516. AI Cost Structures and Pricing, Proprietary Data Sets That Create Moats, and a Clear Method for Determining Which Problem to Solve First (Vivek Vaidya)

Speaker A: This episode is brought to you by Ramp, the spend management platform we use here at TFR. They're offering listeners $150 just to take a demo. We've never had an offer quite like this. Claim your $150 before this offer is gone at our partner link ramp.com partners tfr. Now onto the episode.

Speaker B: Welcome to the podcast about venture capital where investors and founders alike can learn how VCs make decisions and reach conv. Your host is Nick Moran and this is the full Ratchet.

Speaker A: Vivek Vaidya, uh, joins us today from San Francisco. He's the founding general partner at Superset, an early stage startup studio building AI, native and data driven companies. Before Superset, Vive co founded Crux, which was acquired by Salesforce, where he later served as CTO of Salesforce Marketing and Cloud. And prior to that he was CTO of rapt, which was acquired by Microsoft. Vivek, welcome to the show.

Speaker B: Great to be here Nick. Thanks for having me.

Speaker A: Great to see you. Can you tell us a little bit about your background and how that led to Superset?

Speaker B: Yeah. So as you kind of just said in the introduction, I've been a serial entrepreneur. I've always uh, worked at startups except for two short stints at Microsoft and Salesforce that happened through the acquisition of Wrapped and um, Crux M respectively. And uh, I've been working with my uh, co founder at Superset and business partner, uh, longtime friend. We refer to each other as our uh, hetero lifemate, um, Tom Chavez. We've been working together for the last 26 years now, um, oh, it'll be 27 soon actually. And um, so he was the founding CEO of rapt, um, and then he and I, I was the first engineer that he hired at Rapt. And then he and I started Crux after we left Microsoft. And then uh, after we got acquired by Salesforce, I was uh, offered the opportunity to be CTO of Salesforce Marketing Cloud, which I thought would be a great uh, experience. So I took it went from managing a team of about 80 to 90 people all the way to almost a thousand, uh, overnight, um, because marketing cloud was that big and learned a lot. But then uh, uh, the big company machine that is Salesforce was a great company by the way, uh, was just not for me. Things don't move that fast, uh, at a big company as they do in startups. So uh, I decided to move on. And uh, Tom had already left by then and he was contemplating uh, other things, things we both love company Building, we wondered to ourselves, well couldn't there be more than one? And uh, therefore the idea of a studio was born. We were also lucky in that a lot of people from Crux who were then at Salesforce were hungry, um, and itching to start another company or work at another startup. But all of them had grown up right and they were ready to take on bigger roles. So we had three teams, two maybe three teams ready to go. And so we were experimenting with a couple of ideas ourselves. So we launched three companies and the studio came uh, about and so that's what led to the formation of uh, Superset. One of the companies on the first batch was a company called Habu, which became uh, we went through a few zigs and zags as startups normally do, ended up being acquired by Liveramp, this data clean room company. Habu was, and Liveramp is uh, now Liveramp recently got acquired by Publicis, um, and then another company from that batch, Ketch, uh, is uh, a privacy company that is a past series B and doing really, doing really well. So that's the quick background on Superset.

Speaker A: Awesome. So Vivek, how is Superset different from other venture studios?

Speaker B: Well, so every venture studios by and large are a relatively new phenomenon. Right. Um, when we started um, seven years, six, seven years ago, we were I think one of the earlier ones. And every studio has a different model of how they build companies. Our model uh, is we, we start partnering. We have two company formation models. One um, is we start working with founders pre idea, pre seed. So you may come to us and say hey, I want to build a company. It could be as open ended as that. I want to build a company. And uh, um, you say, I've been thinking about this problem, this problem, this problem. I have domain expertise here, here, here. And so we will work with you to shape the idea, help you discover what problem you want to solve, whether it's a real problem or not, and then work with you to build the company. Right, you are the founder, you're the captain of the ship and uh, you run the company. That's one model. The other model is we actually build companies ourselves. So we have our own theses that we explore and uh, uh from that we will start hatching building ourselves. And then we get to a point where okay, this is a real company and then you bring in people from the outside to work with us to work on that company. So it's those two formation models that uh, we operate. I think we're different um, from other studios in the sense that we're very founder friendly in that when we work with founders, they actually get founder level equity, um, in these companies and a lot of autonomy to build the companies the way they want to. We're helpers, uh, we guide, we don't dictate, um, how the company needs to work or build or whatever, but as long as it subscribes to overall company building fundamentals and all of that.

Speaker A: And do you install the CEO as well, Vivek?

Speaker B: That's a good question. So, uh, in the first model where you and perhaps a partner, uh, your co founder come to us, you're the CEO, so that CVO that's already in place. Right. And then our job over there is to assess whether the idea, not just whether the idea is good, whether you as a team are the right team to be building this company or whether you as a CEO are the right CEO for the job. So that's uh, a, um, entry level decision, if you will, that we make on ourselves. And then in the second model, um, it varies, Nick. Sometimes we. Like at Catch, I was telling you about Catching, Tom is still the CEO of Catch. Right. We just launched another company, um, Kana, where Tom is CEO, I'm cto. So there we haven't installed a CEO per se. But then as we kind of, and there have been multiple companies like at Habu, I was giving you another example. We brought in Matt Kilmartin to be the CEO after Tom and I had been in our role, CEO and CTO respectively, for a couple of years. So both models work. Uh, so we don't require a CEO to be brought in on day one. Even if it is a company that we're hatching. We're happy to kind of build it ourselves because the studio provides a lot of that supporting structure and the initial funding is, uh, taken care of by the studio. So we don't need the CEO to do fundraising, if you will, in those very early stages.

Speaker A: And how many of these will you do per year and how many per fund?

Speaker B: Because we get operationally involved, deeply operationally involved in these companies, were not able to do a lot of them. Right. So in a year maybe we'll do three new ones, uh, in addition to all of the backlog that already exists and per fund, uh, we end up maybe doing a roundup and say 15, uh, 10 to 15 is perhaps the right number, uh, from, uh, a fund perspective.

Speaker A: Perfect. Okay. Helpful. And you know, I know that you've been involved in multiple sales in M and A processes. Right. We talked About Salesforce, we talked about Microsoft. Um, I'm curious, you know, to get some distilled learnings and maybe, maybe what's the most important thing you've learned about selling companies that you would share with founders in the audience?

Speaker B: Yeah, I think the first thing is for founders, right, Be very clear about why you're selling or why you want to get acquired. Right? Because it's two different things being bought versus, uh, being sold. Uh, two very different things. And, um, be clear about why you're doing what you're doing. Um, like at Crux, for example, when we got acquired, we were on a tear, actually. Things were going really well. And sometimes, uh, what happens in the company's journey that you become very strategic for another larger company, so it's almost like they cannot afford to not acquire you. Right. And if that happens, great, you're in an awesome position. Um, but sometimes you're also in a situation where, yeah, you want out. It's been six years, seven years, whatever, and you're tired or whatever else, so you want to sell, and that's okay. Uh, no shame in that. But so be, uh, very clear about why you're selling, and then also be very clear in your own heads about what you want after the acquisition, you want to stay on, what's going to happen to the team, what roles will you get. All of that you need to think through and discuss with the acquirer as part of the negotiation, acquisition, uh, process. Um, and the clearer you are in your own head. You don't need to disclose everything, of course, but you need to be clear in your own head, because if you're not, then you may have expectations and they may not be met. Ah, post acquisition. Um, and as I like to say, expectations reduce joy, so best not to have any. Right.

Speaker A: Um.

Speaker B: Uh, but be clear why you're going through the acquisition process and what you want from it, not just for yourself, but for your team, um, as well.

Speaker A: In the case where a company is being bought, right. I think we all want to be in that scenario, right? We're being pursued, things are going well. How do you know it was the right time to sell? Right? Like, the price could be great, the partner could be great, but you're also on a tear. Things are going very well. You probably have venture or growth options available. Um, how did you know it was right time to sell?

Speaker B: Yeah, Nick, if I, if I could answer that with certainty, we wouldn't be having this conversation right now. Um, but, uh, but jokes aside, I think sometimes there are a lot of external factors as well. Right. That play into this. You have to take into account the fact you mentioned earlier you've taken money from a lot of important people, right. And it's your fiduciary responsibility almost to return that money and then some to your investors. So the calculus you have to kind of conduct is, well, I have this option. It's a really, really good option. Of course, things could get a lot better. There's uncertainty in that. So, uh, you end up having to form or conduct an expected value calculation in your head almost as to if things could get really, really awesome three years from now, what's the likelihood in that? And so what's the expected value from that outcome versus what you're getting right now? And then also think about, yeah, your investors and your employees. Because if you've been at it for a long time, your team may be tired. You, uh, may not be, but your team may be tired and you can't do this yourself. So, uh, the question of when do you know it's the right time or how do you know it's the right time? It's a very subjective kind of assessment that you have to do. You also have to think about where the market is going and all of that, and will it continue to be like that for the next one, three, five years? Uh, so it's a judgment call that you end up making. Um, but, but just it's one that you have to make as founder or CEO, what have you.

Speaker A: Yeah, I mean, super relevant for all, all our SaaS friends that, you know, exited before multiples, kind of got rewritten in that, in that space. Um, so back to the building side, Vivek. You know, you've been a cto, you've been a founder, you've been a CEO, now you're a venture studio gp. How would you say your perspective and approach to building companies has changed?

Speaker B: Yeah, look, I've had this view for quite some time now that's been accumulated over the years is building companies. Building technology, uh, is easy in the grand scheme of things. Building companies is hard, right? Because building companies, not just about building a product. You have to hire people, you have to, uh, get results through people. The market has to cooperate. You have to have the right product, you have to know how to sell it and create systems and structures and processes and frameworks allow you to do all of that. So there's that. But over the years, what I've learned, and it's very, very relevant now, Nick, is that, uh, selling, the ability to sell ahead of the curve is m the most important thing. Because especially now that given how quickly we can build, uh, at least initial products, the ability to. You have to embrace. Well, let me say it like this. You have to embrace, uh, this notion of living in the declarative present tense, meaning that you have to sell tomorrow's technology today. So if I'm talking to you as a customer, if I figured out how to solve a problem, I may not have solved it, but I've figured out how to solve a problem that's a pain point for you, then I will sell it to you like I have solved it already. Right. And the reason for that is by the time we get through all the, especially if you're an enterprise, by the time you get through all the negotiations, the contracts and all of that, you have enough time to actually build it. So selling ahead, getting comfortable in selling what's not on the back of the truck is become even more important now than it was five years ago.

Speaker A: And how do you advise? Like the technical CEOs out there, right? There's, there's so many strong, capable, smart, uh, CEOs out there that have technical backgrounds, but they, they don't have a lot of commercial experience. Yeah. You know, how do you, how do you advise them on how to build that muscle, how to sell ahead, how to figure out how to position the product and get the enterprise buyer where they need them?

Speaker B: Yeah, that's a great question. So having been a technologist all my life, I had the same challenge. I would get too enamored with the tech. Oh, look what cool features we have. And look, we've used this cool technology to build our product. Nobody cares. Uh, you don't care whether I use anthropic or, or OpenAI or, uh, Kimi in speaking in today's powerlands, right? What problem are you trying to solve? And does the customer care about that? Is it a hair on fire, painful problem that the customer faces today? So you always, technical co founders especially, just focus on the problem and the value it will value your solution will provide to the customer if they adopt your solution for the problem. Forget about how awesome your technology is or whatever. Well, your technology probably is awesome and that is important, but not when you are trying to get the customer over the line, uh, on just getting alignment on, hey, is this even a problem for me? And will your solution even work? Now the technology part comes into play when they say, well, can't I do this myself? Or what about this competitor or that competitor? Then you bring out your technical Chops and your moats or technical differentiators, but don't lead with them first. Get alignment on the problem that you're solving and why it matters to the customer and what value it will provide before leaning in with the technology. Technical founders are well known to do the opposite, which is the thing that they need to just unlearn, so to speak.

Speaker A: So let's double click on this, right? Like, AI is a blessing and a curse, because in one respect, it helps illuminate, oh, there's all these problems that we can solve. Right. But on the other hand, anytime you have abundance and you can do more, it can distract from focus. Right? So I remember reading this quote that you had. It's about solving the right problem. That matters more than solving a problem. Well, um, how do you know which problem to solve first?

Speaker B: Yeah, I think, uh, one thing that we do, one thing that we do often in our own customer interviews as we conduct them, whether it's for design partners or for early customer acquisition or whatever, we ask our customers, what keeps you up at night? Um, or another way of asking the question, if you could wave a magic wand, what problems would you, like, have gone away? And they'll say one, two, three things or whatever. Then that's not enough. They just told you what's on their mind. And a, uh, typical executive will have multiple things on their mind. Then you follow up and say, okay, and what are you doing about it? And then they'll tell you, oh, I tried this, I tried this, I tried this. And you're looking for problems that keep them up at night, for which they have gum, tape and glue solutions already. And so not only is the problem a pain point, but the solution they have put into place with gum, tape and glue, that maintaining that solution is also a pain point. You're looking for that, right? And then you go in and say, okay, we can. Then you've identified not only something that's a pain point, but also something that they have actually tried to solve themselves. That's very important because now they're telling you two things. One, it's important, and two, I've actually tried to do something about it and it hasn't worked. Right. Uh, so that's what we look for. That's what we advise our founders to look for as well. Is that second order or second degree? Uh, uh, assessment of, um, of problem problem definition.

Speaker A: I like that. That's great. Let's talk more about AI. Right. AI native products, uh, they require a different cost model than traditional SaaS. Um, how does one engineer the cost structure early? And what, what does that look like?

Speaker B: Yeah, Nick, I think it's, we're still so early right now and people are still trying to figure that out. What I'll say is this right?

Speaker A: Yeah, um, Come on, get your crystal ball out, Vivek.

Speaker B: Yeah, I know, I know. I wish I were that kind of person. Uh, people ask me for where do you see this going in five years? I have no idea. Uh, I'm a live in the present, here and now kind of guy, you know. Um, uh, but if you think about large, even medium sized enterprises, if you think about It from the CFO's perspective, CFOs hate unpredictability. So all of this usage models and all of that, and they've been around for the last 15 years since the cloud happened. Right. Uh, CFOs actually don't like that. They want predictable uh, cost. If I'm selling something to them, they want to know that every month it's going to cost them $10,000 a month or every year it's going to cost them $60,000 a year, whatever it is. Right. Uh, so you now as the um, uh, as the founder or product architect, whatever, you now have to decide how you are going to present a uh, almost fixed fee product using technology that you have to pay on a usage basis. Right? So recognizing that a couple of things take into Account 1, not all of your customers will be using all of your product, your entire surface area in the same way. Customer may, for example, Customer may have 10 terabytes of data. Customer B may have 6 gigabytes of data. If you are kind of uh, have a wide spectrum of customers, how do you take advantage of that in your pricing model back up to your uh, customers? How do you architect your systems in a way that you can take advantage of the flexibility of the cloud pricing model or the AI pricing models. One thing that we did well at Crux is uh, we used spot instances for a lot of the uh, batch compute data processing that we had to do and because of that we were able to extract more juice out of the cloud in a very cost effective margin positive uh, kind of way. A similar opportunity exists on the cloud side as well on the AI side as well. Right. Open weight models. There are lots of inference providers now that provide um, support for open weight, uh, uh, models and whatnot. You don't always have to use anthropic, but use it if it's giving you the quality and the value you're after. But you don't always have to. All that means is that you have to engineer systems in a way that you can track, that you can track cost. You have the right bells and whistles or knobs to kind of go here, go there, uh, go the other place, depending on the situation. But from a all in the service of providing a pricing model that's easily understood by the customer. Because the CFO is going to ask you what's it going to cost and if you say, I don't know, month one it could cost you $5,000, month two cost you $20,000. CFOs hate that. Right? So that's what we advise our companies or we try to do ourselves as well.

Speaker A: Vivek, will cost of inference continue to plummet in a way that will expand margins for these startups, or do you think the demand side will continue climbing, you know, in lockstep with cost side decreases, um, such that, you know, costs will continue to escalate?

Speaker B: Yo, it's a good question. I don't know. Um, but I can tell you what happened. What's happened in the past when we were at Crux, right, We were one of the very, very early adopters of the cloud, right? We started in early 2010 and we would use. AWS was the only real game in town. So we started with aws, uh, we stayed with them. Um, now for the first three years maybe every board meeting I would have to present the case for why we are staying on the cloud. Why doesn't it make sense to have our own data center? And for the first M18 months or so we would ourselves wonder when is Amazon going to raise prices? And Amazon just kept lowering prices. Storage was the perfect example. S3, uh, uh, cost just kept going down and down and down and down, right? And so my job became easy. Like, okay, look, here is the comparison we actually had for the first three, three and a half years, every three, four months or so, my head of infrastructure would actually do this analysis. What would it take for us to migrate off the cloud? And we never got to numbers that were even remotely close, that came remotely close to us switching off. I think the same thing is going to happen with infants as well. Costs are going to come down, people are going to build more. Instead of costs going up, they are going to keep coming down because whether it's the hyperscalers or these intermediate inference providers or the frontier models themselves, they're going to see so much usage that they're going to continue dropping prices and they're going to keep maintaining their Margins or revenues or whatever in spite of that. So that's what I think is going to happen. Of course, who am I and what do I know? But that's what happened in the past, so likely it's what's going to happen in the future.

Speaker A: Are you still managing expenses with outdated apps and spreadsheets? If so, it's time for Ramp, the leading US spend management platform covering corporate cards, travel, AP and more. All built to save time and extend Runway. We use Ramp at TFR and it's been a game changer. As our preferred partner, they're offering listeners $150 just to take a demo. No need to sign up. Head to ramp.com partners TFR to claim your $150 before the offer expires. Now back to the interview. Talk to us a bit about pricing. Right, valuation and multiples are fundamentally different for AI businesses versus traditional software. Um, I'm not sure the market has figured out how to price AI companies today. Um, or at least we're certainly not at consensus. Um, what are your thoughts around kind of the shape of AI pricing in valuations, you know, over the coming years?

Speaker B: Yeah, again that's, that's another topic that I don't have any real perspective slash opinion on. Nick. It's uh, I agree with you 100%. It's very, very early days. Right now you're seeing the exorbitant multiples, uh, for companies that fall into one or maybe three different categories. If you look at all of the acquisitions that have happened uh, in the AI space. Take Cursor for example. Great acquisition, very happy for the Cursor folks. Uh, but uh, the reason for that high multiple is adoption. Every engineer, the calculus and it's not deep point, but as the number of developers starts to increase, which they will, the demand for a product like Cursor starts to increase. So it's kind of obvious, right? But uh, if you take another uh, product, AI native product, say agentic finance for example, that same calculus doesn't really apply because it's not like everybody is now wanting to become a hedge fund analyst, uh, or running their own whatever, investment management, uh, firm or whatever. How the market ends up pricing those companies I think will be different than uh, how the market is priced developer focused AI companies. OpenDrowder is another great example. Stripe acquired them. Great. Very good outcome for openrouter, but not completely clear why they paid that much, uh, uh, for openrouter. Time will tell. Uh, um, but yeah, I don't have an answer. I don't think the market has figured out yet. The obvious cases are easy, but even then, do these companies deserve these high multiples? Maybe. But uh, as is the case with everything else, there's going to be hype and then things are going to kind of come down to some accepted normal, uh, in the next three to five years, right?

Speaker A: Yeah, it's a tough one.

Speaker B: Yeah.

Speaker A: But with these growth rates, I mean, we're seeing some wild multiples.

Speaker B: No. And look good for you if you got in any of these companies and especially got in early. Right. Take advantage of it, for sure. Uh, but just know that the music is going to stop and you don't want to be left without a chair when the music stops. Right. That's the, that's the other way of looking at it.

Speaker A: So, so let's talk about moats a little bit. Right. Um, you've talked about data and uh, investors have been making the data as a moat argument for years. And in the early generations of machine learning, in early generations of AI, that didn't really turn out to be the case. Um, why do you think that's different now? And why do you think data will create durable moats in this AI cycle?

Speaker B: I think. Well, the first thing I'll say, Nick M is I'll push back on that point. That data, if you. The data wasn't demote earlier. Right. And I have personal experience. Like at Crux, we uh, had um, architected not just our systems but our contracts in a way that gave us access to anonymized data that we could use to build more products, improve our services, create new products and all of that. And uh, that played a decent role in the overall economics, um, of the acquisition. The trick is how do you create that incentive for the customer and for yourself? The simplest way to do it, it sounds easier than it is, is to have uh, a set of product features that make it so obvious for the customer to say, ah, yeah, of course. This, this makes sense. I'm, I, uh, I'm, I can see the value in, in this, this transaction that you're proposing with, with my data. Right. So, so doing that I think makes sense. Uh, if you're an enterprise, you've always been told first party data is key. First party data is key. It was all about, up until then, it was m. All about how do you use your data to make your business better. Now, how do you use your data to make your business operations better. Right. If you have crappy data, doesn't matter what kind of AI you put on top of won't work. Right. Or it won't work as well, um, as it would, uh, uh, if your data was good or complete or what have you. Nearly every customer conversation I have these days, Nick, is, uh, especially with the larger enterprises, they will say, well, our data is not ready or our data is not of high quality. And that's okay. We have, our AI, has agents, and all of that. We'll give you an assessment of the quality of your data and help you fix it and all of that. Now our data is not ready. Well, unless you try. So you're saying one of two things. Either you're not interested in solving that problem, and if you're not, that's okay. But then forget about deploying any AI, uh, leave alone from us. Or the second, you're concerned about what that will tell, uh, you. Right? Uh, the third option is whether you actually try it out and it tells you what it tells you and then you fix it. Right? So whatever it is, choose one. Um, uh, because otherwise you won't know. So, um, I think the ability to use your own data to get better use of the AI. We talked about costs earlier. There's all this talk about small language models and how you can fine tune open source, open weight models with your own data to create proprietary models for yourself. All of that only happens if you have good data. The data does become your moat, if you will. Um, so again, I think it's a good narrative. Uh, it makes sense, it's a logical narrative. I don't know how many companies are doing it well right now.

Speaker A: Uh, Nick, when you're creating a new company, how do you distinguish between what the Frontier Labs might try and take on and build versus businesses that you feel the Frontier Labs may have less interest in?

Speaker B: Yeah, it's a good question. Um, I'll say this. The Frontier Labs employ probably some of the smartest people on the planet, right? It's safe to assume that that problem of a big company employing the smartest people on the planet is not a new one. Google, uh, has been like that for quite some time. And, uh, earlier it was, and before that, Oracle, you are a young startup and you would be asked, oh, why can't Google do this? Why can't, uh, uh, Oracle do this? And I would say, look, the answer to the can question, can you do something? The answer to the can question is almost always yes. The question is, will they? Should they, does it make sense for them? And then are they going to have their A team working on the thing that you're working on. So working on the assumption that you have an A team, your A team is competing with the C team of Google. And I'll take those odds any day. Uh, or anthropic or OpenAI or what have you. The other thing is, if you have lived, if you're building B2B in the enterprise and whatnot, if you have lived and stood in the pain of your customers as deeply as a good founder has, the Frontier Labs don't have that expertise. Of course they can hire people, but people who want to build companies are not working at the Frontier Labs. You'll get again the, uh, people who want to work at big companies and who want to have that safety, but then they won't be able to move as fast. Uh, I don't worry about it that much. Um, of course the counter example that people often cite is, well, if it were that easy or if that argument were true, then surely the Frontier Labs shouldn't be using any SaaS. Like why do the Frontier Labs use Slack, for example? They just build it themselves. They don't. Why is that? Right. So I think the same calculus, you need to think about it along those lines. Uh, so why do Frontier Labs not build things that they easily can, but they use from third parties? So, uh, the answer lies in like, okay, maintaining and enhancing and all of that becomes tricky.

Speaker A: Yeah, we were just, just the other day we were talking about Whisper Flow and I'm a happy customer and I love it. And uh, they did a really nice job of onboarding me for free. And yeah, graduate graduating me up to paid. Um, and it's a super fast growing company, like ridiculous growth, but at the same time, you know, I hit a key, I record what I want to say. It helps me translate that in an email or, or what, what have you. Um, I'm sure it's a technical challenge to do that, but there's a lot of others that have done it and it's difficult for me to think about the defensibility for that business.

Speaker B: Well, I think it's again, people think about defensibility in terms of technology. Oh, is it easy to build? And let's just assume that it is easy to build anything these days. The moat really is execution. Are you maniacally focused on execution? Right. You mentioned they onboarded you. Somebody actually took the time to sit with you and onboard you as a premium customer. You're not going to get that, uh, from the Frontier Labs. Heck, they don't even claude code. I pay $200 a month for claude it gives me a crappy answer, which is wrong, but I can't do anything about it. Right. Imagine that you're paying for a service, it's giving you the wrong answer and there's nothing you can do about it.

Speaker A: Not good.

Speaker B: Right. So.

Speaker A: Well, um, even in this example, they onboarded me from a distance, like I did it all through tech, you know, and it was seamless and easy and time to aha is super fast. Yeah, um, but it's such a native component of the chat products. It's so native to what they're doing. To your point before, maybe it's their C team that tries to do a competitor. So maybe focus and having your A team will continue to uh, out execute.

Speaker B: Yeah, but see, again, look, uh, the other challenge that these Frontier Labs have, which is the same as the ones Google had, right. You just mentioned, um, works in your email and it's seamless, Right?

Speaker A: Cross platform.

Speaker B: Yeah, exactly. Anthropic is not going to take the time to build that seamless experience across all of the surfaces where you work

Speaker A: outside of their native surface.

Speaker B: Yeah, exactly, exactly. So that's another dimension to this whole problem.

Speaker A: How do you feel about founders sending their data to the closed source Frontier Labs? Um, you mentioned open weight models before. What are your thoughts on open source and when is the right time to start protecting that data again?

Speaker B: So first you have to be clear about who owns the data and who has usage rights to your data. Right. So if you're a founder building in B2B, you don't own the data. Data is the customer's data. Right. So now you have to. So that's thing one. The second thing is, uh, you have to embrace the fact that you will be using Frontier models. There's no way in which you can go and say, I'm building enterprise software and I'm only building it on open weight models. We're not there yet. We may get there, uh, at some point in the future, but we're not there yet. Right. Nobody can take that risk, uh, right now at least. So whatever you end up doing, first understand what happens to your data when you send it. Make sure you have all the uh, zero data retention policies and Boolean flags and everything else, all that checked. Right. And make sure that you are not doing stupid things like you are not sending PII in the prompts, uh, to the Frontier models. Doesn't matter whether It's Frontier or OpenVait or whatever, you should not be sending your PII or a customer's PII out to the Frontier models and there are solutions that you can put in place to help prevent that. Use those. Right. Frankly, um, um, I don't think that the frontier models have uh, infrastructure which, like, oh look, Vivek's product is doing all these things and this data is coming in. Let me carve that out so I can use it for training or whatever. It would be too much of a risk, uh, for them to do that. So I don't worry about it beyond what I just said. As long as you take all these other things, uh, into account. I'm a big fan of open source. I think that um, the models are good and they keep getting better. That said, it's not snap your finger and suddenly you're, oh, I'm starting to use Kimik 3 instead of Opus 5. Uh, uh, doesn't work that easily. You have to have robust eval infrastructure. You have to know which problems you're going to solve with. Who's your inference provider? Are you going to. Really? You're telling me you are a 5 person, 10 person startup solving agentic sales or agentic finance. You're going to run your own infant servers? Really? Bullshit. I call bullshit on that. So you're going to be using base 10 or Nebius or, or Fireworks, uh, or any of these providers that are out there. Well, you have to understand how their ecosystem works as well. Right. So, uh, m. Anyway, so big fan of open source, um, but advice to founders. Know what you're getting into.

Speaker A: Vivek, if we could feature anyone here on the show, who do you think we should interview and what topic would you like to hear them speak about?

Speaker B: So uh, I'm assuming you're talking about people who are still alive right now, uh, because there's so many people from the past that uh, I would love to have a conversation with that uh, would be very, very apt, uh, for this type of show. Um, but I would, you know, as I said, I'm a builder. Right. Uh, and a technologist. I recently finished reading the Infinity Machine, which is the story of Demis Hassabis and uh, DeepMind. It's a great book. If you haven't read it, I would love to have Demis, uh, on this show talking about the whole build out the Nobel Prize and all of that. It's a fascinating. I think your listeners would love it.

Speaker A: Love it. Awesome. Um, I was going to ask you for a book recommendation, but I think we got it. Vivek, do you have any habits or behaviors that are a secret weapon?

Speaker B: No, not nothing that I would Consider a secret weapon. I do have some habits like uh, I have a routine in the morning which I try to stick to regularly. Uh, which is, and it's nothing deep. I just like make my own tea first thing in the morning and sitting by uh, uh, I read a book for the first half an hour of the day uh, in the morning and that helps me. It's not a secret weapon really but it helps me a lot. So uh, that's only routine thing that I do. Um, the other thing that I've started doing uh, the last two years or so is um, to my point about taking care of myself, just exercising regularly. Um, so much so that when I was younger I would look at people who had trainers and stuff and I would be like why do you need a trainer? Why can't you just work out yourself? Right? And now having worked with a trainer for the last uh, uh, two, two and a half years, I'm like what an idiot you were back then. It's so valuable. So uh, that's the other kind of habit that I have which uh, which actually keeps me sane, grounded uh, throughout

Speaker A: the life, allocate money to things that matter most. Right? Yes, 100% health is important. And then finally here Vivek, what's the best way for listeners to connect with you and follow along with superset?

Speaker B: LinkedIn is great. Uh, or you can just email me vivekuperset.com I'm easy enough to find. Um, yeah, I'd love to hear from your listeners on anything really. I have lots more book recommendations. Nick. I'm a voracious reader, uh, fiction and nonfiction so uh, we can have a separate conversation about that. Uh, but we leave it there.

Speaker A: All right, he is Vivek Vaidya and the firm is Superset. Vivek, thanks so much for the time. I got some really good insights and some original thoughts and really enjoyed the chat today. So thank you.

Speaker B: Yeah, I really enjoyed the chat myself. Nick M. Thank you for having me.

Speaker A: Alright, that'll wrap up today's interview. If you enjoyed the episode or a previous one, let the guest know about it. Share your thoughts on social or shoot them an email let them know what particularly resonated with you. I can't tell you how much I appreciate that some of the smartest folks in venture are willing to take the time and share their insights with us. If you feel the same, a compliment goes a long way. Okay, that's a wrap for today. Until next time, remember to over prepare, choose carefully and invest confidently. Thanks so much for listening.