Why Most AI Startups Will Fail ft. Gokul Rajaram

Hey everyone, thanks again for joining us on The Unpacked Pod. Before today's episode, I want to thank Stifel Bank for sponsoring the show. For Jurisdic Capital's fund banking and my own personal wealth management, Stifel is our trusted partner. Now on to the show. Hello everyone, welcome to yet another episode of The Unpacked Podcast. I'm Arvind, partner at Jurisdic Capital.

Hey, I'm Vignesh, a partner at Sierra Ventures, and today we have an amazing guest with us. We have Gokul Rajaram. Gokul has been involved with some of the most storied companies in all Silicon Valley, starting with Google, where he was— he led the team at AdSense, helped build that out. Went to Facebook, helped build out the ads team there and scaled that to over $6.5 billion in revenue. Then he went to Square, DoorDash, and I'm not sure if there's a great company Gokul hasn't touched or been involved with. And after all that, he's been involved with a bunch of companies as an angel investor, including the Trade Desk, interest in Coinbase. And so, Gokul, thank you so much for taking the time and welcome to the Unpacked podcast.

Thank you both for having me. Excited to be here.

Well, uh, Gokul, I know you post a lot on LinkedIn and you're sharing a lot of stuff about the companies you're talking to and working with, but, uh, maybe we'd love to hear a little bit about what you've been working on these days and, uh, what are you up to?

I call myself a company helper, but really I'm an investor and advisor to companies. And I spend most of my time supporting early-stage founders and also mid-stage, but early-stage founders primarily, which is my passion, who are deeply obsessed. What I want from them is obsession with solving problems, building a category-defining companies. And my hope is I can support them, serve them to figure out what the gaps are that they need to help build those category-defining companies.

You probably shared the story a lot, but I think just for the audience, I gave a quick rundown on your background, but your journey is always super interesting. Uh, can you just share for the audience on how you went from, uh, Google to being an angel investor?

Yeah, I think one of the most, um, interesting things that people ask me about being an angel investor— a lot of people start angel investing for profit, but I think it's always good to start things which you enjoy, and maybe it can become profitable or not. So for me, angel investing was because a bunch of my colleagues from Google were leaving. This was back in 2006-2007 to start companies. So it was to support them, that just writing a small check to help them. And then I realized that, oh, as a result of this, you get access to the founder of a company and you get, you know, you get to see how they are growing, what tactics they're using. And then they also ask you for help so you could help them. And so that's what I started enjoying, that process. And that's how basically I started realizing that as an operator, being an angel investor, there's a good reinforcing loop. Because you can take some of the lessons from early-stage companies and bring them to larger companies you work at. And you can help larger companies through lessons you learned. So you can help the early-stage company lessons in larger companies. So that's how I started. And then it became a— it was really a hobby, a passion. And I always say that anything you do, you should first make sure you enjoy doing it. If you do it purely for monetary gains— in fact, I don't think I've made a single career decision for monetary gains. Uh, it's always been first and foremost, will I enjoy doing this? And then second, okay, you know, all of that, if you enjoy doing it and you think you're going to be good at it, I never even negotiated my package at Google. All of those things just, you know, just happened and the company did well and, you know, you do well.

Um, yeah, it's, um, I think that's a really good piece of advice and we'll, we'll probably come back to this a little bit later, especially in this age of AI. It's so new, nobody really knows kind of what an AI-native company really is going to look like at scale. And so it'll be interesting to hear your thoughts on where people should be spending time and how they should be thinking about their, uh, their career. But maybe just to start with, um, would love to hear your view on what are you finding exciting in the world of AI. Are there certain companies, are there certain categories that you've been most excited about lately?

Yeah, um, so if you think about over the last 2 and a half years away— over the last 2 and a half years, I would say even the last 1 year— there are 3, I think, fundamental innovations that have opened up new vistas in AI. One, better document processing. I think until recently, I think you could extract some structured information from unstructured documents, but it was unreliable and limited to simple things. Now you can take a medical invoice, medical record, you can take a legal contract, and you can really extract a lot of not just basic information, but you can summarize insights, you can categorize them in, in powerful ways, etc. So And there are companies not doing just document processing. The doc processing layer has really evolved. The second thing is browser automation. I think a lot of business-critical data and logic resides behind these interfaces that are used by humans. And you had these RPA layers like UiPath and Automation Anywhere that were 10 years ago. These are all legacy, non-API legacy systems that this thing resides behind. And I think it's been very hard to interact and to automate the interactions with these systems. But now with browser automation, it's becoming better and better, more and more robust. And there's a lot of reinforcement learning environments obviously that are coming up to make that better. So browser automation is the second one, which really can help you navigate some complex workflows. And the third one is voice and conversational agents. Again, I think previously interactions which required tactical conversational capabilities like customer support calls, et cetera, but not very effective to automate, easy to automate. But now you see voice and text combined together to really pump powerful chained workflows that you can create. So combination of document processing, browser automation, and voice agents. I feel like companies that can use one or more of these capabilities are the ones I'm excited about, that are using them to unlock capabilities in, you know, mostly at the application layer. Could also be in the security layer. I'm very nervous about infrastructure companies, to be honest. I just posted about them because I feel in today's age, so if I think about the layers, there's a hardware layer, the chip layer, and the infrastructure middleware layer. Then there's security, you could say is part of it. And then there's the application layer. I feel that infrastructure— there's also the model layer, the model layer, then the infrastructure layer, middleware layer. I feel that infrastructure middleware layer is challenging. Why? The models are creeping into the infrastructure layer. They're slowly taking more and more. Second, you are not the infrastructure companies of days past where like developer-focused, they would just build something for developers, sell to enterprises, be happy. Third, that's not anywhere true because increasingly non-engineers are making decisions on what infrastructure to use. Infrastructure, you can see a lot of the— a lot of growth of Supabase has not come from developers. It has come from people who are basically connecting the Supabase code, not even knowing what Supabase is, through the white coding apps and so on. And true for most of the infrastructure companies, Browserbase and so on and so forth. So you have to have a customer mindset, end customer mindset, not just a developer mindset. And the third one is there are going to be crucible moments. As an infrastructure company, you have to change into a consumer company. Look at OpenAI. If they had just been a developer app, what would— or developer API, they'd be a very different company than when they launched ChatGPT. Vercel and v0. If you look at Groq's homepage, you think that's a white coding app, not a chip company. And so I feel like the next generation of companies and founders need to no longer be developer-centric, They need to build a completely new muscle, which is customer-facing, brand-building, storytelling, and customer product-facing. And so I'm not sure that most infrastructure founders I meet at least have that muscle. So I'm basically sitting out for the most part infrastructure layer, even though I'm sure they're going to be great companies because I feel like it's just a very— it has also a lot of leapfrog risk, I feel, the infrastructure layer, I call it. I think a good example is, I think the embedded databases, like the, let's not name any, but I think a bunch of the embeds. And I think those things were standalone for a while, but then it's unclear if the standalone business is how competitive it is with the Mongos and so on who added their functionality, embeddings and so on, the vector databases, sorry. So anyway, I think I'm very focused on the application layer, different kinds, both horizontal and vertical, but using those three fundamental capabilities I outlined. And then security, I feel it's always evergreen. I've been investing in security for 15 years. One of my big initial exits was a company called Shape Security, which was bought for $1 billion by F5. And I realized that there's always going to be security challenges with AI. Like half the things I get now are phishing, social engineering, or like fake stuff. And I think there's a massive industry to really make sure in a better way to automate, basically fighting AI attackers with AI defenders.

We actually had Sumit on the podcast from Shape, which is also a Sierra company earlier on.

That's right.

You know, I think I gotta jump in on this, Gokul. I mean, this is such a bright spot of conversation. I think we're going right into it. Um, I completely can kind of agree with you in terms of the infrastructure because I think, um, uh, if you believe that— forget the chip layer because I think that's a bit of a duopoly perhaps, um, uh, maybe some extent monopoly.

Let's hope for Gokul Rajaram investors that there's a third company there.

All right, it's a, it's a multiple, whatever it is, right? I mean, but then you have the model layers, and I think the model layers have sort of there was a point in time where people felt like, okay, is this gonna be just like telephone lines? Is it gonna be commoditized? But you can see that maybe for the long tail, but I think there are a couple of them that have sort of come up to the fore and then their ambitions are no longer just content with being the frontal for the consumer, but trying to go into the enterprise. And that comprises of building a, we can talk about muscle memory and how consumer-focused kind of companies go into enterprise. That's a separate topic. But it brings me to this notion of application layer companies. And then I think you bring out a couple of key components like what will be the next, you know, with AI, would SEO work the same or with companies like Browserbase and others kind of change the way that ad tech business works, right? I'm sure we can talk about that too. But would you say that the, what would be, your hallmarks of saying that, okay, these are the few things that an application company should get right to sort of win the— using two rights in the same sentence— using to win the right of being in the enterprise and, and, and, and, um, win the dollars, right? Because it is also noisy on the application layer because, I mean, that's right, very noisy, very noisy.

And I think increasingly, especially in the enterprise, I'm hearing that a lot of enterprises are standardizing on horizontal platforms for both the model layer but also These other platforms have built agent building or agent composer platforms. I think Google has something called Agent Space where you can go in and build agents. So if you can do that in Google and there's also Gemini built in all of that, why do you need a third party? So I think the defensibility playbook today, I think you have to, as an application company, you have to maintain an advantage when the infrastructure provider both has visibility into the product, also has a competing product to yours. So I think the defensibility playbook has to include creating, I think, a proprietary reward function. So you have to figure out how you build a domain-specific— I think it's all around evaluation. So how can you build a more domain-specific reward function, evaluation system, and reinforcement learning mechanism tuned to the specific workflows in your target market? If you can somewhat do that and you can get better in your target market by using those, then I think in theory, the workflows that you build are going to get better and better and the accuracy and the evals will get better and better. So I think you have to build something. So I think RL and being able to have a self-improving system is the name of the game. And the only way you can self-improve is if you basically have some kind of reward function on one side that your reinforced learning uses to improve, but then you also have evaluation that is tailored to that domain. With some deep workflows and multiple products that are so custom, so unique, and using custom data that it is harder for the horizontal, or it's not worth their while, I should say, to go after that deep niche that you're going after.

Yeah. I mean, I think we, so interestingly enough, on the day that we're recording this, we just dropped the episode with Sergey from Emma, and they've obviously made a very strong bet to go horizontal. I talked to— actually, I've made a— dropped a blog post, I think about a week ago now, talking about context and why the next generation of AI is going to be systems of context-driven. The question for you, Gokul, is you mentioned that, look, the AI— larger AI platforms are all going to think about going horizontal. And like, if you're going after like an HR function, okay, there's probably a lot of, you know, repeatability and you can kind of make it work. But if you're In those use cases, like if I'm going to go do HR for frontline workers versus HR for knowledge workers versus like, you know, in the UK or EU versus even like, there's so many nuances. Yes. Managing that context seems to be the key piece of it. You're absolutely right. And so as you've talked to the companies in the application layer, like is it smarter to be like, hey, I'm going to own everything HR related and just go deep there, or do you think it's better to be the like more like platform? Like how are you thinking about what is an exciting company in the future? Because everybody can say they're going to build a lot of agents, but I don't know how well they work at scale.

The exciting company in the future is something that truly drives ROI very, very quickly in a very specific use case to start with. I think, for example, if you look at accountants, for example, there is a bunch of companies that are going after accounting because CPA market, there's a lot of people aging out of the CPA market. CPAs are fewer. but only a couple of them, if you look at the product, if you look at how quickly they're able to— what is the exact labor replacement happening? There's only a couple of them that are actually replacing labor. Where— so I'm looking for companies that the ROI is measured in days or weeks, where the sales cycle— you sell the product, you don't even have to sell the product. You just do a— go get on a first call. They're like, this is what it does, I want a demo. Show the demo, it's working, and you don't even charge till there is value in the form of replacing labor or getting things done much faster. Ultimately, that's the proof of the pudding. I think if you can show value in the form of either new revenue or in most cases, lower costs, you don't have to employ people or hire people. That I think is it. And I think that can only be done to your point, to be honest, in a very specific use case to start with. So it's probably payroll for construction workers or something like that. You start with that, and then you build concentric circles around it, uh, to, to expand beyond just that. But I do believe in the value of specificity. I think, uh, taking on something like HR, broadly speaking, is a, it's just like boiling the ocean. And I think you need to first find a tiny thing. Dom, I believe in what Thiel said about you need to just find a niche, dominate that niche, and then go from there. Um, I, I think it's just too hard to try to be everything to everyone. I, I've never seen that company succeed. that's trying to be everything to everyone.

As you've made your investments, I'm curious about two things. One, in order for these companies to be successful, like, how do you balance, I guess, speed from a go-to-market perspective versus depth of product? Because obviously if you're going to go do something like payroll and construction, setting it up for that customer, making sure that, you know, you get this overtime right for this subspecialty, like, There's so many rules and things like that that need to probably be set up by customer. Like, how do you balance speed to market versus depth of product and really solving the problem at scale?

Yeah, I, I don't care about depth as much as I care about that product being remarkable enough to solve the promise that you have to customer. So if you can't deliver the promise to customer, you cannot launch to customers. So you should not go and launch to the customer unless you're confident once the product is working, it can solve that specific problem. It's okay to not be everything. In other words, you don't need to solve every specific payroll problem, but there might be a specific thing. Maybe there's a payroll— maybe it's basically reducing the number of hours it takes to reconcile payroll with your bank records or something like that. That's the thing that you want to solve. You cannot launch, in my opinion, until you solve that. You don't need to solve anything else. Everything else can be manual, but you need to solve that. I remember when Square launched, we were solving a very specific problem for a very specific segment of customers. Which was a single sole proprietor business, and we were solving the fact that they could accept payments quickly. We didn't have insights for them. It was hard to even know, but, but it's okay. We just wanted to get payments and we could get insights to an email to them. We didn't solve for multi-location or brick-and-mortar locations with multiple employees. So you need to solve a very specific problem for a very specific customer. If you can solve that, until you can solve that, you shouldn't launch. But as soon as you solve that, you should launch, even if you don't have the other bells and whistles around it. I think that's a key question. What is the promise to the customer you've made? What is the fundamental core of that promise? And you have to know what that promise is. You should not launch until you can. You can. And even if it means on— the reality is, if the promise is compelling enough, the customers will jump to bells and whistles to onboard themselves. That's the other crazy thing I've seen. They will go through, they will crawl through like all kinds of like thorn-infested marshland to get to your product. And that's the power of— that's when you see it's customer pull or market pull. And the market is like, oh my gosh, this thing saves me so much, or this thing allows me to do so much that I will onboard. And so in some ways, having these crazy onboarding, it tells you, are customers willing to— if you have to make the onboarding beautiful and everything else in order to get the customer to take it, that means that the value prop by itself is not compelling enough for the customer to say, I want this product. Sometimes that's, that's the thing. I think Square actually, crazily enough, Square forced you to enter your Social Security number to onboard. And so we had people from large companies entering the Social Security number as the CEO of this company because that's the only way they could access payment processing. We realized that we had something there when people were entering the SSN because that's the only way they get payment processing for their business.

Yeah, uh, just a, uh, adjacent question about product. I mean, I recently had a founder ask me, so what do you, what do you think is the right time to launch products, the next product, as you think about your product velocity? And that was an interesting question because this is an AI-native company founder, and I think he had this moment of thinking. I mean, my general heuristic in the past would have been, I mean, we're thinking about general SaaS, like, would have been like, if you're getting to like $15-20 million ARR kind of scale, that's when you should be starting to think about, all right, what's my next sticking the fire here, I guess. But, you know, I stumbled a bit because I think my answer to him was, and I'm curious to your answer, is you got to probably start now because if you're already thinking about it, you should, because I think your competitors are and you probably don't have the advantage of 12 months out to think about, do you want to start your next product? I'm curious how you would have responded to that in the AI era. Like, thinking about product velocity and getting multiple products onto the—

it's a great question. You know, exactly like you, I would have said 5 years ago, I'd have said you have to get to $10 to $20 million in revenue before you launch, before you even think about product number 2, uh, today. And so 5 years ago, building a compound startup was a luxury. It was a domain of people who could raise a tremendous amount of money, and otherwise was heavily discouraged by investors, you know, basically. But today, I would say exactly like you said, The time to start your second, build your second product is as soon as your first product has gotten product market fit. As soon as the first product is starting to, okay, you no longer have to constantly— there's product market fit and there's some pull, you need to build your second product. Why? Because one of the most important things, I think one of the biggest moats in software today is relationship with the customer. And once you have a relationship with the customer, you want to basically surround the customer with as many interlocking connected products as possible. To make sure no other competitor can wedge into them. And so I think it's very important for every company today to be a compound startup. I led recently the round in a Series A company that had 6 revenue-generating products at Series A, which you don't typically see. But for sure, nowadays, seeing even seed companies having 2 products and a third product on the way, or pre— I mean, so I think, yeah, the, the earlier rule books, I mean, on every part of the product development cycle, especially this one, have to be thrown out. You you cannot wait to turn 10 million or even 5 million. I think you've got to do it within 500,000 of your first product. You're absolutely right. It's a great question. I have a post called Compound or Die. If you search for Compound or Die, it's like, it's really exhorting everyone to think about what it takes to be a compound product, a compound startup with multiple products.

Let me ask, uh, the, I guess, the corollary, or maybe the counter-argument to that question, which is, look, 5, 7, you know, 5, 10 years ago, if you told me a company had gone from like 2 to 7 or 2 to 8, and that was really special. Like, it's starting to feel like you see it more and more frequently now. I mean, again, it's still special, but you just hear it more frequently. And I'm curious, are you, as you work with the companies that you're, you know, that you're invested in and helping out, do you have a heuristic that says like, hey, this growth for this company that went from 2 to 8, that's real revenue and that's real product market fit. So go, you know, start thinking about product number 2, 3, 4. Versus, hey, you're getting pulled into a lot of, call it maybe experimental budgets, or everybody's so excited about AI, so therefore you're just seeing like market pull. Like, how do you, how do you tease out what's real sustainable growth in this world?

I actually don't look at top line much. I actually think the two most important metrics for a company are gross revenue retention and net revenue retention, not, not top line. If you have very high net revenue retention, which means that your, your customers are growing But if you have— but it has to be balanced with gross revenue retention, which means your customers are sticking. Because if you have low GRR but high NRR, it might actually obscure the fact that most customers are churning and then some are growing really fast. I want to see both. And that obviously, that number is relevant to the type of business you're going after. But I want to see best-in-class dose. And I actually really like companies that are maybe not growing 10x year over year, but are growing 2 or 3x, but with very high GRR, NRR. Which means that once someone— because I think there are many ways to acquire customers. What is really hard is once you acquire them, to get them to stick and to get them to expand. And so what I want to see much more than the top line number growth is that if I'm not able to see evidence of that, I'm actually not as interested. Only third thing I look at is how's your top line growing? So I might be a different kind of investor, but I really care about retention and expansion.

But with how new everything is, I mean, many companies that see that initial growth and I mean, we all see, you know, so Urban maybe you see slightly different lens coming from the growth side. But at Series A, most of these companies haven't really sold for more than a year, a couple years. Like, the buying cycles are so short.

And yeah, and I'm curious, like, you can then look at gross retention at the very least. At the very least, you can get more retention. If these cups— if these people are behaving like consumers where they're basically losing 80% of customers within, then, then that's it. You've got to think of it as a different quality of revenue compared to businesses which have 30% churn after one year, not 80% or 70% churn.

Do you look at token usage or any of the— like, how do you kind of factor in usage data versus maybe the pure kind of buying?

I actually look at usage data more than revenue data because you're absolutely right. I think usage is a precursor to revenue. And so you want to understand when— if they're retained, it's not just meaning they're just sticking on. Are they actually using the platform? Versus, versus just sticking on. So that's the second level. If they're not retaining at all, then you kind of say, okay, pass. But if they're retaining, then you want to understand, okay, what are they doing? Are they building more apps? Are they using it more detailed ways? Because that is a precursor to revenue in the future.

Great. Uh, I think, I think the, you know, I'm just tagging on to your earlier comment around, hey, one of the modes that you see, uh, is customer relationships. And I think that sort of goes into this GRR and RR commentary you made. I, I, I 100% agree. The other aspect that at the growth side, um, because I mean, by the time it comes to us, the companies have scale, they have a lot of customers, referenceable customers, and sometimes suddenly we are thinking about, oh my God, this could compete with XYZ companies at the model layer, this could compete with this not AI native but AI enabled companies who are not slouches, they're just, uh, getting better. And also have customer relationships and a distribution. Therein lies this notion of data and how can you access the data. I'm curious, what's your observations having invested in, you know, early stage and seeing them grow, maybe even in the newer generation of companies? How are they acquiring data? How are they sort of convincing customers? Because customers have wised up. That, hey, their data is valuable. So they don't necessarily want to give it up too. So how do you, what are you seeing, I guess?

Yeah, it's interesting. I think one of the interesting things I'm seeing with data is that you basically tell the customer that their data will only be used for their own model training. So you basically create, but that can only be done if the customer is making a big commitment to you. So if a customer is making a multimillion-dollar commitment to you, then you can say that your data will be in your private cloud and we'll basically do it in your data center and we'll basically host our models and our technology, our models, whatever, build custom stuff in your data center. So it's in your private cloud. I think if the customer is smaller, then it's harder to say that. And then you basically, you know, kind of talk about, you have to make more hand-wavy gestures. I think what you do can say is that what you can do is better than what the models can do because you are training it with their data and not with, you know, general data. I think that's the biggest thing that folks are saying. I think it's very hard to promise. I mean, I came— I've come across a bunch of companies that are, for example, using their data to— I think the larger the customer is, the more they can have leverage on how this data is used and not shared or shared. So I think that's a counter that startups can use. So I would say, but for mid-sized customers who's spending $100,000, $200,000 a year, it's impossible to guarantee the data is private and cannot be used for and cannot be mingled or shared. But one hopes that if the performance, if the remarkability, if the value you're generating is high enough, then they're going to look past it because it's not like OpenAI is guaranteeing that your data is going to be not mingled. So you basically are like, well, if you go to the public cloud, 'You're— it's going to be the same here. You can get something pretty valuable here.' Most companies are like, you know, unless there is some stuff around HIPAA compliance and so on, there's some medical stuff you can tokenize it and so on, uh, so that's private. But outside of that, people are generally, generally getting more comfortable. I think there's regulated industry, financial services and healthcare come to mind, there's more challenges there. So I think those ones, the big companies are going to host these models in their environments. And that's why I think the open source models really come to play here, where you basically have the open source models and you can run your stuff on top of open source models. So now with OpenAI releasing its open source models, I think there's some very, very compelling, good performance, uh, performance and cost rate of things out there. I mean, OpenAI is— I think they're playing, playing a good game here. I mean, they are trying to really commoditize enterprise with their GPT-5 pricing and I think really trying to make it a profitless game for Anthropic and others and really moving the value up. I think you mentioned, Arvind, before that they're trying to move up to become a consumer product and they're trying to commoditize everything underneath it and saying, well, we are going to basically make our money from advertising and subscriptions and we are going to make it so that the enterprise business, at least from a token point of view, is not monetizable for anyone. No one can build a profit pool there. And I bet they're going to keep trying to do that. Commoditize your complement. They say commoditize your complement. So they're like, this is complementary to me, it's not my core business, I'm going to commoditize it. My core business is subscription and, uh, and advertising. So Fiji, who comes from Meta and Instacart, built a big advertising business both places. So you can see that they are going to definitely lean into that.

We'll put a pin in that. I want to come back. I know Arvind had mentioned the, uh, uh, GAO kind of category and that world. But, um, talking about AWS and the clouds, um, we've talked so far about the, the AI-native companies. I mean, we can talk about the non-AI-native companies. Obviously, one advantage they have is the data, but speed and how they integrate AI into their products and their strategy is kind of an interesting challenge. You obviously have a pretty unique vantage point on how companies are thinking about this. How are you finding that companies are— the best companies are implementing AI into their product and their strategy, and not just adding like a copilot or some kind of more quote-unquote simple AI layer on top.

Yeah, I think, uh, I've had, I've had 2 or 3 companies now that reached tens of millions of revenue in the pre-AI world, and then Gen AI hit, and they basically had to make a hard decision around what to do. To their credit, both companies built completely AI-native version of their product, which changed not just the product itself, where they basically built an AI worker, you could say, which was doing the same job as this earlier product was, and also change the pricing of the product to be based on outcomes that this AI worker could generate. I think the challenge was that the amount of money they made from the AI worker was much less than what they made earlier because of competition and so on. But to their credit, they started migrating customers from their old model to the new model. And there was a transition period. Where they basically went through a J-curve, essentially, where they were burning much more money. But I think they're trying to come out of it. So the best companies are essentially cannibalizing their existing business and saying, we are burning the boats on the old business, we are building a new business. The good news is that, um, for more simple applications where, for example, you're providing an AI worker of a certain type, it's fairly easy to build these things. On top of the foundation models, at least version 1. And if you already have customers who are used to using you in a certain manner, you can actually migrate them to this new, uh, more outcome-based, uh, model and product. The only thing is the board has to be supportive of maybe taking a revenue hit and maybe also going to the J-curve. And so I think the companies who haven't done this have about 6 months by early next year, mid-next year, otherwise the window is closed at this point or going to close. So every SaaS company that hasn't moved to becoming— that hasn't moved to having an AI-native equivalent product of what they used to have before GenAI, I think the window will close because customers would have already migrated to a competitor and will not be interested in trying their product. And also, it also shows that if you haven't been able to build the product 2 and a half years after ChatGPT launched, probably not going to build it anyway. You're not going to— any words you make or say are going to be hollow. So, most companies already, and I'm sure you guys have portfolio companies also, and so we all have from the 2020-22 cohort that were suddenly hit with this '22 tsunami and had to pivot. And some have done well and some, they're still finding their way.

Yeah.

I mean, it's interesting. We look obviously for a lot of the very early stage companies, it was either adapt or die. And the ones that adapted, managed to take this on and do well. The question though for you as a follow-up, love the idea of outcome-based pricing. Like, it conceptually makes sense when we're having, I guess, the Monday morning sort of discussion, but then when it's like, hey, I'm gonna go build a product and then I gotta go figure out like a new comp plan for all my salespeople because they've been, you know, talking about selling seats and whatever in the past, and marketing is gonna check How have you seen the companies that have done this well tactically approach this problem? Do they just say, hey, like, I'm going to kind of hive off the old stuff and you kind of go keep doing your own thing and I'm building a thing on the side? Do they just radically shift everybody to the new model? Like, what's worked for the best companies?

They do two things. One is they spend one quarter building this while the old thing is running, the second quarter starting to sell it. So about a 6-month transition. And the third quarter, by the end of 6 months, they basically just say the company is moving all in and they fully announce to employees and customers and move everybody and just aggressively. We are no longer investing in the old thing. But there is, like you said, a 3 to 6 month period where they're testing out pricing, testing out the product, building a product, figuring out the motion of migration as well as new customers. And then they just say, look, this legacy, this revenue, you have to almost think of us as a zero revenue company. Legacy revenue is just to reduce burn. There's no equity value in this revenue anymore, period. All the revenue that you have to think about is this revenue from this new product, and that is what is going to come. Good news is, I think on the pricing and sales compensation, there are companies like Salesforce, consumption-based, Stripe, et cetera. There's enough models out there around how you compensate salespeople for consumption-based pricing, which is only based on what they consume. So there's good enough public comps out there of folks, especially Snowflake, I think is a great example. Not Salesforce, Snowflake, sorry. Great example of consumption-based pricing.

Gokul Rajaram, I guess, you know, We got to switch topics, but we got really deep into many things here. But before we do, you did bring up one topic that's top of mind for me, which is voice AI. And, you know, so maybe get some commentary there. You know, I think just to frame the reference here, it's, you know, at the bottom, you have the models that are, you know, becoming much better, making it much cheaper to sort of time to first sort of token and all that. TTS, STT are all becoming a bit more commoditized. And then you sort of go up the stack, you have these middleware companies that are providing you sufficient— I'm not going to name any names, you know, but there are a bunch of there who are building these frameworks so that you can actually go and build your own agents, right? And then you have full-blown vertical integrated kind approaches, some in healthcare, for example. Um, where do you think this shakes out? I mean, do you think what is a good approach? Is it like try to be vertical specific because maybe healthcare demands it? You need to be the best for a particular vertical which is regulated, for example. Um, or is it built enough so that, uh, in the middleware, so that you can be relevant to multiple folks and then you can have multiple swings at the at the ball here?

I feel the middleware layer is, I hate to say, I think it might be death over time because you're being squeezed by the model layers on one side. Because if you look at Invoys, Cartesia, and Levin Labs, and Sesame now, I mean, they're basically releasing the whole stack and so will others. And so I think you, as an application company, I don't feel you want to pay the tax of a middleware company I mean, just like Google has their own agent builders now on top of Gemini, all the voice model companies will also have the agent builders and so on. And so I think it's going to be an MCP will just let agents and these models communicate pretty easily with data. So I just think that these middleware companies are a— I hate to say they might be a point-in-time company. I think they're obviously very powerful. I've seen many PMs actually just whip up models or whip up agents using these companies over weekend. I just don't know if it's durable because again, the models are just a different beast of type of company than we've ever seen before. They are absolutely voracious and they don't want to let oxygen between them and the customer. They're very customer-centric and customer-centric and they're going to build better and better tools, no-code tools and others. So I feel you either— again, customers, I think you own the customers at the end. That's the ultimate thing. If you're just a middleware company trying to serve a bunch of tech companies, those are never enduring customers because the tech companies are always looking to get better, cheaper, cut things out, et cetera. I would rather serve non-tech. So anyone who has a non-tech customer, I would go after them. I would not go after companies that have tech customers in this environment. It's been— I mean, it's, to be honest, if you look at the lesson from our prior generation, it's been the data companies. I think There's about 7 companies that are at $200 to $300 million revenue in the data space that all basically growing at 5, 10%, not that profitable. Why? Because Snowflake and Databricks have basically become the warehouse and they've made it very hard to build sustainable businesses on top of it. And they're all selling to tech companies, these companies. I won't name any of the companies. There are different analytics layers and so on. And they've never, you know, One of them was complaining to me that, you know, you sell to a tech company and then some engineer wants to be a hero. They just come up with some cool thing and then boom, the tech company just switches from you one day. And I think the same is true for these middleware companies. I can totally see, you know, some company has built on this middleware, but then some engineer says, why do we— why are we paying this middleware company $500,000? I'm just going to whip up something to directly talk to the— you know, because if it becomes really important and if you're paying a lot of money, then you want to just build directly yourself. It's not like payment processing where it is such a different thing with all kinds of compliance and regulatory things where you can't really replicate Stripe. Even if you're paying Stripe $10 million, you still have to pay. You just get second and third sources. But these things, if voice is your core, then, you know, you're right now a lot of people are using the middleware companies because they're testing things out. But as soon as they're making $20 million or paying the middleware company $1 or $2 million, guess what? It's easy enough to get 2 engineers to build the same middleware much more easily, in my opinion.

Yeah, it seems like in many of these categories, the orchestration layer has an interesting kind of future, right? It's like today, yeah, you need some tools to make the process of building the agents better. But over time, ultimately, it's, I think, who owns the context and the customer context specifically.

Customer context, yes. That's going to be key, right?

Yes. Where do you see the role of something like infrastructure though in voice? And the question is, you think you mentioned a couple of names like Cartesia and stuff like that. Obviously the big model providers haven't gone in that direction.

They should not. That's my biggest miss. One of my biggest misses was Cartesia. I met Karan and team. I was very impressed with them. I was like, hang on, this is something that the model layers will come and I've totally missed it. And they're doing very well and kudos to them. I think maybe, maybe voice. It's still impressive, crazy to me that they've gone after video and so on. But voice is still— there are these model companies that are doing pretty well, Eleven and Cartesia, and, and now, uh, uh, you know, Sesame with Brendan. And like, there's 3 or 4 of them. So shoot, I mean, to be honest, if the model companies really decide to go after them, I think they can. But for some reason they have— for some reason they have not focused on it as much. So I think they have— who knows how it's going to evolve, right? It's the next, next week, maybe OpenAI releases something and we reframe it. But for SoOn, I do think it does seem like they have created value. They've created a set of fans and customers who care deeply about these models, and they've also gone up there. They are exactly what I said, they are product-focused infrastructure companies where they have created end-customer good products and they're not just developer-centric. So they have actually tried to own the entire stack end-to-end. If you look at 11 and some of the things they're doing, pretty, pretty cool. I mean, end-customer-facing stuff they've launched. Yeah.

How, uh, how do you see the, these sorts of companies though playing out in the long run? Because obviously now that 11 Labs has gotten to scale, you know, you mentioned a few of the names that are coming after them, and I think you're seeing this in other categories as well, but cost performance are the two vectors everybody's going to be competing on. Maybe not.

I think they're going to also go the consumer route where they're trying to abstract it all out and make it easy for consumers and businesses to have, you know, to create your own songs, videos. Because ultimately, if you look at where value migrated even in the prior generation, it all migrated to the application layer. There are obviously infrastructure companies where you look at the market cap of the application layer companies, it is by far maybe 10x more than the value of the infrastructure company. So I think they are building this as a means to an end. And I bet they're looking at OpenAI and saying, if we grow fat and happy on supplying tokens to developers, that is a profit pool that someone is going to take aim at. We have to build the end customer brand where customers come to us and we have to use this to become the lowest cost provider for this application. And I mean, someone joked that if ChatGPT now were to replace their models with instead of OpenAI, they would use Anthropic. No one would actually care or even know. Because they've built a great brand. And that's ultimately the goal for all these guys, to build a great consumer brand, I think, a business brand that gives them direct relationships. And then maybe if OpenAI comes out with a better model, maybe they switch it out. That's what Cursor did, right? Cursor went with Claude and they were losing a lot of money, but now GPT-5 is there. I bet they're switching everyone to GPT-5, or at least giving them the option. Or maybe they will just charge not for the model, cost it just passed on, and they just charge you $20 a month. To be the software layer on top of the model cost. So maybe that's what the end model is where, okay, whatever, you can use whatever model you want. You— I'm, I'm— it's not going to touch my margin. I'm just a software layer to manage this and be a good IDE on top of this stuff. That's, you know, that seems reasonable. Maybe I don't charge $20, maybe charge $15, but it's 100% gross margin. $15, that's what I would do. And you just choose your model from a dropdown, model cost billed to you directly, Mr. Customer or Miss Customer. And so that's what maybe, you know, the end game is for all of these guys to quickly get agnostic of the model cost of the model itself.

I mean, then that makes sense because the value at the end of the day, I call it a lot of these companies seem to be just getting free on top of inference and you don't want to be a premium on top of inference. You want to have standalone value to the customer. Exactly. We've talked a lot about the change in business models and kind of how these things are going to play out. One thing definitely doesn't change, which is you always need great people. I'm sure the skill set of people is changing though. Curious what you're seeing with your early companies. I think it's two questions here. One, from a founder perspective, are there certain traits, characteristics? Like, what are you seeing that defines a great founder in this era? And then we can come back to the question of what are the both AI-native and non-AI-native companies doing when it comes to finding great talent for themselves?

Yeah, it's interesting. I think There are— there were— there was an archetype of founders that I find is missing now. There was an archetype of founder which was, uh, they would work for— they would work at a company for 10 years and have experience, 10 or 15-year experience at companies, and they're coming out of there to start a company. I don't see that many of them, maybe a few. Most of the founder archetypes I see are: one, kids straight out of college, or maybe a year or two, like literally out of college. Um, second, researchers at the AI labs. And third is repeat founders, basically founders who've just done— founded, start a company again, but not people who've worked at companies for 10 years. It might be because, and we are seeing this broadly, I think mid-career professionals increasingly feel like they grew up in a non-AI world, so they feel ill-prepared to start an AI-native company. Because so I do recommend to many of my friends who are in that bucket that they should go back and become ICs, individual contributors, at AI-native companies to build those skills back up. Because the way you think, the way you operate is all very different. AI-native founders are operating at a level of speed and they're operating almost in managed chaos. I mean, look at it from the outside, you're like, how's this company even working? It's like when Microsoft probably looked at Google in 2003, that's what it probably felt like for Microsoft, which was very waterfall-centric. And Google really pioneered agile and moving fast. That's what it probably looks like for the Googles of the world when they look at something like a Windsurf or so on. How are these kids even moving and moving so fast without any product managers. One of the interesting things is they're delaying having product managers for a while, 20, 30, 50 engineers. And second, the speed at which they're able to build software and launch software and ship software is pretty incredible. Like I said, I think with these middleware tools, one of the things about these middleware tools is that they allow you to ship code very fast. Almost everyone at these companies is a builder at this point. Even the marketers are builders. In fact, I have a portfolio company where there's a GTM lead, and he said he was doing a pair programming assignment with an engineer. I'm like, hang on, you're a GTM lead. He's like, no, I'm a builder now at this company. And so it's kind of crazy. They expect everyone to know and understand the technology really, really, really well and to be able to speak at quite a level of detail. And I think this wasn't the case. So I am seeing also people who just come from a SaaS company, GTM people and so on, struggling because the level of technical understanding in depth that is needed to succeed at these companies is quite high. So technical depth, chaos, managed chaos with very few PMs. And the third one is the speed of— and then 996. I think the level of intensity is kind of crazy. I think 996 has become— I've heard even at 007, which is 12 AM to 12 AM, 7 days a week in the office, right? In the office. I mean, you saw what happened with James Bond here. Exactly, man. You know, cognition, like, you know, that, that the thing that they give everyone, like, notice that we've got to have ADR weeks, and no one's apologizing anymore, I think. And I kind of agree. I think I do feel like early in your career, if you want to build something remarkable, broadly speaking, the notion of work-life balance is overstated. I think everyone says, oh, Elon, they don't realize how much effort Bezos or Elon or Zuck spent and how much of their time they spend in the office. Um, you know, and so we are all getting used to work at home, but I don't— I think back, back to the office, that's the other thing I'm seeing. No one, no one is, is doing remote. I mean, everyone is not just remote, I mean, the best AI companies are 6 days a week in the office. One of our companies, one of my companies, I think we— it was hard. I think the problem is that after about 15, 20 people the 21st person starts pushing back. And so I think it's going to be interesting to see how you can scale to 50 or 100 people with 996. And so I think there is definitely— there's— because you no longer have the same equity motivation and small team motivation that you have, that you had in the first 20. You no longer have this tight-knit team. And so the 21st, the 30th, 31st person, we'll see how that goes. But yeah, 996 has become the new mantra.

Yeah, it's really interesting because actually I was having this debate or conversation yesterday, uh, and the So much farther with the same, with fewer people as well.

And less capital also.

Yeah. So the pressure to go hire the 21st person is actually not as high as it used to be. So maybe many of these companies haven't really, haven't really thought about it. But I guess, are you finding that in any of the companies, especially the ones that are getting to scale, the value in going and getting that, you know, mid-career professional, somebody that's kind of seen it? Because the reality is that I'm sure every company will go through growing pains at some point. Seems like it's a natural thing that company will have to go through growing pains. Having somebody that's seen it, at least from the outside, seems like a prudent thing to do.

Yes, what they're doing is they want these people to join as ICs to start with, to prove that they can, they can do the work. Because I think I— and I agree with that. I think at Square, for example, after one or two people, almost everyone we promoted to director of engineering or director was all from, from within. And so the best companies do promote from within. So maybe they hire 1 or 2 people from the outside as exemplars to show leadership and so on. You know, but I think many of the promotions they wanted from within. So they want experienced people to join as ICs for a year or two and show that they can do the work in that culture, in the environment, with the tools in a native setting, and then they promote from within. So I have seen like many of my ex-colleagues have joined as ICs and I think they will get promoted. They joined as ICs, they were leading 20, 30, 50-person teams. Joined as ICs, one of the large platforms or other AI-native companies. And they're enjoying it actually, because they also feel like there's not as much pressure to prove yourself. If you come in as a leader, you have extra pressure to prove yourself because now you have these people who are super hungry, they know the stuff, and you're both learning this stuff and trying to be a good leader. Now, as you only have to do one thing, which is learn and do that, and they feel confident and they feel happy, especially because this is the first time in history where leverage from technology might be higher than leverage from people. Because now with technology, you can go much farther. I remember many years ago, I brought it up to our VP of Product at Google. Hey, can we get promoted without managing people beyond L6 at Google? He said, no, you— product manager, you need leverage from people. And so there was no way to get promoted beyond L6. You had to become— to L7, you had to manage people. It's kind of crazy. There was no step. And now I think you can actually, as a PM, you can accomplish 3 times as much as you can earlier. And I think the best people are trying to figure out— we're all trying to figure out how do we get leverage from— and managing people is not that much fun. You learn after managing hundreds of people, you would rather not manage people.

So I hear you, um, less emotions to deal with, I guess, but, uh, and more code, I guess.

But, um, and also Dunbar's number, right? There's a Dunbar's number concept where after a certain number, you, as Vignesh said, You can keep the organization small and spans of control are also increasing. Earlier you would see people managing 5-person teams. Those are all gone. Those roles are being eliminated left and right. 20-person teams. If you're managing, you be a full-time manager, manage 20 people. Managing 5 people is not a full-time job. You're just doing 5 one-on-ones, 5 hours a week. What else are you doing? You better start writing code. Otherwise be an IC manager or you manage 20 people. That's a full-time job.

Now, just to push back on the archetype of people that may still be able to succeed in the AI-native kind of land? And this, I'm gonna map it onto my favorite topic, which is GTM. I think GTM in AI-native is becoming so important just from a perspective of like, how do you rise above the noise, right? So is it your channel partnership knowledge? Is it your, how do you—

Prove it as an IC, prove it as an IC. That's what I would say. Company would say, great. But I'm not going to give you a team. Prove first that this thing works. Prove that your channel partnership knowledge works in my setting. Come in and show it to me. Have the confidence to show it to me, and then you can build a team around this. So I think I agree with you. I think, and the people who are confident are going in and saying, great, I'll prove it to you. And that's what I would recommend to people. Prove it as an IC because you've already led teams, etc. That's not, that's not a question. What is a question is Can this model, can your learning, can your abilities succeed within this environment? Can you, can you have an impact as an IC? Because I think you, you earn respect and credibility and trust first as an IC for most companies, and then you grow up. I've seen myself, the companies I earned the most respect and trust, where I did, even though I was a manager, I was able to as an IC do some work that showed people that, yeah, this guy is not just a talker or spreadsheet jockey, but he can actually get something done.

What are you seeing in terms of kind of new roles? And are there certain— again, for the founders and operators listening, are there certain kind of, I guess, non-negotiable roles in an AI world? I think, you know, is there like a prompt manager or something like that that you absolutely have to be a great company in the AI world?

First of all, I think the one important thing to note, I think I forgot to mention, is that what they're looking for in talent talent is also kind of weird. I think they're looking for AI-native talent. They're looking for these T-shaped people who are full-stack thinkers, not just engineers, but they can do product, they can do some design, et cetera. So it makes for very confusing hiring. Anyway, all that said, I think one of the interesting roles I'm seeing is what I would call an AI ops person or something like that. And AI ops person, you could argue, They basically are using, you know, they're basically taking all the prompts needed by the sales team, improving their prompts for lead gen. Basically, you know, they're creating browser use tools to then crawl LinkedIn and basically create like, you know, lead lists and do lead and account search. They're doing deep research on like prospects to figure out, you know, what the org structure is and do account enrichment. And so they're also like, they're basically doing, building a suite of AI tools used by the GTM teams and even using those tools on their behalf. So I think this is going to be true in every function where you're going to see a centralized, because yeah, you can get the salesperson to create their own AI things, but it's hard. So what is happening is the teams are creating an AI, there's AI engineering roles. There's literally a team that will be an AI engineer who's basically for the engineering team, figuring out what is the— evaluating all the tools. Should we create an AI, you know, AI engineer? Should we create an AI MCP server? I've seen actually people saying, well, we want people to self-serve, so let's create our own internal MCP server so that different orgs can just ping that to query it. And then you connect Claude to it or ChatGPT or something like that. So there's someone who maintains an MCP server of all the internal documentation. then there's someone who creates, uh, tools for the ops team. So I think you're going to see AI tooling as a very interesting, uh, thing emerge where each function has their own little AI tooling team or person. And so I think— and good news, these are emerging just like RevOps and sales have emerged over the last 10 years, 15 years. Now you're going to see this AI ops, AI engineering also emerging. You don't need to be the engineer, you just need to be curious and know how to use these tools well. And I think it's a very, very interesting and lucrative, in-demand, I would say, over the next 5 to 10 years.

It's funny, I think you're actually even seeing venture firms— I don't know about you guys, Arvind, but we are definitely like, like, in order for us to succeed at scale, we absolutely have to have these tools be effective and useful for us. And so I think even outside of the companies, even the venture firms are starting to do this.

Um, we got to talk about one role that has been, you know, thrown around quite a bit, which is forward deployed engineer, right? I mean, I'm curious to hear your thoughts on this. I mean, I think this term apparently is only reserved for Palantir because apparently they created it. Um, you know, I think it's a glorified way of saying it's services, uh, that is needed to take AI and deploy into the enterprise. Um, of course there is fringe cases that gets productized thanks to these, uh, FTEs. But do you think that's a sustainable role over time? Or do you think this is, we are in this phase of like help customer get the most value out of your AI and you need that to get your GRR and NRR in a place? Curious, where do you lean on that?

Great question. I think you need this role when it is not clear to the customer what they're buying. In other words, I think enterprise customers still truly don't know how to get the most out of AI. So you need— if you just throw a product at them, they're not going to be able to know how to use it. You need somebody who can bridge and translate what you're selling to what the customer is buying, and you need that connective tissue. And so I think all the large— even, even if you take the most basic example of customer support, which is the— which is, you could argue, the most used thing The big companies and the customers, AI customers, all have agent PMs and forward-deployed engineers in a space that is pretty well understood. Even there, to translate what they have and what the customer needs, and let alone like more complex things where, you know, where the value is harder to show. Even customer support, which is just— it's replacing a headcount of a customer support agent. But even there you have— so I think whenever you have to translate that I think you need a person right now because the reality is most Fortune 1000 companies don't know how to use AI beyond giving all their engineers Cursor and, and, and, and giving all the employees ChatGPT and trying to automate customer support. So I think till— and the best companies will use it, use it, use it till, till they are able to productize it away. I think Palantir at some point, till 2016 the number of forward-deployed engineers at Palantir were more than the number of engineers. And so that flipped when they started building, taking all the learnings from the customers and started building into a product that is easier to deploy. Right now, you're absolutely right. I think there is not a— AI is not a product sale, it is a solution sale. So when it's a solution sale, you need FDEs because it's a solution to a problem. I think Salesforce is simply a product. Salesforce cannot truly tell you what the ROI is on Salesforce. I promise you, there's no way. And so it's a product, you use it, you're supposed to use it, you just enter yourself. No one can say if it's working or not. You pay a certain amount, it is what it is. And maybe I'm sure even there, there's, they're customizing it, but here you're truly trying to deliver a solution that meets the needs of the customer and is also being priced on outcomes and so on, which is a secondary thing. But I think whenever, as long as there's a solution, you need solution engineers. So you're right, it is a glorified name for solution engineers. Who are basically taking a product and converting it to a solution, translating between what you are selling and what the customer is buying.

It's a really funny, I guess, dichotomy between the AIOps engineer and the forward deployed engineer that you talked about. It's almost like you're using the AIOps engineer in like a way that the customer thinks about the forward deployed engineer.

So good.

In some way. We are unfortunately coming to the end of the podcast. And so we will want to make sure that we I don't take too much of your time. There's one last section that we always like to ask about and we like to talk about, uh, very ironic topic considering, uh, we just talked about work-life balance, but ask it anyway. Um, in the short amount of time that you have to decompress in this AI world, um, we do think it's important to like be mentally recharged and find ways to, you know, come back and just be as strong as you possibly can mentally. How do you kind of, uh, decompress, Gokul, and What's the best way for you to stay charged?

Instead of talking about what strategy I used to decram it, I'll tell you one tip I use myself. I've seen that for me to be productive, I need my mind to be uncluttered and to be focused on what I'm doing. What that means is whenever something comes into my brain, and there's always like crazy stuff that's coming, I need to not have it in my brain. So I use Apple Notes as my second brain to just jot stuff down. I have folders for different things. And it's available on my phone, it's available on the laptop anywhere. So I just throw it down, it's out of my head. For example, here are the tasks I need to work on, or here's something I just remember, here's a tweet I need to post. So that is my number one strategy because I realize that something is in my head. I'll always be— even if I'm watching a movie or playing pickleball or taking my dog for a walk, it'll be in my head. So the number one thing for me that has given me more mental focus and clarity is to not have it in my head. and Apple Notes is my second brain that I use to take out of my head. So that's what I use really. I mean, there are other things I do obviously to decompress, but without that, I don't think I'll be able to truly decompress because things will stay in my head. So I feel a cluttered mind is the enemy of being able to truly recharge.

Yeah, I think that's 100% right. And I think for all of us, the more clarity we have, the better. I think if you go back to that founder question, clarity Clarity of thought is something we talk a lot about and the founders that have that.

And the best founders I've seen are able to have absolute clarity. Absolute clarity. If you talk to Tony, Larry, now whether or not they're right or wrong, only future proves, but they are absolutely clear on what the company should be and where it should go. That's the number one. Absolutely right.

Well, Gokul, thank you so much for taking the time. This is an awesome, awesome episode. I'm sure people are going to get a ton of value and just enjoy hearing your feedback and your thoughts on how the markets are shaping up. So thanks again for joining us and thanks everyone for for listening to the podcast today.

Thank you. I have strong opinions weekly held. Follow me at @gokulrajaram on X. Thank you, Vikbenj. Thanks, everyone.

Thanks again for joining the Unpacked podcast. Before we wrap up, I want to thank our partners at Stifel Bank. From general banking to venture debt, they have been incredible partners to my portfolio. Please remember to subscribe and we'll catch you next time.