Sarah Tavel on AI Startup Moats, Product-Market Fit & Venture Capital | AI Rabbit Hole 2025

Hi Sarah, it's great to be here with you today. Thanks for having me. let's just start simple hoping you can I know there's a lot of startup operators in this audience. I'm hoping you can give us a brief overview of your career and then talk a little bit about what types of companies you invest in and what types of companies Benchmark is looking for. Sure. So good to see everybody. I'll keep I'll keep the intro quick. I was very lucky to start my career at a firm called Bessemer as an analyst. Lowest rung of the ladder. So if you guys get cold calls or cold emails from somebody, that was me. I would like to say not too long ago, but it's a little bit longer than that. and then I was just super lucky to find a company called Pinterest very very early on.

and I was one of the very early users and ended up loving the product so much that I was lucky to lead or co-lead the series A in Pinterest and then it kind of my my love of Pinterest continued and I ended up joining as a utility player became the first product manager at Pinterest was there leading product on a bunch of different teams for a few years and then just felt the call of venture capital again really loved partnering with founders ended up joining Greylock as one of the general partners there on the consumer team and then got to know the benchmark team and realized there's something very special there and joined benchmark as one of the GPS at benchmark and so now been here about eight years actually almost next week will be eight years congratulations and you know very I think we as a firm are very focused on what I you know actually Peter Thiel described as companies that escape competition you know it's what I wrote a blog post.

It's like not about being number one, it's about being number one by a mile. Like how do you become a dominant player that tends to like companies with with real network effects? And so I've I've been very oriented towards consumer networks, marketplaces, but now as you can imagine in the AI world thinking about vertical applications of AI, how do you dominate those markets? We're as a firm very focused on the early stage kind of our aspiration is to be the first board member to the most ambitious founders of our generation and so that's that's what we we really focus on. Got it. very exciting. I'm sure a lot of founders will be emailing you after this. so I was also a product manager in kind of the 2010s era when you were and something that I've noticed is that this current AI boom seems pretty different from the 2010s kind of marketplace era.

one thing I've noticed is that the AI companies do seem to work a lot more. tell me if you agree with that or not. so I'm hoping for you to share some of your observations. How is this different from, you know, that 2010's tech boom era? How is this this boom different? well, you know, there are a lot of ways in which it's the same. Like there's there's there was something very obvious in that kind of early era. I I think of it as a SAaS era where there was know obviously it it ended up it was SAaS first and then and then kind of mobile which which led to a lot of a lot of the kind of consumer wins. but like there was something that was like very obvious about the opportunities that founders were going after. there was kind of almost this copy and paste from perpetual licensed software businesses to SAaS and there was very much a a land grab like it was you know an intense focus on going after these opportunities you know Salesforce was the SaAS version of of a CRM there was you know name your vertical name your kind of functional part of an organization and there was a new company that was going to become a dominant player in that space and and there was something also about how the markets started to really expand like it went from being that a software market was like the US market to a software market being this global market and today like there's a very there's at the application layer at least there's an obviousness to a lot of the opportunities that AI unlocks and also that I think the market is even bigger for AI companies

because instead of just going after software spend they're going after a kind of human capital spend, the amount of money that people spend paying other people to do the work. But there are like definitely some differences. one is like, you know, everything that happened in 2010 I just feel is amped up now. I mean, there's so much more venture capital. There's so many more companies going after those opportunities. And so it means that like the intensity of these teams and how hard they're driving is definitely another level because they just realize and I hear this from you know our portfolio that this is a window and we've got to run hard through it and you know it is a question of like you know there is something about the intensity and what what what comes out of the intensity of an organization and the shipping cadence like if you can ship better faster like you are going to outrun your competitors and so you see that like the six days in the office if not seven days just intensity of the culture but the other big thing is that there's way more existential risk to the companies that are that are being founded right now like in that early era and again I'm thinking a lot about it's mostly B2B stuff right now.

So, I'm thinking about kind of the SaAS era. You know, SaAS, those companies were these magical businesses. They they were systems of records. And what that ends up meaning is that once a SaAS product is deployed in an organization, it becomes incredibly sticky in that organization. It's very very difficult once you start working with a a system of record to move off of it. and and that's why you have net revenue retention for these SaAS businesses that you know 120% net revenue retention if you're a SAaS company is like good you know and so there's like an annuity stream that comes from having a new customer when you're a SaAS business that is very different right now than what we're seeing in the AI companies like the AI companies you know by and large there's like two forms of existential threat or really it comes it comes from all directions.

you have you just have there's a lot of experimentation right now of course and so there's just a higher level of gross revenue turn like logo churn that you would expect versus a SaAS company but there's you know the switching costs are way lower I mean you see this playing out now with you have cursor you have Windsurf you have you know co-pilot I mean there's and then there's like you know Devon ends. There's there's just so much competition and to go from cursor to co-pilot to Windsurf it's it's it's like 45 second switching costs. And so, as you might imagine, like there's the competition just across competitors, like lateral competition is very very high. And then you also have this question that looms over so much of the space right now which is what happens with the foundation models like you start you know you hear it if you think about from first principles it would make sense to me that over time the foundation model companies go higher up in the stack and now at this point they've all talked about doing so and you see things like you know Anthropic's new product with claud code starting to come into the space, starting to try to own the developer in a way that they haven't really owned before.

And so there is this question still despite the very high valuations for a lot of these companies, which is that will it still exist in its same form a year from now? And for some companies, they're very stable ground, but a bunch of others, it's it's still very unclear. Yeah. So, yeah, maybe expand on that a little bit. How are you thinking about investing in companies like that? You know, like OpenAI, Anthropic, Google, they might all be getting into and competing with Cursor. so how are you thinking about that when you invest in these companies now? you know the well I'll I'll caveat it by saying that like the halflife on any opinion in AI it's like you know you have to be very careful what you say because it it's you know no one would have predicted deepseek like you know there there just these unexpected things keep happening in this space and so it really is I I think it really is like part of what I think about is that it has always been important the founder that you you partner with and that you back.

but you know in some ways it has you could it with a SAaS company it was like if it was working like it was going to continue to work in some way and then like it was a question of how big the market would be, what the pull was and then obviously whether the founder could help it get to its full potential but there was a lot of gradient in between. Now it feels like the the strength of that founder and their ability to be adaptable, their ability to to be paranoid and optimistic is incredibly important because if you at any point right now as a founder rest on your laurels, you could be surprised. even open AI again like with DeepSeek that was a big surprise. and so I think that like right now primary is the founder, the intensity, the adaptability, the learning mindset, the paranoia, the ambition of that founder and and then secondly it's like is there going to be some kind of acrruing asset some something that becomes an unfair advantage to that f to that company over time and and the the tricky thing with that and I see this all the time when I meet with founders is that there are a lot of times where something that was hard today for them.

And so they might associate a founding team might associate a lot of IP with some difficult work that they had to do upfront. But the the the risk is that as these models get better and better, it becomes easier and easier for that next competitor to come into the market and compete with them. And so you can't rest on the laurels of anything that you've done so far. You have to constantly have that paranoia, have that, you know, recognition that the ground is shifting very, very quickly underneath our feet. And, you know, only the paranoid survive right now. I I really like that take on founders. It's very interesting. I'm curious like what is separating the AI startups that you're seeing that are really gaining product market fit and traction versus those that aren't? Is it the founders or Yeah. Yeah. I mean I think there's like a there's there's there's a few there's a few recipes and I've I've written about them.

and so I'll talk about them and and and certainly though it's not as McKenzie says MECE it's not mutually exclusive collectively exhaustive. so the first kind of area that I've seen be very successful and this has been one of these you know I I first wrote about it actually with with Uber and what was happening in that in that era which is this idea that if you're ever able to build a product that is 10x better if not 100x better than whatever the substitute is in the market and provide that cheaper by recasting the cost structure in some way you have no problems growing. And so I'll give a couple examples from our portfolio or maybe just in the interest of time I I'll give I'll give two. so like one is a company called DeepL that we invested in. So DeepL, you know, gosh, you wouldn't look at the name, you wouldn't look at the description of this company and think that it's going to become a multi-billion dollar software company, but it is it was kind of a pioneer in AI machine translation.

So they have without question the best ability to translate between languages using AI and they built they have their own kind of model to do that. And you know I was when I was at Pinterest actually one of the first jobs I had at Pinterest was to make Pinterest able to be localized. You know when when the engineers of course when you first build a product like Pinterest you never think it's going to become a global product one day. And so we had to extract all the strings and then I had to find translators on oDesk to translate Pinterest, you know, and like so we would export the strings, we would send them over, they would send us a CSV back, we'd upload it. It was like a ridiculous process. And and so as you might imagine, like if you are a company, like there's so much friction to hiring a translator.

There's you managing them, finding them, the cost of it that you don't actually translate everything that you could translate, right? And now it is instant. It is so much cheaper than working with a human. And it opens. It's one of these things that once you realize how much friction holds back a market and you remove that friction, the market expands dramatically and the company is able to expand dramatically. And so that's that's one example. Another example similar is a company called HeyGen that we're investors in. And so HeyGen, it's another one where you look at the website and you're just like how big could this be? Like it is who would ever think that you would hire a human to actor or do it yourself to talk about your product on your website or do your learning and development or have a product demo like it just hiring an actor or doing it yourself or you have to hire a production studio.

You have to do all the work. If you tweak anything in the script, you have to go back. It's so painful, holds the market back. And so, HeyGen uses AI. You do a 30- secondond video of yourself and then you can change the script. You can you can do so many things and it and it's the same dynamic where a market that was held back by the friction of finding and hiring people just gets expanded dramatically. And so I think like that is a timeless recipe. There's a lot of companies going after that. The other thing that I've written about is just this idea that I think a lot you know whenever there's a paradigm shift in you know facilitator catalyzed by technology you tend to go through this call it a schemorphic era like an era where the same mental model that you have been using for the past paradigm you apply to the new paradigm.

And and to me that has been software you know deploying software that employees adopt in their companies to to improve their productivity and like you know the the software that's that's kind of the SaAS products that we all love you know know to this day we they're the household names they are per seat revenue businesses that makes it better for people to collaborate with each other make them more productive all those things and you know the co-pilot type idea is is in that same mental model. And so the first wave of companies that we saw in the AI world were very much this like give software to people to adopt it and then get the leverage from AI. I think that this next paradigm and I've written about this is this idea of selling the work instead. And so instead of actually having software that employees use, you just are able to automate the full work product.

and so we have a few investments in this vein as well. You know, we're investors in a company called Sierra and the customer support side. So instead of a ticketing system, it is a product that does the tickets. we have a company called 11X that I'm on the board of which is on the kind of AD SDR kind of side. again automating some of the outreach to to to candidates. I'm sure we're all getting a lot of those emails now. I know I am. and so that is that is another threat. Of course, there's a lot of different opportunities beyond that, but those are two very clear very strong economic value propositions for that are that are facilitating and really catalyzing adoption. All right, final question. We'll keep it quick because you already said you don't like predicting the future, but do you think we're going to hit an AI bubble in the next in this next year or what's going to happen in the next six to nine months in terms of VC?

Oh, I wish you know that is again like it it is funny. It's like one of those things where I keep on expecting the market to dry up for foundation model companies and they keep on raising larger and larger rounds. So, I I've given up predicting this this future. All right, it's anybody's guess then. Thanks, Sarah.