Ep 94 - 2024 Rewind: AI, Entrepreneurship & Life Lessons

Speaker A: I wanted to give you a sense of how this year has been and talk through the highlights. We love talking to people from all walks of life who are doing something interesting.

Speaker B: Welcome to another exciting episode of the Aarti and Sriram Show.

Speaker A: Hey everyone. Welcome to the Aarti and Sriram Show. This is going to be a special highlight episode of our best content of 2024. If you're a new subscrib. Thank you. Um, over the last couple of weeks we've had a lot of new, uh, folks joining us. Uh, welcome to the show. For folks who are new here, this is a show where we host conversations. We host guests from all kinds of fields, from startup founders, builders, creators, entrepreneurs, uh, but also we've had authors, we've had sports, uh, personalities, you, uh, name it. We love talking to people from all walks of life who are doing something interesting. And we generally tend to focus on optimistic conversations, on life, on creativity, on productivity and how to win, how to be successful at what you do. So whatever your plans are for 2025, I hope these segments, um, across Silicon Valley and stories, across AI and across advice and self help, I hope these are useful for you in some ways. Um, with that, I won't keep you from, uh, enjoying the rest of the episode. Happy, um, holidays. Happy New Year. Hope to see you in 2025. I hope you get a good holiday break. I hope you're able to all build something. It's my favorite time of the year to go tinker on things and learn something new. So I hope you're able to do something that you love and enjoy these holidays. Thank you.

Speaker B: You have been a, uh, uh, trusted execration to many of these books, right, Jack, Uh, Mark, because you built out, uh, advertising at Facebook for many years. In fact, you know, just uh, for fun fact, Gokul helped recruit me into the ads world at Facebook. And then immediately after I joined, he left.

Speaker C: So I knew ads was in good hands with you joining. So I worked this down.

Speaker B: We didn't insert that. Michael joined me. I'm like, I took it personally. I took it personally. I was like, you know, the week after I started Google, I was like, I'm out. Right? But we, I want to get to that later. But I'm curious if you look back upon these executives that you work for and then also before that you helped build AdSense at Google and really, I would say pioneered how a lot of monetization works on the Internet. Let's walk through each of them. Right? Like Larry, Sergey, uh, you know, the senior executives uh, at Google, um, they've been studying, they've been talked about a lot. What did you observe from them up close? What were times when you had interactions, were able to influence them, change their minds or, uh, they influenced you?

Speaker C: It was interesting. I think Google was the seminal point in my life in some ways as a product person because it was the most fun, I almost say the most fun I had because I was an ICPM for half of my time at Google. And I always feel, uh, being an individual contributor is when you really have your hands on the product. As you start becoming a manager, unfortunately you get one layer removed. And I had the most fun when shipping products myself and in particular for Google AdSense. Sergey was an executive sponsor. In fact, the idea came from him because he basically realized very quickly, as soon as AdWords launched in 2002, he realized that, hey, just like AdWords is all about taking or AdWords or search is all about taking keywords and figuring out web pages that match the keywords, we should do the opposite. We already basically taking web pages and figuring out the keywords that match them and finding ads. And uh, one of the most interesting lessons I had, which I didn't internalize till a few years later, was when he nixed something we had built for a long time. He used to attend. We were heading towards launch and um, I'm dating myself here in January of 2000, in June of 2003 when we were going to launch Google AdSense and it was March and we had built the way AdSense worked is websites apply to be admitted to the AdSense program. And that's what we built and we were building. And so there was a big approvals queue. We had this ops team that was ready to approve websites based on a long list of criteria. And we basically were reviewing it. And then Sergey walks into one of the meetings he was attending. He was like, what's his approval queue? It's done. We proudly said, yeah, we built this great queue to approve any website that comes in. He's like, why the hell do you approve, do you need to approve websites? We're like, what do you mean? You know, you can't let any website with NSFW or violent content, all kinds of stuff, get onto AdSense to Google to run our ads. Why not? He asked. What do you mean why not? He's like, look, didn't you consider the fact that they could apply as under a benevolent website and then after you approve them, they can change the website content? You're like, no, that's True. Uh, and then he was like, you know, what you guys are missing is the number one thing that the website cares about is making money fast. The first thing they apply and I want to get them to see their dashboard immediately and see that they're making their first dollar, first dollar two. Within a few minutes they put the JavaScript on the website. It should show. And what you're doing is you're saying, okay, now it's going to the approval queue and you might hear back, who knows when you have them right? Then show them the money right then. So I want you to kill the approval queue. I was like, what do you mean kill the approval queue?

Speaker B: Yep.

Speaker A: Huh.

Speaker C: Kill the approval queue. So that was, I mean, we killed the approval queue and it worked because what happened is he actually said that Google in general is built on this philosophy where even AdWords, you could actually back in the day set up an ad with for a keyword and it would start running immediately because there was an instant psychic gratification to seeing your. You would search for that keyword and see your ad and for the first hundred, two hundred impressions it would run and then it would go into an approval queue to make sure there was no like copyright or other violations. And the same thing happened in that sense once you get. Because the reality, so I think it really solidified a key thing around product led growth in Miyane is that you cannot have barriers to the aha ah moment. As you guys know, the aha or the magic moment for any product is when the user gets hooked after a certain experience, right? In Uber it was watching the driver come to you in doordash. It was when you get the first delivery every Facebook it was getting to 10 friends and so your news features populated. So you cannot have any friction, any barriers to getting to the aha moment. We never thought about it that way, but that's basically how it was. Sergey just had a beautiful understanding of products intuitively. Where Larry was the technologist. Technologist. Every feedback we got from him was why is this so small? Why aren't you thinking bigger? Why aren't you thinking bigger? Why aren't you thinking bigger? Well, Sergey, it was all about the. He really, I mean he had just incredible product sensibilities. So I learned a lot from them in different ways. But that magic moment thing was something that stuck with me because he killed a lot of work, but he made m the product significantly better as a result.

Speaker B: I'm curious how Google was in that era because you were a little earlier in your career um, you know, you were not the mythic legend. The only few people in Silicon Valley who go on first name basis where you don't need the last names. Goku is one of them.

Speaker C: Somebody says Sri Ram and Aarti are two others.

Speaker B: There are a few Sri Ram's questions.

Speaker D: Right.

Speaker B: But thank you, thank you. But you're a little earlier. Walk me through how it was to interact with Larry and Sergey. Right? Like how was debating, how was discussion? Did you feel like, hey, you know, I could challenge them and how would you do?

Speaker C: It was pretty amazing because it was Larry and Sergey for many meetings and then Eric would join them for some meetings. And so it was sometimes two and sometimes three people. Our, uh, VP of product, Jonathan Rosenberg, and then my direct boss Susan. They were, I mean generally, I mean there was really no barrier between them, uh, between us and Larry Sergey, just like in, in the days of Facebook, I think when I was there and you were there, you could just directly go in and talk to Mark. So they were super accessible. They were just sitting very similar to the setup that we had at Facebook. You could just go and talk to them. And uh, what was crazy was that uh, Sergey would sometimes be, they had set up like uh, bikes in one of the rooms. There was a bike and Sergey would be biking and we would be presenting or talking and they didn't like one thing. I realized that, you know, I think leaders don't like formal presentations, good founders, because they realize that formal presentations hide sometimes imprecise thinking. So they really many times would, okay, stop this presentation. Let's just talk about what you're building. So I was kind of nervous initially because I was used to presenting formally. And then slowly I lost my fear because, uh, I realized that they truly want to just hear the unvarnished truth. And they never shied away from tough, tough news. I mean that's the thing. What they didn't like was like you giving good news and it shouldn't because you always, initially I was trying to sugarcoat stuff for them. Then I realized that they were too smart to be bsed. And so you just give the bad news upfront. The other thing I realized was that just like Mark, they basically had the most risk tolerance of anyone in the room. They were willing to risk it all, amazing number of times. And we all were much more risk averse than them this founders. And I think that's the thing about Greg founders. As the company kept growing, they were willing to make massive bets and massive risks. When they bought YouTube, we were like, oh my God, this company, they had no monetization. It was under massive lawsuits from ICOM, and they paid 1.65, uh, billion to buy it. And Google, it was a good chunk of Google's market cap. And you were like, what the heck? Again and again, they were willing to bet the farm on things that. And that's a hallmark of great founders. I think that's founder mode, to be honest. More importantly than anything else in me, where they're able to make. I think that's a challenge that, um, a, uh, CEO who's not a founder faces that they don't have because they feel themselves beholden to shareholders, while the founders, they remember it was nothing a few years ago and they are perfectly willing to roll the dice. I think Elon does the same thing. It's perfect. Willing to roll the dice again and again and make company. Making a company destroying bets in crazy ways.

Speaker B: Yeah. A couple of stories. One, Brian Armstrong came on our podcast a little while ago and, uh, he talked about how coinbase were there some really dark times during, uh, one of the crypto bear cycles. And he talked about how, um, he thought to himself, you know what? Worst case, I built the company from myself in my laptop in my bedroom before I can do it all over again. And I just don't think even the best hired CEOs, and there are some really good ones, and I think sometimes in Silicon Valley, um, on Twitter, we tend to, uh, maybe dis the hired CEO, but even the best of them can't do that because they fundamentally haven't had that experience. There's just something unique there. I want to talk about Mark and you to Facebook, but, you know, one of the things that always struck me about Mark Zuckerberg, uh, is when you are interacting with him, I agree with you, uh, that you could have a very open conversation with him. But I think the subtle, interesting thing that he did, and I don't think a lot of the others, uh, uh, uh, great CEOs sometimes do, is he would tell you very, very quickly where he stood on the spectrum of the nation you were talking about from, um, all the way from. I don't care what you're working on, and sure, I'm going to indulge you. Please feel free. Go, whatever. Go have fun. Go do it. Thank you for the status update. All the way to the other end of the spectrum, which is there is a piece of paper in Delaware which says, I own this company and run this company, so please do what I say. Right. And the spectrums in between all the way from, um, uh, I let you overrule me, but you better be really sure all the way to, I'm CEO. And I thought all this. Zuck was very, very good about telling you where he stood on that spectrum. So you knew, even as maybe a intern or somebody new to the company, you knew the framework of how to engage with them. And I always think about that in terms of how to engage with somebody who might have a big power level disparity with you.

Speaker C: I think it's a good point, though I do disagree that other CEO, I think he was exceptional. Larry and Sergey were also very good. You knew exactly where you sit. I mean, you could talk to anyone at Google. There was, I think, the look, I think the best way to engage with all of these great founders who want to hear the truth is just tell them the truth and be ready to. I think the other one is you got to really hear and listen to that feedback. One of the best examples I think was the foundation of most of Facebook advertising is something called custom audiences, which you probably remember where I think we. And, um, that actually came about through Mark. He's a learning machine. He's a learning machine. What happened is Mark Pincus at Zynga was one of our biggest editors at Facebook. So Mark Pincus was talking to Mark Zuckerberg and he was complaining about Facebook ads and saying, why the hell can't I reach my customers? Because Zynga was a very whale oriented company, uh, like most gaming companies are. I want to reach these whales and target them. Why can't they reach them? M. I know you have them. And, uh, Mark. Mark Zuckerberg came to us and said, why the hell can't you target his customers? So that was the orientation. We were like, oh, shit, okay, he can target. They can't target their customers. They just need to tell us who their customers are. Initially, those customer IDs. And because Zynga already knew the customer IDs, because they were on the Facebook, they were a Facebook app. But then over time, it was other ways to identify customers. Email addresses, phone numbers, et cetera. But that was the whole genesis of custom audiences. And now it's a foundation of all of Facebook ads. And it also was copied by Google, Twitter and everyone else. And that was all because of Mark Zuckerberg. When I first met him in 2008, he had no idea about advertising because it was a pure consumer company. I met him in 2010 and 2011 again. When we joined 2010, he had absorbed so much about advertising and the first year I felt. So he was a learning machine like no one else. He just learned at a rate that was incredible. So much so that his idea was what underlay, what Facebook ads was. But I think at the high. So you've got to listen to him. I think I realized after that experience where I'm like, he's not just saying it to say he truly means it. And he actually said, I want you to figure out a way. And he would remember it. He would remember it because he also had a great memory. Almost like it was incredible. I don't know if it's like one of those, um, what do you call it, Those camera like memories he would remember the next time he saw me. What do you think? Because if you try to not. If you try to not address the feedback he gave you last time, he would remember and he would bring it back to it. I think that's the other thing that, um, you just got to be straight up. You can't hide, you can't dissemble, you can't BS them. You've got to just take their feedback and be direct and say, here's why it's not working. And he would. I forgot the. I think there's one or two times where I convinced him what he felt, what he thought was incorrect, but it was not my opinion. It was using facts, using data and using. Using data. That is the only way to convince them. I think otherwise. If it's their opinion versus yours, guess what, their opinion is going to win.

Speaker B: What would you say that as a state of the art is given, where elements are, where do you see capabilities lacking and where do you see capabilities headed or say over the next eight to 12 months?

Speaker E: Yeah, absolutely. Um, I think there's two schools of thought. If you look at applied AI projects, whether they're kind of projects in large companies or startups like ours, there's two approaches. One is driving for automation, right? So really look at specific types of work in the economy and try to automate that away. Um, the second would be augmentation, like Michael Jordan calls it, like intelligent augmentation. There's two approaches. So the first one is to take classes of workflows where you can completely automate that away. So if you look at the coding side or the building side of enterprises, that would be idea of kind of automating database migrations. Uh, in the selling side, it could be automating kind of the Long Tail SDR or email outreach. So that's one direction of application. The other is supercharging and augmenting existing folks like Cursor for top engineers or rocks, hopefully for enterprise sellers there. The idea is to decompose the job that these individuals are doing into kind of basically kind of some modules or many jobs and automating those away. So when you look at applied AI companies, there's kind of single shot automation approaches and then there's the augmentation approaches behind that. To kind of answer your question, what's the state of the union? What we're learning is the alpha is in data being the system of record and earning the right to manage the data. Where we can work with the underlying base models to either augment or automate a certain job is where we're seeing um, value, immediate value for customers. Um, the second is being extremely customer centric. Um, and I'll share where I made a mistake. So I was chief growth officer for a public company, Neural Lake. Uh, I got obsessed about this problem because we won and we lost and I kind of had a visceral understanding that the CRM, um system is no longer the essential system that businesses run on. And there's a generational enterprise software platform to be built. And last year we put in our own money, kind of did uh, the pre seed and that point we built a system of record. And then when we started the business in February on the system of intelligence, our agents we took this kind of the build and they will come approach. We tried to build an agent, we tried to do all the work and what we realized that the base models are getting better, but we just didn't have the data to actually kind of complete the work. So we decompose our agent down into these sub modules where we kind of took the different approach which is hey, let's just break it down into many jobs to be done and try to kind of automate that away. And so that's kind of um, My state of the union is base models will improve over time. Applied AI companies can't wait for the base models to get to singularity. If you're applying today, try to build a system that owns a right to manage the data layer and then in the interaction layer just proactively engage with customers and build with them and pedal downhill. And that's the approach. Uh, we broke all of our Silicon Valley systems muscles and said okay, we're going to switch to shipping multiple times a day, only weekly targets, no monthly quarterly targets, and we're just going to work with customers. So we're about seven months in. We work with some of the best um, companies, we try to make them better and it's a very pragmatic approach to build a true um, kind of application kind of in this agent era. Is that helpful?

Speaker B: Uh, yes, I think if. Can I just go to details a little bit? When you mention uh, agent tech CRMs, uh, when you mentioned say cursor, one of the things I think about what makes like a cursor Devin, one of these products work is there's such a clear feedback loop on what is right and what is wrong which makes it better. Like coding is deterministic. Uh, you're in the repl and the model can learn and just gets better and better. And when you look at other use cases which have really taken off, for example say support and with folks like what Brett Taylor and others are doing, there are some elements of that I'm curious about. When you talk about CRMs, how does it play out into these discrete tasks and how do you basically know when you're doing a good job when you're not doing a good job and converging

Speaker F: on

Speaker B: accuracy or whatever the metric might be?

Speaker E: Yeah, that's a great question. So in our domain, um, the traditional CRM system was such where the human did all the work, they got all the data in. Once the data is in there they would drive analytics workflows up on top of it. And then in this kind of new world there's going to be a class of software which tries to do the work and the work very pragmatically. Sriram, um, is if you're a customer facing professional and there's literally tens of millions of them out there, the three jobs that you do is you plan, you figure out, okay, off all of my customers, who should I work on? Second is you research or prospect. Hey, for these customers, what's going on? Who should I reach out to? And then you engage, which is okay, I'm going to reach out to them, I'm going to do a call and then you uh, kind of loop that. So if you look at a calendar, Sunday night to Monday morning is essentially planning, um, Tuesdays are for research. It's called pipeline generation. The rest is mostly engaging with customers. That's kind of an abstraction. So in terms, and then this whole game is not 100% precision domain. Uh, so the idea is how can we help you gain time, 20 to 30 hours um, a week so that you can support more customers. So in our domain our primary optimization kind of target is time saved. And the way to save time is you decompose the planning, research and engagement down where we can give you, we can do 80 to 90% of the work and we do those actions where you don't have to complete um, the complete work. Right. So when we started off, because we're close to Scott and uh, we also built an agent which tried to do the whole work but in our domain that model did not apply in uh, our domain. So we kind of decompose it down. So going one level deeper, what does Rocks Agent do? So let's say you are a kind of enterprise account executive at Ramp. Let's say you have 50 customers. You get a swarm of 50 agents. There's one per customer. That agent every Sunday night will essentially give you a note. Hey, here. The customer you should work on. And this is because of all of your internal and external context. So that saves about three to about eight hours of work. Next day you're going to sit down and do research. The agent queues up that work for them. But the human does trigger that. Like we have taken this approach where there's immense brand risk. If you kind of spray and pray today in our domain, uh, it has to be thoughtful. You got to present the band. So we'll do that work for you. That's already saving about 200 hours kind of week. And that's kind of what's converted the next is. Okay now we're in a meeting, we prep all your calls, we support kind of live transcription and we do the follow ups and we can do multimodal. So in the morning you get a rundown on audio. Hey, here's how the day looks like at the end you get a recap and a look forward. So that's kind of our domain where the customer support like some people like Decagon, which is amazing, or like Devin and Cursor, that didn't apply. That model breaked and we kind of took on this.

Speaker D: What we're doing is proven better new. We're deconstructing any product we admire. I did this. Bill Farmville. We, we are all doing this whether we admit it or not. We are taking what works for us, what speaks to us, deconstructing it to its smallest mechanics, to the atomic units that we can get to.

Speaker G: Mhm.

Speaker D: And then we're copying those. We're not making them better. We're not reinventing those. We're copying those legally. Okay? And that is your, your right is to copy it legally.

Speaker B: Right.

Speaker D: And you're not trying to get respect from your fucking peers. You're doing that. Go Work in some big company. You want to be a founder, you got to be like me. Ready to be disliked and dis and scorned and laughed at. But you copy and you become the best at copying. Picasso was the best at pica.

Speaker G: Mhm.

Speaker D: Picasso was the best at copying and tracing before he was the grass artist. Steve Jobs was the best at copy. I am impressed with people who can copy like a master. Okay, so first you copy like a master. Second, you do better. Better is not what you think is better. That's uh, new. Better is what 10 out of 10 of the users of that product say is a fuck. Yeah, right. Okay? What you think is better and they don't. That's called new. Okay? Get your ego the fuck out of this. It's new. That's new. Okay? And it will fail. And you isolate new. In my book, you, the amount, how you apply this framework, what percent? You say to yourself, what percent is new? That changes based on the platform. Okay? When it's a new growth platform like mobile was. Supercell is a master of copying.

Speaker E: Mhm.

Speaker D: Ilka is a master. He's better than me. He's better than me. I've proven better. New. He took Farmville 2. They copied it on a mobile. It was proven. I mean, literally, if you look at it, you cannot tell the difference. They, they took our little white mailbox, the half driveway, the house. They said they copied it, okay? And I say this with admiration. They said proven is FarmVille 2 on mobile? On web. Better. We're going to make it work with a touch with, you know, uh, pinch, with shorter time frames on crops.

Speaker E: Mhm.

Speaker D: No new. The new was mobile. Okay? And it crushed it, right? It was called Heyday. When we built farmville, it was proven and better. There was no new. If you can do proven better and 0% new. That is amazing. Poker. Zynga poker was proven because we took, we copied a poker game. We didn't change a thing. It was janky as fuck, right? Better was no download because we didn't care about security. Which do you want? Download. No download. 100% of people take no download. We know it because you lose 50% every time they click, right? They vote with their clicks. New was a picture of your Facebook friend, right? A real person. And new worked. What do you know? New worked, but that was proven better.

Speaker A: New.

Speaker D: It's much tighter than people think about. And what I'll say was Zynga PMing was that became not just our culture, but our religion. Um, and we turned that into what we called our blue sheets. And when we had our product roadmap meetings, that was church, okay? We were doing God's work. This was our religion. This was deep in our. This is our souls coming out, okay? That's what we were doing. And it was working. And we were getting feedback loops every day and every week. And we had this beautiful, beautiful product roadmap room. I spent all my time crafting these blue sheets. You could not change a single thing on these. This was our Torah, okay? Our roadmap meeting room, which had been originally a porn DVD store that we found. It was an empty storage room in the bottom of the Dejaro building. But it had ceilings that were 30ft high, no lights. You don't need them in, uh, a DVD store. And we had moving whiteboards. They moved this way. They moved this way. And I sat in the middle with the entire team, the product leads of engineering, design, and that was my operating table. You know, this was like 1895. And it was the, you know, whatever you call it, like the teaching hospital. And is how I trained new people. 35 people sat around the outside of the room. That's how I trained new people. They sat and listened. They couldn't talk. And I sat there with my team, no middle managers. Uh, and this is all I did. This is how, you know, Bing had me measure my calendar. And if I was ever spending less than 50% of my hours on these roadmap meetings, then we're redlining. I'm not investing in our future. That's why I had to say to you. Why am I here talking? Because I was talking to you at the expense of my player.

Speaker B: Right. I want to ask you because. Just to jump in, because you were talking about Zynga, and I would say the phenotype of Zynga product managers, another distinct part of the phenotype, is their obsession with data and metrics when it comes to growth. And I would say in some ways, Facebook gets a lot of credit for their growth team and the growth accounting. But I would argue that it was Zynga, which was the real originator of thinking. So explain how that came about.

Speaker D: Okay, well, Facebook was not a data company, okay? They had no interest in data. They were an intuition company. And, uh, Zuck was working off his intuition, and he probably has, you know, better intuition than maybe anybody else. And so it was pretty good. Um, although people forget pretty much every new feature they ever launched failed. As you know, other than the like. Right? They tried recommendations, they tried events. They all failed.

Speaker E: Right.

Speaker D: Pretty much everything. Um, their own products they launched all failed until they did Facebook.

Speaker B: Somebody should go down a rabbit hole on Facebook. Beacon. Uh, uh, oh yeah, you could have

Speaker D: a whole podcast on what products ever made. I think the only product they ever launched that got anywhere was threads. And what guess what they did Proven better new. Right? They proven we're going to copy Twitter perfectly.

Speaker B: You know, stories, the same example Stories, the exact same example. Like I was at Snapchat when Instagram launched Stories. Uh, but talk to me about data.

Speaker D: Okay, okay, so here's what happened. We had to be better than Facebook because we had to work harder than them to survive. Every other week we'd get some email from them saying they're deprecating or changing a channel. Remember, we were buried inside Facebook. We, we didn't control, we had no persistent navigation. You know, they would move where you found farmville because they wanted to promote events. And now apps were gone, they weren't on the homepage anymore. Um, so we had to work a hundred times harder than Facebook did to keep our engagement and our retention. So we had to be 100 times better than them. So we were a big data company long before them. Uh, Kadir Lee came in as our cto. No one knows that Zynga had amazing world class engineering. Kadir Lee, Scott Diehl, they were in the same class as, you know, Sergey and Larry. They brought in the best and the brightest and I gave them free reinforcements. We over invested Zynga. Never. Zynga was cash flow positive every day since I started. And we wanted to have the biggest chip stack. Uh, we raised massive amounts of money. We spent it on big data on data centers. And what we would do is we would use the data to convince Facebook to try features and change bad features. So we would, we were the ones, we never, they never gave us a single edge, ever. We were negotiating on behalf of the whole ecosystem with data. So we would come in, we would a b test whatever it was they were doing and come in and show them the data. And originally it was like Jared Morgenstern, who, I sold him on the idea of user pay. He became the user pay guy. But you know, I kept walking in to Facebook and hang out and I'd have lunch with Zuck once a month and then dinner and I'd show them the data. I'm like, dude, here's how much money you can make on user pay. And you know, I'm trying to convince Twitter and Elon right now what they can make on user Pay, So we had to be better than them. Eventually they started to get the data religion. Years later, not till around, um, you know, their IPO or post ipo. And then intelligently, they started hiring my whole team. So the guy who built all of our data and analytics, Ken, what's his name? They, they hired him. You know, they hired our cfo, who was, was a brilliant guy, Dave Wehner, who was really the first consumer Internet CFO who was data driven. Right. They, they eventually hired, you know, all of our PMs and, you know, even our game designers. So that's why you met a lot of people at Facebook. We were the training grounds, not just for Facebook, but for, you know, eBay and PayPal and. And, you know, John Donahoe came in to Zynga and he said, you know, our number one recruiting target is always Zynga PMs. Um, so anyway, yes, that's, that's how that came to be.

Speaker A: Um, you talked about talent, and I sort of want to double down on that because I think from m. What you've written, what you've talked about talent, whether directly or indirectly, something you cover and focus on quite a bit. How do you find good talent? How do you, uh, spot them, fund them? Uh, you do a lot of work behind the scenes on this, but how do you find good talent?

Speaker F: Well, I hope you guys haven't found everything. I try to stay behind the shadows as much as possible.

Speaker B: We are outsourcing.

Speaker F: I wrote this a couple years ago that got very popular, uh, which is the average age everywhere we do is increasing dramatically. That's good. You kind of accumulate knowledge and you're making decisions with better, um, informed decisions over time. And you have, uh, pattern matching and other things. But the really bad parts are two things. One is the burden of knowledge is increasing. Um, Benjamin Jones and others have written a lot of great papers about this over the years. Matt, uh, Clancy, who's an independent researcher now, Matt's written really great things, uh, about, uh, the burden of knowledge increasing. Um, but no matter which field you pick, the date or age of first achievement is increasing dramatically. Um, and that's not just true in, like, science and math, but it's also true in film. Uh, it's true in the government. Um, the average age of a congressperson is growing something like four and a half months every year. The average age of a, um, uh, somebody who runs a university is growing similarly. Um, if you look at, uh, the NIH directors or who are PIs at NIH, basically everywhere that we see, uh, People are getting older before they have responsibility. Um, and that could be good. Um, but if you turn back the clock, before, it was young people that did all the most impressive things, from Watson and Crick, uh, to, um, the Macintosh team, I think the average age was, like, 23 or 24.

Speaker B: But the founding spot, the Founding Fathers,

Speaker C: that's what I was going to bring up.

Speaker F: Founding fathers, that's probably the best example possible, is Washington was one of the

Speaker D: few that was over the age of 30.

Speaker F: Everybody else was younger, and that's, like, the greatest founding story possible. But I think on the talent side, there's two separate areas to think about. One is how do you evaluate the right talent for you? But also two, how do you identify young talent, uh, that hopefully will get to run the world 5, 10, 15 years from now? Uh, and outside of meter, I spend a lot of time on the former, a lot. And in that, I believe in the entire approach of outbound rather than inbound. And I just think inbound is wrong for a lot of. I'm happy to get into.

Speaker B: You know, by the time this podcast comes out, you know, you'd had this amazing video with Cristiano Ronaldo. And I bring him up because he's somebody. I remember him running out of the pitch in Man U when I was a teenager in India, right? And we've seen many bright stars in many sports who show up at 17, 18. A lot of them flame out. Either they don't have the skills or they can't deal with the lifestyle. Something happens. Uh, but Cristiano still around, still looking ridiculously fit, right? Like, uh. Uh, you know, I. You know, maybe he's lost a step or two, but, you know, 20 years kind of dominated the sport. And we're talking about Yamal, for example. You know, maybe we would have preferred that he doesn't score against England. Uh, but the guy's like, 16. I can't remember what I was doing at 16, but definitely not taking on the pressure of playing for Spain on a stage like that. So when you think about young prodigies and genius who kind of come up, what do you think is important to make sure they can then suddenly handle getting millions of Instagram followers and being there? The pressure and also the temptation. What. What separates the people who make it and the people who don't?

Speaker G: I think, first, the family support system behind you is very important. I think that you look at Cristiano, his mum is pivotal in. In what he does. Always a reference point for him. And, uh, if you watch My podcast, it's coming out with him is he talks about his mum on there and a situation in the Euros that everyone saw, but his mom was a big part of it. But family, I think, is huge. The way they help shape you. The way we were talking before we came on about family, creating, driving individuals, and again, that consistency and building the character, that's just going to put you in good stead for lifestyle, uh, a, ah, lifespan. Also the team that is built around them. I think Cristiano Ronaldo, he was the first athlete that I ever saw. Now LeBron's doing it as well, but, uh, Cristiano was the first one I physically saw in my own eyes build a team around him. So he lived a couple of doors away from me, and I went into his house one day. I should walk in and say, yeah, what's going. Are we up to? And I saw like, five, six guys. No, no, five, six guys in. Sitting down in the front room. I said, chris, who are these guys? M. Man, who are these guys? He says, yeah, that's my, um, physio. That's my, um, uh, that's my chef. That's my doctor. I was like, bro, uh, we got all this at, ah, the training ground. He said, no, but these are my ones. And I was like. So he built a team specifically for him to become the best player. And you roll on. Two years later, he left Man United as the best player in the world with a Ballon d'.

Speaker B: Or.

Speaker G: It's no coincidence who Ronaldo became, because he prepared like, he. He put the pieces in place for him to become this. And he's obsessed with details. He's obsessed with getting the edge physically.

Speaker C: Uh-huh.

Speaker G: Getting the edge mentally. He reads a lot of books. He listens to a lot. He doesn't miss anything on social media. He's across it all. Like, the guy is always looking for an angle. And I said to him the other day when I was with him at his house, bro, you scored 900 goals, man. Like, relax now. Surely, like, it's over, man. Like, come on.

Speaker B: Uh, he's like, eat a pizza. You know, put on some.

Speaker E: You know, eat some.

Speaker B: Yeah, like.

Speaker G: And he's watching his food after we're eating lunch, and I can see him meticulously watching his food, almost weighing it in. He's like, in his brain. It's like, I'm sitting there going, this guy is not. It's not normal. But I think any person, any people I've met in football, in sport, in business, they are not Normal people, the ones who are the super elite.

Speaker A: And you can tell. You can tell. That's the thing that sets them up.

Speaker G: Oh, yeah, yeah, definitely.

Speaker B: You're playing for Man U, right? One of the most historic, you know, franchises in all of sports. Right. Like, I grew up in India, you know, um, half the country was either a Manufan or Gunners. We don't want to talk about that. But, you know, you know, the Red Bulls are iconic. And then one day, you know, you know, you get a little bit of, like, you know, clot on your arm, which means something, and you get the pressure and the pressure. How did you handle. Handle all of that? Like, what did that mean for you?

Speaker G: Yeah, I mean, like, you say, the history at the football club could be overwhelming. Like, you listen, you see the stories and the success that they've had and the big names they've had there. But I always just thought that, um, I've been given the armband. I'm captain for the way I am and the way I've been up to this point. Uh, you don't change. You continue doing what you're doing, you maybe tweak a couple of things, but you don't make wholesale changes. Because I've been given it because I'm doing the right things. One of my biggest things that I thought about when I was captain, my role is to make other people play better. Uh, my role is to make other people feel more comfortable. My role is to make other people feel they have a responsibility for the team. And so that might be on the pitch for encouragement, that might be in the change room, having to pull somebody and say, listen, you need to kind of, this isn't the way we do things here or. And in different ways, like, a lot of people, some. Some leaders are, uh, shouty and aggressive. Sometimes you can do this, have the same response with a joke, but it's got a tinge of, you know, that's going to keep them thinking for a bit longer or. Or do it in public, or it's just reading the. Being able to read the room and read the individual, how you deal with them. And I think those are big traits of captains that have to be right.

Speaker B: What made you wake up one morning and said, well, yeah, you know, I'm going to try and make Fusion happen in my, you know, small college room.

Speaker H: Yeah, I think me, generally, I have low activation energy to get things done. So when a friend of mine, Olivia, was talking to me about this and she was showing me her setup, and I was like, dude, if she can do it, why can't I? I met her at this place called Edgesmeralda which is like a pop up city in California. And uh, we were talking about this and then immediately after talking through I just started looking into it. So I went to this forum called fusa.net and just started reading and I was extremely confused. I had no idea what was going on. But then like just slowly feeding this into like just googling things, like using AI for things. I was slowly getting a better picture and then I was like okay, like I'm just going to do it, I'm on a gap term, I don't have a job, um, like there's nothing like stopping me. Um, so yeah, I don't really have a better answer than I was just like I just want to make it, I just want it to exist.

Speaker B: And so, so, so damn cool. So cool. Okay, all right, so now you want to do this and I would say you know, years ago somebody want to do this and maybe a lot of people now the way they would do it is okay, I'm going to start googling things, I'm going to collect like a Dropbox or Google Drive folder, I'm going to collect all of the assets, the information. But you sort of been, I would say sticking everything into Sonnet. Uh, I would say why? And maybe you can kind of show us what you've been doing also.

Speaker H: Yeah, of course. So the, so I want to stick all this in Sonnet for the sake of context. Um, because of my lack of context I want to have like almost like a very good EA who's very good at building hardware or has the knowledge of building hardware, maybe not the actual like technical skills by hand. Um, and so what that actually looks like is if I go into like clock projects, I built this for the fuser, um, I would dump all information, email conversations I had with people over the Internet. Um, these are very long chains of conversations. I would just take everything and dump it in here. Um, and what this allows me to do is I can quickly retrieve things from almost like second brain. So I can just be like um, this person, what did Aiden use for his vacuum? Um, and he could immediately just do really good rag on it based on all this information and just give me answers very fast and save me a ton of time instead of just searching on superhuman or Gmail or whatever. Um, and of course it's a lot slower because there's so much knowledge but yeah, I can just get instantly all this information uh, and it just pulls it out Based on the best ideas, uh, or the best, uh, recommendations. Another thing that my friend Olivia taught me was from fuser.net, she ended up scraping this entire website and converting it to embeddings and she made this thing called openfuser. So what this is is literally just a queryable openfuser.com so instead of browsing like all this form, like I don't want to go through this shit. So I can just go like what is the best uh, uh, bubble counter to use? Um, and it will like instantly query it and not only find like an explanation, it gives me a link to it too. So um, here I can read about like uh, different types of bubble counters and it has like all this like knowledge just there and I can just quickly like skim through this without you know, having too much context and things.

Speaker A: Amazing. It was very cool.

Speaker H: Yeah. Anyway, in addition to. Yeah, in addition to just using Claude, you can just build your own things. Like it, like this is kind of the power of software is that you can, you can just take information and digest it into like such a queryable form and extremely fast too. Um, I know she has a public repo for this and she's been an incredible help uh, with this whole project. So things like this are very helpful. And also puts files and stuff here. Um, so just very simple interface, nothing complicated about this, just a streamlit app, but it works so well.

Speaker B: Wow.

Speaker H: Okay.

Speaker B: Um, okay, so. Wow, this is so cool. Uh, and I can see myself definitely wanting to do this for much of other things, maybe not building a fuser but back on the future process. Okay, so if you look at your tweet, you know, you start ordering these parts, they start showing up. Um, so walk us maybe through, you know, what was going on, you know, maybe week by week what was kind of the process like and maybe what were the hardest, you know, most challenging areas.

Speaker H: Yeah, so I think week on week, so I would just start with like fuser, how to build or something and then just like open, you can see I opened like every single link on here. Uh, I watched every single video. So first thing was just digesting everything like into my brain before doing it into Claude and I actually wanted to know what is going on myself and then just kind of reading these setups and like seeing what's going on. So this is like very inspired by my setup or like I. Sorry, I was inspired by this setup at least for the vacuum chamber. And so I kind of mimicked a lot of this. So I just like control. A control C and put it all into Claude. Um, and then went from there. And then I was like, okay, what parts do I need? So here they have some certain materials, but a lot of these materials are actually kind of wrong. Um, and so what I did was with Claude, I would just talk with it and ask it, okay, what materials look right? What don't look right? Um, so, for example, like this, certain nylon spacers. So I just dumped in the McMaster part piece here and it just gave me all these details. Like, I didn't understand what any of this meant. So this would just break it down for me and explain it and then I just go on McMaster and then I would just start looking up things. I'd be like, okay, this looks right, I'm going to buy this. And then, yeah, like, this is extremely intimidating, by the way. Like, I like, if you're just like a software guy looking at this, you have no clue what's going on here.

Speaker A: So I would just like back to what you mean. For me, uh, cloud and GPT has been like, build websites, build apps, do specific sort of pointed problem solving. But very much as an excellent. I could write the code myself, but it's just like, solves a few steps for me so I don't have to do the boring JavaScript, like that kind of thing. But if you want to ask me to hardware, like, similar to you, I'd be like, I have no. Like, I'm looking at this list going, I don't know which one's.

Speaker D: What.

Speaker A: What does that even mean? Ah, like, what should I do with it? So it's amazing that you're able to like pick from Claude, go to McMaster website, figure out. Okay, yeah, this sounds about right. Go get that and then go back to cloud, I'm guessing, and just like, uh, stick it all together. Mike, I think some of the sort of a continuation of what you just said, right. I think one time you had mentioned how you basically have the spreadsheet of all the seed that had done really, really, really well like at like, seed stage. And, um, you know, from there on missed opportunities once you'd invested all of that one. It's amazing that you track it and you track it in a religious way. I love that. I just, you know, as like a data nerd, I just love, you know, the fact that you just track the whole thing. But what do you see as like, patterns, anti patterns, or is there no such thing? Or is it just. Is it like a bit like the end of, you know, this movie Burn After Reading where it's like, what have we learned from this? Nothing at all. And it's just sort of like chaotic. What is, what is the lesson there?

Speaker I: So first of all, how, how do we go about this? Um, so I got intrigued by the fact that Warren Buffett and Charlie Munger used to read the annual reports of every company and they'd spend five hours a day reading. And what you realize is that part of why they're good investors is they have a temperament advantage, right? So like in a day by day basis, you can't justify reading every Fortune 500 annual report because there's not a specific investment tied to that activity. But if you read that many reports over time, you have a giant lattice of mental models about what greatness looks like and what mediocrity looks like. And so I thought, well, why don't I do the same thing with startups? And so I started to create a database of startups, um, where you would have made more than 100x on your first check. And I was like, okay, hm, what if I decided, you know, just like Buffett and Munger are kind of like these trained spotters when it comes to understanding Fortune 500 companies, I want to be a trained spotter about understanding greatness in startups, right? Uh, like I want to understand it better than anybody in human history has ever understood it. Right. And so, so I would study these companies. Um, but to answer your question, Arti, um, I would say that the high order bit is there are two types of capitalism. The kind of capitalism most people think about is like normal corporate capitalism, Fortune 500 companies. And those companies are trying to create moats and they're trying to leverage their competitive advantages and they're trying to do all the stuff that you learn in business school that you're supposed to do. Michael Porter Five Forces, you know, all that stuff. And uh, you know, when you, when you bet on one of those companies or when you run one of those companies, you're betting on its ability to persistently compound longer and more intensely than most people think it will. And I realized that startups don't do that at all. Like, startups don't have a moat, right? They don't have anything. All they have is the founders and their ideas. And so what I realized is a startup capitalist is a different kind of capitalist. Like, a startup doesn't create value by persistently compounding. They create value by changing the subject. So they, they show up out of left field with something radically different. They can't be reconciled with anything that's ever come before. The problem with a lot of startup frameworks is they study failed startups and then they say, why did they fail? They failed because there wasn't market demand. They failed because the founders didn't get along. They failed because of X, Y and Z. But why do startups succeed? Why do they have outlier results? They succeed when they do something radically different. That changes.