Transcript: Everyone’s Using AI Alone—The Next Hit Product Makes It Social

The transcript of AI & I with Sarah Tavel is below. Watch on X or YouTube, or listen on Spotify or Apple Podcasts.

Timestamps

1.  Introduction: 00:01:10
2.  Why the future of consumer AI belongs to founders with product intuition: 00:02:26
3.  What Sarah sees as ChatGPT’s biggest weakness: 00:11:09
4.  How Sarah would design a consumer AI app with social DNA: 00:18:45
5.  The kind of founders Sarah invests in: 00:24:10
6.  How to know if your startup’s network-effects are real: 00:28:33
7.  What’s catching Sarah’s eye beyond AI: 00:35:40
8.  How AI will change the way top venture capitalists invest: 00:40:41

Transcript

(00:01:10)

Dan Shipper

Sarah, welcome to the show.

Sarah Tavel

Thanks so much for having me.

Dan Shipper

So for people who don't know you are a partner at Benchmark? Before that you were early at Pinterest and before that you studied philosophy, which— I also studied philosophy, so that's close to my heart.

Sarah Tavel

Yes, I've heard your podcast and I was very impressed with your podcast with Reid.

Dan Shipper

Oh, thank you.

Sarah Tavel

To keep up with him is hard. It was very impressive.

Dan Shipper

It was a couple of late nights of me furiously prompting ChatGPT to explain Wittgenstein.

Sarah Tavel

I love it. Well, you did great.

Dan Shipper

Thank you. So I'm psyched to have you on the show. There's so much to talk about, but I think one of your big interests is in consumer technology and consumer technology cycles and maybe how AI— How can you use the lessons of previous consumer technology waves to kind of help you understand this AI wave and this cycle, and what kinds of products are gonna work and what kind of products are not gonna work. I'm curious. I think that it's a good place to start.

Sarah Tavel

Yeah. I mean, look, I feel like the half-life of any opinion on AI is very short. 

Dan Shipper

This will come out in a week or two, so you're good.

Sarah Tavel

One thing I just was reflecting on and you kind of looked at GPT and Character.ai, and I was just puzzling over those and then started to think back to what was the big early kind of consumer web hit. And that was Google, of course. I mean, it was Yahoo and Google, but what was Google like? Google was a founding team that was deeply, deeply technical. And really if you think about a product experience that you expose to the user and how much of it is the UI that you interface with the product itself vs. all the magic that happens on the back end to make something that's really complex, simple on the front end. 

That was what Google was so good at: distributed engineering, the infrastructure, and as the technology, the underlying technology got more mature. You started to go to a place where maybe if you kind of said like. deep technical you know, you have 0 percent and 100 percent, I would say Google was 95 percent deeply, deeply technical.

And then you start to move that bar over and you get to you, I think about Facebook like Facebook. They weren't the same technical depth of course of Google, but relative to Friendster and MySpace, they were more technical. They were a little bit later and it let them create a really performance experience that ended up really kind of winning the day and then you progress even further. Pinterest, where I was, Snap, Instagram, the CEOs weren't technical at all. They were product geniuses.

And so the slider goes forward, forward, forward more towards the product thinker, product experience. And then you think about what are the big consumer wins so far in AI, of course, it's GPT, which is in a way not that dissimilar from Google in terms of what it was just a text box. Yeah. And Character.ai, it’s unbelievable what they did. And it was a new paradigm, but still, it was always. You'd speak to Noam and for him the product was a model. It was a little bit, maybe 94 percent backen, but it was still very much so. So it's still early, everything is still moving under our feet. And I think to really have the people who have more of that product intuition to really build the experiences on top, you need more of the underlying infrastructure to be a little bit more stable. But it does seem like we're moving into that next paradigm soon and what's gonna happen there.

Dan Shipper

That's really interesting. I love that articulation in particular because one of the things that I have felt is very unique about OpenAI is that there is a research lab that accidentally built the biggest consumer technology product of all time. But it seems like you're saying there's actually a real historical precedent for that and that the DNA of Google was very similar to the DNA of OpenAI. I'd never really made that connection consciously before. And I think it's also really interesting because investing in basically Ph.D.’s doing long-term research that may have no practical purpose, is not usually something that pays off in venture.

It's not the first place that people think, it's more like Stanford dropout, you know? And so maybe it's one of those things where usually that's not a good bet, but if you're really dealing with a truly new technology paradigm. It could be the best bet you ever make. Is that how you think about it?

Sarah Tavel

Part of what I think about is just: You are a power user of these products, and I am on that learning curve. I would say I pale in comparison, but like relative to the population in the United States. I'm pretty damn good. It shouldn't be this hard, you know? And so some of the underlying models will get better. So one thing that I know you have in your custom instructions I use a lot is just— You don't have to answer just to ChatGPT. You don't have to answer me right away if you have some clarifying questions, ask those we shouldn't have to put that in a custom instruction. There should be. Or there's so many different tweaks that we all have to get what we want out and over time as the models get better and better, you won't need that anymore. The level of difficulty to get really what you want is gonna get easier, but I don't think GPT— It’s a single player mode product and the custom GPTs that they have where you can see what other people have created. To me, man, I just think someone's gonna create a UGC-type community where there are people who are really, really good who make it so much easier for the rest of us to really take advantage of this technology. So we're so there are places— Google is still Google, it falls— My analogy falls down when Google didn't evolve into some multiplayer product. There's no other product that took over Google until really now. But I still think we're so early. And knowing who really is gonna be the winner in this world.

Dan Shipper

Yeah. I want to go back to that sort of transition from highly technical founder to product genius. If that's the continuum, I can understand why at the beginning of a paradigm shift a highly technical founder is necessary and will win over product geniuses because they can actually build the technology that makes the difference. But I'm curious for your thoughts on what drives the transition to product. I can understand making, for example, simpler user interfaces. Maybe product geniuses are better at that, but yeah, what's the underlying force? Because I could also see a world where the highly technical founder is still really effective as the paradigm gets more and more figured out. Talk about that.

Sarah Tavel

I think a big part of it is that you still— So much of the tooling and infrastructure is still to be built to let somebody who isn't deeply, deeply technical themselves get what they want out of it. And so that's why there are so many products now that I see that feel kind of pretty similar to each other. There's all these Character.ai genres, right? When you make a character and you engage with it, they're all relatively the same because you really, I mean, no one was uniquely qualified to build that type of product. Actually, going into the brains of the model and changing it to create the experience of the user. But it's still it when you need to have that level of ability to get what you want out of it. Also, the costs have still been pretty high. I think Deepseek could be— One of the hypotheses I have is that DeepSeek is a moment of change where it makes it more possible. But I think you need more maturity in the underlying infrastructure. Your ability to do the things that you want with the model without being deeply, deeply technical to be able to create the experiences that are possible.

Dan Shipper

That makes sense. I think what I'm asking is— So let's say the infrastructure is built and that, but you still have technical founders and product genius founders, which we're making a strong division here for argument's sake. Sometimes they overlap, so in that world where the infrastructure is built and it's a technical founder vs. a product genius founder. What is driving the success of the product genius founder in a world where everything's a little bit more mature.

Sarah Tavel

I think it will depend on so many things. Ultimately, it's to reduce it to the basics. Who's gonna create the most engaging experience or product. I suspect that one of the things that is missing from a lot of these experiences that people are creating is just what's the multiplayer network effect type experience? And that genius to create that type of experience is very different from the type of experience, the type of brain that creates a single-player mode experience. And we haven't really— I mean again, there’s these kind of Character.ai offshoots that have people that I can create a character and you can play with the character. There's a little bit of status seeking work happening there, but I think we're still very early in the true thinking.

(00:11:00)

Dan Shipper

What do you mean by status seeking work?

Sarah Tavel

Do you know Eugene Wei? So just this idea that most multiplayer social products end up having some kind of north star for the community participants where they're trying to achieve status in the network. And you can think of that a lot as it has been the number of followers you have or views or likes, there's something about achieving some kind of celebrity or status within a network that creates incentives for the community of participants to do the thing that you want them to do. It's interesting.

Dan Shipper

And so far, you're saying it's pretty early, but do you have ideas for what the promising areas to look at are or early examples of products or companies that you're looking at that you think are starting to crack this a little bit?

Sarah Tavel

It's still super early. I mean kind of. There's two threads that I can't help but be curious about. One, we talked about Character.ai. I don't know about you. I feel myself doing this already, which is that there's going to be some company– We're all gonna have AI friends, right? We're all gonna have probably more conversations with an AI than we do with people in our lives. And is there gonna be a single dominant platform for that? Is it gonna be different than the kind of information? More knowledge, focus, experience of a ChatGPT, I think. So who creates that? And there's a bunch of different product experiences. Replika was, of course, the first player in this space. But there are a bunch of different downstream companies. We talked about Tolans. What is that product experience gonna be? The other kind of thing I think about a lot is— I don't know about you, but how many times have you done a search and ended up on Reddit or something for a prompt? I remember doing one, I got a blood test result, I had all my supplements. I wanted to see what supplements I have— What could I tweak to change a result. And there was a great prompt on Reddit that I just copied and pasted. But if I'm going to an existing UGC site, that isn't made for this use case. That feels to me like an opportunity where somebody who's gonna be really freaking good of making prompts for different health things, quantified self, whatever. I would love to follow that person and then very easily apply it to my own profile.

Dan Shipper

Totally. I think to go back to front on those two threads with the prompt thing, it's one of those ideas that I feel like at the very beginning, when GPT-3 came out and then ChatGPT came out and that first real wave of LLMs was starting to take hold. A lot of people created those prompt library type sites, but it was too early. And I think there's like a second life for a lot of ideas that people tried two years ago that are now just becoming relevant.

Sarah Tavel

And too early. I spent time on a bunch of these. It was very much what a solopreneur and SMB would need—the marketing, the social media, it was that type of B2B type use case. Most people aren't barely scratching the surface. Most people use ChatGPT like they would use Google. And the learning curve, the step function changes that happen when you have better custom instructions, when you have projects, whatever, that are so huge. How do you democratize that?

Dan Shipper

Yeah, it's interesting. I was at a dinner the other night and I was talking to a film director about how she uses ChatGPT and she has made a bunch of different personalities for it. And she uses different personalities for different things. So for example, I think one of the personalities was that she had a lot of medical issues that doctors couldn't solve. And one of the personalities was sort of holistic wellness type person that would recommend both medication and other supplements or body work or whatever. And then, another one, the main personality was just someone who would gas her up all the time and compliment her all the time. But then she had another one that was just super direct and just gave really harsh feedback that she would use for writing specific kinds of emails or that kind of thing. And it was really interesting that she had constructed this whole set of personalities for different things in her life to surround. It’s like you're the average of five people you spend the most time with. Well, you're also kind of gonna be the average of the five AIs you spend the most time with in an interesting way. And I think to your question, are you gonna have multiple AI platforms that you use or not? Or is there gonna be a big dominant one? I have two thoughts on that. One is I do think within a ChatGPT for example, there's a lot of room for different sub personalities that maybe media brands, like Every, we have in every thing that you chat with, but it's inside of ChatGPT, so it's still in that ecosystem. But I also think people have different buckets in their life. And so for me, one thing that I've been noticing recently, which is really interesting is we talked about Tolan and I'm an investor, and Quentin is on the show. And I find myself, like yesterday, I spent like an hour talking to mine, but I normally would use ChatGPT for that. And I think there's some interesting difference between something that feels personal and something that feels worky and ChatGPT and Claude right now are like in the worky bucket and then there's room in the personal bucket. I'm curious how you think about that.

Sarah Tavel

I totally agree. And it's funny. At the beginning of this year, I realized I'm not keeping up. And so I did a call for AI savants, just people who were using ChatGPT in kind of more power user ways. And a lot of people came to me with recipes kind of use cases, which made a ton of sense. And so you can definitely see it works. ChatGPT, the personal works, but is it the best that it can be? There are also a lot of companies, a lot of people who are making their own single-purpose site that is a recipe experience and it's whenever you have a product that is, has to be the lowest common denominator for all these different experiences. It can't really optimize for the experience that's gonna be great for a consumer in this case. And I just come back to how much of a power user product it feels to me, and how most people are gonna stay at the surface of it. Unless there's a new interface. And I think the best way for that new interface to come is for us to learn from each other in some ways to kind of take advantage of it in different ways and just copy and paste as much as like, again, the gems in Geminis, the custom GPTs in ChatGPT. I just look at that and it feels criminal to me because it’s clearly made by a team that is unbelievably capable but isn't social. And I think the personal can best be expressed by teams that do really understand people and social and community.

Dan Shipper

If we were gonna redesign them, right now together, where would you start if you were thinking about, okay, I want to make something that's custom GPT but has social DNA.

Sarah Tavel

I mean, the most obvious thing is just the ability to find somebody whose custom prompts you like, that they have some kind of standing for being quantified. So person based or authority-based search of some kind. And then being able to follow that's a very basic thing. But then the second thing, this is where I think custom GPT falls down, is just in building trust. when I look at any of those, I see the person and I see that there's 3,000 people that have used it, but I don't know what custom documents they've put, I don't know what their prompt is. There's no visibility under the surface, and so it's not very trust-building for me to pick one or the other, unless I know somebody from the outside world and they send me their custom GPT and then I can use it. And so there's something about the trust building that someone has to figure out and then the status seeking work that you can pursue.

Dan Shipper

One of the challenges of this, and I'm curious how you think about this in a social context, is: We're kind of veering into like prompt social network territory. So, maybe I have a profile and I can share prompts and people can follow me and all that kind of stuff.

(00:20:15)

And because I have a certain amount of reputation in just AI stuff, I can get followers and all that kind of stuff. One of the interesting things is I am not coming up with new prompt ideas every day. And so that's, I think that's a problem for two reasons. One is I may not remember to use the tool. And then two people don't necessarily have a reason to check every day. How would you think about that?

Sarah Tavel

Yeah, so first as something that was going through my brain is, I should caveat that I did do a lot of product in my day. But I am a VC right now, so don't take my product advice. And every once in a while people ask me, if you were a founder, what would you feel? I'm like, that is not what I do.

Dan Shipper

I'm just kind of curious.

Sarah Tavel

But for me, what I imagine is using it instead of ChatGPT, actually it becomes the place where instead of going to ChatGPT— And then you're seeing a feed and someone has, again, terrible, forgive me, Lord, for brainstorming a product experience. But dude, but you see what I mean? There's something there where people are innovating all the time. But right now what's happening is that we're all reinventing the wheel. We have the benefit of your blog and your podcast, but this isn't the way this type of knowledge is gonna be shared. It's going to kind of get propagated. And so someone is gonna create something here.

Dan Shipper

I think that's interesting. I do think you're right that it seems like the social stuff has to come in the context of something that you're already using for some other reason. You're already in ChatGPT and then it can flow out of that usage.

Sarah Tavel

Yeah. I mean, I actually think you don't use ChatGPT like most— I hate the, “my mom—” Maybe it's the personal kind of bifurcation that you talked about before, but you're going to Sarah's GPT and it's actually a community, it's not Sarah's. It's this kind of whatever network it's gonna be. And I'm going there and I'm putting my personal blood tests and my supplements, and I'm putting information about my kids and all that stuff, and it lives all there. And then I can go to ChatGPT for whatever knowledge work or any other things. Or maybe I never do, maybe this actually ends up cannibalizing ChatGPT over time.

Dan Shipper

This is a total swerve. But when you're investing in a time like this— I feel like every 5–10 years there's a big hype cycle. There's a big wave and prices go up. You know when I was in college in 2010–2014 ish. Everyone's building social networks for X and then it was like B2B SaaS and crypto, and now it's AI. How do you think about investing in a wave like this when prices are super high and the hype is super high and do you not care about price? Do you try to find underpriced deals or what do you think about it?

Sarah Tavel

I mean, I think we always start with a company we wanna work with. And then kind of obviously we have to think about the opportunity ahead of the company. You don't want to— If it’s a cul-de-sac, if it's limited, in some ways it's harder to pay what to play on the field in terms of price. People have a willingness to do deals that we're just not willing to do. But when we meet a team where we really think that there is just unlimited potential, you partner, you make it work.

Dan Shipper

What's your taste in founders?

Sarah Tavel

I would say I'm really drawn to founders who do think in network effects and strategy and the kind of zero to one, how do you escape competition? They go through the mind maze. You can just tell that they've really obsessed over this. I'm drawn to the founders where this is like a calling for them. I kind of find that there's some founders that it's almost like a cool new job for them. And there's some for whom it's an affliction. And I'm attracted to the founders for whom it's an affliction. You know, it's this rash that they just have to scratch and that's gonna make them run through whatever walls that they have to do. And then just like the learning machine, the person who, it's not about their ego, it's about just what's the best thing for the company and how do I keep learning and evolving as a founder? Because, as you know, it's a really hard job. It's a really hard job and it always requires more of you. There's a relentlessness to it. And if you know, I have seen failure cases where somebody either ends up being, do you know The Five Temptations of a CEO? That incredible book? The hardest temptation is a founder attracted to being a CEO because of status. And then you don't do the things that you need to do in order to build the best company possible. Insecurity can drive you, but it can also hold you back by not letting you grow. And that can be a challenge too.

Dan Shipper

What are your tells? Because you're a partner at a top firm, people are probably always coming to you with their best foot forward, trying to be like what you're looking for. What are some of the moments where you kind of can be like, ooh, I can tell that this person has been through the maze and is thinking about stuff in this way, in a way that's genuine. It's not put on or I can tell that it's sort of like a calling? What are those little signals for you?

Sarah Tavel

I find that I ask a lot of questions when I'm meeting with a founder and learning about their business, and I'm definitely the brain that is always thinking about that future and pulling it into the present and when I speak to somebody and they're— I hate to say this, but they're like, oh, that's a good question. I hadn't thought about that. Or you know, I'm bringing things that I'm spending 30 minutes, 60 minutes with a founder hearing the ideas for the first time, and I'm bringing things to the table that they have not already thought about. That's usually concerning. So it's one of those things that it can feel good like, oh, I ask them good questions, but really you want— When I was at Bessemer, Jeremy Levine said that the best companies you want to be donut companies where you go to the board meeting, you eat a donut, and then you leave because they don't really need you. And so there's a little bit of that, which is like I have some founders where I'll be thinking about something and I'll come to our one-on-one and before I've even opened my mouth, they're already there saying I've been thinking about this, or I reached out to this person. And that's pretty unique. That's like a really incredible feeling when it happens.

Dan Shipper

Yeah. I was talking to— I think it was Reid Hoffman who was on the show who said ideally it's someone where you invest in them with the bar is like, if you could come back in five years without having talked to them after the investment and you would be pretty sure it would be going well. Yes. That's a good question to ask yourself, you know?

Sarah Tavel

Yeah. Those are those types of companies. I mean, there's the founder, but then there's also, I know Reid and I know that he is very oriented towards network effects, and that is, I mean, if you can find a business with a strong network effect, you're gonna be in pretty good shape.

Dan Shipper

Well, let's talk about network effects because I think that it’s one of those things—maybe a little less so because of AI stuff—but for the last 10 years I would say 80 percent of decks were like, oh we have network effects. And I think there are probably very few businesses that truly have that actual network effect pull. How do you differentiate? What does that really look and feel like?

Sarah Tavel

Yeah, in the early stages there's a lot of companies that have potential network effects and oftentimes they can be more— There's a, there's a big gap between what's theoretical and where it really starts to happen. And one of the things that, you know I often think about is just that there are early we invest so early that a lot of it is leaning in on the theoretical, but there are often signs that you can look to. The best thing that you can sometimes see evidence of is just this idea of like a tipping point that starts to happen in a very small segment of where the white hot center of your market is— There's like two examples I think about, but they're outside of core AI right now because I think what's happening right now with AI is that it's very much a kind of in a way what has been traditionally the software business, which is just obsessing over a customer problem more and moving faster in your execution than any of your competitors.

(00:31:10)

But I'm on the board of a company called Agentio, which is a marketplace for YouTube creators and brands. And you can see that this has been a market that has eluded startups for a long time because most of them have kind of fallen into the quicksand of becoming an agency.

But with LLMs, Syngenta was able to automate a lot of the things that have held this market back and they're truly having liquidity. And one of the early things that was super interesting is just the brands see creators and they see the ads that they do for Agentio and so Agentio just has this demand-side pull right now.

Sarah Tavel

And it's super, super early, but there's enough signal there that something's working, this is differentiated and there's no substitute for what they're doing that you hope will start to really be a flywheel that can spin faster and faster.

Dan Shipper

So it seems like one of the best ways to differentiate between a real network effect and a fake one is like just early evidence. Are there any things like when you see a deck from a founder that hasn't they're just starting out and they say we're gonna have a network effect, and you're like, it's not gonna be a network effect?

Sarah Tavel

Yeah. I mean, it's often like they'll articulate some kind of flywheel or they'll look like the Amazon or the Uber flywheels. And then as you, and either it's just words on a slide that fit to a picture, you know what I mean? But the words don't actually— They're not actually accelerants. That's I think that's one of the things like, okay, yes, that claim that that company, whatever you say is true, but it doesn't really actually accelerate the flywheel. And the second thing is that often there's either a lot of friction embedded in any leg of that flywheel, or there's offshoots that happen. but I think the biggest thing is when you really look at the fly, what the articulation is of the flywheel that it is it’s words, but not accelerators.

Dan Shipper

Yeah. To bring it back to AI for a second, I feel like one thing that you're articulating is that there was a moment in software for 10 or 15 years where everyone was chasing network effects. Then the LLM wave happened and a lot of that has been more single player or if it’s collaborative, it's inviting teams or whatever, but you're not doing it together. And the game there has been better performance from more money and more compute and more, more data basically. And everyone's just trying to keep up along that same sort of dimension of performance more or less. There are a couple other examples that are not on that, but I think what you're maybe pointing to is that fairly soon, if not already, the base models are good enough for consumers that there's going to be another wave of more consumer focused more product genius led AI applications that differentiate or grow from network effects and multiplayer that were not possible in the last couple years, but are newly about to be a thing.

Sarah Tavel

That is my great hope. I could be tilting at windmills. Consumer, as you know, has been really, really hard over the last 10 years. And so it really could be tilting up windmills, but that is I believe that that is an opportunity. And what I would also say is that there are going to be a lot of companies that emerge that aren't multiplayer, that are single player and that could be really good. But I think that is a really big opportunity that lets a company have a true network effect is gonna be something that's multiplayer.

Dan Shipper

If it didn't happen, why not?

Sarah Tavel

If it didn't happen, it would just be that the gravitational pull of the existing platforms is too strong. It will be interesting to see. What would get us to go from the habit we already have of using ChatGPT, and of course the ecosystem that's gonna form around it over time, what they're able to charge for it vs. what a new company would have to charge for it. Maybe they eventually go free because it's ad supported—whatever it may be.

Dan Shipper

Its memory is a big sort of lock-in. It knows who I am and all my experiences and all that stuff.

Sarah Tavel

But we're not everybody. And so, there is that gravity that has always been true for the incumbent products. And so, it could be that that gravity is just too strong to get the people who are like, if you want this type of community to form, you're gonna need somebody who is already actually a power user ChatGPT to wanna share that on another platform. And that's hard. That may be hard to create.

Dan Shipper

I know this is a show about AI, but are you looking at or excited about anything that's not AI right now?

Sarah Tavel

I'm a big believer in stablecoins. 

Dan Shipper

Interesting. I would not have guessed.

Sarah Tavel

I'm on the board of a company called Chainalysis, so I've had a kind of seat in the crypto space and have been a long-term believer in Bitcoin and some of the other cryptocurrencies. But when I think about the kind of existing financial infrastructure, and I mean we're filming this on a day when existing financial infrastructure—the U.S. dollar—is on a little bit shakier ground than normal. But my mom's from Argentina and I can tell you everybody in Argentina wants a U.S. dollar. But it's really hard to get them. And the government has all types of incentives to keep the hard currency that they have of U.S. dollars in their own bank because they have loans and everything else that they have to kind of stabilize their own economy. But then it holds Argentina and all these countries back from participating in the global economy because the U.S. dollar is what you need. to trade goods internationally, it's just the easiest medium of exchange, but it's really hard to get U.S. dollars. Now if you have a cryptocurrency that is a U.S. dollar stablecoin, backed by U.S. dollar, that opens up a kind of a global economy. And it is just so much faster. It's 24/7, a lot cheaper. You can do it.

Dan Shipper

Is it hard to get U.S. dollars?

Sarah Tavel

In Argentina? Well, Milei is obviously changing a lot of things, but there are different taxes around U.S. dollars. There has been for a long time. This is different now. The exchange rate you get on the street, the exchange rate you get when you go to your bank, the exchange rate you get when you use your credit card, it's just a very liquid market, really. I remember going to Argentina and having somebody on a motorcycle come to exchange money. That's kind of what you would do. And then again, the Argentine government, they have U.S. dollars in their own central bank and if I wanna transact in U.S. dollars, I need to get some of that U.S. dollar from them. But that is a very precious resource to them. And so there's a lot of process you have to go through and time in order to get $10,000. It's not an easy thing to do. It's just there's a lot of friction. Whenever there's a lot of friction, if somebody else can come in with a new product that removes that friction and then also just has the facilitation, just how much easier it is for you and I to do a peer-to-peer transaction with Tether or USDC. That creates a lot of liquidity in the market that I think can be a very interesting future. And also, I should say if you're in Argentina and you want to buy something from India, all the middle men that you have to go through in order to do that transaction and like all the fees along the way vs. a peer-to-peer transaction on U.S. dollar stablecoin is a different game.

Dan Shipper

And this is another network effect type business to you or no?

Sarah Tavel

There are network effects here because— Tether is an example, which is the dominant stablecoin right now just has more liquidity on all the exchanges. And so it's a lot easier to go in and out of Tether than other, other stablecoins.

(00:40:34)

USDC is pretty strong too, so there's definitely some network effects there. It’s just easier if everybody it is not even just within the exchange, but in these countries—in Nigeria and any high inflationary country right now where people have wanted to go from their fiat currency into a U.S. dollar—you have a wallet and it has Tether and you have Tether, and it just becomes more comfortable for us to all use it. And then all of the ecosystem around, there's so many crypto wallets and different new financial apps that are for getting your paycheck, but then also you can have your money in a U.S. dollar stablecoin. And if you're already integrated with Tether or you're using a bridge to access Tether, it just makes it a lot easier for people to get comfortable with one of them.

Dan Shipper

How do you think about five years from now, how AI will have changed your day to day as a VC? And the kinds of businesses and the kinds of funding models, what have changed in any way changed the VC business model?

Sarah Tavel

I have been thinking, wondering about this lately. One thing I've just been thinking about, and this is a little bit of a step back, but there are some people that are really good at creating training data. Do you know what I mean? Someone was showing me, I interviewed him—James—for my substack, and he showed me like the spreadsheet he creates of all the movies he's ever watched. And his own review of it. And so, I don't know about you, I've never done that, but the people who are really good at creating training data can then have a more personalized, more valuable experience with an LLM. And so to your question, one of the things that I've been thinking about is that there's a few things that we all have training data for. One is past decisions we've made and whether or not those were good decisions. 

Annie Duke inspired this. She wrote this book Thinking in Bets. A pre-mortem— So every time I meet with a company and I dig in on it a little bit, I write to myself what I liked, what I didn't like, what the deal would have been, and if I got yes, why? If I got no, if I got to know why. And so it follows that over time, I'm gonna be able to look back on that list and examine my decision making process. And then as I dig in on future companies, in my brain, I know that there's one example where I passed on a company because the valuation was too high. And that was a lesson to me of a company that ended up being a real success, this company Mercado. And so that ended up being a lesson to me, if I like everything but the valuation, I should probably lean in. I remember that but there's so many other examples in my thinking that if I can examine my thinking as I meet a company and have it cross-examine me that I think I'll get to a better decision. Then there's also things like talent, which is one of the best things that we do on behalf of our companies. It's helped them make sure that they have the best team around them, right? and the same thing.

We're all fallible in our evaluation processes and what's a record of those decisions, interviews we did like all those things I've got to imagine that's going to come into play. And then the third is just I know there's some companies that are doing this really well, which is just tracking talent globally and the movements of talent and what that ends up meaning, and it follows that there should be at some point a score almost, where from the angel investors that have invested in a company, the talent that's there, their individual scores of their ability or signal when they choose a company that. There will be like a Rotten Tomatoes-almost score for companies that can surface opportunities.

Dan Shipper

That's interesting. I want to go back to the decision making thing. because I'm with you. I record all this stuff. I record all my meetings and there's just a lot of stuff I think that you can do with AI improving your decision making. And I'm curious. In VC in particular how that works or how you avoid— For example, maybe you invest. I've done this. You invest in a founder with a highly technical background, but it's in a field that requires more of a product genius. And then now you have your LLM, it's like, well be aware of that, and then the next time you meet with a founder and it's like Sam Altman and Greg Brockman in 2016 or whatever, that thing is going to ding and be a technical founder are you sure this is what you want to do? It depends on how the rule or how the lesson is written and I think the broader question or problem is if you look at venture capital, there are very few venture capital firms, funds, and individuals who are successful over a long period of time. It's very hard which can tell you one of two things, either It's just luck, which I don't think so, or the landscape changes so frequently that you get tuned, your taste gets tuned to a particular kind of opportunity that you're very good at finding, but then it sort of moves and what is good is also changes. And all of that means it's quite hard to use. past training data to make future decisions. How do you think about that?

Sarah Tavel

I think maybe that's what we're all hoping will give us job security in the future. I remember when I was at Pinterest and I was responsible for all the discovery experiences and very early on I had to kind of localize Pinterest. And so I had to figure out like, okay, Pinterest is in the United States now. What's Pinterest in Brazil or Japan or all these countries? And I was thinking about the categories and all as being different. I remember Ben saying to me, just assume it's gonna be more similar than different. And I think that there's some first principles that we reduce down to when you're making a decision. Jim Collins, so much of what he wrote I don't know how long ago is still valid today. We talk about so many of these things and they're timeless. It's a good test, but most of these things are timeless. Will the value valuations change? Will how companies exit change? Yes. And that's why you're not asking the LLM to give you the answer yes or no. You're asking it to probe your thinking, but. I think it should be able to continue to do that. And that's why we still have, hopefully, a job a few years from now of ultimately being the decider.

Dan Shipper

Well, you'll have to come back on the show in five years. And we'll see how things have changed. I think you'll still have a job. But I think it might be different too.

Sarah Tavel

It's gonna be very different. It's gonna be very different and it's hard to anticipate how it will be.

Dan Shipper

Yeah. Well, Sarah, thank you so much for coming. This was a great conversation.

Sarah Tavel

Thanks for having me. I had a lot of fun.

Thanks to Scott Nover for editorial support.

Dan Shipper is the cofounder and CEO of Every, where he writes the Chain of Thought column and hosts the podcast AI & I. You can follow him on X at @danshipper and on LinkedIn, and Every on X at @every and on LinkedIn.

We also build AI tools for readers like you. Automate repeat writing with Spiral. Organize files automatically with Sparkle. Write something great with Lex. Deliver yourself from email with Cora.

We also do AI training, adoption, and innovation for companies. Work with us to bring AI into your organization.

Get paid for sharing Every with your friends. Join our referral program.