done my work in a different way. Maybe the next thing I want to do is like actually reach into my applications and work with me. So I want it to come in and I want that Duolingo agent to come and say, hey, you're reading in Spanish. Let me go and pre-translate the words you don't know so you can you can actually read everything. Or maybe you're reading in English. Maybe I should like build a little quiz for you. And I'm just going to replace words that you should know with Spanish and just kind of force you to do it in your natural rhythm of your work.

52:24So you can learn as you go. And then maybe the third thing is now I come back to the Duolingo application and maybe it knows something about what I was trying to do. So it says, hey, today I'm not going to give you a lesson on how to order food at a restaurant. Today I'm going to give you a lesson on how to how to respond to questions on a podcast. I'm going to give you a lesson on how to buy a car. That's clearly what you're trying to do. So, and in this way, I think, so there's one definition of agents, knowledge, skills, assignments, and soul. That's like one way to think about it. But another way to think about it is we're going to take every application person on the planet and we're just going to enable them to act like a human right next to you and Duolingo.

52:57Like if you put it in this terminology of like, think about language learning over the years. Language learning used to be, you'd have to go to a tutor and Duolingo like made this massive shift of like the tutor is in your pocket. You don't have to go. You can do it right before you go to bed. You can do it anytime you like. And I think the next step is actually, it's not just in your pocket. It's with you everywhere you are. And it's like you took your your physical language teacher and had them work next to you. What other applications does that make sense for? Like we spend a lot of time on students and, you know, Grammar is obviously a very popular product with students.

53:28You know, professors went from lecturing to posting videos of lectures. That was a YouTube generation. or what if professors were like sitting next to you while you worked? What if they were marking up your work? You didn't turn in your work and wait for a grade. They were sitting next to you. I'm not going to wait for you to fail and give you an F. I'm going to tell you now you're on the wrong track. You should read this thing. You should do this a little differently. And now I can, I can help you get there. What if every application felt like that way? So, so anyways, there's a technical definition of agents, you know, knowledge, skills, assignments, and soul.

53:56But then I think there's a, a little bit more of a conceptual definition, which is we're building digital humans and they, and they, they can represent, sometimes they represent other humans, sometimes they represent an application. They can represent lots of different things. That's what we're bringing with us. But Shishir, given those examples, like I think about what Grammarly has been for me over the past seven, eight years. Yeah. Like it's been that, like it's been that. Yeah. So like I, I, those four checkboxes, like I would consider Grammarly that person. Yes. That's looking over my shoulder, like on my Gmail.

54:30Your grammar, your grammar teacher that's looking over your shoulder. Exactly. Yeah. So now, now imagine it wasn't just your grammar teacher, but like, are we just now putting a fancy name on it? Because like this, that's kind of my point is like, we've, this has been around, like, this is a thing. Do you see what I'm scratching at? I do. And I think the word agent in particular, I've, I've now been through multiple generations of the, of the web. And, you know, we've been through, we've been through many different periods of, of labeling. And so I think I've, I've learned to be a little bit immune to the label may change, but the concept is what matters.

55:02Like the word agent, I don't love the word agent. I mean, to be honest, when people say agent, I picture the matrix. I was just going to say the same thing. And so it's like, that's not a positive association. So I don't love it. But on the flip side, I think it just helps us have vernacular as we're going through things. And so we've had a lot of debate about, should we use that word? Should we use a different word? We've got a stack of other words we could use. So I don't know that it should be a user-facing word. But when we're talking on a podcast, we're trying to understand the concept, I think it's helpful for, you're going to have a lot of guests that talk about agents or some concept of it.

55:32And I think it's helpful. And I think there was like, if I go back for a while, it was pretty unclear what an app was. But is the core assumption of an agent is that it's doing the work for you rather than next to you? No, I mean, I would argue that being assistive is like Grammarly is like next to you. Yes. And you would definitely consider that an agent. I would definitely consider an agent. Meaning I don't have to, like, you don't think that the, in order to definitially accept agent, it could. It has to be autonomous. Yes. No, you don't think so? No, I don't think so at all. Okay. And I, you know, that's, that's what I mean by, I think it can be reactive.

56:08It can be like, I have to ask it a question. ChatGPT serves as an agent. It can be assistive. It sits next to me and it, and it, and it works alongside me, or it can be something I can delegate to. I think it's easy to think of agent as digital humans. and from that perspective, I'd say, you know, is a human only useful if they are like fully automatable and you can never talk to them? It's like, no. Sometimes you want a whiteboard right next to the human. Sometimes you're going to sit down and script this out and you want to script out together. Sometimes you want to ask a question and you want to have a discussion like this.

56:38I don't think, I think defining agents as only ones that can go away is incorrect. I do think it's a magical thing that we can do with a lot of these agents. We can give them assignments that are long running, you know, open-ended, so on, because they're digital humans. That's how I. And is what you want to basically go after is just try and tee off on the Microsoft suite? Like, is this the fullest expression of actually trying to go after it all the way? I don't, I think the Microsoft suite is a, is a great suite. I've, my first job was working on Outlook. I love the suite. I think that definition of suite is about to dramatically change.

57:12And when people think about the tools you need in a productivity suite, you know, I, I just, I think that set of tools, I sort of refer to them as plumbing. You need them. I think it's rare to have a company that hasn't picked. You have a Microsoft suite or a Google suite or so on. But I kind of think of it like having desk chairs. That's an important thing, but it's actually not the productivity layer that you spend real time in. And so in my mind, we are building an AI native productivity suite, but I think it's sort of at a different layer than that. I think people that use products like Coda or Grammarly or Superhuman or so on, they view it as the next layer of what you do when you're really ready for that next level of productivity for yourself or for your team.

57:54Yeah. And I mean, I don't know if you can answer this or not, but like, is the idea to go buy a bunch, build or buy a bunch more cars that race down the highway? I mean, it could. I mean, I think we'll - What areas would those be? Potentially. Yeah. I mean, I don't think - You don't have to answer if you don't want to. No, no, I think it's fine. I think we are, it's clear to me that there's an opportunity to redefine that productivity suite. Slides. There's a wide set of things that we get asked about that we don't have today that we, or maybe people work around with one of our tools, work artifacts, communication tools.

58:28There's actual, you know, what would people define a SaaS application that we get used for a lot. So many of those are possible. I think in many cases we can do it in a great partnership. I don't think we have to buy everything. I think there are certain agents or surfaces that are somewhat fundamental. And I think in those cases, the type of innovation we're going to do is only possible if we are deeply embedded with them. And that's why I chose with Superhuman to take an approach of buying rather than just partnering. But I think we're going to do a lot that is partnering and helping people build cars on the highway.

59:06And I think that's a, I think it's gonna be really powerful. And by the way, it's pretty unexpected. I don't think people think about Grammarly that way. I mean, you think about Grammarly as a tool that does one job. And the idea of thinking about Grammarly as a platform that can run many agents is I think it's gonna be a big mind shift for everybody. It's gonna be really exciting. When you think about making that transition successfully, like if you were to write a memo taking the other side, like if you were to steelman how this is wrong, which I'm sure you've done in 10 ways to one, Yeah. I like premortems.

59:37Is there a way, like, how do you still man that? I mean, the risk, anytime you take a product from a product to a platform, there's a risk. And I think the, I've worked on enough platforms now to have seen both sides of that. Because one of the things to remember about platforms is to be a good platform builder, you have to let go. Yeah. Like if I go back to my YouTube time, YouTube was clearly a platform. What did that mean? it meant that people like the positive way of looking as other people could build their businesses on top of YouTube. Sal Khan built Khan Academy, Mr. Beast built the Beast Corporation.

1:00:11Like, I think these were massive things that people could build on our platform. It also meant that we didn't have control over them. We couldn't say that's a good idea. That's a bad idea. And I think you have to be willing to let go and kind of adopt that. It also means that you, your ability to focus that community is a little bit up to the community. You know, you asked earlier about how YouTube evolved and where the zoo and safari and so on. Like the fact that music and gaming took off on YouTube was no choice of YouTube. Totally. It just happened. And, you know, we could like steer all we wanted.

1:00:49We tried really hard. How come people aren't building Netflix style shows on YouTube? It's like, I don't know. They want to build music and they want to build gaming. And I was like, that's what they're going to do. And then like, we would have to push to do that. And so I think when you take a product like Grammarly, that is like very clear, single purpose, ICP, you know exactly what you're hiring it for. And you say, I'm going to turn it into a platform where Grammarly is one agent, it's one car on that highway. You'll know why you're hiring that car, but what about all the other ones? You're going to take some risks.

1:01:17Some people are going to build ones that are exactly what you wanted. Some people are going to build ones that are a little bit different. I think that's really exciting. And I mean, it's sort of similar to, you know, what happens as the phones opened up applications and some of the applications we could immediately picture. And many of them we couldn't. And like, I get excited by that. Like there's going to be a bunch of things people are going to build here that are just completely different than what we've pictured so far. I gave a few examples of what I can picture, but you know, there's going to be so many that we haven't pictured.

1:01:41But you lose a little bit of control and you have to be okay with that. Yeah. Maybe like a side random question. Yeah. Is like, you strike me as someone that's very ambitious. That ambition seemingly feels limitless. You started a company, then you start another company, then you go and you're building companies within companies, then you get acquired and then you become the CEO of that company. Like, do you ever feel like a victim of your ambition? Meaning like if you were to talk to, if I talk to your wife, would she say like, like, is there a trade-off to ambition? I am, I'm a pretty hardworking person.

1:02:11So there's clearly a trade-off there. I think the, I don't think that's a, we're on the grit podcast. We can talk about grit, but I think that, I think the, I don't think you can do it without being pretty focused, without being all in on something. So when you have an idea, you jump all the way into it. And there's a trade-off there. I would say, I think I bring that same trade-off to other aspects of my life. So I don't think it, I think it's an attitude that isn't like work versus personal. Like I think, I think, you know, I have two wonderful daughters. I have a wonderful wife. I think my daughter's got that same treatment for, I view them as products.

1:02:46I view them as, you know, part of my job was to turn them into great adults. I think they're pretty good examples of that so far. But you had the option, let's just say, of potentially selling Coda. Maybe it's to Grammarly or somebody else. You could have raised money, but you could have also just sold it and done something. I don't know. You're like, money is not an object. It hasn't been for a while, right? Like, hang out with, I don't know how old you're, like, you don't hang out with this. I don't know. I don't know. That's like, is there a trade-off there? Like, do those things even cross your mind?

1:03:17You know, it's, I think what my wife would say is I would be bored to hell. Like, I just don't, I don't see the world that way. Like, I think, I think of it as, you know, when you have, let me back up a minute. When we, when we started Centrada, Vinod made this requirement, you had to move from Boston to Silicon Valley. And at the time, I was very resistant. And I was in Boston, Boston seemed like a great place, seemed like there's no need to come to Silicon Valley. And since then, I've become a complete Silicon Valley convert. And I think that this is not really about the place. I think actually, there are many places now that feel like Silicon Valley.

1:03:52But one of the characteristics I always tell people at Silicon Valley is this is one place in the world where if you're sitting down at a barbecue or dinner conversation, so on, and somebody starts complaining about something. Somebody says, how come this car doesn't work that way? How come nobody built a product that does X? How come somebody in that room is thinking, you're right, I bet I can fix it. And it's like a very can-do place. It's a, this is just a group of people that are, surrounded by can-do people. It's like, why can't cars drive themselves? It seems like we can't. Why can't we just send things up to space and bring them down and not have them crash?

1:04:31Like, why not? It seems like any of that's possible. And I think once you get that bug, that's not like a, I don't view it as a job. I view it as an attitude. And so you don't really stop. I mean, it's not like, you know, so what would I do? I'd pick a different idea and go see how we could make that happen. And this idea is, I think, more interesting, more central. Like I think we get to redefine how digital humans work for the world. I think that's, I think that's, I think that's worth doing. I think it's a lot of fun. Um, so I didn't, yes, there's a trade-off obviously, but I didn't, I don't think of it as like, for me, the idea of like sitting at home and I don't actually know what I would do and not working on any ideas wasn't really the choice.

1:05:13So it was really a choice of like, is it this idea or a different idea? Um, right. But you could have, let's just say in a bizarro world, sold to Microsoft and been, I don't know, fifth from the top or whatever. And the weight of the world is not on your shoulders. Like you have, you know, it's like a different type of lifestyle. You're vesting. But that's another question is, is, is being CEO Grammarly versus running the YouTube team or so on. Are they like dramatically different in terms of, you know, pressure responsibility, so on. I think they're just different flavors, the same thing. I don't think they're that different.

1:05:44Um, so, you know, I think you think the pressure felt the same of like being, I don't know. There's different pressure. There's a, it's definitely different. And it comes down to this, like yes versus no culture. Like when you're, when you're the, when you're a senior executive at a big company, you're surrounded by people who can say no. And there's only one person who can say yes. And that creates a certain dynamic, but it's a lot of pressure. I mean, I definitely worked hard. I mean, there's no, there's no question. And there's also just a lot of overhead and there's a lot of like things you have to do to be part of this company.

1:06:14and when you're running a company to CEO, like it's this nice part of like, nobody can really say no to you and a lot of people can say yes. But at the same time, like nobody can like anoint you and nobody can like guarantee a path the way it can happen in a big company. It's kind of on your own that way. It's just different pressure. But I don't think it's, I think you're, and usually what drives your pressure is the pressure to see an idea succeed. And so while I was running YouTube, it was, oh crap, like if we don't do a good job at this, like my fear was this product was, like I was saying, the viewership was through the roof.

1:06:50Like my fear was if we don't turn this into a real business, product is going to disappear. And there were, even at the time, there were already a billion people using it. It's like, that would suck. Like what a letdown for the world. And I think that like, that's the pressure. The rest of it is like, yeah, can Eric and Larry and Sergey, can they like relieve or like cause more pressure, less pressure? I think it's kind of a wash. You have a board of directors and so on. It's like the pressure comes from, you know, when you have a mission you're trying to achieve and you're close, like the pressure is like, what if I miss?

1:07:19And it's all the details that matter. It's like it's not. And and you have to be you have to be motivated to do that. I don't think it's that different. And is the pressure of missing really just like this product needs to be in the world? Like, let's just say for Coda, if you're like, I'm at the 10 yard line and I got to get this football across the line. is that pressure to get the ball across the line solely around just, I need to get this product into the world because what a shame if people are using Microsoft instead of Coda, for example. So sometimes it's that sometimes it's competitive, but I think the cases where it is more interesting is the cases where it is categorical.

1:07:55Like I'll go back to my, to YouTube time. So I showed up, I wrote this paper. Why aren't at, why don't, why does intelligence feel like the Superbowl every day? One of the ideas of skippable ads. obviously that's the thing YouTube is very well known for now the skippable ad format it took us three years to ship it and part of the reason was the world thought it was a terrible idea the sales team had a standing rule that Shashir's allowed to come speak at the sales conference but he's not allowed to talk about a stupid skippable ads idea it just seemed so dumb you let people skip ads they're going to skip 80 % of the ads revenue is going to go down by four-fifths why would you do that?

1:08:32and turned out that there was a way to do it that actually made a lot more money and it all worked out fine. But I'm fairly convinced that if we hadn't stuck through it, if we hadn't taken that insight and stuck all the way through it, it just wouldn't exist. Like it's not like YouTube would be big or small or small. I don't know, but like that format, that idea wouldn't exist. And I think that's a shame. I mean, for me, that was like, it was a motivator. I think of advertising as the grease of an ecosystem. It's like, how do you get products in front of people? You need to get it in front of people in ways that maybe they don't, they don't know they need it yet.

1:09:08And if you have a more effective way to do that, like the world just works better. Like people get to the products that they, that they care about better. And it mattered to me. And I was pretty convinced that if I didn't do it, it would go. It's just an interesting framing of how you feel pressure, because often the way that guests describe their pressure to me is a failure, right? Pressure to not fail. Yeah. Because then you're a failure, right? But then you like did not deliver on your word and people are going to think you suck. Right. Yeah. Or pressure to not let down the people that you hired or whatever.

1:09:39Yeah. Right. Like those are usually the articulations of how people describe pressure. Yeah. I mean, it's interesting. I mean, those are pressure too. I'm not. Yeah. So I'll give you, there was a moment. So I was like two years into my time at YouTube. Things were not going great. We were still, we were growing, getting the profitability, but it was kind of chaos. us, we hadn't crossed that threshold and we were still seen as the first mistake for Google. And I got recruited by the Facebook team and they wanted me to come work on this job. It was, you know, in retrospect, quite interesting. And I went to one of my advisors, a guy named Dean Gilbert, and I said, what should I do?

1:10:15And he asked me to make three lists. I said, by the way, this was a, I would highly recommend this technique. He said, the first list is your current goals, whatever your timeframe is, the next six months, 12 months, 18 months, whatever it is, what are you trying to achieve? And that could be work. It could be in the business. It could be in life. It could be personal, whatever it is, write down your goals. Second one was write down the subset of those goals that wouldn't happen if you left. I called it the marginal list, meaning that if you left, I went from YouTube to Facebook, which of those goals just wouldn't happen.

1:10:53And then the third list, and so like second list is a subset of the first and third list was, and which of those things fast forward 10 years, which one of those things would you give a shit about? And that was like the, I call it the purpose list. So you think about goals, you have your marginal goals, your marginal utility, and you have purpose. And that was like, that was really hard. Like you look at this third list, you say, what, you know, what, what really matters to you? And so I made this list and my list of goals was long. I had a bunch stuff we were working on, I had five things on my marginally trailer list that were like, these were, these were ones that I'm pretty convinced if I left, they would just go away.

1:11:27Like Skippable Ads was one of them. I was, I was quite certain if I left then, we hadn't shipped it yet. I was the only one that still - No ads are getting skipped. That it just wouldn't exist. The product just wouldn't exist. And there were four others too. Some were people related. There's one that was sort of skills related for myself. I felt like the skill I was learning that if I left, I probably would never build that skill and so on. And the third list was, which of those would I really care about? And I think four out of the five were still in that list of like, if I look back in 10 years and I said, I left that, would I feel really bad about it?

1:11:57And that list had things on it. And Dean's instruction was, if that third list is empty, you should leave. There's nothing on it that really matters. If the second list is empty, you should definitely leave. Like if everything is just going to happen anyway, then don't worry about it. And to be honest, a little bit of why I left YouTube. I don't mean to say it in a bad way, but I felt like the marginal impact I was having on YouTube was starting to fade. And I felt pretty good that the team was able to take it to the next level. And I think that's been true. I mean, I think that Susan and then Neil have done a fantastic job.

1:12:27And I don't mean to diminish any of their accomplishments, but I felt like the unique things I was doing were no longer those things. It's time to go work on something that wouldn't exist. I go look at Coda. Would Coda exist if I hadn't committed to it? Definitely not. And that product was not going to be invented just out of monkeys typing on typewriters. It wasn't, it wouldn't exist. Yeah, but couldn't you have done the same lateral that you did with Google TV to YouTube within Google? Like, oh, like, I don't know. I don't know. Yeah, Sundar offered. He said, go run the Google Apps team. And I don't think it was, I thought about it and I decided, and this is back to, we were talking about earlier of like the conditions for invention.

1:13:09and I think there's a lot of people that look over and they say, oh my God, that big company, they have this huge bank balance, they have all these users, so on. They're obviously going to be able to win this category when they put their mind to it. And I think sometimes that's true and there are definitely examples of that. And then sometimes there's not. And I think there's some patterns to it, by the way. I think I came to the conclusion that I could not have invented Koda inside of Google. I don't think it would have worked. There are conditions sometimes where the innovation requires, I'll call it capital, and sometimes where it requires insight or change.

1:13:42And maybe I'll give an example. So a famous example was Google Plus versus Facebook. This was an example where, or maybe I'll start the other way. Well, let's do Facebook versus Instagram. Or Facebook versus Snap, actually. So Snap starts growing. Zuck decides, you know what, the stories format's real. We better do something about it. He starts four teams, one in each of the four Facebook units. He says, go build your version of stories, or build your version of Snap into the stories format. And, you know, the four of them ship. One of them really works. Instagram really works. And that was a case where he was able to - He had already bought the Instagram.

1:14:26He had already bought Instagram. And Instagram shipped stories. Yes. But so did WhatsApp, so did Messenger, and so did Facebook. All four shipped, right? People have forgotten that, but it actually still exists in all those products. It's like, you don't really, I never use it in WhatsApp or Messenger. In Facebook, you kind of use it, but it's a different comment that people use Facebook a lot less now. But Instagram worked, right? So that was a case where he marshaled a bunch of people in capital and said, go build this thing. And it worked. On the flip side, Google tried to do the same thing to Facebook, and it did not work.

1:14:58and we built this Google Plus team and we took some of the best people in the company and we put them on this project and we said, go invent this thing. And we just thought you could put sort of an infant number of people and capital on it and you could go compete in social and just turned out to be wrong. It wasn't true. And I think the reason was like one was really a feature of a product that could be mimicked. And the other one was a rethink of how a space works. and that just requires different types of innovation muscle. You can't throw capital at it. In fact, throwing capital at it gets you all the wrong things.

1:15:32Like you'd much rather have a five person team working on it than a 500 person team working on it. And in those cases, like when I wanted to build Coda, we started with six people and we were six people for a long time. And like, that's what you need. You need that group of people. And if we're trying to do that and simultaneously figure out how to not screw up Google Docs and Google Sheets and so on, I don't think we would ever build anything. And there's lots of examples of that. So in my mind, there are cases where the big company can apply capital. And there's cases where it's actually completely counterproductive.

1:16:06And what you should really do is let a much smaller team that is unchackled from the safety net of the big company go give it a try. And for Koda, I thought that was how we had to do it. Aren't we in a fascinating test case, speaking of Zuck right now, where he tried the AI llama stuff within Meta? Clearly didn't work. uh okay whatever it didn't work in the way that he wanted it to yeah so he buys scale for 16 billion and now he's you know putting out 100 to 200 million dollar offers per person yeah to come and join this team like he's gonna try and he's gonna try again yeah he's gonna try again yeah and it didn't work internally in that same way like he couldn't put his best people of the google plus version yeah i'll tell you one of the things i admire about zuck is i mean his his level of So if you are the big player, you have to find those moments where you can apply capital to a problem, capital or distribution.

1:16:59In this case, it's capital. He's applying to a problem. And sometimes you have to do it in a way that seems outrageous. I mean, we think that this is a big bet. I mean, think about Satya's bet with OpenAI. I mean, investing$10 billion in a company at that stage was like unheard of. All of us were looking around. It's like, what is OpenAI? I mean, now it seems like so obvious. But in retrospect, I think that is a case where, for the most part, it worked. I would argue Llama is more of a success than people think. We use it heavily at Grammarly. And I don't think it's, I think people underestimate it.

1:17:35But it clearly wasn't enough for him to not shake things up. It wasn't enough to not shake things up. And I think he saw a bigger prize out there and he's going after it. And, you know, Jacques has done it multiple times. In Instagram, I think what he did with Oculus. I think there's multiple times that he's done it. And we'll see if this one works or not. I mean, I wouldn't bet against him. I think he's, I think he's, the thing I most admire is his willingness to have a view and put his money where his mouth is. It's insane, right? I mean, the like, the, I don't know if the rumors are true, but the 200 million that he spent on the guy that ran AI research at Apple, that was like triple Tim Cook's salary.

1:18:11And they're like, we're not matching this. And he just hires him. Yeah. Yeah, I mean, the thing I would remember about that is I think people think about it in dollar terms. I think that the media reports that If you think about it in dilution terms, it's very different. And I think whenever, maybe this is another line of logic. When you're hiring a senior executive, I think you have three ways to think about compensation. So one is you think about compensation relative to their alternatives. Like what, in that case, I think you're talking about Daniel Gross. Like what was his alternative? Like get to outpay.

1:18:37I mean, he was running a very successful venture fund. He was building a new company with Ilya. Like, you know, he had, that's what it, that's what it costs. The second is comparison to his peers in the company. And that, keeping that balance is really hard. And Zach clearly threw that out the window and he's going to have a big problem when it comes to refresh time for the rest of his executives. And the third one is marginal impact on the business. And that should be your test. Like when you're buying, like when I'm buying a company, when I bought Superhuman, like we paid X percent of the company for Superhuman.

1:19:03And my math was, does it increase the value of the company by at least that amount? And if it does, then that's a good deal. Like that's the way I think you should think about it. And you should think about hires the same way. Like, would I go, like, if you're hiring a new executive into a startup and you say, would I give 1 % of the company to this person? In my mind, the right logic isn't, you know, is this like the first one, is it better than what they would have gotten elsewhere? We call that BATNAF. Like, what is this better than their best alternative to a negotiated agreement? Is it, how does it compare to the rest of the executive team?

1:19:32That's the next test. But the most important test is, do they increase our chance of success by 1 % or more? And like$100 million sounds like a lot to the rest of us because we're comparing it to BATNAF. Sure. We're comparing it to this is what Cook paid them. If you look at it the other way and said, I think the total of all those offers is still 0.1 % of Meta's market cap. Like if I were sitting in Zuck's shoes and saying, does all of this increase our chance of success by 0.1 %? I mean, of course, he's lucky to be sitting on a trillion dollar market cap. But, you know, does it do that? Like, I mean, if I was sitting on his board, I'd say, yeah, that sounds great.

1:20:07Now, the fact that he did it mechanically is offers to individuals. And he also like completely raised the stakes for every other company while he did it. You know, that's that's a tactic that I don't know was unexpected. But the math of like, yeah, of does it does it increase your chance of success by 0.1 percent? That doesn't seem that crazy to me. But the interesting thing about what Zuck is doing relative to what you did with Superhuman, for example, is he immediately the market cap either goes up or down the next day after that's announced. He's doing these things. Yeah. Right. And so there's this ruthless public judgment that comes with every decision that he's making at that scale.

1:20:43Whereas for you, we'll find out. We don't know. We'll find out. And so you're relatively insulated from judgment for quite a while. Yeah, private companies have an advantage that way. And I think that when I think about founder-led companies, to be honest, I didn't look at the stock price for Meta, but my guess is it didn't move. Or maybe it went off. I didn't move. But Mike, wouldn't it be amazing? I don't think anybody reported that way. wouldn't it be amazing if he made offers for 0.1 % of the company and the stock moved up by 0.1 %? Nobody would have even noticed. The stock fluctuates by that much every day.

1:21:14But I think nobody's looking at it that way. He acquired a company. He did it person by person, but he acquired a company. I think his view of the world starts with, I'm going to act like a founder and the market's going to reward it. And I think it's one of the advantages of some of these, and to be clear, when I say founder, I think Satya acts like a founder. I don't think it necessarily literally means founded, but there's a mindset in that view that says my job is to increase the long-term value of this asset. And I'm not going to start by how I'm going to be judged that afternoon, even the next week, even the next year, I'm going to start with my, my list of like, these are the things I'm trying to get done.

1:21:52And like, I'm going to be judged over time. And, you know, sometimes I'm going to be right. And sometimes I'm going to be wrong. I mean, Zach's been right about a bunch of things. He's been wrong about some too. I mean, it's not, Not everything they've done has been a success. But I appreciate the ambition of it and the put your money where your mouth is. But I also think people are misunderstood. Like if you're a capital allocator and you have a multi-trillion dollar market cap, you can now make bets like that. And you're kind of like, you're kind of squandering it if you don't. Well, and isn't it worse to not make the bet and then miss the ride?

1:22:23Absolutely. And that's why we're in this crazy time because there's like a lot of that FOMO that's kind of driving what are some of the most competent leaders ever, like Satya, Elon, Zuck, Sam, like these are serious people that are making ridiculous, like they're getting loans for data centers. Like Zuck's getting a loan for a data center, you know, like this is like crazy big bets and they're going for it, like going for it, going for it out of fear of missing out. And they're all like one upping each other. And you get like a new model that comes out the next, It's insane. Yeah, it's Gold Rush.

1:22:57It's insane. Yeah, I think it's really exciting. I mean, I think it's, and I mean, so far, I think it's not exactly like a Gold Rush that hasn't paid off. I mean, this is not like, I view it, there have been past Gold Rushes that were rushes, things that didn't pan out at all. This one is clearly working. I mean, it's like the, as defined by. The level of usage of AI is skyrocketing. This isn't like a small trend. You can look at your own personal life. You can look at the stats for any of those companies. Look at what Anthropics is doing, what OpenAI is doing. Hopefully you consider us in that category with Grammarly.

1:23:28But like, I think it's like, it's obviously working. So this is not really that, like the mechanism of the bet, like do you take debt to buy into a data center? Do you give offers to individuals? Like these actually, they seem like big bets, but in reality, I don't think they're really that risky. Yeah, working as defined by users though. Meaning if the CFO of Google, when you were at YouTube, defined working as users, he would have said, you're crushing it. But he didn't, he defined it as your P &L, right? And if you define it by your P &L. And for some of the companies, even the P &L, I mean, obviously, NVIDIA has found a way to turn it into profit.

1:24:00And we will see about some of the other companies and the rest of the layers. I do think there's some skepticism on whether foundation models have the right economic incentives of the team that invests all the money in the foundation model. The team that accrues the value of it, we'll see if that tradeoff is clear. But in a macro sense, would you invest in this space? I think you absolutely would. But the thing I'd say about Zuck is Facebook has a long history of investing in technologies that monetize completely different ways. And I think that, you know, like every, almost every website out there uses the same React library to lay out the website, you know, and Facebook open sourced it and takes enormous value out of it and builds, monetizes it through a social network that has nothing to do with React.

1:24:42and I suspect what's actually going to happen with Lama and what they're doing, I think it may look very different than how other people are monetizing AI. So I wouldn't be surprised if it doesn't feel like a frontal view of what people are seeing as AI there. It's just funny. Like you wake up one morning and it's like, you get it, you see a tweet. It's like, oh, don't bet against Zuck. And then the next morning it's like, don't bet against Elon. And then the next morning it's like, you can never bet against Satya. And then it's like, you know, Paul Graham is like, Sam Altman, like never bet against him.

1:25:09You know, it's like, what's going on? Yeah, yeah. I mean, you asked further, there is a list of people they'd bet against. This list is all pretty good. Totally. Yeah. I appreciate you doing this. Really fun. Yeah. I really enjoyed it. Are you hiring? We are. Absolutely. Like anything in specific, anywhere in specific? Grandly.com slash careers. Primary offices for us are San Francisco, Seattle, New York, Berlin, and Kyiv in Ukraine. We're actually kind of the most sought after employer in Ukraine. It's kind of interesting dynamic. But everything, engineers, product people, marketers, salespeople, we've got spots open for everybody.

1:25:51I asked the same closing question. When you hear the word grit, what do you think of? Well, we covered a lot of different types of grit already. Sure did. Sure did. Maybe I think of three things. So I call it idea, grit, problem solving, grit, and work grit. Maybe I'll describe each one. So in my mind, idea grit is you have an idea and you can see it when nobody else can. And you stick through it even when the first signals aren't great. Like for me, my best example of that was probably Skippable Ads at YouTube. I mean, it took me years after joining to even ship the thing, much less when it became successful.

1:26:27And I think that's sometimes you have that grit. Even starting Koda, like it took us four years to ship Koda. It was a product that took a long time to get to a thing where people understood what we were building. and that's four years of like every dinner party people being like what are you working on and you know user studies for people like i don't get it that that that type of idea grit many of i think we look at these products as overnight successes one of my favorite examples figma figma was we were figma's first customer um coda was uh figma's v1 product wasn't collaborative like it literally wasn't a multi-user product you know dylan and and evan would come over and they'd work with us and so on.

1:27:06And we now look at it and Dylan's level of grit, like a similar story, it took him about four years to ship their first version. Like people have completely forgotten that it looks like an overnight success. So that's number one is idea grit. And I think people really underestimate it. It's one of the reasons why I think some of these startups kind of have to be done outside of the big companies, because you just can't, the pressure and idea grit is way too high. Like you just, you just expect it to produce in six months in a way that's really hard to do elsewhere. That's number one. Number two is what I call problem solving grit, which is you've got a problem.

1:27:40How hard are you going to look at how to solve that problem? And this is almost the opposite of idea grit. Like you don't have the idea yet. How hard are you going to push to find the solution to that problem? And I think there's many cases where people focus on speed of decision making or, or so on. And they, and they, they miss that often the right way to solve a problem. It's just not going to be apparent in that first discussion. It's going to take a little bit to see, okay, that's the solution to this problem. Maybe I pick a YouTube example. I got to YouTube. It was like, obvious we're spending way too much money on music, but the way to correct our balance with the music industry wasn't to get music off the site.

1:28:16That wasn't going to work. And so we had to rethink our way of working with the music labels. And there were a couple of things that came up in kind of unexpected ways to completely reset that balance. And the third one is the most mechanical is work grit. And I just think when people think about, I'm a, I'm a, I'm a hardworking person. And I think people underestimate just the day-to-day work grit that it takes. One of the things I have a friend named Des Treanor, the Intercom co-founder, and one of his statements, he said that really sticks with me, is your team's productivity is upper bounded by yours.

1:28:48If you're like days behind on your email, you don't respond to Slack, you're never prepared for meetings, so on. Like you can just expect that behavior from everyone else in your company. And on the other hand, if you're on top of your shit, then you can hold them to that account. And I just think when we talk about grit, it's like stick to an idea, be willing to work out a problem until the right solution emerges. Then there's just day to day, like respond to people, call them. When a person is having an issue, talk to them now, not next week. When a customer is having an issue, call them now, not next week.

1:29:21Like that kind of grit, I think is dramatically underestimated, but very evident in all the great leaders and founders I know. that's in the pantheon of one of the best grit answers I've ever heard. Thank you. Hey, I appreciate you doing this. Thank you. That's it for now. If you liked the episode, please leave us a review or go back into the archives where we've done more than 200 episodes with some fantastic folks. This podcast is a Kleiner Perkins production and I'm Juven. Thanks for listening.