Special Season: Inside Summit 2024 Next Gen GenAI Panel ft. Tavus, Multimodal, Dynamo AI, Canapi Ventures, and GenAI Collective

Hi everyone, so excited to sit down today to get started on our INSIDE Summit miniseries.

We're so excited to have had our first ever all-day conference event just last week in New York City with over 120 innovators and founders in In the room where it happens. If you want to hear more about our vision behind Inside Summit and some of our learnings of hosting our largest scaled event yet, listen to our opening episode of this mini-series. But what you can expect next is a recap of the conversations that we had in the room from the actual guests themselves. You can expect our opening fireside with KJ Sidbury, the partner over consumer investments at GV, our next-gen GenAI panel, which was curated by the GenAI Collective, our Future of Platforms panel, which was moderated by the associate editor at Forbes, and two more. Claudia, you want to fill our listeners in?

Yeah, absolutely. The future of direct-to-consumer with some incredible brand founders, operators, and agency partners. And then we ended the day with a recap episode, um, our fireside chat with Kashish Gupta, co-CEO of Hightouch, a previous podcast guest. So without further ado, Let's open the door.

How many people have thought about AI in the last week? Okay, cool. How many people actually think AI has an impact in their day-to-day lives or business they're running? Okay, okay, this is good, this is good. Just pulse checking a little bit about interest. Um, let me first introduce our esteemed panelists here. To my immediate left, we have Lulu Chang, principal at Canopy Ventures, um, and we have Ankur Patel, founder CEO of Multimodal. We have Vaik Mukundan, founder and CEO of Dynamo AI, and Hassan Raza, founder and CEO of Tavis. Uh, I'm gonna get— let them one by one give a little bit of an intro on their career path to where they are today and their founding journey as to what their company actually does, because they will pitch it better than I can. So I'll start here at my immediate left. Lulu, if you could just share a little bit about your path to Canopy and what you're focused on.

Sure. Um, so great to be here. Um, thank you all so much for having me. Insight Summit, it's an awesome event. Um, so I joined Canopy Ventures 3 years ago. We are, largely speaking, a B2B fintech fund. I would say that we define fintech less as financial technology and more as technology that sells into financial services. Reason for that being that our LP base, in addition to your traditional institutions and family offices, also comprises about 70 banks across the US. So the whole idea is that there's a dual value proposition, dual mission to the banks. Hopefully we can make you some amount of money, but also, by the way, here's this innovative roster of companies that can improve your AI capabilities, your fraud stack, your cyber posture, what have you. Um, and then on the other hand, to our portfolio companies and prospective companies, the mirror image, right? So here's strategic capital. But oh, by the way, if you want to sell to, sell through, partner with a financial institution, we can help you do that. So I've been there for 3 years. My path there was a little bit circuitous. I started my career in comms and in marketing. I was at Squarespace where I was largely responsible. I was a celebrity handler. Honestly, that was my job. So we worked on Super Bowl campaigns, on Grammy campaigns. And then I went to Blockchain.com where I also was in PR, and that was saying no to every journalist ever about any comment about blockchain or about crypto. And then after business school, I went to McKinsey for a couple of years and then landed at Canopy.

Hey everyone, I appreciate you all being here for this. So yeah, my background stems mostly from financial services at places like JPMorgan, Bridgewater, I ran a macro hedge fund for a bit, sold a startup to Vista Equity Partners. So my journey has been kind of much more in the financial services space and then, and then much more in the machine learning space last 10 years or so. But the founding story for Multimodal, and we automate certain types of workflows in banking and insurance, actually stemmed from my last startup, Glean AI, where we were doing basically smart vendor spend intelligence off of invoices. So we took data out of invoices. We had time series analysis and benchmarking. And what we realized was that that technology, large language models, right, the thing that powers ChatGPT, is incredible at automating the work that people do around documents and making decisions. And we are building multimodal with that frame in mind, right? Take away the administrative work away from people that they don't need to do, empowering them to be much more productive than they've ever been.

Hi all, I'm Vaik. I'm the CEO of Dynamo. Like one good news I wanna share is like we just hired our 50th employee today. And so as a company, what we focus on is we help enterprises, financial institutions identify the vulnerabilities like of AI, like where do like AI models like hallucinate or like what are like the privacy and security risks and how can these enterprises now deploy in a safe and responsible manner. My founding story was I did my PhD at MIT specifically around how do you train AI in a privacy-preserving way, and my thesis ended up becoming Dynamo.

Awesome. Yeah, great to be here. I'm Hassan. I'm the co-founder and CEO at Tavis. And Tavis, we're building foundational models to create human digital twins, right? So we see a future where there's an extension of you that can exist, that can in video form, in Zoom form, and we're building the models to do that. And so my founding story is one of like a mixture of startups and tech. So before this, I was at Google. I also ran a product studio for a while. And the idea of Tavis actually came from helping a number of startups, you know, create their first sales funnels. And they would go and they would create these personalized videos that would work really well. But they were really difficult to actually produce. 200 videos took them 13 hours to record and send. And so naturally there was an opportunity to automate that entire stack and we used AI to do it. And that's, that's our background.

Awesome. Awesome. So thanks, guys. I know the audience in the audience is— everyone's curious about AI. We obviously have a ton of founders here with AI experience and repeat founders. But before we get to AI as a subject, I know we have a lot to cover there. I also know we have a lot of builders and operators and investors and potentially future founders in the audience. So I wanted to hear a little bit maybe from each of you about how you approach the fundraising process, 'cause I know people are thinking about that. And maybe we'll start backwards. So Hassan, move over here, which is more about how you approach fundraising for your company, for an AI company versus a non-AI company, and what that looks like in this current market and maybe at the time that you're fundraising.

Yeah, so we're about 4 years old. So funny enough, whenever we were first doing our seed, AI was a thing, but it wasn't the thing, right? So it was, back then it was having to convince people that there was a future here. And it definitely was more so investors that took a leap of faith in us and saying, hey, like, you know, this AI thing, like it's gonna be real. No, no, no, you don't have to do crypto, you can do AI, right? And versus now it's very different where like everyone, all investors are like all in on AI. AI. Either way, you have to navigate noise now, right? Where there's so much focus on AI, where everyone's doing AI, you have to navigate that noise versus 4 years ago where there was less people to believe in you that this is going to be a big thing. It's been a little bit different in that, but building great products get you great investors either way.

For me, I was still doing my PhD when we raised our seed round 2 years back. So we raised around $20 million in seed and Series A now. But I would credit a lot of that to San Francisco. I think it is the best place to— if you're a tech startup founder, it's the best place where you can learn from other people. You have access to a lot of investors. And then we decided to move to New York where Canopy invested in our A round. After hitting like product market fit. So kind of like after my PhD at MIT, moved to San Francisco, was there for 2 years, and then came to New York. We also did like Y Combinator 2 years back, so that also helped with the process.

Shots fired at New York VCs and founders. Lou, you'll have a chance to respond. Lou, go ahead.

Yeah, so with Multimodal, we actually haven't raised yet, and I think we're building this a little bit different than I've done in the past. So just, just a little bit background. This is my third startup. AI has been around for a long time. I think people forget that. I think they look at AI as being this net new thing. But, you know, we were using machine learning 10 years ago at my first startup, had zero premium attached to it because no one really cared. It wasn't hyped. I, you know, I joined startups that were all in machine learning space past 10 years, and none of them really got a premium. I think that's only really happened in the last 2 years. I think venture capital is a really great accelerator, allows you to accelerate a motion that's emerging and really put more fuel on the fire. You know, I think you could also raise too much too soon, and then that could insulate you from having really good capital efficiency. And I think right now, if you look at public markets and the way that they're looking at companies, capital efficiency is really important. Being able to have a really strong foundation and then raising venture capital on top of that to scale that motion, I think is pivotal. And that's the way we're building it. I do think though, and this is where it gets tricky, is that AI is seeing a little bit of a frenzy in the same way that the dot-com companies had in the internet age. So I don't know how that plays out, but it's a great time to build. It's a great time to invest. Yes, some companies will get it completely wrong. Some investors will get it completely wrong, but some really great businesses are going to be built in this time.

And Lulu, I'd love to hear a little bit about your thoughts on fundraising from the venture standpoint for AI versus non-AI, but also would love to hear your rebuttal against Vik. And as a New York VC, what is the landscape like here versus—

Technically a DC VC, believe it or not. East Coast? Yes, an East Coast VC. I agree up to a point, right? I think that San Francisco definitely is a hotbed of innovation, but there are hotbeds of innovation everywhere, and our fair city is certainly one of them. So to each their own. In terms of, you know, the AI frenzy, as Ankur called it, I think that from the investor seat, that's really exactly what it is, right? It's very, very easy to slap.ai at the end of a URL and say, now we are an AI company. And you see this intersection of AI and climate and AI and cyber and AI and all of these other things. And you just don't need AI in a lot of these, in a lot of these business models, right? I mean, I think that there is, a premium that is given to companies that are truly changing the way in which— it cannot simply be a process workflow change, right? It has to be a fundamental shift in the way that we think about things and the way that things are actually done. But making things just marginally more effective or marginally more efficient, that's been around for a really long time. And to Ankur's point, I think venture, as all things, is highly, highly cyclical. You know, you mentioned dot-com. This to me feels a lot more like crypto a few years ago. And so, you know, I think that there's a lot of— there's an equal mix of real excitement and a recognition that the next billion, $10 billion, $50 billion company is certainly being built right now in the generative AI space. But similarly, that road there will be littered with the carcasses of many AI-adjacent, sort of AI in scale air quotes, companies that simply won't make it because it wasn't a real AI company to begin with. And so I think our job as investors is to pick out the wheat from the chaff and make sure that we are doing the due diligence that we can as non-AI experts, largely speaking, and differentiate.

Yeah, no, it's fascinating. I think that diligence that you're speaking to is what a lot of companies are doing as well in terms of deciding whether or not to adopt AI solutions in their stack. And that's where I think for each of you guys as founders, all of your companies have sort of an enterprise angle to it. And I would love to learn a little bit about how you've approached that enterprise sales cycle, what sort of resistance you're seeing from large enterprises when you pitch your solutions to them, and what that sales motion even looks like. Because we do have a lot of founders who are also focused on enterprise. Maybe we'll start with Ankur and we'll kind of popcorn around.

Yeah. So I think the typical mantra you have for startups is you want to move fast and break things. And that's the exact opposite of what you see in enterprise, right? Which is they have a lot of brand reputation at stake. They have a lot of great business at stake. The last thing that they want is to introduce GenAI too quickly. It's immature and it leads to brand reputation hit, right? Or hits like operationally and then they're not liable, especially if you sell into banks and insurance. And so you would need to have really high customer empathy. And understand what it's like to be in their shoes. They'd rather move slowly, adopt once they have enough confidence from a legal regulatory compliance standpoint. And then there's, of course, the employees themselves that need to understand how to adopt AI, right? So there are multiple layers that need to kind of move in that direction. That tanker takes time to move. And so meeting them where they are, introducing it in a way that's a little bit more methodical, that's really important. And that's what we see is these cycles are long. It's hard to get in the door, but if you're a good partner to them, you will break in, right? With the right product. It just takes a lot more time and you cannot, certainly not, break things as you go, right? You're going to be completely blown up otherwise and you put your customers at risk.

Yeah, I think like for the first year we didn't even sell. Like we were just like making sure like a lot of like our research was like we could actually make it as like a proper robust product. Which becomes much easier for enterprises to adopt. One thing which I've learned is enterprises don't want to spend time telling you, "Okay, make this change, make that change." They really don't have the time. They just want something that's an accelerant for their business and that's of immediate value for them. So one thing which I've also learned is once you've got one or two big customers, the process just becomes super fast. So the moment we got Experian and Lenovo, getting like a JP Morgan CD, BlackRock was super faster. So I think like, like good words from the, from a large customer helps with the next customer.

Yeah, yeah, I definitely agree. Social proof there can, can make a big difference. I also think it's, it's important to meet them, uh, right at where their pain points are, right? So if they're afraid of hallucinations, which is what all these enterprises are afraid of, it's like, hey, they deploy this to customers, employees internally, um, build the tools that help them, you know, get over those fears, right? So for us, that means like, hey, like, we have moderation that's automated, we have consent that's automated. Like, actually be forward about the potential risk and say, hey, this is how we're tackling them. The other part is like actually filtering for the right enterprise customers at the right times. There are enterprise customers where there's directive from the top that, hey, AI is core part of our offering in the future. And those are a very different type of customer than those that like AI is sort of a nice to have. And in identifying early on where they fit can help a small startup, you know, a Series A seed stage startup, really figure out where to put their energy and where not, right? If this large company is going to take 2 years because everything takes them 2 years, well, maybe less so the focus, right? But if there's frontier C-suite or the people who really want to make a big impact with AI within their company, that might be where you want to focus instead. And there's certain industries where that's more true or less true as well.

Yeah. And Lulu, I'd also love to hear how you're advising your portfolio companies in terms of that enterprise sales cycle. And then we can maybe even unpack, because you guys had talked about some customers being more resistant than others, which verticals within enterprise are you seeing more adoption? What do you anticipate even in the future? Where is there resistance? Where has GenAI even been effective or not kind of disappointing for some of these enterprises. So maybe we'll start with Lulu and go around.

Sure. So as I mentioned at the top, you know, our LPs are banks. So you want to talk about long sales cycles. We're sort of exhibit A for that. And our banks range from some of the largest banks. So it's JPMorgan, it's US Bank, Ally, Truist, all the way down to your smaller regional and superregional banks. And so they're all different flavors of enterprise sales cycles. And I think one of the keys really that we have found is selling the customers what they want, maybe even more than what they think they need. I think that sometimes as founders, we find that a lot of our companies are very opinionated in the type of solution that they actually want to give to the bank. But the bank has been very, very clear in their RFP, or in their request for anything, that this is actually the solution that we think we want. And you could be right all day long that that's in fact not what you want, you should want this instead. But I think that is a losing battle and you lose a lot of time trying to convince your buyer to come to your side instead of actually just meeting your buyer where they really are. We have at Canopy what we call a vendor management playbook that really clearly illustrates that the hardest thing about selling to a bank is that you think that you're there and then it's actually another 6 months and another 6 months and another 6 months. And so really knowing where in the lifecycle of the sales journey you actually are and not allowing yourself to get really discouraged discouraged and think, oh, yet another set of compliance rules and regulations that I have to get through. That's really key. And, you know, banks are one of the most regulated industries in the world, but that's the same with healthcare. It's the same with any large industry, right? So that's what we'd say.

Awesome. Thanks. And then Ankur, maybe we'll go along as well, but just which verticals are you seeing kind of more adoption? What areas has maybe AI fallen short as well?

Would love to hear a little about that. Yeah. So, I mean, multimodal, we're selling mostly into regular industries, primarily banking and insurance. So the way I look at it is which areas of, for example, insurance are more likely to adopt some GenAI technology versus not. So you have to pick your lanes. I'll just give you some insight. So underwriting is core to what an insurance company does. It's, it's so near and dear. There's a lot of legal regulatory compliance scrutiny there. Even though carriers would benefit from GenAI, even though they're eager to adopt it, they are much more apprehensive about moving on that quickly. So it's a terrible lane for you to bet on in the first year or two of a relationship with a large carrier, with any major insurance player. Whereas if you look at certain other workflows within insurance, like I'll give you an example, P&C claims processing, so auto claims, workers' comp, There are paradigms to go and automate some of that using GenAI or any sort of prior-gen AI. There are things like express claims. So you have an auto claim, it's a de minimis amount, let's say under $5,000. If you have enough supporting evidence, you can go and automate that. And so there is a framework, right? There's less risk for the insurance provider. And so you can go and introduce your software there. So picking the right lanes, getting going there and then moving to what might be a bigger opportunity later down the line, that's really critical. It's sequencing the way you open up that customer to bigger opportunities in due time.

Oh, it's me.

So specifically, like, what do you, in addition, like, want to know that would be nice to add some new insights?

Yeah, so I guess with Dynamo specifically, I know you guys are looking at a bunch of different verticals. I know you're looking at financial services, you're going to healthcare for different, enterprise verticals. Which customers have been sort of the most resistant to adopt and which customers do you think will maybe will adopt and then later see the value? Maybe which ones haven't already? And then how have the conversations there been?

Yeah, so we primarily sell to like a lot of like the Fortune 500 companies. Yeah, finance is like kind of like our primary vertical. One thing which I've learned is like healthcare is like a different beast of itself. For example, I don't think I would ever want to do that in the near future. Unless like someone already exactly knows the playbook for doing it, right? But in general, I think like Fortune 500 companies are like moving like much faster than like how we saw them like move last year because they know that like every board meeting people are like talking about GenAI. How do you have some sort of like application that's going to improve productivity or like how do you have like a chatbot? So I think It's actually a good time to do a startup because of the fact that enterprises have to move fast and anything which you can do to empower that would make a difference right now.

Yeah, I'd agree actually. I think if we rewind to 3 years ago, convincing enterprises was actually harder. Now there's these directives from the top that, hey, we need to be an AI-first company. So it has become easier in that respect. I also agree that I think that even for these large companies, there's actually multiple customers within them. And figuring out which internal customer is best plays to your strengths. For example, even for us, whenever we're working with a healthcare company or financial services, OK, they might be afraid to deploy this to customer-facing products. But internal products are still somewhere where they're more open to try new things. So getting your entry there and then spreading from there and getting that trust built up to say, OK, this worked It worked really well for an internal use case. Now this, this, let's deploy it to customer use cases. A great way to get in. Um, the other thing I'd say is finding a champion, whether it's C-suite, whether it's someone else in the org that really believes, uh, in the future, or like a future that you build with them, can go a long way. So finding that C-suite, finding that PM, finding that VP that really, really loves, uh, what you're building or sees a future that you can help facilitate And really working with them to sell it internally can be a massive win, especially for a startup.

Yeah, those are some great points. And I think this is kind of a nice transition to where we're going next with this, which is I would love to hear everyone's thoughts on AI broadly. And I think this is where we can start to put on our like Nostradamus caps here and think about which areas of AI become commoditized, maybe from the enterprise angle, which areas from an enterprise perspective will, uh, end up becoming sort of more beneficial for them, whether it be open source versus closed source. What are you guys seeing in the landscape of AI moving forward? And, um, maybe what areas do you think scare you? What areas do you think maybe improves your business as well? So we, we'd start with a little bit maybe with Hassan, what you're talking about in terms of employee productivity improvements as a solution versus is it actual— like, is that how enterprises are measuring their their value creation, slash, where does this go when these models improve and improve? And what do you think about that?

Yeah, so I definitely think that, like, the issue with a lot of AI development today is that it's sort of general, right? And I think that as we see, like, enterprises adopt it, they're adopting solutions that work for them. So the application layer actually ends up being super important. I know that Historically, you know, especially in the past couple years, like, oh yeah, yeah, application layer companies, you know, why should they exist? But actually, like, the reality is, is that healthcare company or a financial service company, they're more likely to buy the solution that's tailored to them, that answers their specific needs around, you know, hallucination prevention, all these elements. And so I do think that there's a lot of value in domain-specific, niche-specific application layer companies. Companies, um, that really target the key pain points that these companies are facing. And I say this as like a foundational model company. We're not necessarily focused on a specific industry, um, but we, what we do see a future where we're powering the application layer companies that will build these niche applications, um, or the applications are specific to each industry, and that's where the value will be created. Um, and I think the open source versus closed source debate. I think that there will be plenty of areas to improve and really own both in closed source and open source. So I don't really have a— I mean, yeah.

Yeah, I guess I have to ask, as a foundation model company, it's sort of the elephant in the room. Are you worried about foundational models sort of becoming commoditized, or what do you think about that position?

Yeah, so I think that naturally all tech, like all core tech to some degree becomes commoditized. But there's always that, like, for example, even now we see that, like, for at least for what we do, open source is at least 1, 2 years behind what closed source reveals. And I think that will— that lead will always remain. But it's also just important to focus on where the value in model improvement is. Like for us, like, yeah, like we'd be happy if there is like a core part of our pipeline for foundational model that for like for our video models that is open source is better. It's like, great, now we don't have to focus on that. Now we can focus on improving Y over here where there is actually more value to be had. So we think that the progress in open source actually also accelerates the progress in closed source because you can just focus your energy on the really hard problems that no one's solving yet. And then also, I think there will always be domain-specific specific needs that require a different angle in research. It may be that, for example, for our pipeline, you may have great face track that's now open source, but it's like, okay, great. You still have to do emotional context and all these different things that require a different angle. Either way, I think that we'll all win.

Yeah, no, that makes a ton of sense. We'll work our way around here. So, Vaik, in terms of future of enterprise adoption, mass adoption of AI, what do you think needs to happen fundamentally at the AI level?

I definitely think it's like starting to become more of a data play right now. Like, yeah, there are like a lot of like good models, right? Like open source, closed source. But I really think like the value now which people are starting to realize, it's like in the actual data they have. Why is JP Morgan special? Because they have like trillions of like petabytes of like data over the years. And that is like helping them like, say for example, build their own like internal like financial GPT model, for example. So I definitely think data Then the other challenge which we are seeing is like you want to like be 100% sure on like any external-facing application if something is like fully deployed in a responsible manner. Meaning like you don't want like an analyst to know if these two companies undergo a merger, what would the price of my stock be, right? Or you don't want like a way to decipher someone's like social security number through a chatbot. So it's really important. So like when we bring awareness, we just show like how like a lot of different models fail in different contexts and like how can like now enterprises like solve these problems using the second part of our solutions to make sure they don't have to worry about like something where like it could be like a lawsuit or like some sort of like defamation case. So it's really important to understand like how you are using that data. And I think that is what is going to play a big role. We've seen what happened with like LLMs when it comes to like text. And I definitely think the future is going to be audio, video, and like images, like the multimodal aspect of AI.

Makes a ton of sense. We'll just work our way around here.

So yeah, so I think the foundation models even today, if nothing changed, they're really, really good. I think you could build incredible businesses if you build really great application layer companies on top of the existing foundation models. But foundation models, you know, thanks to Hassan and other players, they're going to continue to get better. I do think that, you know, if you think about how do you get really great AI companies out there, some of the incumbents, the ones that have great distribution, are going to be the major winners because for them AI makes their product suite much, much better. Right. And it's going to be very hard to compete, outcompete them. Like a Microsoft adopting AI capabilities in their entire suite for productivity, for general purpose business applications, they win that. I don't see how you break into that. It's going to be very, very hard. I do think that for very more like niche problems, niche areas where the bar is, is much higher, you need specialized applications that are built like, let's say, in healthcare or banking where the bar for good is very high. You can't have it hallucinate because you have financial lives at risk. You have legal and compliance issues. I think building a deep verticalized application layer for that audience is going to be really important. And I think that's where the big opportunity is, building application layer companies built on top of the foundation models that, you know, basically a couple of handful of players are releasing. That's where a lot of the value is. Now, that is, that is the future that we collectively can build and partner with. I think there are only going to be a handful of companies that really succeed at the foundation model layer because ultimately it's a half a billion to a billion dollar, even some sometimes per year game. It's only a handful of players that can compete in that sandbox. But the application layer, I think that's where a lot of innovation can happen. And frankly, all of us have access to that, right? Whether, whether it's open source or closed source. Now it's about dreaming of what type of application layer company do you want to build on top? I often describe Perplexity as being one of the best consumer application layer companies out there. Because they don't have their own models largely. They're using OpenAI models, but they've built an incredible search experience, basically outmaneuvering a player like Google and OpenAI. So, you know, turn to examples like that as a good playbook for how to build a really great application layer company leveraging foundation models that are present to all of us today.

Makes a ton of sense. And Lulu, from an AI perspective broadly, this is where as an investor you can kind of look long-term. What do you think needs to happen in terms of mass adoption here?

Yeah, so, you know, one of the things that we're excited about in order to get to the application layer, in order to get to the foundational layer, you know, we're looking at a lot of the data infrastructure to, to Vaik's point that really undergirds a lot of this. So if you think about the way that Splunk came to be or Snowflake came to be, right? This was just a vast explosion of data and this is happening all over again. It's a different kind of data, right? So now you're looking at vector databases, you're looking at RAG pipelines. And so we've been really excited by the idea of this new, really base layer, right? The infrastructure that powers a lot of this and allows for banks, for other regulated industries to ensure that they are sharing their data in a compliant, private way. Right. And you can do that through federated learning, or you can also do that by ensuring that actually the source material itself is kept where it needs to be. So, you know, I think getting that dealt with is going to be really important. And I think that that ultimately allows for AI to do what I hope it will do, which is really grant a lot of access in a way that it hasn't, that we just haven't seen democratized in a long time. So obvious example, right? If you have really good AI underwriting that can actually eliminate a lot of bias, it can actually spit out really fair credit decisions for a vast swath of the population that previously has been very underserved, we should probably rely on models to do that better than a human underwriter. And I think that that's where things get really exciting.

Yeah, that's awesome. And keeping it on the future theme, I think before we go into audience Q&A, I wanna have a little bit of fun with this last question. This is off the script, guys, but the panel is called Next Gen, Gen AI. And I would love to just know from an investor perspective, from the founder's perspective, if you weren't building the company you were right now, what would you be building? And I think, 'cause in the audience we have a lot of founders and operators and future founders, may consumer focus too, not just enterprise. We can kind of like pull out all, you know, remove all the filters. I'm just curious what folks would be, would be doing. And you might need a minute to think about it. So also, audience, please do think about some questions too, because we do have Q&A coming up. But maybe Lulu, what would you invest in or what you yourself would maybe build in for if you weren't doing what you're doing right now?

What would I invest in if I weren't investing in Gen AI? What else would there be to invest in?

Or what maybe what company would be emerging that you, you that you haven't seen yet that you'd love to just invest in? In Gen AI too.

In Gen AI. I don't know. It's, I feel like all these companies are gonna exist in some way or another.

If you could be the first check-in.

If I could be the first check-in to some Gen AI company that would truly feed me on demand in a way that I wanted to be I'm always hungry, so where my mind goes, like, I'm pretty sure that exists and it's called Uber Eats. And I just am like thinking about that right now. I don't know, like, I think that there are, again, like interesting consumer applications of Gen AI that are not, that we just haven't even dreamed about yet. I'm not creative enough to answer this.

You know, I think for me, having been a public markets investor in the early part of my career, I still feel like investing is one of the things that's so essential, investing well for your long-term financial stability and like that financial life that you wanna live.

Live.

I don't think it's a solved problem, as crazy as it sounds. There are a lot of like wealth management apps, some that are gamified to the point where they're actually disrupting you like in your financial lives. Um, I think that really great financial advisors are worth, worth every, every dollar you pay them, but there are also a lot of financial advisors that are, are not that helpful. So I think this is an area where Gen AI, if it takes the basically learnings, the advice, the, the frameworks that the best investment managers and financial advisors have and then make that available broadly to consumers, that's a massive win. Now you have, you know, existing distribution that you have to compete against, but I just think that's a killer application because I think, you know, finances, financial lives are so important. I think we think about health and finance in different ways. I think health people realize is important. Finance, I don't think people pay enough attention to, and I think there's a huge opportunity to build that, build, build there. I think it's a very hard problem, You have to go against existing distribution, like I said. But if I weren't building this, if I had tons of capital and time, that would be a really, really interesting area to go and build today because I think it's going to have massive societal impact if you build that right.

So when I angel invest in companies, it's usually hardcore AI-based tech companies. If not, I'm recently really starting to like fashion tech. A lot of cool products since I moved to New York have been emerging. So fashion tech, if I were to build my own company, then it would definitely be something in sports, if not AI.

Yeah, I mean, if I wasn't doing Tavis, my team says I like pain, so I'd do something even harder. I'd do something even deeper tech. I'd do something that's like a 20-year horizon, maybe like space tech or something. Building AI companies is fun but hard. Maybe we just want to make it really hard. So space tech, probably.

That's awesome. And so I think we can now shift to audience Q&A. And if folks have questions, just feel free to raise your hand, and I'll just call out some. Oh, yes.

Oh, great.

OK, awesome. So anybody?

I'm just curious on the right team size to make a model. Because at Midjourney, I think they had 8 employees at Midjourney, and they're building the best image model. And then OpenAI has— perhaps thousands of employees. Where's that right team size that you have sufficient kind of quality of the team? It's like there seems to be a huge span from 8 to 1,000, the best— in the church of the best text models. Just curious your thoughts on that.

And maybe—

so thought, basically, to reiterate the question, just thoughts on basically the team breakdown on building these models. Like, it sounds like Midjourney has about 8 employees. OpenAI obviously is like a huge team. But even maybe we can expand the question broadly. How have you thought about team size in terms of building your own companies too and hiring for AI?

Yeah, so the first 30, 35 people in my company were like machine learning PhDs or like privacy experts, like hardcore engineers, like ML people from like MIT, Harvard, like CMU. We always wanted to build like a product where we can clearly differentiate from like others. So I was never a fan of like building like a sales team till we hit like clear, clear PMF. And we just like hired our first partnerships person a few months back. So even in the 50, we have like 1 or 2 people now doing this. So I think like a good technology is what we've seen that like clearly differentiates from others and it's actually providing value. So that's been my, that's my take.

Ankur, thoughts on this?

Yeah, I mean, we're about 30 people now and I think The size of the team, I definitely believe in keeping it small. I think that you can move much faster with a smaller team, and 30 is not small necessarily, but I think there is a lot of advantage in keeping it small, keeping it a team that can move on its toes really, really fast. So I'd say the perfect size for an AI startup is right around the 15, 20 mark where you have just enough people to to make an impact. Like, obviously Midjourney is doing great with 8. Like, you know, it's mostly just engineers. And I think like the team size matters less than the actual team itself. Building a team that has the same mission, that same values as you, that is really excited to work is more important than the actual amount of people. I personally like keeping it a little light though.

I mean, I think my take is, you know, today we're about 20 people. I think smaller is better in the early days because speed is everything. So by being small, you reduce the cost of communicating within that team and you could just move super fast. I do think though that depending on the type of business you are in and the vertical you're selling into, you do need to add headcount and you need to add organizational layers as you go, right? So like in an enterprise software selling into a bank, you're going to need a ton of customer success at some point, right? Because every single enterprise sale is custom in a particular way, even if your software can do the majority of it out of the box, they will need that handholding, they will need that support. These are long, lengthy sales cycles. So having different functional areas as you build your organization become really, really important. You cannot operate with an 8-person team selling into Fortune 100 banks, right? You're just never going to make that happen. But prematurely building out a team in ways that you don't need, I think that's really, that's really great, which I think Vaik spoke to. It's like, get the product right and then build out the sales org, the go-to-market org as you go. And then eventually you need great leadership, right? And so thinking about how to get great leaders becomes really important. And people don't realize this, but a lot of the leadership team changes as you scale your company. So your first leadership team may not be the leadership team that you have during later rounds and when you go public. So managing that really well becomes a major challenge. But I would say early on, being lean to reduce the communication costs and moving super fast is everything.

Yeah, and Lulu, from an investor perspective, are you seeing startups emerge with smaller teams because of all the operational leverage from AI?

I mean, it really depends. I agree that it's less about the size of the team and really the quality of the team. I don't think that there's a silver bullet answer.

Makes sense.

We have time for maybe one or two more questions.

Great panel, guys. A lot of chatter about generative AI, whether it's due to augment more or replace more. For the builders on the panel, what do you see your solution augmenting and what do you see it replacing? It could be people, it could be existing stacks or technologies, whatever comes to mind.

Yeah, so interesting for us, like, we see a bit of both, right? I mean, we say that we're building foundational models for building digital humans, right? And so So naturally, you know, the first sort of use case of that is how can you extend your likeness into digital sphere? Like how can you have your personality, your face, your voice go and, you know, perform tasks for you or be your digital self? So naturally that's one thing that we see. But the other side of our product is actually used for bringing a human touch to something that otherwise was going to be automated anyways. Right? So it's like, so it's replacing, you know, workloads that otherwise didn't have enough humans to be able to operate them. Right? So it's like, okay, like maybe customer service that's like super personal and feels very connected was limited by how many humans you could put at it. Well, now you can instead have AI plus a really awesome, you know, model that looks and sounds like a human. Human taking on some of that workload. So for us, it's a little bit of a mix. There'll be augmentation and there will be some replacement. So yeah.

I always feel like AI only improves the productivity. It can never replace a human. I feel like it can definitely make us humans better in our decision-making or just makes a lot of things faster. So say, for example, we have a compliance product for AI that makes sure if a model is safe or not. So this is like really helpful for like legal teams and lawyers to take better decisions, but you still need like a human touch and that's always like making people better.

Yeah, I mean, it's a really interesting question. I think most of what people do at work cannot be automated fully away. Now, are there certain tasks that they do that can largely be automated away? Yes, those individuals are usually very well-trained, they mean a lot to the organization. Their role changes for sure. But generally, you're not replacing an entire person and what they do in entirety. Right. And a lot of the things that, like, for example, my company does multimodal, we're doing part of what a loan officer would do or a claims adjuster would do or an underwriter would do. And by the way, a lot of that is welcomed by these individuals because it's largely administrative in nature, reviewing documents, validating information, making some preliminary decision. They'd rather not be doing that. They'd rather be doing more complex work. So absolutely, the role of these individuals changes, but we're actually enhancing their productivity, right? Allowing them to do more with less time. That does have some implications for the labor market in the sense that you might need fewer loan officers or fewer claims adjusters. So it does have societal impact. But generally today, where AI is, it's actually supporting people to do more work in the same amount of time that they have.

I mean, I really resonate with this meme I saw on Twitter, which is, I want AI to do my laundry and taxes so I can do art and poetry. And right now AI is doing a lot of my art and poetry, and I'm still doing my laundry and taxes. So would love for that to change. Do we have time for another question? Yes. Anyone else?

So this is kind of a corollary to that question, but I'm curious how you guys are using AI tools within your own organizations? Those are as well, like we talked about how AI can replace job functions for some of your potential, or not replace job functions, but augment job functions for some of your customers, but within your own orgs, 30 people, 50 people, 20 people, how does that work out?

I think we use a lot of like Slack, like Notion, like Salesforce, like Einstein's AI. It's just pretty like, Really cool. And a lot of like Copilot definitely makes like the engineers like move way faster.

Yeah, same thing. It's the same set of tools that all of, all of you would normally use. So having tools to help coders, programmers write code more quickly, do review more quickly, search for information more quickly using things like Claude or ChatGPT or Perplexity, auto-drafting emails, enriching data, for go-to-market campaigns and such. So it's enhancing the things that we would normally do, but just, you know, be able to move a little bit faster. I think that we, like the type of work that Multimodal does, it's automating what I would say is like very high volume activity. So it's very bespoke, very like specific to certain things that our customers would do. But generally internally at Multimodal, we're consuming like what I would say is like more mainstream AI tools are broadly available to each of us.

Yeah, I mean, I'd agree. I think we all use very similar, like, you know, all of our engineers are using Copilot to, you know, and ChatGPT, Claude to engineer faster. I do also agree that like it is more so an augmentation. We're not replacing software engineers, we're just making them more productive. So a lot of our AI tooling is things that just make our team members more productive. Right? It's things like Copilot. It's things like, you know, using Perplexity. It's all these tools that just help us do our jobs better, faster. Using Claude to help me write that reply email to investor because I really don't want to. Things like that really just, it helps you unblock and that mental block that might exist. One of your first, like trying to draft out an email, you're like, oh my God, I don't even know how to start. Well, great. Now you can use Claude to start, and then you then review it and like you say, okay, this is great, I can take it from here. And that's sort of how Copilot works for engineers as well. Like, yeah, you can ask it to, to write a piece of code. It probably won't be perfect, but what it did is it unblocked you. And that's really where we're seeing a lot of, a lot of use for us.

Makes sense. In my day job as a product manager, I turn on Microsoft Copilot in Teams so I don't have to pay attention to any meeting and it just transcribes all my notes. Do you use anything in your sec?

We're starting to increasingly. So we're trying to use AI actually to source more. So we're sort of feeding models, hey, this is, We're looking for this stage, having raised less than this amount of capital, based in these geographies. What can you find? There are a growing number of funds that I'm aware of that are really training models on their own investment memos. So as you continue to collate and gather data, then you feed all of that in, and then instead of spending the week of actually writing an IC memo, it just spits one out. And then of course, you're going back through and cleaning it up. But I'm not saying anything new. Automating the tasks that are sort of the worst parts of our job and allowing us to do the more human touch element. So doing things like this and actually meeting founders and things that a machine cannot do, it's what we're trying to free ourselves up to do more of.

Sweet. So huge thank you to our panelists and the Insight Summit for having us. Uh, Prithvi, who was around here, my co-chapter lead with the intros, uh, as he mentioned, we are GenAI Collective. So we have, uh, basically this is all that we do. We essentially our positive-sum community and trying to basically collaborate with more such communities and have the conversation around AI here in New York City. So please talk to one of us if you guys want to co-host events together. We have a research roundtable coming up as well as an executive dinner and a bunch of more events throughout the summer, and we're building out our calendar throughout the year. But thank you all for joining. I'm going to pass it now to Claudia for some announcements, and hopefully we'll see you guys around at lunch.