Sarah Tavel: Will Foundation Models Be Commoditised? | E1149

1. 0:00 – 0:51

   Intro

   1. STSarah Tavel0:00

      Right now, it feels like if you need a model that's on the frontier, you're gonna be closed source. I focus on the application layer because I do think that's where a tremendous amount of value ends up being captured and created. You just see the tremendous amount of investment that has to happen right now in order to progress these models. At this point, it's kind of compute constrained. Each successive model is going to be more and more expensive. That suggests a world where you're gonna have an oligopoly.

   2. HSHarry Stebbings0:26

      Ready to go? Sarah, I am so excited for this. I always love our chats. So first-

   3. STSarah Tavel0:44

      I love it.

   4. HSHarry Stebbings0:44

      ... thank you so much for joining me today.

   5. STSarah Tavel0:46

      No, thanks for having me as always, Harry.

   6. HSHarry Stebbings0:48

      Now, well, I want to start, for those that do not know,

2. 0:51 – 9:11

   Background

   1. HSHarry Stebbings0:51

      how did you come to be at Benchmark? I- I love a good story, Sarah. So like, did they call you up and they're like, "Hey, join Benchmark?" Is it over dinner? Like talk to me, what was that courting process?

   2. STSarah Tavel1:00

      It's, um, you know, for... It's a very serious decision for us when we bring on a new partner because it's, you know, it's a s- very small group. Right now we're five general partners. And so it is a bit of a, uh, uh, it's a very deep getting-to-know-you process. Wh- how it started initially was actually, um, Peter Fenton reached out to me (laughs) and we grabbed a, a coffee at Sightglass. I was at Greylock at the time, and he, he just mentioned that for whatever reason, uh, they, they, uh, wanted to, to get to know me and, and, and, and kind of talk about what it would be like to partner together. Uh, and I, I said no (laughs) . Um, I, uh, I just didn't, you know, I was ha- like Greylock was just, it's a great group of people that been nothing but great to me. And I'd been there for, you know, a year and a half or so, and it just didn't feel right, uh, to me at the time. And so we kind of parted ways, and then they smartly had, uh, Rich Barton call me. I had met Rich at a prior event. And Rich, you know, had been a, a... Rich is the CEO of Zillow, um, amongst other things. And he called me up and we talked about it a little bit 'cause he knew the, the Benchmark crew very well, and he basically told me, he's like, "Look, what do you have to lose by spending time with the team, right?" Like, "If you wanna be great in this business, see how some of the greats practice the business of venture capital." And when he said that way, I just felt like, you know, he was right, that what did I have to lose by getting to know the team? And then as I did then start to get to know the team and see what felt so different to me about the way Benchmark practiced partnering with founders in this like very small, equal, uh, very committed way of doing things, it was just one of those things that once you saw it, you couldn't unsee it, you know? And that's, and, and so ultimately, thankfully, stars aligned and, uh, they felt the same way. And so it's been, it's been, uh, almost seven years now.

   3. HSHarry Stebbings3:10

      I promise we're gonna get to AI, but just, I've just got a, such a man crush on Peter, okay?

   4. STSarah Tavel3:15

      He's amazing.

   5. HSHarry Stebbings3:15

      So just like let me r- he is so amazing, his brain is just majestic.

   6. STSarah Tavel3:19

      Yes.

   7. HSHarry Stebbings3:19

      You've worked with him now for a number of years. What makes Peter Fenton so good do you think?

   8. STSarah Tavel3:24

      I, I mean, how do I... Okay, do, how much time do we have? Uh...

   9. HSHarry Stebbings3:29

      (laughs)

   10. STSarah Tavel3:30

       You know, Peter has so many incredible skills. One of the things that you just see for Peter is a, just a relentless learning mindset. Like he is always reading something new, talk- you know, listening to a podcast, internalizing all that information, lo- you know, taking... Right now I think he's taking graduate courses in some, some subjects that we don't even, you know, that is not even relevant. Like, it's not like an AI, uh, class. It's, it's kind of further afield. And so he, he has that incredible orientation, the, the curiosity that drives, I think some of the best to always be learning. Um, and then, and then he has just many superpowers. You know, the one that you always, uh, can't help but feel, and, and I think we've all lear- every time I am talking to a candidate for one of my companies and, and trying to close them, I feel like I, I channel, uh, everything I've learned from Peter because his EQ around people, what motivates us, what drives us, uh, what, what the f- what holds people back, and, and how important that is in the, you know, the arc of any company that we end up investing in, this watching, helping those founders become the best version of themselves. I think he's, he's probably not one of the best, but the best at doing that.

   11. HSHarry Stebbings4:59

       I remember Peter telling me that the best VCs are a combination of hyper curious and hyper competitive in one.

   12. STSarah Tavel5:04

       Yes.

   13. HSHarry Stebbings5:04

       And I, I always remember that. Final one before we do move into the meat of the show. You've now been doing venture for a long time, between the first in, obviously with Bessemer, then the second with Greylock and, and now Benchmark. What do you know now that you wish you'd known when you started in venture?

   14. STSarah Tavel5:20

       You know, I think like if, the thing that comes immediately to mind for me is just the importance of the why now. You know, the... It's such a cliché question, right? Like, oh, there's always a slide when you're, when you are pitching a, when a founder's pitching, which is like the why now, but the, the pointedness of that why now, how real that is, whether it's a technology catalyst, whether it's something, you know, some other current crypto coming out. That why now story is, is really so important because as I, I wrote about in one of my posts, like what you just realize is that when-... you have a strong why now, to me, it's like this strong current that just pushes the company forward. And it means that there are so many things that as a founder, there's, you can, you know, we always say that, uh, being a founder and going through hyperscale is all about making new mistakes, not redoing mis- not, not making the same mistake twice. And the resilience that your company has of making, you know, of being able to make mistakes and it almost doesn't matter, um, because the current pulling the company, uh, is so strong, that, that is just something that once you see it, you, you can't unsee it. And, uh, and then the g- the best founders take great currents and maximize that opportunity. But when you don't have enough of a real why now, it just feels almost like you're, you know, you're in a boat and you're just having to paddle really hard and not make a lot of progress.

   15. HSHarry Stebbings6:57

       Do the best founders not will markets into existence? If there's not a why, no?

   16. STSarah Tavel7:02

       I don't think so. I would love, I would love to think so, but, but... And, and just to be clear, why now doesn't necessarily, um, mean a trend. It can, it can be an opportunity that's catalyzed by a new technology, like we're seeing now with AI. But without that and just a new idea in an existing market, you, you're, you're having to, you, it is a force of will and are very, very difficult to, to really create big opportunities that surprise.

   17. HSHarry Stebbings7:38

       I think one thing I find hard about it, honestly, Sarah, is, like, the why now needs to be sustaining as well. And, you know, so I looked at, like, I did BeReal's pre-seed or seed round-

   18. STSarah Tavel7:46

       Yeah.

   19. HSHarry Stebbings7:47

       ... uh, whatever their first round was. And I was like, the why now is authenticity. The, the need to feel like-

   20. STSarah Tavel7:51

       Yeah.

   21. HSHarry Stebbings7:51

       ... you can be yourself in a world of depression and anxiety on social media. The why now was quite clear and I could intellectualize myself into it. Wasn't really a sustaining why now.

   22. STSarah Tavel8:02

       Yeah.

   23. HSHarry Stebbings8:02

       And I find that hard. Would you advise me on anything on that?

   24. STSarah Tavel8:06

       Well, I think part of, you have to contextualize that desire of the authenticity, which I totally agree with, with the countervailing current, which is TikTok being, you know, a, a black hole for people's minutes, you know, Insta- Instagram.

   25. HSHarry Stebbings8:23

       (laughs)

   26. STSarah Tavel8:23

       Like, all the other competitive forces, like, what, what happened with BeReal in my estimation, is that they did capture that why now of the authenticity, but they could never earn the right to enough minutes for a consumer because the consumer was then having, you know, the dopamine hits of, you know, the most addictive atomic unit we've ever had in social, which is the, you know, short form video, you know, powered by an algorithmic global maxima feed. Like, you're competing against that. And so if you have that authenticity why now, you know, that's a little bit against this, like, avalanche. Um, that's pretty, that's pretty, pretty difficult. Tidal wave should be a, I shou- instead of o- of avalanche, but you get the picture.

3. 9:11 – 23:37

   Value of Selling the Work in the Application Layer

   1. STSarah Tavel9:11

   2. HSHarry Stebbings9:11

      I, I totally get the picture, and I feel like you should be TikTok's CMO, doing banner adverts-

   3. STSarah Tavel9:16

      (laughs)

   4. HSHarry Stebbings9:16

      ... you know, the black hole of minutes really just captures that consumer's zeitgeist. Uh, listen, I, I want to discuss AI today. We mentioned kind of, you know, the TikTok algorithm there. Most people talk about AI today, and they talk about it being this sustaining technology. I'm really interested to start there. Do you agree with that as a positioning, as a sustaining technology?

   5. STSarah Tavel9:37

      There's a lot of truth to this, and I think part of where this, um, meme has come from, you know, everybody talks about AI as something that, that advantages the incumbents, um, and when it advantages the incumbents instead of the disrupters, which are startups, then it's a sustaining technology. And there's no question that, I, I have personally never seen in my career a time when it has been more true than that it's been this race, you know, it's always a race of the incumbent to get innovation before the startup gets distribution.

   6. HSHarry Stebbings10:13

      Mm-hmm.

   7. STSarah Tavel10:13

      And it has been, uh, it has been so easy for incumbents to innovate, because it has effectively been for this first wave of AI use cases just implementing, you know, an API from OpenAI, right?

   8. HSHarry Stebbings10:27

      Mm-hmm.

   9. STSarah Tavel10:28

      And so they're, you know, if you are Notion, if you are Adobe, instead of having to adopt a new produ- product to do image generation or a new product to do, kind of, summarization of, of content, you can just use Notion or Adobe and get access to that technology. That's very much a sustaining way of using the technology. And so, like, the way, the w- the way I divide the world then is that, if it's existing employees and their existing workflows, that, you know, AI technology is going to be sustaining, and, and it's gonna drive tremendous market cap for the incumbents.

   10. HSHarry Stebbings11:09

       Mm-hmm.

   11. STSarah Tavel11:09

       But that doesn't mean that there aren't disruptive opportunities for startups leveraging AI. And I think that the difference is that you have to really change the mental model that w- you and I have been trained to kind of think about startups for a very long time, and there's kind of two, two, two veins to that. Uh, the first is, you know, we're so used to software, thinking of software, and application software in particular, as this, like, productivity improvement product. So you and I adopt a new software product, let's take Notion again, that lets us collaborate with our peers in our company, that makes us, you know, uh, just more productive in the work that we have to do. Then you think about AI and, like, it's, it's back to increasing my productivity, so now instead of having to write a whole paragraph, I can take my bullet points and expand it, and boom, you know, that makes me much more productive. And that has been all of application software for the, the last 25 years. But if you realize that what AI enables-... is actually a very different unit of work that you sell, which is, which is doing the work. And so you're almost a software company that looks like a services business, that is able to sell, like, the f- the full work product, the o- the outcome, as opposed to selling software that an employee has to learn to use and then gets a productivity boost from. And this is very disruptive to incumbents, because incumbents are used to thinking about selling per seat, and, and pricing per seat based on the cost of the head count. But if instead you're selling something that doesn't require a seat, that, that is, like, a very disruptive opportunity for startups.

   12. HSHarry Stebbings13:03

       So a couple of things I just have to dive on here. You mentioned kind of selling the work and being able to do that. You know, we obviously had Sam and Brad on the show, and he said very openly, "The models are not good enough," simply put today. Are we anywhere near a situation where for enterprise workflows, it is able to sell the work end to end, where we do not need to do the work itself?

   13. STSarah Tavel13:24

       It's a spectrum. There are certainly use cases that can be automated right now, and we see, we see a lot of them. Um, you know, you have, you, you know, I met a company the other day that is, uh, automating HR ops. You have, uh, companies in recruiting and sales, like, all different facets where there is specific types of work, and it's kind of... I always think of it as, like, unbundling the employee. So what are the different work products that an employee has to do?

   14. HSHarry Stebbings13:53

       Mm-hmm.

   15. STSarah Tavel13:54

       Like, definitely there are work products that are automatable right now. That is not to say that, you know, if you're Sam and your, your brain is thinking about the next, the next horizon, GPT-5, to work beyond that, there's no question that as the foundation models improve, we're gonna see the ability to take on more and more complex tasks. But even now, there's still work to do. Then there is, like, this question of having a human in the loop, and that's the, the way that you see some companies bridging the gap right now, where they have n- m-... You know, it's either an employee in the loop. I prefer the model where you have your own employee as, like, the, the AI software provider in the loop that is doing the QA work that bridges the gap until the models are able to do it on their own.

   16. HSHarry Stebbings14:45

       I totally get you. I, I have to touch on the element you said there about kind of doing the work instead of, like, selling on a per seat basis. Do we not just see Cooley or the lar- uh, PWC, KPMG, any of these kind of large firms that sell kind of traditional services to businesses, do they not just adopt the work creation tools, but then charge back to their clients the same hourly rates and just become better businesses themselves?

   17. STSarah Tavel15:09

       The- there's certainly... You know, I actually think about mar- marketplaces, there's lessons to learn there. You have to divide the world between the incumbents and, like, the hungry non-incumbents that want to become the incumbents, right? New technologies more often than not are best served by- for the group of hungry up-and-comers that want to become the new incumbents. And so you're gonna see a class of companies that realize that if they're early adopters of this technology, they're able to come in with a very different pricing structure. They're able to do things faster than the traditional firms. And so they're gonna be the ones that are hungry and, and g- and having a cost advantage that they pass on to their, their customer that lets them grow their market share.

   18. HSHarry Stebbings16:00

       There's this common trope in venture which is like - and you hear it the whole time - "Ah, the infrastructure layer, super exciting. Application layer? No, no, no, it's a fool's game." I always hear this. And I'm just intrigued listening to you there. I'm like, "That doesn't really sound aligned to you." How do you determine where value is in the application layer and where it's not? And where would you advise me, "Actually, Hari, there's, there's treasure here and there's not here"?

   19. STSarah Tavel16:25

       I just am a huge believer that the application layer is gonna drive most of the value, because what you have to imagine is who owns the user over time, and who... If you own the end user, you're able to provide more and more value to them over time and capture that value. There's... You know, and we could talk about does- what happens to the underlying models. There's certainly just incredible intense competition. Is it gonna be an oligopoly? Is it gonna be d- You know, we can, we can talk about those, those subjects. But, like, I focus on the application layer, because I do think that's where just a tremendous amount of value gets, ends up being, you k- captured and created.

   20. HSHarry Stebbings17:07

       So Brad said the question is, are you excited by a 100x improvement in OpenAI? If you are, then you will be a sustaining technology.

   21. STSarah Tavel17:18

       Yes.

   22. HSHarry Stebbings17:19

       If not, Sam eloquently put it, "We will steamroll you." (laughs)

   23. STSarah Tavel17:23

       Yes. Yes, yes.

   24. HSHarry Stebbings17:24

       Do you agree with that, and do you think about that when investing in the application layer and the defensibility of it today?

   25. STSarah Tavel17:29

       Oh, absolutely. I mean, I, I, I think all the companies that we invest in, they will only get better as the underlying models get better. And they'll be able to take on more work over time. They'll be able to, you know, f- uh, take, you know, have better and better margins, better, because they won't have to have a human in the loop if they do. Uh, that is, that is part of what's so exciting about what's happening right now.

   26. HSHarry Stebbings17:54

       I love this. This is, like, optimistic. I get, you know, bored of the pessimistic view on application layer. But one thing that does worry me is, like, you're right, um, and you've said it before about user experience being better for a lot of the application layer startups and the incumbents who are trying to kind of embrace those technologies too, or those products too. But they're not, like, 5x or 10x better. They're like 15 or 20% better in a lot of cases. But the distribution of incumbents is so strong that Microsoft can just bundle it in-

   27. STSarah Tavel18:22

       Yeah.

   28. HSHarry Stebbings18:22

       ... and then enterprises will adopt. How do you think about the...... distribution advantage of enterprise, versus the user experience advantage of startups, and what ultimately wins?

   29. STSarah Tavel18:32

       Yeah. I, I think this is exactly the challenge that the first wave of AI startups have had.

   30. HSHarry Stebbings18:38

       Mm.
4. 23:37 – 36:06

   The End State of the Model Landscape

   1. HSHarry Stebbings23:37

      get you. One thing that's hard for me also is differentiation. W- there are so many AI customer service tools.

   2. STSarah Tavel23:43

      Yes.

   3. HSHarry Stebbings23:43

      There are so many AI sales agents. And I'm like, and they're brilliant teams. And I'm like, "Ah. Shit." (laughs)

   4. STSarah Tavel23:51

      Yeah.

   5. HSHarry Stebbings23:51

      (laughs) I'm not, I'm not sure. How do you think about separation, differentiation between the 10 players in each, and, and where you wanna place your bet, or not?

   6. STSarah Tavel24:01

      This is, I think this is the single hardest question right now.

   7. HSHarry Stebbings24:05

      Mm.

   8. STSarah Tavel24:05

      You know? Because, it is true that there's a land grab, it feels like. And part of the challenge also of this question, Harry, is that, w- some of the response for a lot of companies is just to raise bigger and bigger rounds, so that they have the resources they need to, to do-... you know, the kind of to buy the GPUs or whatever it may be. It is about the founder that you invest in, now more than ever, and I think it's just you're, you're backing a founder whom is, has that competitive energy, and high urgency, and aggressiveness, and ambition that will let them navigate a very competitive ecosystem and emerge, you know, the victor. But that, it's going to be a very competitive next few years.

   9. HSHarry Stebbings24:55

      First, how do you feel about adjusting your mental model of what you expect from a company at a certain stage? Because, you know, if we look at Benchmark traditionally, or series A traditionally, which is obviously where Benchmark is at home, uh, it's like 20 on 100, say I'm just kind of bastardizing, but kind of average. When a company has a certain set of foundations or metrics, it has something, it's at series A. Now, so many companies have nothing but great teams in AI, and it's 20 on 100. What we used to expect is now different. Has it been, how do you think about changing your mental model of investing for a world of AI?

   10. STSarah Tavel25:33

       It is very rational in a way, which is, is the opportunity size bigger? So let's go back to what we were talking about before, which is selling work. Part of the opportunity with selling work as an AI startup is that the market for that is, could actually be 10 or 50x bigger than if you were selling software. Why is that? You're selling a 95% productivity improvement instead of a 10% productivity improvement. So you're basically selling against the cost of the headcount, as opposed to a productivity improvement for that headcount. And when that happens, and, and it, by the way, it also has an easier go to market because you're not asking an employee to adopt something new. You're selling a package of work. And so when that happens, you actually open up the aperture of how big a market can be. If you're opening up that aperture, then it's actually a rational decision to spend, you know, to, for the valuation or the entry price of an investment to be higher. There are plenty of companies that you could be talking about where they're not selling work and they're selling a productivity improvement, or they're selling, you know, an underlying, you know, technology, and that, that becomes a little bit more difficult for me to, to rationalize.

   11. HSHarry Stebbings26:55

       Another thing that I can't get my head around is like the dilutive nature of a lot of these companies.

   12. STSarah Tavel26:59

       Mm-hmm.

   13. HSHarry Stebbings26:59

       Which is just, they are cash machines, and I'm just worried, Sarah, that I'm going to put in whatever the check size is, 5 million, 10 million, whatever it is, and it's just gonna be diluted to shit. (laughs) The ............................ have also just got more blunt over time. Uh, how, how do you think about that and the increased dilutive element that's inherent within what seems to be this next segment?

   14. STSarah Tavel27:22

       It's, you know, every strength has a corresponding weakness, right? So you could tell yourself a story, and it remains to be seen whether the story is true, which is that if you're raising money, if you're one of these companies, and why would you raise a lot of money? I, I hope you would raise a lot of money. Let's take the Cognition round, uh, la- lately. You're raising a tremendous amount of capital for a company that I don't know if they have any customers yet, but I, I suspect it's very, very early. Why would you do that? Well, you have to believe that there's, as an investor I would think, that there's almost a self-fulfilling prophecy of giving a very talented team a lot of capital so they can then invest that capital in GPUs, train their own model, and have a competitive advantage over any other company. And so there's kind of that upfront dilution, that if you, if you accept that, then, and believe it will be invested in a way that creates a moat for the company over time, then you're going after a tremendously big market that has a, a very rare moat because of the capital required to, to even be on the field fighting that fight. And I think that's the only way that I can rationalize a little bit what's happening right now. Otherwise, it just feels like there's, um, you know, and us, us VCs, we always get into this, which is the FOMO capital deployment mindset, which doesn't, doesn't always lead to, to, uh, to good things.

   15. HSHarry Stebbings28:52

       I'm, I'm throwing the grenade in here. I feel like we haven't learned our lesson at all, and I'm, this is where I'm like the grandpa and...

   16. STSarah Tavel28:56

       Oh, we're goldfish. We're goldfish. (laughs)

   17. HSHarry Stebbings29:00

       I know, and, and so do you worry that you fall, like not you, but like, 'cause we have to be prescient not to fall into the FOMO field and remain disciplined. It's something that Benchmark did incredibly well over the last few years, actually. You guys were one of few firms (laughs) which did retain discipline. How do you think about that in this generation?

   18. STSarah Tavel29:22

       What we always try to do is just remember what the, the ends game is here, which is ultimately building a company that can become a independent, enduring company, and imagining what that might look like in the future, and pulling forward that future into the present to think about the dynamics of the company, and our, does it have the type of dynamics that lead to those types of outcomes? Usually, I, what I think about is that escaping competition idea. You find those types of companies, something with a network effect, as a, as an example, you should lean in hard as a, as a founder, as an investor in that opportunity because they're very rare. And when they work, they'll lead to, to very large outcomes. That's a, that's very much a picking your battle type, uh, orientation. Like, our way of thinking of things at Benchmark is that each of us makes one or two new commitments a year, and so it is this kind of stars aligning, sometimes feels like almost an unreasonably high bar to getting to yes, but then when we commit, it is like the full partnership committing behind that company, somebody whom...... is only taking on one or two new investments a year, doesn't have, like, a team to whom they're, they're delegating, you know, any part of the job. And, and, and, and those are the ones where we pick our battles. Otherwise, it sometimes feels like it's more a capital deployment game, as opposed to, like, really focusing on the investments that you're making and how to, how to make those really big.

   19. HSHarry Stebbings30:59

       Does the nature of network effects change in a world of AI? When we think about a world of selling the work and not the s- the service or the tool, um, does the nature of network effects change?

   20. STSarah Tavel31:11

       B2B software has very rarely had network effects, right? What, what it has had instead is more economies of scale, you could think of it that way, where you build a product for a very specific use case, you execute to, like, on the go-to-market to, like, grow that, that, that product into the hands of a lot of users. And then you are able to add more and more features over time to that product, so that you are able to charge more, that gives you the advantage that as you charge more, you have, you know, positive net revenue retention, you have an, a go-to-market that can afford to, to scale, and have better and better efficiencies. And that, that is, to me, most of what has been the B2B software, uh, kind of way of building really outstanding value. Um, that's been different than... And I think that's still the case for AI startups that are doing a B2B use case.

   21. HSHarry Stebbings32:11

       You mentioned very, kind of, specific use cases there. You wrote about, you know, HeyGen, DeepL, ElevenLabs. I actually spoke to (laughs) , I spoke to all the founders before the show. I told you, I have far too much free time.

   22. STSarah Tavel32:22

       Amazing.

   23. HSHarry Stebbings32:23

       Uh, yeah, and, uh, it was actually Mati at ElevenLabs who said, uh, "I would love to hear Sara's take on models' focus on one modality versus a model doing all modalities-"

   24. STSarah Tavel32:35

       Hmm.

   25. HSHarry Stebbings32:35

       "... and how you see that playing out." I'd love to hear how you think about that.

   26. STSarah Tavel32:39

       In the podcast that you guys did together, I think one of the things that Brad said, Brad said Sam was so good at was almost keeping the main thing, the main thing, right? Like, knowing the thing to get really right. And it's very clear that for OpenAI, that is, you know, the progression of GPT, for them, from, from three, to four, to five and beyond. And that means that the other types of models, you know, you can imagine audio, the image, video, those are gonna progress, but they're not gonna progress with the same level of focus and ambition as the core foundation model that OpenAI's working on. They, they might progress becau- as beneficiaries of that core work, but it doesn't have the same level of focus as somebody like Mati and his team at ElevenLabs that's focused very specifically on a very specific use case.

   27. HSHarry Stebbings33:33

       I totally agree with you (laughs) . When we think about, kind of attaining product parity, people always think that, "Oh, actually, with that focus, you're gonna be so far ahead." I worry that actually it takes incumbents less time to reach product parity with startups than we think or give credit. Do you think that's right, or do you think actually with the focus that they have, they do present such a headstart?

   28. STSarah Tavel33:54

       Well, this is back to what I, you know, the, kind of that thing I said before of, like, what percentage of value comes from the foundation model?

   29. HSHarry Stebbings34:03

       Mm-hmm.

   30. STSarah Tavel34:03

       And I think, you know, we could talk about DeepL. So DeepL, you know, it's a, just a tremendous company. And so, so much of what they've built is the foundation model, the, the, their own model around language. But they've also built a tremendous amount of workflow to integrate with, uh, the, with, for, for their customers, you know, to enable a local kind of vocabulary list, to, to be part of what- whatever workflow or product that they, that someone's creating, and making sure that it's a seamless experience. That's, so that's kind of crunched. And so even though it started maybe where all it was was actually deepL.com, where you're going and you're just, just like Google Translate, copying and pasting co-, you know, copy from one, one language to another, they have gradually built more and more value on top of that model, so that something coming out, you know, in OpenAI, they're, they're just not gonna build that same level of workflow, the same level of integration. And so it's going to be... There's, there's, there are pl- there's plenty of work to be done that's beyond the model itself.
5. 36:06 – 39:15

   The Challenges & Future of AI Models

   1. STSarah Tavel36:06

   2. HSHarry Stebbings36:06

      Why does it not get cheaper to train models? I thought the whole point was we got more efficient with time, not more expensive.

   3. STSarah Tavel36:12

      The belief is that, yes, some of the underlying costs are gonna go down, that the chips will get cheaper, will get better, the research is gonna improve. But it's just more and more hungry for more compute. I al- You can almost think of it as like a highway, where you keep on adding more lanes and it just (laughs) more, more cars want to go through. I think we just have a situation where it is a race, and part of that race, in order to be ahead of everybody else, is gonna be this very expensive...... process of more, you know, more and more specialized chips, more and more investment in that. More investment on the power side, which is actually becoming a real constraint given how power consumptive this, you know, not just the training but the inference itself is going to be. To train is gonna just get more expensive. I mean, it's just gonna be a bigger and bigger lift each time we do the, the new, the new step function. The, the cost to the end customer, I think that should continue to go down, and that's what we're seeing, is that it is just getting more and more competitive, the price continues to go down, and, and so we're all big beneficiaries of that right now.

   4. HSHarry Stebbings37:21

      How do you think about the open versus closed argument? Like, again, I just think we see so many start open, go to closed. Is that the natural evolution of the environment? How do you think about that?

   5. STSarah Tavel37:31

      What I believe right now is that it, uh, if you just assume, okay, people say that each kind of s- next step function in, uh, these foundation models is like a 10X cost increase right now to do the actual training, and what happens when you do open source? You're, you're making that tremendous investment, and then you have to believe that if you're open sourcing the product, you're going to be paid, you know, uh, uh, you're gonna be, you're gonna earn the value, um, for that open source. That is, that's very TBD right now. And so, what it feels like is that if you want a model that's on the frontier, that's gonna be closed source. Now, uh, l- look, every, I, as I mentioned, every month it's, feels different. With, with what, uh, Meta's doing with LLaMA, that may actually fundamentally change the game. If Meta is willing to make that huge investment in the, in the underlying training, and then open source that model in a way that may seem economically unrational- not rational in the b- in the short term, but is, is, is right for them in the long term, that changes the game. But, but right now, it feels like if you want, if you need a model that's on the frontier, you're gonna be closed source. There may be some, some, uh, use cases that people have where they don't need to be on the frontier, that what is open source is, is powerful enough, and then the ability to, to do what they want with it gives, you know, makes up for any difference. That, that you can imagine that, that, that story evolves as well.

   6. HSHarry Stebbings39:13

      What worries you most about this

6. 39:15 – 46:02

   The Role of Competition in the AI Industry

   1. HSHarry Stebbings39:15

      space, Sara?

   2. STSarah Tavel39:16

      I think what you s- we talked about earlier, which is just competition. You know, like, how there are obvious opportunities, and, and they're huge, and it feels like very large companies can be built over time. And at the same time, it also feels like there are more companies than I've ever seen pointed at every opportunity, or m- many of these opportunities. And so, what tends to happen, back to this food delivery wars discussion that we had earlier, is that it ends up requiring a tremendous amount of capital, because you are fighting tooth and nail for every percentage of market share, and someone emerges victorious. Uh, and it may be that there's a lot of companies that end, you know, it's may it be that there's not a winner take most opportunity in some of these segments. But those are the ones where they create the most value for all shareholders, and I do wonder how that ends up shaking out.

   3. HSHarry Stebbings40:19

      You mentioned the competition. It's also competitive to win those deals and get in front of those founders at those rounds, and beat the other 10 term sheets. Maybe more broadly outside of AI, how does Benchmark win so effectively (laughs) ? Is that, like, a swarming of partners? Is there, h- how do you think about what it takes to win as a group in a partnership?

   4. STSarah Tavel40:41

      You know, you (sighs) would be best served asking the founders that question. What I think about is that we just have a fundamentally different product that we're offering, right? The idea of Benchmark, what it means to take, to, for us to make an investment in a company and partner with that founder is that we are not delegating any part of that core work of that partnership to an internal group of consultants. Like, there's no... And what I experienced when I was at Greylock and saw is that Greylock had, when I was there, the best internal recruiting team, and I remember one of my companies, Sonder, uh, was doing a CRO search, or a head of sales search, and I was like, "Oh, great." You know, and I threw over the search to our head of talent, uh, Jeff Markowitz, and what I realized is that then Frances, the CEO, is forced to play this game of telephone, where he is talking to Jeff about r- uh, the recruiter or, you know, some different candidates. Jeff's talking to me, I'm talking to Frances, and you realize that actually that model of partnership is more about scaling the f- the GP than it is the founder. Like, in that case, I didn't have to have the weekly recruiting call. I wasn't expected, I didn't think it was my job to have the weekly recruiting call on my calendar, to be in sync with Francis on every CRO s- like, candidate that he was speaking to. And, and, and so it saved me time and let me have meetings with other companies or do whatever I need to do, but it was worse for the CEO.

   5. HSHarry Stebbings42:22

      Is it not just providing a better quality product? And I don't mean that rudely, but, like, Jeff Markowitz is obviously a, a specialist in recruiting, is an expert in his field. He probably provides a better quality recruiting product than, than you do or than I do or than, than any GP in the world would do. Is it not about actually just providing a better quality product?

   6. STSarah Tavel42:41

      It's not something that is, like, you, you, you think about in a vacuum. It actually is how well do you know the CEO? How well do you know the execs on the team? How much do you care...... about getting the right person, the best person possible in that role, than just having the seat be filled. How much, like, do you know through the process so that you can really help close the best candidate? All those things, they're, they're not possible with a, a specialist recruiter. Even if they're the best at their job, they just have so many other clients that they're responsible for, their, their orientation isn't like an owner, in the way that I feel like an owner of any of the companies that I invest in. But it is just a consultant where they wanna get the job done. And so, it's a very, very different mindset. And then I think part of what we do, and what I've realized, is that I have done so many reps of this now, that why does an executive recruiter become so good at what they do? It's because they do it a lot. And that's now what I have done. I have helped recruit and close so many candidates on behalf of the companies that I work with, that it has become a superpower for, for me, and it's the same thing for all of my partners, is that we are collectively... W- what we aspire to, is to collectively be the best recruiting firm out of any other place, because that is what we do day in and day out for the f- for the founders that we work with.

   7. HSHarry Stebbings44:14

      What's your favorite winning story, Sarah? When you review the many deals that you've won as a team, was you ever like, "Ah, I remember that one." I'm fascinated.

   8. STSarah Tavel44:25

      I, I could give two sides of the spectrum. I could talk about the companies where it is just very clear that there isn't competition.

   9. HSHarry Stebbings44:34

      Mm-hmm.

   10. STSarah Tavel44:35

       That there's such a strong connection with the founder, and, you know, there's such a strong belief in what they do, and they feel that, and they feel the, the strength of the relationship, and they feel the commitment that it comes with our model, that you don't... Even though there is competition, you don't feel the competition. Then there are those cases, certainly, where it is that benchmark group effort. And, and that's part of what's unique about our model, is that we are an equal partnership. And so, what that means is, you know, if my partner, Eric, makes an investment, and that investment's successful, I benefit in the same level as he does. And same if I make an investment and it's successful, then my partner, Chathan, is, is the same beneficiary of that economic reward. And so, we actually really do orient, then, towards the team. Like, when, when one person makes an investment, all of us are making an investment in that company. And in the same way, then, if one of us... You know, if we, we find a company and we all feel like we should invest in it, it can be a group effort, sometimes, to, to make that interest felt strongly with the founder, and, and realize that kind of collective effort that will be possible for the founder moving forward.

   11. HSHarry Stebbings46:00

       Huh. (laughs)

7. 46:02 – 58:03

   Reflecting on Missed Investment Opportunities

   1. HSHarry Stebbings46:02

      That was pretty cool. When you've made an investment and it hasn't worked out, what did you not see that you wish you had done?

   2. STSarah Tavel46:11

      I think it's, it's a couple of things. It's, it's often, um, the why now that we talked about before, that just, in the same way that you believed in authenticity, we can sometimes, you know, you get, you, you get into those ideas, and you, and you believe there's a why now, excuse me, that may end up being ephemeral, or not quite as a, as acute and real. It doesn't have quite the momentum that you believe it had. And I see that, that scenario play out.

   3. HSHarry Stebbings46:42

      Some here said that you say, "I'm not gonna be a founder's cheerleader." How do you think about, you know, the VC being the cheerleader? And if that's not the case, what you should be?

   4. STSarah Tavel46:52

      Yeah. You know, what I always think about is like, my... When, when we invest in a company, and what I, what I feel most strongly, is like, I am there to help you build the best company you possibly can build. And the means to that end is helping you, as a CEO, grow to be the best version of yourself. And if you're just coming to the board meeting and, you know, being a cheerleader saying, "Go, go, go," not asking critical questions, not being truth-seeking, not thinking about, "How do we, you know, pull that future into the present of, like, well, what are the things, not just to hit the quarter this year, but to be able to scale with velocity four quarters from now? How do we start setting our company up for that success and pulling that future into the present?" If you're not asking those questions and you're not, you're not asking, "Oh, I think I see one of the executives starting to have challenges scaling. Like, how do we support her? How do we make sure that, you know, she has mentorship? And if that doesn't work, we actually find somebody new?" Those are the types of questions that, if you're just a cheerleader, you don't push, and each time you do push, I hope that it leads to 5% better decision here, a 5% faster decision there, a 5% better candidate there. And those small differences end up compounding.

   5. HSHarry Stebbings48:26

      Do you feel the quality of board membership is high? Founders always bluntly berate it behind the scenes. Often VCs berate it behind the scenes. Do you feel that the boards you're on are good, and that board membership is good?

   6. STSarah Tavel48:38

      I think this is one of those very self-fulfilling prophecies, which is that for both the venture s- the VC side and the founder side, I meet founders all the time who either had a great seed investor and see how great it can be when they have a great partner, or in a past company, they either had a great partner or the opposite.... they, those founders have the wisdom to know that having a great partner really makes a difference. And, and if you believe that having a great partner can make a big difference, you will have that bar high, and you will find somebody who will do the things that we just talked about. If you go into the process and you think they can't add value, i- it's a self-fulfilling prophecy. And at the same time, like, you know, w- for the work that we do at Benchmark, like, if we believe that f- board members don't do any work, I mean, they don't make a difference in the out- in the outcome of the company, even if it's, you know ... And let's just be clear, uh, the... (laughs) 99% of the work is for the founders and the team that they built, but there is that, that small difference that does come in at really important moments and can end up compounding. If we didn't believe that, then we're not ... You know, we would change the Benchmark model entirely, because our model is built not to scale. It's built to make a small number of very focused investments, very committed relationships with the founders, and then from there, have an impact on the, on the eventual outcome.

   7. HSHarry Stebbings50:16

      Final one, and then I promise we'll do a quick fire. But one can't always win. In the cases where one doesn't win, why, why is that? And has there been a change in how you approach the deal founders as a result of not winning a deal?

   8. STSarah Tavel50:31

      What I think about is, there are kind of two broad cases when we don't win. One is, we ... You know, uh, there's a couple examples that I think about where there's actually a selection bias that happens, where what we offer is, we want, you know ... W- we wanna partner with ambitious founders who are willing to be vulnerable, recognize the things that they don't know, and, and, and help push them to build the, be the best versions of themselves, build the best company they can build. And there were some founders ... I remember one company I spoke to at some point, the CEO told me, he's like, "You know," and, uh, it's easy for me to say this now, but he's like, "I, I want to partner with you, but I really think I need the platform." And that was a great selection bias decision for me. You know? Because we ... That, that's just not our model. The other, you know, example of that selection bias is a founder whom ... Just didn't want a board member at that point in time. W- we're not c- ... You know, that's, that's fine. Like, that's ... I, I ... That's a great decision for that founder, and it's not the, the comp- that's not what we offer. There are certainly other cases where we lose, where it's, you know, "We're, we're not ..." You know, it's, the valuation just gets ahead of us. The, the round size gets ahead of us. We, um ... Maybe a founder feels like there's not quite the belief in what they do, although I find that very rare. But, but by and large, there's just a selection bias, I think, that happens a lot that ends up meaning the difference between these, these outcomes.

   9. HSHarry Stebbings52:23

      If you break the model, what do you break the model most often on? Price? Board member status? Um, what would it be?

   10. STSarah Tavel52:33

       Uh, uh, certainly the round. It's, um, you know, we ... Uh, uh, w- there ... Uh, s- in my seven years at Benchmark, I think I've seen one example where we broke the model on, on board, and it was for a small (clears throat) crypto-related company. Without question, that is our product, so it's very hard for us to break the model there. But we ... All the time, like, we ... You know, we do play the game on the field, and if there's a partner, a founder that we wanna partner with, we will figure out a way to make that work.

   11. HSHarry Stebbings53:06

       Huh. That's so interesting. When you've lost on price, do you ... In hindsight, has it actually turned into a good company?

   12. STSarah Tavel53:14

       You know, I can't actually think of a company where we've lost on price. I can think of situations where we haven't engaged because we knew that the situation was just gonna be, uh, not really part of our model. So as an example, these large, you know, multi-hundred million dollar investments in some of these new models or, or companies that h- require a lot of training, those are ... You know, it's ... And, and for all those cases, it's just too early to know.

   13. HSHarry Stebbings53:46

       Final one. How do you determine when to pay out versus not to?

   14. STSarah Tavel53:49

       It comes down to, do you believe that this is a company that can escape competition? Do ... You know, and it's a founder who's going to navigate through all the competitive dynamics to, to escape competition. There's, there are dynamics in the product that they're building, whether it's a network effect, or some strong mode or economies of scale that let it, again, escape competition. In those cases ... Someone said this to me once and it really hit me, is like, "If you like everything but the price, you pay the price."

   15. HSHarry Stebbings54:20

       (laughs) I love that. Um, uh, o- one person said to me the other day, a really good one, which was, "If you're ever happy to take less, don't do the deal at all." Which is like, if you are going for 15 million, but the founder says, "I really would need you to take 12 and a half," don't do the deal. Like, you should not be wanting to do less, ever. Do you see what I mean?

   16. STSarah Tavel54:42

       Yeah, I do see that. I think of it the other way, which is that, if you are trying to use price to get you comfortable with a deal, you probably shouldn't do it.

   17. HSHarry Stebbings54:54

       Yeah. I remember Peter Fenton said on the show, that pri- use price as a litmus test for your own conviction.

   18. STSarah Tavel54:59

       Yes.

   19. HSHarry Stebbings55:00

       D- do you guys do the same for reserves? How do you think about reserves?

   20. STSarah Tavel55:05

       We don't think about reserves.

   21. HSHarry Stebbings55:07

       ... period?

   22. STSarah Tavel55:09

       Barely. We're very oriented towards our initial investment, and in that moment, you know, we are more ownership sensitive than probably most other firms. But then, after we invest, we are 100% aligned with whatever objective the founder has. And so, more often than not, that has meant that we invest almost nothing after our initial investment. Of course, we'd love to invest, uh, more after our initial investment, but I don't believe in the idea of pro rata, you know? I actually think that pro rata, I, I have started to, uh, use a term with some of the companies that I work with of, like, earn pro rata, which is that after we invest, like, what I've seen happen is that the business model of the venture firm ends up creating a lot of challenges for a founder as they raise subsequent rounds, because they're trying to bring on a new partner. The new partner has ownership, uh, requirements in order to take a board seat, and so you, you kind of optimize for that new partner, and then you have all these other people who aren't involved with the company anymore who are demanding their pro rata. And the challenge with that is, that it just ends up meaning more and more dilution, really for the founders, who, who can't invest in subsequent rounds. And so, I just think that that is not the right level of accountability. If people are creating work, then yes, please. Like, you know, I, I would hope that the founder would want, you know, would, would take the dilution for that, for that work. But more often than not, it creates conflict that's unnecessary, and it's just very different than our, our, our model, which is like, "Okay, now that we've invested, we are 100% on the side of your, your side of the table, and let's make every round as successful for you as possible."

   23. HSHarry Stebbings57:04

       So, I'm the same, and I had a call with the founder recently, and they were saying, "Hey, uh, we've lost our head of sales. Numbers are not looking good. It's hard, Harry." And then, I went to the board, and there are many multi-stage funds there, and it was a completely different view, and I called the founder up afterwards, and I said, "What is going on?" And they said, "I need their reserves, dude. Like, they've got a lot coming in. I can't tell them the real story."

   24. STSarah Tavel57:26

       Huh.

   25. HSHarry Stebbings57:26

       Obviously, right? And you're like, "Huh." It creates this imperfect relationship of information flow.

   26. STSarah Tavel57:32

       Yeah, and I feel for that founder, because what is the foundation of the board to, you know, board member to founder, uh, relationship? It's trust. And if you don't feel like you have that person on your board who's one of your investors, who you could, you have that trusted relationship with, and you can be vulnerable with them and he- and use them to navigate those types of moments, that sounds extremely lonely and very difficult.

   27. HSHarry Stebbings58:01

       No, I agree. Um,

8. 58:03 – 1:05:49

   Quick-Fire Round

   1. HSHarry Stebbings58:03

      but listen, I wanna do a quick-fire, 'cause I could talk to you all day.

   2. STSarah Tavel58:05

      Okay.

   3. HSHarry Stebbings58:06

      So, we were talking about boards there. What's the best board you sit on, and why that one?

   4. STSarah Tavel58:12

      That's a very unfair question to ask, uh, somebody-

   5. HSHarry Stebbings58:15

      Yeah, I know.

   6. STSarah Tavel58:15

      ... who has multiple boards. (laughs) Um, what I think about is, uh, probably just the, the relationship I've had the longest, 'cause, you know, one of the things I always, we always say at a f- We always, it's like this cliché that if you aren't embarrassed by the CEO that you were six months ago, then you're not growing. Uh, I think the same thing as a, as a board member. Like, I am growing and learning how to better partner with the founders that I work with every time, every board meeting, every interaction. And, and so my first investment at Benchmark was this company, Chainalysis, where we, you know, w- I was the first board member for the company. Uh, we led the series A. Uh, this was, I think it's 2017. And it's just been such an incredible journey, because it's both getting to see the founders grow, like, tremendously over the arc of time, but then I also have benefited so much from, as the company has scaled, seeing all of the new challenges that come up, and helping kind of navigate and learn from, from their growth. Uh, I just feel very grateful for that journey.

   7. HSHarry Stebbings59:23

      What's the biggest miss, and did you change anything on the back of it?

   8. STSarah Tavel59:27

      Without question, Ethereum. I-

   9. HSHarry Stebbings59:31

      Wha-

   10. STSarah Tavel59:31

       You weren't expecting that. I can see. (laughs)

   11. HSHarry Stebbings59:34

       No, I wasn't.

   12. STSarah Tavel59:34

       You-

   13. HSHarry Stebbings59:34

       At all.

   14. STSarah Tavel59:35

       You know, what, well, Ethereum, I mean, I remember I was at Greylock, and the ICO had just happened, I believe, or just closed, and I was reading more about what's possible with smart contracts, and it just blew my mind how the world was gonna be different in the future. And I didn't know what the time horizon was, but I, I, it was just this, this crazy moment of realizing the world was going to be different with this new technology, the kind of blockchains and smart contracts, and, and then I didn't act on it. And that is, uh, uh, it, there are so many times when you just take something that's ubiquitous for granted, and you don't act on it. And I think when you have these moments of, of insight, where you realize this is gonna be fundamentally disruptive, you have to find a way to act on it.

   15. HSHarry Stebbings1:00:32

       What was the most memorable first founder meeting you've had, Sarah? The one you're like, "Ah, I remember that one the most."

   16. STSarah Tavel1:00:38

       Without saying who the founder was, there's one founder I remember meeting, and I was just like, "Wow. I have never felt this force more strongly, that this founder was gonna run through whatever walls that they had to run through." They were just, like, so deeply and intrinsically motivated that they were gonna do whatever it took to make that company successful.

   17. HSHarry Stebbings1:01:06

       Did it turn into a good company?

   18. STSarah Tavel1:01:08

       A remarkable company.

   19. HSHarry Stebbings1:01:10

       Well, there we go. You were right.

   20. STSarah Tavel1:01:11

       (laughs)

   21. HSHarry Stebbings1:01:11

       What have you changed your mind on in the last 12 months?

   22. STSarah Tavel1:01:14

       ... maybe I'll go off piste here and say-

   23. HSHarry Stebbings1:01:17

       Sure.

   24. STSarah Tavel1:01:17

       ... uh, anti-Semitism. Uh, you know, for me, honestly, like a year ago, you, I just didn't know. I didn't know how real, uh, how, how real and present it was. And, and there's so many crosswinds happening right now. There's so many... There, there, there's... It's such a complicated topic, but to see it manifest right now, metastasize, has been really shocking and, and surprising to me.

   25. HSHarry Stebbings1:01:46

       Uh, uh, going off piste with you on this one, I'm like, it's such anano, anathema to me, because it's like such a strange concept, (laughs) respectfully. Uh, like, have we, do you think they've always been like anti-Semites who've just been quiet and now they have the chance to be anti-Semitic in public? Or, do you think it's like actually people who weren't, but it's just kind of, I don't know, pro-Palestine marches, and so it's kind of like just join in?

   26. STSarah Tavel1:02:13

       Uh, uh-

   27. HSHarry Stebbings1:02:13

       Do you know what I mean?

   28. STSarah Tavel1:02:14

       ... I'm sure. And, and look, just if you're pro-Palestine, it doesn't mean that you're anti-Semitic. But there is definitely a confluence of, of these two things happening, which is that there are people, and, and this is a lot of what people talked about, uh, during Trump's presidency, where suddenly the things that didn't feel like you could say had cover, and, and it let the kind of the vocalization of these, of these things happen more. And of course, when some people feel safe to say something that was once n- regarded as not the right thing to say, then it makes it impossible for more people to say it and believe it. And so there's certainly a lot of that happening right now. Um, but then also, like we're all, we're all like vic- you know, subject to osmosis. And if a lot of things get repeated, then we start to, it starts to change people's belief systems, and I think that's part of what's happening right now too.

   29. HSHarry Stebbings1:03:14

       Are you worried Trump will get reelected?

   30. STSarah Tavel1:03:16

       I am. I'm, um, it's, it's... If you were to, I'm not a, I'm not somebody who follows the polls very closely but certainly, what it feels like is that it's almost an inevitability, and that's a very scary thing for me, um, 'cause I care about our democracy. Uh, but it, it, it seems to be the direction that we're going in.

Episode duration: 1:05:49