 Thank you and welcome everyone to the panel on investing trends in GenAI. So in terms of logistics, we'll do a 20 minute sort of quick answering of questions which we most commonly get and then we'll open up the floor for questions. So we've assembled a set of panelists who focus a lot of their time on GenITBI, both in terms of investing in GenAI startups as well as serving on boards of GenAI startups. So with that, I'll start off with a quick intro about myself. I'm Avi Bharadwaj. I'm an investment director at Intel Capital. Intel Capital is Intel's venture arm. We invest roughly between $350 to $500 million annually into startups. Typically focus on series A and series B startups, so slightly later than the foundational stages, although we've selectively invested at the seed stage as well. In terms of where we invest, roughly half of our investments are in cloud software and the remaining half is in hardware investments. Yeah, so with that, I'll pass it off to Jocelyn. Hi, I'm Jocelyn Goldfein and I'm a managing director at Zeta Venture Partners. We are a boutique sub $200 million fund focused on the seed stage and focused exclusively on AI. We're named after the Zetabyte and have been doing AI and data infrastructure investing exclusively for the last 10 years. So we were maybe foolishly early to this market, but we've seen a lot. We've seen the first generation, the second generation, and now easily the third, if not fifth generation of AI. We'll see. We usually invest $2 to $4 million checks to be the first institutional capital into a startup, generally anywhere from inception to kind of pre-series A, and very focused on both applications powered by AI, as well as the tools and platforms that are enabling companies to build these models. Hey, everybody. I am Erica Brescia. I'm a managing director at Red Point Ventures. We have been around for almost 25 years, so I guess, OG in the venture world. I'm an investor out of our ninth early stage fund. It's a $650 million fund. So we do both seed and series A and some B, and then we also have a growth fund that does B and C. So we can follow partners all the way through their journeys. I am particular and focused on everything infrastructure, developer tools, AI below the app layer and security. I am very fortunate to get to work on a company with Jocelyn, which hopefully we'll talk about a little bit later. Before Venture, I was CEO of GitHub for two and a half years. We built and shipped co-pilot among other great tools, but all anybody wants to talk about now is co-pilot. During my time there, and I was a founder before that in the open source and infrastructure space. So recently moved over to Venture and try to bring that perspective, having been a founder and operator and now a VC, to impart all the things that I didn't understand as a founder to folks that are on that journey now. So over to you, Shomak. I'm Shomak Ghosh, partner at Boldstart Ventures. We have been around 13 years, so a little less long as a red point, but what we focus on is basically the earliest stage of backing founders and that's all we do. So we don't do any series A's. We are the first institutional check to all of our companies and that is usually two people with an idea, maybe some Figma mocks and then that's when we like to partner with them. All we do is enterprise software and given that focus, as you can imagine, of course we look at a lot of AI native companies and also AI enabled companies to just look at different trends. So it's something that we're quite excited about. We're investing out of our sixth fund right now and just to let you know the type of nerds that we are, the fund size was $192 million, $168,111. And so the default router IP address for all of you out there. So that is what we like to do and that's what we embody. That's awesome. Thanks for the intro guys. So maybe to kick things off, I think at a high level we talk a lot about Gen AI as one big blob, but looking in greater depth, I think there are different stratifications within that category. I think there are the foundational sort of hardware providers above that. We have the foundation model creators. We also have the model ops category and then applications. So maybe could you talk a bit more about where you've seen investment dollars flow in the last couple of years maybe and then where we will see investment dollars flow in the next year or so? I'm happy to kick it off. Well, I guess the investment dollars to where it's flowed has certainly been the foundational model. I think it's actually been amazing to see. I think what's also interesting is like the geopolitical landscape where you have some knock-on effects. For example, no one has done this yet, but somebody should. Like the open AI of India is kind of out there and waiting to be funded. But I think that's where traditionally right now most of the money is flowing to. And I think we'll see kind of how that plays out. There's certainly going to be very large companies that are built and have already been built in that space. But at the same time, we're still trying to figure out what those business models might look like, especially just given that the massive amount of capital that they're ingesting early, we're just going to see how that plays out. But that's certainly been where at least I've seen the most. But Erica, if you want to. Yeah, I think we are one of those firms that has written checks into a couple of foundation model companies that we're really excited about. One is Mistral, which has been in the news quite a bit this week just released an incredible new model. Yay. Go Mistral. And also Poolside, which was, we spun it out of Red Point actually, so we were able to get in at a privileged valuation, I will say. That's Jason Warner. He was the CTO of GitHub when we built co-pilot and building a new foundation model for code. So I'll put a plug in there for those two. But I think what's important to note is the reason that all the capital obviously has been flowing into those companies is because they're very, very capital intensive. It's our view as a firm that there will not be hundreds of gigantic foundation model players that went over the long term. We think there will certainly be more than one. And you know, I've been on the board of the Linux Foundation for eight years. I guess I forgot to mention that. But I'm very, very bullish on open source in particular. And so I think there's a lot of excitement there. But I think a lot of the foundation model investing, at least for the general purpose models, at the early stages is probably mostly behind us for a while. We certainly as a firm are very focused on that middle layer of how do you tune and host and optimize and secure and make use of these models in the enterprise. And then we also look, of course, at the app layer. And I like the way that you distinguish it between AI native and AI enabled. And I'm sure we'll come back and talk about companies in those two categories. But I think if I think about focus and where dollars are going to be flowing, I see a lot of interest at that application layer now. And where are there opportunities to invest in companies that can truly change the model of how work gets done? And so we've gone kind of category by category and looked at industries where there's a tired incumbent that will have a very hard time completely rearchitecting their application to be AI native. So think spaces like ERP and procurement and even CRM. And thinking about in the age of AI, how can you completely reinvent this category of application? And I know a lot of other VCs are thinking about that. So I think next year we'll see a lot more investments in those types of companies. Yeah, I'll agree emphatically with what Michael panelists have said. And maybe just to add to that, I think I saw a pitch book report around the middle of the year saying fully 60% of the venture dollars that have gone into AI have gone to foundation models. So like it's a huge proportion. And when you see these kind of eye popping headline numbers about investment in AI, you know, I think a lot of founders have sort of the idea that if they just get a dot AI domain and or put, you know, name AI on their pitch deck that that VC pixie dust, we will just fall out of the sky. I think VCs are not just more discerning than that. But I actually, I think that even the set of VCs, I think everybody here is very much like a thought leading VC. There's a set of VCs who are just following the herd and trying to pattern pattern match what has a lot of hype right now. But even those I think are are not blindly investing behind behind AI. It's not quite that easy. And I would say that apart from the foundation models, I think there's still a lot of confusion out there, especially among generalist VCs of what is a good investment. And I sort of fully agree that like, I don't think now is the right time to go start another foundation model company. I, you know, please feel free to prove me wrong. I wake up smarter every time I meet a founder who teaches me something new. It's the best experience. But I don't yeah, they're so capital intensive to build chances are you're not going to start with a seed amount of money. You know, but I think it will look more like the cloud where there's a where there was a dozen or so entities, organizations that are deep pocketed enough to make this huge capital investment. There'll be enough that it's not a monopoly. I too am a huge believer in open source here and open source model alternatives to proprietary. So yeah, my attention is very much more on the applications and then also the tooling. I think we're still in the early innings where people are figuring out everybody's playing with these models and building prototypes and people are just starting to get them into production and figure out what it takes to have these systems in production. And so what are the developer tools and the DevOps tools to make model based products, AI native products. So I think those are really rife with opportunity for entrepreneurs. I think VCs right now for the last 12 months have been a little bit more excited about the picks and shovels and a little more confused about the applications. What I can kind of add to what my teammates here have said. We had some early headaches in Gen AI. We had some application companies that built up a huge head of steam by basically building a thin wrapper around GPT-3 before chat GPT launched and before everybody understood just how much of their application success was coming from the underlying model as opposed to what they'd built. And those companies did this. And so I think that spooked a lot of VCs actually off of AI applications on the theory that oh, the foundation model providers are just going to commoditize all of them. And I think we're starting to come out of that. As Erica said, like attention is turning back to applications now that the tooling category seems pretty well funded. But I think VCs, many of them, especially those who aren't focused exclusively on AI, are still confused about where there's good opportunities for applications and where there's risk of commodity. So expect if you're building an AI application to have to explain really clearly where your differentiation and moat lie. I think that's the question on every VCs mind about AI applications. Yeah, I was going to say, I think the number one reason after not having a credible team, which is always like obviously table stakes, but the number one reason that we passed on AI companies in the last year was a lack of defensibility or our inability to see how you were building a defensible business. And part of it was thin wrappers or part of it was our belief in how quickly the technology is going to advance to be so evenly available that you're not actually building anything that will be defensible for the long term. Yeah, and let me just tag one more thing on that. So the foundation models themselves are one source of competition. Another one, which Erica alluded to is that an incumbent, if you're building the smart CRM, like Salesforce is going to be shipping all these features in the next six months, you just roadkill in their roadmap. Like don't be roadkill in the incumbents roadmap if it's like a very wide awake incumbent who's pursuing this stuff aggressively. And then the third category to worry about is have I seen 30 other entrepreneurs with exactly the same pitch deck as you, which I'm sorry to say in the last year, I cannot count the number of pitches we've seen about either Gen AI powered marketing, Gen AI powered sales or Gen AI powered customer success. There's no doubt that those are huge markets and low hanging fruit, but absolutely everybody has their eye on them. And so you have to stand out with a really novel product idea in those categories, or with like an idea to address a different category that not everybody else in the world is thinking about. Because if I've seen the pitch 30 times, you have to work 30 times harder to persuade me. Now you don't have to persuade me, it's a good market, but you do have to persuade me that you're going to win it. Yeah, those are all great points. And I think the way we see it is very similar as well. I think all the very large foundation model companies are probably already they exist today, and they'll be one of the handful of winners who will emerge. And we feel that the application layer companies are the ones where a lot of the investment dollars will go moving forward. And I think it was Tom Tungus who came up with the statistic, which is that in the cloud wave, there have been three companies at the infrastructure layer that are worth $2 trillion. But at the application layer, there are 200 companies that are worth $2 trillion. So as VCs, we are probably more likely to find winners at the application layer, even in this problem in this genai wave. So given that, and given that we spoke a bit about like defensibility in this right in the application category, how do you guys think about like, how does defensibility stem from given if everyone is using similar foundation models and probably similar data like is data still the still where more comes from as it workflows like there are many hypotheses. Yeah, so defensibility is a really fun thing to talk about this time, right? Because everything's shifting so quickly that frankly, we're all still learning what is defensible as founders and as and as investors, right? I think there's a couple of interesting things to consider here. So one, traditionally, I've been an infrastructure and security investor, right? And so all of my previous investments have been that the last two investments that I've done have actually been application layer. And that's been very interesting. It's not been if like one of them is a marketing data analytics company. If you had said in 2023 that show make was going to be investing in mark in MarTech, I would have said you're batshit crazy, right? But what's happening right now is interesting things in terms of, you know, one, what's happening with the data integration layer, what's happening with, you know, the so-called modern data stack, the data stores that we have, right? There's these enabling technologies that have been built. And now what you can start to do is you can train models in very unique ways. And so for example, text to sequel, it's something that's really exciting. And if you can fine tune that to a very specific instance, then you can start to deliver a feature that helps out a customer in a big way. In the marketing case, maybe it's someone saying, you know, I'm trying to move from, you know, a certain type of website analytics to a different type of website analytics. If you train specifically on that, right, you can do the data transformation and start to show the value from the product that you're building a lot sooner, right, without that. So that's really an interesting thing because what's the moat there? Well, it's not necessarily that you train this co-pilot. It's not necessarily that you built this problem. It's that you figured out how to integrate those together to deliver a new experience. And then on top of that, now you're starting to build the workflow layer even more aggressively alongside the customer, right? So that's kind of one way that I think about is like, how are we going to approach common problems in a new way? And then how can we start to expand them out more quickly? By the way, I've looked at probably 50 to 75 testing, observability, fine-tuning, self-hosted infrastructure companies, and I'm not investing any of them because I don't find any of them to be defensible. The challenge is like, whenever we look at it and we kind of look into the future, we're like, well, yes, these are going to be problems. But actually, when you talk to the CIO, it's like, Bristol Myers Squibb literally went out and paid open AI a bunch of money. I don't know how much. But just said, hey, just give me a reserved instance so I don't have to worry about you training the data or anything like that. And then just let me go at it. And so now they have Bain working with them to go spin up various applications. But this is something where like the infrastructure doesn't really matter to them because they just got delivered to them. And now they're starting to do different applications and different use cases on top of that. And so that's kind of how right now we're thinking about. But I'm curious. I mean, you've definitely made some investments. And I think that's the cool part is like, everyone's going to have different viewpoints at different times. And you just got to find the right investor that believes in that vision. Yeah. I think one interesting note on that is the stage of fund might also influence how we think about investing. To your point in that, we probably all three of us look at all of the same deals. We see all of the same companies are probably 85% overlap, which is really fun. And we might all come out with different conclusions. One thing for me is because we have a bit of a larger fund, in some categories like this, I've been like, look, maybe there's an opportunity to build a winner. But what I see is two dozen teams and at least five of them seem really, really sharp on paper. And I can't figure out which one of these teams or approaches is going to win and then markets evolving too quickly. I'm going to try to be real helpful to those five teams and build a relationship and see who breaks out and see if I can go do the A. And I think it'll be interesting to see in 2024 and probably 2025, the number of preemptions when we start to see breakout companies in certain categories that we've been tracking with interest, but haven't been able to pick a winner in. Back to the defensibility point, I think there's a couple of ways that we've been thinking about defensibility too outside of just like categories. One of them is privilege access to data to your original question. I do think data is everything right now. And people who have very unique access to data sets, which are very hard to find, by the way, because it's mostly the big companies that have that and they're doing useful things with it already. But I think that's one area of defensibility that we look at. Another one is like for an early stage company, they have some kind of unique access or relationships that will allow them to get really far ahead very quickly in a category where they can start using like RL techniques to improve their own model. So there's just so much better that they might have a shot at just always being out ahead. Again, really, really tough to do that well. And it's not like we've made a bunch of bets on that basis, but it's something that we do keep in mind. I'll give you an example. We invested in a stealth company that's in a one of these like incumbents who is very tired, who we think is finally disruptible. And this team, the guy was a former CEO of this company at one point. So he knows very well the space and knows every single key customer on the planet and can go in there, has them as design partners, can completely reinvent the experience from the ground up, has a very specialized knowledge of the space and what it takes. And we think with the product that they're building will be able to drive enough early usage that they'll get so far out ahead that it'll be hard for others to catch up. But again, it takes a unique team, unique relationships, unique team, like domain expertise in order for us to want to make a bet like that. Yeah, that makes sense. So I think a lot of the factors are somewhat similar to what they were before Genai, right? So I don't think Genai is this magic pill that solves all of your defensibility problems. You still need to have a strong team. You need to have a lot of the same components that existed before as well. Maybe a last question that we often get and then we'll probably open up to the audiences. Are there like specific milestones that you would think of and particularly for folks who are planning on building an open source based startup? Like how do they start thinking about it? What should their like 30, 60, 90-day goal be like? And what should be ready before they start talking to an investor? Well, I think there's plenty of investors who invested inception. So in some sense, you may not need more than two guys and some figment mocks, right? Or two people of any gender. But seriously, I think that in open source in particular sometimes what inspires a founder to start a company is that they are already the creator and maintainer of an open source project. And when they see momentum, they realize, hey, maybe there is a commercial opportunity around this. And so that can be the inception phase story as well. So certainly with open source, there are enthusiasm of adoption. And I think the thing that VCs pay attention to that maybe is not 100% already obvious to all of you is that it's kind of, we care a little bit more about the slope of the curve than the absolute position. So it's not how many GitHub stars do you have. It's like how fast are you adding GitHub stars? It's not how many pull requests or how many discord members. It's how fast are they piling in? Like when you see that kind of knee in the curve of rate of adoption, that will get VCs attention. And if that's happening on a small scale, it will get the attention of early stage folks like me. And if that's happening on a much larger scale, that will absolutely get you the attention of the red points and who invested the A and B and beyond. And certainly I would say when I think about the investing that we're doing, most of our applications are not open source. Almost all of our developer tools and DevOps and data infrastructure, I would say 70, 80% of them are open source led. And that's because it's such a powerful distribution advantage for any product where the selection, not necessarily the budget, but where the selection is owned by a technical decision maker. And we talked about defensibility and we should absolutely, and we talked about defensibility from go to market. And I would say that a really enthusiastic community is an incredible source of defensibility because it's one of those distribution superpowers that Eric was talking about. So I would say it's that. It's the passion of your community, but also think about the total size of market beyond your passionate small community. And if the problem you're solving for your community really applies to a very large number of people, or maybe a small number of people today who are working with generative AI, but foreseeably a much larger number in the future because Gen AI is a rising tide, then you should really, really think about that. Recent open source investment that Zeta just made is into Guardrails AI, which is the creators and maintainers of the Guardrails open source project. Shreya started this as a side project. She had come from the self-driving world, been at Apple, was a founding engineer at Predabase, was really happy at Predabase. It was a great place to be, had no desire or intention to leave it. But that, but she started Guardrails because she had this kind of safety mentality from her work in self-driving. And it, and safety has been such a big part of how on earth are we going to get these things from the lab to production? Like Bristol's Meyer Squibb does not have to worry about production right now. Their infrastructure needs are all met because they're still in the lab. Once they go to production, they're going to have a whole new set of problems to figure out. And so the early feedback on Shreya's project was that people really needed this. Not today in the lab, but, but before they could get to production, they needed something like this. And when she saw that kind of, that head of steam, that pull, it compelled her to leave a job that she loved to start a company because it seemed clear this company had to exist. So I think that's like a, you know, like to me that's the quintessential open source founding story. Yeah, I'll add maybe a little bit to that first love guardrails and Shreya, a very, very impressive and congrats. Watch this space. Yes, eagerly. I would say a couple of things. I think number one, just on the open source front, the feedback loop that you get from open source is also a huge competitive advantage in addition to the defensibility, right? You just learn so much faster what's resonating with people once you start to see that pull. And I think that shouldn't be underestimated. To your point about what we're looking for, I actually have a whole talk that I will try to finally put online. I gave it last week on what investors are looking for in open source companies at the Series A, but I'll say a couple of things. Number one, to Jocelyn's point, what we're really looking for on the momentum front is 20% month over month growth is best in class for an early stage open source project. A lot of people will tell you 10%. 10% doesn't usually move the needle because the base is too low when companies are that small. If you have hundreds of thousands of users, sure, 10% month over month growth is super impressive. But if you're an early stage open source project, you really want to look for 20%. The other key things that we look for is who is using it and what are they using it for? Like how important is it to them? Because just usage, if it's not by people who are going to bring you into a core product in an organization that might eventually pay you for it, doesn't have to be the person that's adopting it. But it does have to be into organizations that are going to use your product in some sort of production use case because we want to see the path to monetization. Gone are the days of 15 years ago and the early days of open source when you could just get users and people would quote, quote, figure it out later. We want to understand the business model, what's going to be open source and what's not, how big the TAM is, and also whether or not people will pay. So we'll go spelunking through your GitHub repo and try to figure out who's contributing, who's using it, what are they doing it for, would they be upset if we took this away from them? We're looking for those signs of actual market potential of what you're building. So I mean, we back companies when it has zero stars sometimes. So it's I guess the question to what we're looking for is going to be much more around end user pain, why are you open sourcing? So there's actually, I'm not going to throw shade on, but there's a calendar app that has open sourced and apparently they're doing fairly well. But that is not how I kind of think about open source. I think about open source is something where it helps developers to look under the hood to see what's going on, to read the documentation, to be able to understand how it could benefit their pain points that they're trying to solve. And then that makes sense to open source because then you're allowing them to test it to iterate alongside you and then start to build for more of their pain points as that goes on. So one I would say is like, don't just open source for the sake of open sourcing, do it because the end user really benefits from peeking under the car hood and being able to see what's going on there. So that's one aspect to it. The other thing I would just say more broadly is we're talking tactical things here, but one of the most exciting things right now is just directionally making bets on what's going to be a big thing in the future. So one thing is, you heard from all of us here, we're very excited about open source. So directionally that is a thing that if you can find something that aligns to that and benefits from it, then that is something that could be quite exciting. Another one is like, I'll give a good example, like a Pinecone where I'm not an investor, maybe somebody else is, but Pinecone is a vector database, right? And they're all the rage right now because everyone's using embeddings and saying, oh, we need a vector database and things like that. Pinecone started in 2019. So that means like, the term of LLMs wasn't even around then, like how what people were using wasn't around, seeing the future there is why we get so excited back with founders, right? Because we're taking, we're building ahead of where the curve will be. So people will start to see it in the future. And that's really what I would say to everyone here is to kind of think about what are those really big directional bets that you believe? One that I hold in belief is like, eventually we'll have a bunch of agents running different things. So if that's the case, now start to think about, okay, well, how can I build for that future two years out? So maybe with a bunch of agents, right, you have the same problems as, as API's proliferate. So you're gonna have API security issues, you're gonna have how do you switch context between those agents? How are you going to, maybe that's a deployment problem, right? How are you going to deploy and correct them as they're going and doing this, hence guardrails, right? And so these are the sort of things that I think like, if you can make these directional bets, whether it's open source, whether it's proprietary, whatever the fuck it is, right? Like it's just like, let's just get something that we're very excited about in terms of the future of where it's going, and show us the vision that you want to build for two years out. And if we're wrong, maybe we're wrong, right? But at least we'll have done something very impactful together. If we're right, it will be a freaking massive company that we'll have been able to build because we'll been building longer than anybody else. Wonderful. Awesome. With that, I think we have maybe 15 minutes remaining. We'll open up the floor for questions. So I'd love to dive on this one. So first of all, I think VCs are, VCs who are focused on AI, not journalists, absolutely are attentive. And I actually think you don't need to look far to find that at least in the case of life sciences, quite a lot of big tech and LLM players are paying a lot of attention to the impact on bio too. You've seen Google DeepMind release their really seminal work on protein folding. Like it's very, very clear that this is the future of all the natural sciences of bio of chemistry of physics of material science and God willing climate and sustainability, right? Everything related to that. And that this technology is a really, really both classic machine learning technologies, by the way, let's not sleep on the, you know, the great linear regression, which still can do a lot of good work and in all kinds of things. But, you know, on everything from weather prediction to protein folding, scientists equipped with these models can do 10 fold better work. You know, we've already seen work that's been hailed as that would be Nobel Prize winning work if it were done by a human being. So I think that for anyone who's paying attention to AI, that name's absolutely out there. Now VCs need to look at the intersection of two things. One is, is there a really good fit between a problem and the technology? Can the technology solve the problem? The other is, is your market window open in a timeframe where you can make enough revenue to raise the next round of capital to become a valuable company in the lifetime of my fund? And so what might want to send you to the NSF instead of to venture capital is if this is still a science project, is if it's not in the sense of I don't know if it works, maybe in the sense of I don't know if it works. VCs will back things that have some invention risk sometimes. But we don't like to back things where there's no visible market. And so I will say that as excited as I am about the potential of AI across all of these natural sciences, the one place where we're investing in doing so heavily, we've made three or four investments in our current fund is at the intersection of life sciences because the pharma market window is open. Pharma's been working with machine learning and data science for quite some time. They have already been using it. They are ready to buy software and models that help them with drug development. And we are backers of a company called NABLA Bio. You can find that came out of the church lab at Harvard. These are basically the inventors of the protein language model. So exactly your point that transformers, just as they are capable of predicting the next English word in a sentence, they are capable of predicting the next antibody in a sequence. And so now you're talking about the product insight, about what is the product based on. So we know this technology can model the real, can model nature in some interesting way. Although you may need to create data to do that. One of the things that's exciting about NABLA is the data to build the protein language model does not exist on the internet. You can't scrape it off the internet. No, nobody else has. It's truly proprietary in the sense of they're creating it in their own wet lab. So you may need to come up with the data set that allows you to model nature for the end consumer. But you've also got to figure out what is the product and what is the business model. If you want to go to VC, if you want to go to NSF, do your worst, right? Like just to just create it for the sake of creating. But you've got to figure out a product and a business model. If you want venture money, if you want to build a venture-backable company, then product and business model has to be top of mind, not just invention. Just in the interest of time, can we take other questions and then we can come by? I mean, poolside is definitely looking to solve this over time, which is why we put quite a bit of money into that company. I think a couple things. One thing I will say is, you said Shomik, show me something two years. Honestly, with my fund size and the type of returns, I'm looking to invest in companies that could be worth a trillion dollars. And that doesn't take a two-year time horizon, right? That takes to me like a five to seven to 10-year. And they won't be worth that then, but at least they'll be trending in that direction. When I think about poolside and its full potential, the idea is, yeah, sure, now it's like your pair programmer, right? But over time, it becomes a conversation about what you want to build and what you want it to look like. And that could be in any language, not just English. I think the translation stuff is getting very good. But how that is influencing my kind of lens on how I look at other investments is, okay, if I believe that that's the world of the future, if I think me who is not a developer is going to be able to build some incredible software with poolside in five years, which I firmly believe will be the case, then what does that mean for everything below that interaction? What developer tools do we still need? What needs to exist? How do we think about legacy code bases versus just building new applications? Like for how long are we going to maintain some of these things? And so whenever I'm evaluating investments, I'm really looking through that lens and through that time horizon. And I think, you know, in some cases, perhaps the size of the fund in the stage at which they're investing might influence the lens of how they think about those things depending on the outcomes they're looking for in the time horizon of those outcomes. Yeah, and maybe to add to that, I think there is a new wave of startups and founders, which are one, two developers, some of them are citizen developers, some of them are not actual developers in the standard term of a developer who can build apps like, hey, upload a picture and you will get a passport picture, passport quality picture as the output. And these companies are generating millions of dollars in revenue, right? So there could be a wave of companies that are not venture fundable in the standard venture fundable sense, but could still be profitable businesses that people can run for decades. So that's another wave that could happen in the next decade or so is what we are seeing. Yeah, I've sort of a personal view about this, which comes from working as a software engineer in my whole career before coming into venture, which is I totally believe that LLMs are going to transform the practice of software engineering that what a software engineer is or does in 10 years. Like that software engineering 10 years from now will look like punch cards does to people who write Python, you know, Python will look like punch cards to people in 10 years. I don't quite believe that the software engineer job is going to go away because everyone can be a software engineer. I still think that there will be a role for people who build products, but perhaps the product manager, designer, engineer, perhaps those roles will all converge. Like right now we need three separate people to collaborate because the skill sets are have such a high learning curve each one. Perhaps if the learning curve goes down, you know, we'll have the all in one product builder. I don't know what exactly that will look like. But one thing that I really noticed along the way was software engineers have kind of a unique superpower compared to people in most other job functions compared to lawyers or recruiters or accountants, which is that if we don't like the tools we're working with, we can stop and rebuild our tools. Like I can't tell you how many times I've learned this lesson over my years, but like the first time was when I was like, you know, a bright-eyed, you know, teenage summer intern at Netscape. And like we had grown exponentially between 95 and 96 and the build was broken every day. And half the product team stopped working on the browser and stopped and made build tools for like two months. And then we got back to working on the product once the build was green every day. And we had continuous testing and like a bunch of great open source tools actually came out of that. And so it's always kind of like hit me that like most people in most job categories don't have that power. Like it's software engineers in blacksmiths that's kind of it who can rebuild their tools if they don't like their tools. And so what's really inspiring to me is the idea that someday people in every job category are going to have that superpower. And it will mean that what a product category is is going to look really different when people are capable of equipping themselves with the tools they need. But I still think great that like it's just going to be a beautiful future where people will be able to be a lot more ambitious about what a product looks like. Maybe one last question. Yes, please. In terms of do we look at? Yeah, I mean, we certainly, we certainly do look at various legal legal things going on. I mean, you know, crypto is an interesting one just because I'll be honest, like two of our fastest growing companies right now are crypto security and crypto infrastructure. And so it's like I think just in terms of markets, they evolve very quickly, right? And things change. And so like legal aspects, I mean, there's always people like Uber has been in the news more than anybody for kind of breaking down barriers in terms of some some legal things, right? I think the question is obviously like, what can you not do because that will injure people and maybe things that you can't. But like, if it's not legally allowed, like we're probably not going to back it is what I would say. I was just going to say, so I think the types of problems within crypto and even like crypto is a big bucket, there's like crypto infrastructure on the one hand. And on the other hand, there were companies taking other people's money and spending it as their own, right? So I think that level of disparity is probably unlikely to happen within AI. Although AI has its own issues of copyright violations or creating content that's probably legal in certain countries and so on and so forth. But if you were to look at the Venn diagram of things within the AI sphere and try to identify things that could violate certain regulations, I think it's a much smaller subset than it was in the crypto case. So having said that, I think at least I can speak for myself and our fund like we definitely continue to do due diligence. Like even prior to Gen AI and even during the Gen AI boom, the quality of due diligence hasn't changed quite substantially. Sorry, Erica. Yeah, I mean, I was just going to say, I think, you know, when we released co-pilot at GitHub and trained it on GitHub code, right? Like we were looking at our terms of service and we were looking at a lot of existing case and copyright law and things, but a lot of the regulation didn't exist. So what I would say in a lot of companies have been very successful being created in environments where they were pushing the boundaries of regulation. I think Uber and Lyft are famous examples of that, right? And so like I do think there's a difference between like being irresponsible and how you approach due diligence and being willing to take a bet on something where the regulation is unclear and rapidly evolving. And I'll tell you, I just give you a specific example, looked at it, an investment the other day where this company was using voice diagnostics powered by AI to ascertain whether or not somebody might be impaired for physical jobs, because a lot of workers' claims come out of impairment, right? Either lack of sleep or drug or alcohol intoxication. And my very first question to the person who sent it to me is I don't know the space very well. It's not a space I've invested in. I was like, well, is this legal? Like what are the what are the rules around that? And so I mean, yes, I think any responsible VC tries to ask these questions and dig into it. But also sometimes there are opportunities with regulatory arbitrage that make a lot of sense to invest behind as well. By the way, just one thing I'd like to add is so in some of these areas like medicine or medical field or something, we may not know what is allowed or not, but we can say that for patient of care standards, this is probably something good. So let's just go figure out the legal stuff as we continue to do this. But that also is a good example of again, if you're looking towards the future, especially as these models get multimodal in terms of they can ingest images and text and things like that, start to think about the first wave of radiology, ML for radiology or ML for ultrasounds or stuff like that, they came, they went, some of them exited to GE or others. But now it's actually like, we have the tooling to do that a lot better than what we had before. The first company that Iangel invested back in was in like 2016 or something or 2015 that was doing ML for ultrasounds and echocardiograms. And now that same business would be a whole heck of a lot, it's still not easy, but a lot easier in terms of building that technology to do that. So that's really where it's like, the main thing I think we can leave you with is just like, again, there's no one business model that's the right way, there's no one approach that's the right way or anything like that. It's actually really just trying to see the future and then work backwards from that into steps that you can de-risk. And as you de-risk those along the way, everybody here will be able to fund that in different ways. But if you're trying to do something that is kind of happening right now, well, then that's usually where the most others are there. And then it's really hard for us to understand differentiation. So you could be the smartest person in the world and we could love to work with you, but just maybe really hard for us to go and do that because there's just so many other players. And so that's where I would just really say, like, think about those things, think about the technology infrastructure blocks that are there, what they enable, and then start to go build that future. And we will get super excited when you come and show us that because that's what we get jazzed every day to get up out of bed and do. Thank you. I think we are over time with that. Hopefully this was a helpful session. Thanks. Thanks everyone for coming. Thank you all. Thank you.