Pod Insights Summation with Auren Hoffman Gokul Rajaram of Marathon - How AI Will Reinvent Ad Platforms About this episode Gokul Rajaram is a founding partner at Marathon Management Partners and former product leader who helped build Google AdSense and Facebook's advertising platform. He serves on the boards of Pinterest, Coinbase, and The Trade Desk, and previously led product teams at Square and DoorDash. In this episode of World of DaaS, Gokul and Auren discuss: Building massive advertising platforms at Google and FacebookHow AI is transforming content recommendations and adsThe transition from operator to ven...
Transcript 10,105 words · 2 speakers
Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.
0:02 Auren Hoffman? Hello fellow data nerds. Welcome to World of DAS. I'm your host, Orin Hoffman, CEO of Incubate and GP of Flex Capital. Discover more episodes, get weekly data as a service news, original content, articles on data and more at worldofdas.com. That's world of das, D-A-A-S, dot com. Hello fellow data nerds. My guest today is Gokul Rajaram. Gokul led product teams at Google, Facebook, Square, DoorDash. He was the creator of Google AdSense. He helped scale the Facebook ad platform, led Caviar and merchant products at DoorDash through its IPO, and he currently serves on a number of boards, including Coinbase, Pinterest, Trade Desk, and earlier this year,
0:45 Auren Hoffman? he launched a venture capital firm called Marathon Management Partners. Gokul, welcome to World of DAS.
0:51 Speaker 01 Orin, thanks for having me. It's so exciting to be here.
0:54 Auren Hoffman? I'm really excited. Now, you're part of building two of the biggest ad platforms, Google AdSense and then Facebook Ads. What are some of the core principles that translate into these different advertising models?
1:06 Speaker 01 It's an interesting question. You know, I've seen now over the last 20-plus years, a bunch of ad platforms, been closely involved with two of them, and then with Pinterest and Trade Desk, seeing different kinds of ads. I think one thing I've learned is that to be successful, an ad platform has to balance both user value and advertiser value. This means that your metrics, your leading indicator of success can't just be advertiser-focused, they also have to include user focus. And I think one of the most interesting things there is how you select which ads to show, which is targeting and running the auction, and also, where do you show these ads? How do you make sure that you're not overwhelming a piece of content with ads? So where you place ads, ads placement, campaign goals that you want to serve.
1:48 Speaker 01 For example, if you started serving in search, if you started serving ads that are not relevant to your search, but relevant to your identity, it would look very, very incongruous and probably would not be interesting. And then similarly, the ad format also matters a lot. Obviously, the powerful thing about, again, search ads was the fact that the ad format looks exactly very, very similar. Similar to the S E O ad format or the organic ad format. So all of those things together, targeting, campaign creation, campaign goals, inventory ad format, all of them have to be consistent and really in keeping with what the broader platform is trying to achieve. So Facebook's ad system, good example. The initial ad system of Facebook when I started at Facebook was actually very much like Google Ads.
2:32 Speaker 01 In fact, I think some engineers had just taken Google Ads, they read some papers and just copied Google Ads. So you could enter keywords and you basically just hit on keywords. And what is crazy is that it doesn't make sense. Why? Because in Google, when you bid on the keyword accountant, what do you get? You get people are searching for accountants and so your ad shows. But in Facebook, when you bid on a keyword accountant, there's no search context. You are actually targeting accountants. So we even at Facebook before I signed, we had built a system to take your Google campaigns and port them over to Facebook with one click. But of course, that doesn't make sense because you take the keywords in Google and the same creative. We had to rebuild everything from ground up, looking at Facebook's unique value, the context, everything had to be changed. And so that was the
3:17 Speaker 01 real lesson. You can never copy what works because what works in their platform, their context will not work in yours.
3:23 Auren Hoffman? And really what you're just saying is this is content that the user wants to see. And the reason why the Google Ads are so powerful is they're actually things that I want to click on. They're relevant to what I want at that moment. And a lot of those Instagram ads are actually things that I'm excited about. I can't tell you how many shirts I've bought or hats I've bought or something from random Instagram ads. I may be because I'm just doom scrolling while I'm bored on a conference call or something. It's more exciting than that, but it works.
3:57 Speaker 01 It works. Exactly.
3:59 Auren Hoffman? It still seems that content recommendations are still not there. When I get content recommendations, even on Netflix or Amazon, which have so much data about me, they're often pretty subpar. I don't get good movies. I don't even get good music on Spotify anymore that I really want to listen to. Why is it so hard?
4:18 Speaker 01 I think part of it is because in the past, till recently, systems have not been able to preserve context on everything you've done. But now with embeddings and generative AI, this is why if you look at the Facebook revenue growth, it's been so dramatic this past quarter. Because from all that I hear, a lot of the innovations in Gen AI, all the stuff that Llama, et cetera, but even using third-party models, Facebook has really dramatically transformed both the advertising platform.
4:44 Auren Hoffman? It's not just like an old collaborative filter or something. It's very, very based on what I'm interested in right now.
4:51 Speaker 01 It's looking at everything you've done. And so you can maintain context of a million tokens per user, and you can see everything they've done and be able to choose much, much, much better.
5:00 Auren Hoffman? They're maintaining a million tokens, Chris?
5:03 Speaker 01 No, I don't know if they're maintaining a million or not.
5:05 Auren Hoffman? Okay, well, that would be amazing.
5:08 Speaker 01 You can do up to a million. I have no idea how many they're maintaining, but essentially their ability to show content, both organic and ads, has dramatically improved. And I think this is going to be the case across all platforms. Every content platform that's not using Gen AI and deep learning and basically reinforcement learning is going to be falling behind. So they have to be. This, this is the reason. And this is why I think Facebook is able to make this kind of CapEx investment, because there is a massive path forward for advertising. I think the acceleration they'll see in the next few quarters.
5:39 Auren Hoffman? Amazon, Spotify, and Netflix, I mean, these are all smart companies with super smart engineers that have massive capital. They can do lots of investments. I think their recommendations still completely suck. Why have they not somehow moved to what Facebook is doing?
5:56 Speaker 01 I don't think you can just use off-the-shelf models at that scale. You've got to have people who basically are inside your company who are doing research. So I think a lot of the FAIR, Facebook Artificial Generation Research stuff, Fair Labs, that stuff has been applied. I don't think they've built native Gen AI capabilities the way Zuck and Facebook invested in it over the last few years. So I think they are building it now, but they are a little bit behind, but hopefully they will also invest. But I don't think you can just use, for that scale, on Netflix or on Amazon, I don't think you can just use OpenAI or Claude off the shelf. I think you need another layer at the very least, because you have so much data, you have so much data, you've got to figure out what the right token size is, or the right token window, what the embeddings are. It is much beyond what a normal company would use.
6:41 Speaker 01 It's a different beast.
6:43 Auren Hoffman? To me, part of this thing is personalizing the product. When I listen to a song like the DJ College shout out in the song, I would love them to shout out me and my friends in the song. I'm listening to it if I'm working out. That would really pump me up. If I'm watching a superhero show on Netflix, seeing me and my family walk by as extras when the meteor is coming, that would be really exciting. That would get me engaged. That's so easy to do. Why is the personalization not happening?
7:17 Speaker 01 All the things we talked about were text stuff around recommendations. Now we're talking about injecting video and making it seem that we're talking about audio models, all of these things. So I think it's going to happen. The thing you mentioned, the scenario is pretty incredible, but I think in the next five years, these kinds of situations are going to be commonplace. And if they don't do it, they're going to be risking someone else doing it.
7:38 Auren Hoffman? Someone will do it for sure. Yeah. If you were advising OpenAI on how to do ads, and obviously they have to think about it completely differently in this chat interface on ChatGPT or perplexity or whatever, how would you help them design a system?
7:54 Speaker 01 Good news is they have a lot of great people who've joined recently, leaders who are going to design it. But one of the most interesting things over these AI systems is Google search has your intent in some ways and Facebook had your identity. With these AI systems, they have both intent and identity. So it's pretty amazing because on one side, people are using them for commercial queries, to find recommendations for commercial content. For those, it's a pretty straightforward thing to actually add useful ads in addition to commercial recommendations, just like Google search has done. That set of queries, you can do that. And for the rest of the queries or the rest of your engagement with these apps, you basically have their identity. Because as many people have done this experiment, if you or I were to enter into ChatGPT or Claude, tell me, draw a picture of me based on all you know about me,
8:39 Speaker 01 it is going to draw a pretty, rather pretty good picture.
8:42 Auren Hoffman? Yeah, I've done it. Super cool.
8:44 Speaker 01 They know as much about us as maybe a Facebook or someone because they've inferred it from all the things you've done. So given that, shoot, you have identity and most of Facebook Ads are based on identity, on your implicit associations with organizations, with loyalty programs, with restaurants, et cetera. So you have that set of ads you can show. I think the big question for these assistants, for these kinds of ads is going to be where to show them, how to introduce them into the experience. I know some of the larger AI apps have been experimenting, and a bunch of them have pinged me over the last few months, and we've discussed different things, I think. Gmail was a great example. Gmail, because I was involved a little bit in helping them think through advertising on Gmail. I don't know if you've ever clicked on that in Gmail, but Gmail actually does quite well in advertising.
9:29 Auren Hoffman? I thought they did, and then they took the ads off. I don't know if I've ever even seen an ad in Gmail.
9:34 Speaker 01 Last five years, I myself am not sure, but till about five years ago, they rotated content and ads into the same slot. In other words, it's very important, if people realize that a certain slot is always going to be for ads, you basically start ignoring that. So it's very important to start showing useful content, eighty percent of the time or fifty percent of the time, the remaining fifty percent show ads. I think this is going to be very important.
9:56 Auren Hoffman? X does that too. So they'll have, here's the people you should follow, and some of those people are sponsored, but some of those people are actually algorithmically the people you should follow.
10:05 Speaker 01 Exactly. So I think that's the best practice, where you have to intermix, interleave content and ads in the same spot. And that's what I think web publishers didn't do well, where they would always put a banner ad on the right, and none of us can ever remember back in the day what the hell the banner ad said, because we all got ad blinds quickly. So the key is to not get ad blind and prevent it from exactly what you said, interleaving content in the same slot. Never have a slot be, they would have a notion of a slot, in fact, an ad slot. Every slot can have some content and some ads, you just don't know when, and it's determined by the algorithm or the AI. But that's the most important thing, to make sure there's no ad spots. That's why in Facebook, I think the beautiful thing is, as you're scrolling, you never know whether an ad is good.
10:47 Speaker 01 It's not like every five piece of content, it's a smart feed.
10:51 Auren Hoffman? Yeah, it's just part of the feed. What I like about Facebook is that the ads are good, and it's content that I want to see, just happens to be paid content, but it's content that I want to see. On X or Twitter, I upgraded to not see the ads. You can pay X dollars a year to not see the ads. And the main reason I did that is because the ads were terrible. I just never saw an ad that I wanted to see, so I just decided I would pay to not see them. But I would happily want to see ads if they were good.
11:19 Speaker 01 And look at what Google did. I was just looking at Gmail to refresh what they did. Of course, what they've done is they've put ads as emails. Their ads are actually just stacked sponsored. It's the ultimate blending of ads and content together.
11:32 Auren Hoffman? World of DAS isn't just a podcast. World of DAS membership is a private invite-only community for founders and executives of important data and AI businesses. It's a place for off-the-record, high-trust conversations focused on the biggest questions data leaders are facing. The community is about peer learning, curated intros, and access to private events. Everyone in the group is building or leading something meaningful in data. If that sounds like you, apply to join at worldofdas.com. That's world of D A A S dot com.
12:13 Auren Hoffman? Exactly. If they show me an email that they know I want to see, that would be good, as long as I actually want to see it. Is it spam or is it content? This is always the debate.
12:24 Speaker 01 And that's when users, I'm sure Google searches ads and so on. So it truly knows a lot of stuff about me and is able to surface really interesting offers for me and ads for me as mail content. So that's a good example. I think the right ads in Gmail should be email offers, and that's what it is.
12:39 Auren Hoffman? Netflix, they're mainly just promoting their own shows. That's essentially an ad. So they're always promoting their own shows, and they've gotten to a point where they just only promote their own shows, and it's just not good for the user anymore. So they think of it like that ad there. But the number one Google rule that you had is you have to do what's in the best interest of the consumer.
13:02 Speaker 01 The reality is, every ad system in history, from eBay's sponsored ads to Facebook, if you, once you start injecting ads into a purely organic setup, engagement will degrade. It's an axiom almost. And so the only question is, how is you want it to degrade? Mark Zuckerberg used to have an engagement budget every year as to how much engagement degradation he was willing to permit. And so the ad team and the newsfeed team had to really stay within that budget. So every company should be measuring. And probably is measuring through holdout groups and so on. So you have holdout groups of people who never see any ads. And so you understand people who see ads versus that holdout group, what the engagement degradation is. And then you want to hold that degradation to a certain amount per year. But then you do other engagement enhancing things on the organic side to counterbalance that.
13:45 Speaker 01 And so you're always had this tension between advertising and organic. It's a good tension to have. You need ads to monetize and grow the business, but you also need the organic side, the content recommendations to get better and better.
13:57 Auren Hoffman? If I'm an ad buyer and I want to buy ads, if you think of things like X, Netflix and Spotify. In those platforms, the richest people opt out of ads. They can afford the premium version of Spotify, the premium version of Netflix, the premium version of Hulu or whatever. And then they don't actually see the ads. And as an advertiser, I might want to advertise to the richest people. The amount of opportunities to advertise to these very wealthy people has gone down. What would be your advice to the advertiser?
14:29 Speaker 01 I think the interesting thing is, even though people want to address the richest people, the reality is there's not that many of them. Most people do want to hit the mid-market and below.
14:38 Auren Hoffman? Yeah, if you're Coca-Cola or something.
14:40 Speaker 01 Exactly. So I think it's probably a smaller set of advertisers who care. And I bet that even for the premium tiers, there will be ads introduced back again if they're not today. If there's one thing I realize is every company, you either die, as I come here, you're long enough to become an advertising company.
14:56 Auren Hoffman? Every company is an ad tech company, essentially.
14:58 Speaker 01 And a fintech company. I never imagined Netflix would do it because they were so against it for so many years. But now that they have capitulated, I know this is an axiom of, again, every company is going to be going out as a company, just like the AI systems are.
15:12 Auren Hoffman? You have a great tweet about how the future of AI is shaking out. Do you think most of these categories are winner take all? Or do you think there's going to be a plethora or like a group of winners?
15:23 Speaker 01 I think almost in every category, unless there's a very strong network effect, there's going to be a few winners per category. We're already seeing that in the foundation model. I don't think it's going to be a monopolistic situation. There's going to be multiple winners. That said, I think it's likely that the number one player in the category will capture fifty to seventy percent of the value. The number two captures twenty to thirty percent of the value, and then everyone else together captures ten percent. So I think there'll be three to four companies, but there is importance to be number one or number two, because I think the rest of the players are going to be fighting.
15:52 Auren Hoffman? Today, companies are extremely promiscuous. We're constantly moving between the model providers and really looking at quality, price, continuum. Even there, we don't want to be too overwhelming to one. But I could see if they start memorizing the context more of what we're doing, that we're going to get a lot more locked in. Is that, you think, the ultimate strategy?
16:15 Speaker 01 I think it depends on what the company's ultimate goal is. There is a theory that the reason OpenAI dropped the price of G P T five so much is that they almost want to burn the boats, and they want to say, look, we are going to make enterprise almost a zero-profit pool business over time.
16:31 Auren Hoffman? You already saw some people move over from Anthropic. Even if G P T five was hundred percent is good, maybe it was hundred twenty percent, but maybe it was only eighty percent is good, but it was five X cheaper. I'm going to likely at least try it out.
16:44 Speaker 01 And then the profit pool shifts. So your profit pool goes from enterprise to the consumer side, and then you have ads and subscriptions monetizing it. Now, an interesting question for me is, is there a profit pool in enterprise? And if so, is it at the foundation model layer? Or is it the application they were building agents and adding more value and so on? It might be the true pricing strategy and so on, because Google and others have other things they can do. They're trying to almost raise the ground in some ways for these token-based models and token-based pricing strategies. And say, look, that's not going to be, we are signaling that this is going to be a zero-profit pool thing. Anyone who tries to make sustainable profit of just selling tokens, any model, that's not going to be the case. We are going to systematically eradicate the profit pool. And that's what happened in some ways in ad sense.
17:28 Speaker 01 There is a saying that says commoditize your compliments, and so they're commoditizing their compliments by just saying, yep, that's complimentary. Our core business is this.
17:37 Auren Hoffman? So the core business is B to C, essentially.
17:39 Speaker 01 It's B to C, exactly. And topic, I bet a chunk of the revenue comes from. Enterprise today, obviously, from Cursor.
17:47 Auren Hoffman? They're going to have a billion dollars from Cursor or some crazy number like that.
17:50 Speaker 01 Exactly. So the question is for them is going to be how they adapt to this.
17:54 Auren Hoffman? Replit.
17:55 Speaker 01 Replit, exactly. All of the V zero, Replit, Cursor, all of them.
17:58 Auren Hoffman? Bolt dot new, all that stuff. We're investors in many of those companies and a lot of them are becoming a lot more promiscuous. They're even using some of these open source models as well.
18:10 Speaker 01 They have to because otherwise you are negative or very low gross margins. So this then changes the axis of competition. It's brilliant. I think historians are going to look back at this period from twenty three to thirty probably, just the wave of innovation and the business model innovation is so incredible. Chat G P T was such a masterstroke in retrospect because it just showed OpenAI that they don't just need to be a developer focused company. In fact, I always say now, any infrastructure company I invest in, I want the company to not just be a defensive company, building deep tech, moving slowly, getting enterprise customers to have the opportunity to play offense, build a consumer brand. Versal, great example, right? I'm going to invest in Versal. And what a brilliant thing. They just went up the stack and made V zero. So I want an infrastructure company that has the potential
18:56 Speaker 01 to do that, like an OpenAI or a Versal, and build that.
18:59 Auren Hoffman? We're also investors in Versal. One of the thing of all these companies, if you think of Versal, perplexity, Replit, Bolt dot new, Cursor, they all have these incredible founders. They're all super technical, but also they've got like this dual thing. They're also super go to market, super good at marketing and mass marketing. Perplexity C E O, he's like a master at mass marketing, at getting into the news cycles, at finding a way in. Obviously, Sam Altman is like the master at it.
19:30 Speaker 01 I think the nature of companies is changing, just like the nature of people that companies want is changing. What a time to be alive.
19:37 Auren Hoffman? How are what people want changing?
19:39 Speaker 01 There was an article in the New York Times that talked about how CS grads are having tough time finding jobs. I got a degree in CS and I feel like thirty years ago, you could do a couple of internships and yeah, you get a job. Today, what companies are looking for is full-stack thinkers, really someone who is not just an engineer, but can be a product person, can think of business problems, can go up and down the stack, is proactive, can really be a barrel over time. And they are looking for people like that, more T-shaped people who can go deep into something, but also have a broad perspective on things. And I think just doing the coursework and doing an internship or two is not going to be enough today. You've got to build a product-thinking mindset as an engineer. So you've got to really be a full-stack. And then you've got to be AI-native. Companies are being weirdly
20:24 Speaker 01 specific in some ways, and many of those are almost unreasonable. They are trying to find people who have truly built with AI and lived their lives with AI and have built their own personal AI stack and AI agents and so on. So showing over the last two or three years, they were not sitting back and doing coursework. They were embracing this wave. They were doing stuff proactively, et cetera. So proactivity, being able to be on the cutting edge of things and be able to be full-stack thinkers. It's hard. I mean, I feel a mid-career professional, they're all having these moments where they are looking back, looking at themselves in the mirror and saying, should I just take a step back and go be an IC now and relearn some of these skills? Because what I learned so far, what made me succeed in the last ten, fifteen years, is not what's going to get me to succeed in the next ten to fifteen years.
21:09 Auren Hoffman? You mentioned proactive. If someone was somewhat smart and just proactive, that puts you in the ninety-ninth point something percentile. Like, that's all I need. It's so rare to find someone who's actually proactive.
21:22 Speaker 01 Exactly. Agency. The two things you have to look for in any role is problem-solving, ability to just problem-solve, just be fearless of problem-solving, and then high-agency. Someone who you don't have to micromanage. He just come to you with problems, with solutions.
21:35 Auren Hoffman? But those skills have always been rare. And it's not that easy just to like, learn how to be proactive.
21:41 Speaker 01 But AI now makes it easier to amplify those skills. Proactivity, you can now have more ways to measure, okay, this amazing tool that exists for two and a half years now. What have you done with it? Have you been, you were using it earlier, there were harder things. Now you can actually say, tell me about five tools you've used outside of chat G P T in your A S.
21:58 Auren Hoffman? Show me what you built.
22:00 Speaker 01 For me, what you built, exactly.
22:01 Auren Hoffman? You have been somewhat vocal about disagreeing with Paul Graham, Brian Chesky about this founder mode essay. Where do you disagree with them?
22:10 Speaker 01 First of all, it's a great framing that Paul Graham came up with around founder mode. Hands on leadership is very important. It's, but it's a double edged sword. I think whether it helps or hurts a company depends on the context and the stage of the company. So where does it help? It's obviously very critical, as you know, as a multi time founder in the early stages of company. You're setting the tone. You're basically being super involved in every single hire, every single decision, every single product decision. And when I look at Google, for example, I was there early and Larry and Sagar, even till about few hundred people were setting the tone every single day in the company. They were basically super involved in it. That level of involvement is needed. Other time it's needed is when the company is either tackling a high stakes problem or even an existential threat to it.
22:52 Speaker 01 Or when something is not going well, something meaningful and meaty, the company is not going well. Those are the times. So I remember when Google Plus was launched, Zuck used to have these sessions called lockdown. Whenever he perceived that Facebook was under some kind of threat, he would basically light a sign outside his office that said lockdown. It would light up and it was almost a beacon for all the engineers at the company to congregate. And he would give a speech saying for the next X weeks, there was no timeframe till we have built all the things we need to build and shipped all the things we need to ship and got into place where there's no longer this existential threat. The whole company is in lockdown. No one's going to work on anything except this.
23:27 Auren Hoffman? I've never heard that anecdote. That's super cool.
23:29 Speaker 01 So when Google Plus launched, even though ultimately it didn't do an existential threat. He became the PM. He's like, here's the things you're going to build. And here's the teams that are going to build it. And I was on ads, but we still contributed. And we had our ads teams, but even those engineers jumped on. And for, I think it was six or eight weeks, no one did anything except these features. And working directly with Zach and he was the PM. And that's hands on. Now, I think on the contrary, if you're hands on in everything, it basically signals to your leadership that you don't have confidence in them and you can't delegate. And then many times you get involved in things where you're not the best in the room at that. So I think you have to understand what your superpower is. Zach today, the AI moment is again a lockdown moment for him. He's, as you can see, taken ownership fully.
24:13 Speaker 01 But when I was there, I mean, a lot of the ads and sales stuff, Sheryl Sandberg would lead. And she was excellent at it and he would defer it to her and her team. And periodically he would get involved, but it was more we would pull on him versus he would push. And I think you've got to have a partner like that. And part of it is having trust and having a partner whose superpower is complementary to your own.
24:33 Auren Hoffman? You've talked a lot about these number twos that are complementary. Why else is that relationship so important?
24:39 Speaker 01 Every company I've been at, now that I think about it, has had someone like that. Larry and Sergei had Eric, Mark Zuckerberg had Sheryl Sandberg, and Jack Dorsey had Keith Revo. DoorDash, Tony Xu had Christopher Payne. Each of them, on one side, gave them a superpower because the reality is every person has a superpower. Even though they are T-shaped, they can do a lot of things. They have a superpower and they have something they really want to do on a day-to-day basis. This helps them.
25:05 Auren Hoffman? And clearly for Zach, it's product.
25:07 Speaker 01 It's product, it's engineering, it's really the vision of the company. But then there's an infinite number of things that need to run in order for that product to sustain, including monetization, the business side, regulations, all of those things. So you need somebody who has an owner mindset, even at that senior level, someone who can get into the details, someone you trust. And interestingly, I think Zach has said this a couple of times, someone you'd be willing to work for if the roles are reversed. In fact, you could argue that Eric was the CEO of the company, actually. And Larry and Sergei theoretically dotted line into Eric or worked for Eric, though they were a triumph over it. And I think Zach has said the same thing. So you almost need a person who's equal. And I've seen the same at Coinbase with Brian Armstrong and Emily Choi. They are so complementary. Brian really runs product engineering and Emily runs a huge chunk of the company.
25:52 Speaker 01 I suspect that all of these folks, it works, there is mutual respect. They're almost like peers and the reason is they have complementary superpowers.
26:01 Auren Hoffman? I have a theory about this. I'd be interested in hearing your thoughts. Usually there's like a number one and a number two, and the number two is very complimentary to the number one. They have very different strengths than the number one. Sometimes what happens is the number one retires or whatever, and then the number two takes over the company. And I think this is universally bad because the number two is complimentary for a reason. That particular company needs the number one's strengths as number one.
26:28 Speaker 01 Steve Ballmer and Gates.
26:30 Auren Hoffman? Yeah, exactly. Sheryl Sandberg, I think she could run many companies, including JP Morgan, but I don't think she should have run Meta. I don't think she would have been the great number one for Meta because she was there as the complimentary person. This is my crazy theory. I'd love to know what you think of it.
26:47 Speaker 01 I actually think Sheryl can run any company she wants, including Meta, but she would need a complimentary product person who's just as good as her that she respects. And I think knowing her, she would hire somebody like that.
26:57 Auren Hoffman? If you think of Tim Cook, Tim Cook was complimentary to Jobs. Now, obviously, he's done an incredible job as CEO. But I still feel like Apple would have been better off if it had more of the product person as the leader and someone who's super operational as the number two.
27:14 Speaker 01 Look, I think Uber has proved it wrong so far, at least where Dara has had some amazing product people as his sleep tenants. So I think it can work.
27:23 Auren Hoffman? In any case, it's hard to succeed a founder in any scenario. So usually it doesn't work.
27:28 Speaker 01 I mean, they are one of one. Each of them is one of one.
27:31 Auren Hoffman? That's a good point. Now, I'd love to ask you a question about boards. You're on a bunch of boards. I've been in many boards in my life, including many I've run, and I would say pretty much all of the board meetings I've ever been to, including all the ones I've run, have been subpar. I've not done a good job of running them. What have you learned about not only being a good board member, but having actually a good board meeting?
27:53 Speaker 01 So to be a good, good member, first of all, you need to care about the company and the industry. You need to care. In other words, I don't think once you join the board of a company where you don't care about the problem they're solving, the customers they have, and that you're not willing to be proactive about learning about this industry and sitting through this. So I think, for example, I love Procter and Gamble as a customer, but if Procter and Gamble were ever to invite you to join the board, I don't think I'd be a good board member because I honestly don't care about that problem specifically to be proactive with what they're facing. I always advise people who call me, hey, should I join this board? Do you really care? Have you ever in your life, or are you going to be proactive about thinking about this company and the problem? So that's one, and I think this is a good one for you to test prospective board members. Do they care about this? What have they shown? Just like we say, show me evidence that you're proactive.
28:38 Auren Hoffman? And by the way, if you're a B to C company, do they already love your product? So they're already into it.
28:42 Speaker 01 Exactly. If you're X and you go and find someone who's never used X, might not be a good executive or a board member. The second thing you learn is that as a board member, you have two or three jobs. One is governance, making sure the company is running well. The second is fiduciary, where you are a representative of the shareholders. Shareholders in some ways have put you into the seat and you are kind of the nominal boss of the CEO and the managing team. And the third is strategic. The second and third are the most interesting ones. First, you've got to hold the CEO accountable. You've got to make sure the CEO is the right person in the seat. The last one is the thing that you do most often, which is strategic. You've got to be asking the questions, not around day to day operations and financials. What's happening this quarter or this year?
29:25 Speaker 01 But is the company set up in light of the competitive environment, dynamics and everything else going on to thrive over the next few years? What are the challenges around the corner that the company might or might not be looking at? And so when I first joined boards about sixteen, seventeen years ago, I was very directive in my feedback. Let's do this. And I quickly realized that that doesn't make sense because. The executive management team is running the company, not me. So my goal, and ultimately I'm going to hold them accountable, they have to make the decisions. I always now ask questions in a Socratic way. Have you thought about this? Have you considered this competitor? So Socratic questions that really still frame the dialogue and discussion. Rather than having an opinion or something.
30:08 Speaker 01 Having an opinion and you try to gently nudge and make sure they have considered enough things. So the right board meeting, in my opinion, is a three to four hour meeting for public companies or later stage companies. I think for younger companies, two hours is more than enough. Three to four hour meeting where the first hour you start with an executive session where the CEO gives CEO and the board meetup. The CEO talks about what's on their mind, fresh off, good and bad, what's going well, what's not going well. That's about thirty, forty minutes, then the CFO and the leadership team joins the board meeting. And then we talk about the financials, which are obviously important. How the company typically it happens around the quarterly earnings, so how the company has been doing. Quarter looks like what the year looks like, all of that. So you get done with that, planning for the year. So that's about one and a half hours.
30:52 Speaker 01 After that, you spend two hours, two topics, one hour each or forty five minutes each, which are strategic topics. One can be product centric, one can be business centric, could be channel partnerships, could be a new product, whatever the cases. And those are the topics that there's a lot of time spent ahead of time discussing what those topics should be. And typically I like one product and one business, typically led by a product team. And you also get to see other folks in the boardroom. You want to make sure obviously the execs and everybody also want to see the next level and you have the discussion. So really, those are things that pre-just center a little bit ahead, but really the goal is to not present, but as much as discuss. And really we open the board and answer the challenges, problems you're facing and have a discussion around this.
31:32 Speaker 01 The board needs to really come much more than any session, those two sessions really with a point of view around what questions to ask, how to challenge folks, how to bring the diverse set of experiences the board members have to each of these discussions. And some board members will be more actively engaged in the product one, some will be more actively in the business one. But you want, again, if you feel your board members can't actively engage in one or the other, you don't have enough, then you probably need different set of board members.
31:57 Auren Hoffman? The only company I know that does this is Meta. I'm sure there are others where they actually have two boards. They have a board of directors and they have a product board. Some of the people are on both. Drew Halston or Charlie Songhurst are on both the product board and the traditional board of directors, but they're different and they have different goals and different objectives. And I assume they have different topics that they dive into and the CEO is going to get different things out of them. Have you heard of other companies that have something like this?
32:26 Speaker 01 I've never heard of any other company doing it. That's pretty impressive that they have two different boards. I didn't know this, to be honest, till you mentioned this. I'm impressed with that. I think most companies getting the board together once a quarter and then there are all the committee meetings and so on is hard enough. So running two boards, obviously there's overhead on the company. You already make sure it's worth it, but I'm sure it's worked well for them with those kinds of folks in there. But yeah, there's product and business. And then finally you wrap up with feedback from the board on the meeting, on everything else you've seen and heard for that day with the CEO initially, and then you have an exact session without the CEO at the end. It's like cleanly, many of the boards I'm on now, we've gotten to the structure where CEO plus board, thoughts, unfiltered thoughts, CFO comes in, CFO managing comes in, talks about the quarter of the year,
33:11 Speaker 01 two or three maybe, but two is better, forty-five minute sessions, and maybe three, sometimes thirty-minute sessions. And then you end up again, everyone leaves, CEO and board, and then the board alone, giving feedback to the lead independent or the lead director to summarize and communicate back to the CEO and management team.
33:27 Auren Hoffman? Now, you recently launched your fun marathon, but you've been an angel investor and you're super prolific. You've been in hundreds and hundreds of really interesting companies. How has the venture landscape changed over the last couple of decades?
33:41 Speaker 01 Fifteen years ago or eighteen years ago, when I started doing investing, it was very clear you raised the friends and family round. Back then, there were no seed investors. You raised around from early-stage VCs, then you went to later-stage growth VCs, et cetera. Now, there ain't any swim lanes. Everyone swims in everyone else's lane. In fact, if you think there's swim lanes, you're living in a different world. And I've seen pre-IPO quote-unquote firms do seed rounds, and I've seen seed firms do like series D rounds, elite series D rounds. So I think- There is no swim lanes. It can be very chaotic and there can be a lot of temptation to change.
34:14 Auren Hoffman? And the strategy is changing constantly as an investor. You can't be like, oh, I have this strategy and stick to it. You have to be very fluid.
34:21 Speaker 01 But it's also very dangerous because strategy drift means that you end up doing many different things. And so you've got to understand where your true north is. What is the core of things that stays constant through all this change? Because if you don't have a core, that basically means that you have no strategy at all. And I think that is challenging. There has to be something that is consistent across all your investments, something. Doesn't need to be the stage. It has to be something. It could be the domain, could be something. I think everyone needs something because otherwise you're just rudderless. You're just capital chasing the best deal, which I think is hard. I think it's possible. In fact, I always look back at my angel. I think of vintages. I have vintage years where I've done thirty, forty, fifty investments, not as many as how many you're doing. But I did more than thirty, forty, fifty.
35:06 Speaker 01 I'm like, I could barely remember the names of the companies in those years. You're doing one a week. And then there were years when I did five investments and I really remember them because I was proactive with those years. Some of them died, some of them did well, but I just had a deeper relationship with those companies and I felt more satisfied with those. The speed at which things are moving and capital is being deployed. It's almost like capital has been deployed and it's hard, at least from the outset. So when you're in, you kind of see there's a method to the madness. But the other thing you realize is that you've got to really make sure you build your own conviction around the company. You cannot outsource conviction to someone else because everyone is playing their own game. If, for example, you say, oh, multi-stage platform fund is leading around, I should invest. Guess what?
35:47 Speaker 01 You better know that if they're leading a seed round, their goal from that seed round could be very, very different than your goal as a seed investor potentially. And so you've got to be very, very careful and thoughtful. And so you've got to build your own conviction. Only then maybe you learn. And so I've many times made mistakes of firm X is leading the round. So, okay, you know, I was on the fence. Okay.
36:06 Auren Hoffman? Yeah, Sequoia is in, I should be in.
36:08 Speaker 01 Literally, you go from saying a no to the deal to a yes. And I would be like, hang on. And then you look back saying, I was a no.
36:16 Auren Hoffman? I've made that mistake many times in my life.
36:18 Speaker 01 You only make that mistake and you learn from it.
36:20 Auren Hoffman? You recently raised a fund. How was that process? What'd you learn from like the process of raising the fund, pitching LPs and stuff?
36:30 Speaker 01 The biggest surprise I have is the importance of patience. Unlike a venture firm where they move fast, I mean LPs are thoughtful. They move at a certain pace and many times they want to build a relationship for fund two or fund three. It's almost like an enterprise sales funnel where you have folks and then some folks could go for fund one. But the vast majority of folks you meet want to see a certain strategy and deployment and all of that. So the team working together before they say they'll invest. And this is true even on the investing side, I've realized as an operator, I always felt able to launch product fast, see impact, move fast. So you're used to certain cycles, reasonably one or two years, even months now, weeks with venture capital. As you know, oh my God, I mean, when your whole life is around making decisions that don't pan out for years.
37:15 Speaker 01 And increasingly, I don't use any markups and so on as any markers of success. It's all about whether or not the company achieves a liquidity event.
37:24 Auren Hoffman? And that could be like 10 years later or more.
37:26 Speaker 01 Which is 10 years. I invested in the seed out of Figma in 2012. It took 13 years, which is why it is a marathon. It takes 13 years from seed to IPO. If you're lucky, some companies have been on 15 years.
37:37 Auren Hoffman? Are there things you look for in founders that most other investors overlook? What's the money ball?
37:44 Speaker 01 The biggest one I look for is can I see these founders operate outside of my pitch? In other words, I'm very, very nervous about only getting to see founders during the pitch they make for me. Because I feel every all behaviors are hidden, amplified, et cetera. And you don't really see the true nature of the founder. So if the only time I'm able to meet a founder is when they're pitching me, I feel that is a dangerous time to invest.
38:08 Auren Hoffman? But any good founder, if they're meeting you, knows that they're pitching, even if it's not during a quote unquote pitch.
38:15 Speaker 01 Exactly. So what you do is you try to become the confidant. And so my goal is to try to build a multi-month relationship with the founder ahead of the pitch. And one of the best ways you can do that, if you can make an introduction to a customer for them, a real customer. Those are the two things even before you invest, that just instantly breaks down the wall. And then actually, if you make an introduction to a potential customer, you can sit in that customer meeting. I realized that recently you have the privilege actually sit in and listen to how they pitched to a customer.
38:43 Auren Hoffman? I just did that in a company where I introduced them to Anthropic. And then I could see, okay, well, is Anthropic actually interested in this product? Are they wanting to buy it? And obviously, so that would also increase my conviction in the company. And then you're adding value and all the above.
38:58 Speaker 01 And when you make an introduction to an employee, the same thing, you can ask the employee, how's it to work at this company? What is the experience like? And so you get two very different perspectives. So trying to really not just get my one perspective, so trying to really have other. And that's what been the difference as an operator angel. You don't have much time, but as a venture investor, you have more time to think about and do these things. That's why I'm always impressed that you're able to do both on a new yourself as an operator and also as an investor. Time is the key thing. The more companies you invest in, the less time you have per company.
39:29 Auren Hoffman? You've actually talked recently about why people in leadership roles should try to like, maybe move back toward being more individual contributors. I can see how that would seem like for many people, a step backwards in their career. How do you engage with them when you're giving people career advice?
39:45 Speaker 01 Good news is it started out by a couple of the folks who worked for me proactively saying they are exploring this. And over the last few months, I have seen ex-colleagues, friends proactively move from leadership, which is manager or manager, manager roles to IC roles, mostly at AI-native companies. I think the key twist is at AI-native companies. I think they were prior to that, working, not working at AI companies. And like you said, it does seem counterintuitive in us to start with, but it's a right career move. Why? Because most AI-native companies, they want to make sure the person can do the work. And they are generally suspicious of people who are leaders and managers without necessarily doing the work. And so signaling the eagerness or ability to do the IC work and doing the IC work.
40:29 Speaker 01 But you have leadership experience, you, I think, have a really good chance of becoming a leader again at this company if you want to, if you can do the IC work.
40:36 Auren Hoffman? Managing people is almost the opposite of doing good work. You're starting to see, there was this ratio where I was seeing many companies where it'd be like one manager to five people or something, and now you're seeing one manager to 20, 25, 30 people.
40:50 Speaker 01 Exactly. Span of control has increased dramatically. And this is the first time in history where probably we're getting to the point where leverage from technology is greater than leverage from people. You used to get as a manager leverage of managing people. Now you can manage AI agents and you can manage technology, do the work of people.
41:07 Auren Hoffman? You get more vendors, you get good at selecting other vendors to help you, and that's better than having a person.
41:13 Speaker 01 Exactly. And as you know, I think from managing people and running companies yourself, managing people is not all that it's touted to be. And so the more time you spend managing people, you actually realize that if I can do IC work, if I can be impactful, if I can be productive, I can get paid well, I can be an IC and I can do great work. Leverage is what is missing and leverage is what is now unlocked through technology and AI.
41:36 Auren Hoffman? There's this relationship between like product engineering, data science, design. It seems like it's changing a lot. How do you see those relationships evolving?
41:46 Speaker 01 I think the actual job of each of the product engineering and data science folks is still the same. So product is still the what and the why. Owning that obviously in a collaborative way with engineering and data science, engineering does the how. And the data science is the, I call it the inside engine, which is always not just building dashboards, but coming up with new interesting insights around pricing, around markets, around hypothesis testing, et cetera, hypothesis of you testing, those will stay the same. What is changing dramatically is how many and what that ratio should be within a team of product to engineering to data science. In other words, and how many people who should do those roles. For example, I have companies now that say engineering is no longer the bottleneck because they can move much faster. Product is a bottleneck.
42:29 Speaker 01 So instead of having one PM to five or six engineers, I'm going to have one PM to two engineers.
42:34 Auren Hoffman? Also, the product people, they can vibe code, they can get things done, they can use vendors and outsource things, etc. They can spin up a contractor if they need to.
42:44 Speaker 01 Exactly. The role of the product managers now goes far beyond. I feel some of them have been put into service as designers.
42:50 Auren Hoffman? You get in a bolt on new, you can just spit it up.
42:53 Speaker 01 You can spit it up. Yeah, you can do AI prototype super quickly. And then you have one designer shared across multiple teams that's making sure that you're not doing something crazy, but you're staying within the design systems they have. Even front-end engineering can be done with these prototyping tools now.
43:07 Auren Hoffman? And of course, there are those engineers who kind of act as product people too. So maybe it's hard to know. There's a tension of how to know like what you need.
43:14 Speaker 01 Exactly. So that's a skill that if you're a college student, the skill, you want people to build this comfort with taking any of those roles. I feel product engineering, designer, I shouldn't say fully interchangeable. I think there's backend engineers who do complex system stuff, which are not. But front-end engineering, design, product and data science, everyone should try to get all four of those skills. At least figure out how to be possible at two or three of them and have a superpower in one of them. That's the example when you're talking about college students and so on, early career folks, people are looking for people who can do multiple things. We used to all look for full stack engineers back in the day, the unit one, but now it's full stack thinkers, product people who are comfortable just thinking about the design.
43:56 Speaker 01 We're all comfortable interviewing customers, doing insights. So got to be comfortable doing those.
44:01 Auren Hoffman? Yeah. And they've got to be AI native.
44:03 Speaker 01 Got to be AI native. They've got to be able to work with AI tools. Exactly. From day one.
44:07 Auren Hoffman? All right, a couple of personal questions to ask all of our guests. What is a conspiracy theory you believe?
44:12 Speaker 01 I don't really bind to conspiracy theories in the traditional sense, but I do think that there are systemic blind spots in how people interpret events. For example, the way the media and public initially fully dismissed the potential origin of COVID-19, right? People are like, oh, it wasn't made in the lab. It wasn't a lab leak. It wasn't just about science. I think a lot of it was politics, group think, and people weren't intellectually curious to try to understand.
44:36 Auren Hoffman? What's your theory about why that happened? Why did so many people dismiss was somewhat the obvious thing?
44:42 Speaker 01 It became a political thing to express your opinions. There's almost a strong peer pressure. If you express a dissenting opinion, you'd be quickly shut down or canceled in some way. It almost was a sign of the times in some way, you could say. And only now that we are actually seeing real evidence with, I think, a bunch of the health authorities have released that it's probably a lab leak. And I don't know whether it was a government knee jerk thing, but then everyone felt compelled. And if you weren't, you're going to be canceled because you're questioning. It was uncomfortable. So that's, I think, a recurring pattern you see. I think knee jerk dismissal of uncomfortable ideas. And so I've always tried to say, I think, any consensus narrative probably should be questioned. It's just become too consensus.
45:23 Auren Hoffman? And before I get to our last question, I noticed that behind you, there's a Settlers of Catan box behind you. Are you a big Settlers fan?
45:31 Speaker 01 Settlers and Carcassonne, two favorite board games.
45:34 Auren Hoffman? Okay, I love both of those as well. Why do you like them so much?
45:38 Speaker 01 They are strategic, but they're also fun. They kind of allow you to think long term within the context of a game. So you're not just thinking transactionally each play of the game, similar to poker in some ways, you're kind of building your hand over time. But it's a fun game and it's also something that young kids can pick up. So it's a great family game.
45:55 Auren Hoffman? I played with my wife and kids last night. I played Settlers. We had a great time.
45:59 Speaker 01 Great thing, kids love it, adults love it. You can play it at different levels. There are movies you can experience like Inside Out, you can experience it at five different levels. It's one of those games you can experience at five different, young kids, with adults, with colleagues. I mean, many companies have Settlers of Catan nights and so on.
46:13 Auren Hoffman? Absolutely. When we were in person, we would always be doing them.
46:17 Speaker 01 Well, next time we meet in person, we should play Settlers of Catan, my friend.
46:19 Auren Hoffman? That would be awesome. I would love to do that. Our last question, we ask all of our guests, what conventional wisdom or advice do you think is generally bad advice?
46:27 Speaker 01 There was a obsession with work life balance. And I think early in your career, when building something transformative, balance is a luxury you can't afford. The reality is that building a great company and mastering a craft. Requires intense, maniacal focus. And I think you see every single person who achieved extraordinary outcomes, don't just clock out at 5:00 PM. You probably have companies, and I have companies, the AI companies are 9:00, 9:06, 9:00 AM to 9:00 PM, six days a week in the office. And some of them are 9:00, 9:07. Actually, it's called 007, where it's like midnight to midnight, seven days a week. Some of those things are challenging, but that's what it takes. I think a better analogy is you have intense periods like that, and then you have renewal periods, and then you have intense periods and you
47:10 Speaker 01 have renewal periods. But early in your career, especially, I think to build something extraordinary, you can't do it without extreme amount of focus and cutting out some of the balance stuff.
47:19 Auren Hoffman? I think what you're saying is almost conventional advice now, at least for our people in the startup team.
47:25 Speaker 01 For our 100,000 tech Twitter friends, basically.
47:28 Auren Hoffman? Exactly. All right. Thank you, Gogo, for joining us on World of DAS. I follow you at GogoR on X. I definitely encourage our listeners to engage you there. I knew it was gonna be great, but this has been awesome and a ton of fun. Thank you.
47:41 Speaker 01 Thank you, Autumn, for having me.
47:45 Auren Hoffman? Thanks for listening. If you haven't already done so, please subscribe to World of DAS wherever you consume your podcasts, YouTube, Apple, Spotify, and more. And please help us get discovered by leaving a review and check out worldofdas.com. That's worldofdas, D-A-A-S, dot com. And of course, connect with me on Twitter, at Auren, that's A-U-R-E-N. Would absolutely love to hear from you.
Previous The LM Brief: Synthetic Data in AI Next The LM Brief: Navigating GDPR Compliance Outside the EU and UK Page updated November 1, 2026 · transcript and summary are machine-generated