Adeo Ressi on Mastering AI, Speed, and the Future of Startups

So Adeo, let me start by asking you, could you please give us a quick high-level intro to what the Dassault Group does? And as we delve into this topic, how the company, the company's approach to using AI has evolved since the journey started.

Yeah, hi everyone, my name is Adeo Ressi. I'm CEO of the Dassault Group. We're a a company that basically is modernizing venture capital. Venture capital, as we know, it's been stuck pretty much in the dark ages. It uses, you know, literally technology from the late '80s and early '90s, spreadsheets and email to run, in many cases, billions of dollars of AUM. And it seemed like it was only time that someone tried to come in and modernize the asset class, creating a common platform for venture capitalists to raise money from limited partners and an equally common platform for them to manage that money and deploy it into startups and follow up with the startups, get updates and the like. So we built, unfortunately, that sounds like a pretty obvious idea, but it was a lot further from reality than people realize because a lot of these VCs are really stuck in the dark ages. So we needed to do 3 things, build training tools and then also provide access to capital. The, and we did all that with a foundation of ethics because a lot of the VCs that had built their systems on, you know, antiquated technology really also had bad ethics. They would invest in things that didn't advance the case of humanity in a positive way. And so we said, look, we're going to start with a foundation of ethics where these managers treat founders with respect. They treat all other investors with respect. They look to have outcomes that are good for humanity. And so our ethical core is called the Mensarius Oath. Everyone we work with agrees to the oath, and then we provide the training so that they know how to build and run funds correctly. The tools, the platform for them to operate the funds in a successful way, and capital. We provide them different ways to get money from limited partners to invest in their funds and in turn take that money, invest in startups. So that's what we do.

Nice. Thank you. And when it comes to adopting and implementing AI, I mean, I know that I've been a firsthand witness of kind of how you guys implement AI in the company. So can you tell us more about how you've been trying to adopt this new technology and what are the benefits that you guys get from that?

We were definitely an early adopter in AI for sure. We were using ChatGPT, LLMs, when they were very beta. You know, I have friends who created some of the core concepts behind the transformer architecture and LLMs in general. And so we saw them coming along and we were hearing about stuff that was happening in the lab where, you know, oh, look, we think it's passed the Turing test. And I'm like, oh, that's interesting. But of course, the lab at that time lagged behind what consumers saw by 1 or 2 years. So this was kind of at the end of the pandemic, and we started using LLMs internally to do things like helping with writing, ideating on different concepts very early on. But they were rough, like it would write something that was pretty terrible. So it might take a process that took an hour and get it down to half an hour because it would help you do some checking and maybe get some of your writing improved, but you couldn't really rely on it to write stuff and, and use it. You would either have to edit, have it edit your stuff, or, you know, write stuff that you would have to heavily edit either way. So as an early adopter, We kind of were one of the first people to integrate it with APIs into our platform, one of the first people to make tools on it for others to use. And so I would say that in terms of our adoption right now, we're a market leader of using AI for everything. And we use it internally to make our processes more efficient. We use it. Our main corporate site is now an AI-driven site. You go to the site, there's an LLM, you talk with it about what you need and want, and it answers based on training data that we've accumulated. So it's trained on what we offer, on what, on venture capital. And so our corporate site is an LLM, our marketing is an LLM, and we use it in our products. So managers use our products and get LLMs built in. We have, and, and other AI built in. So for example, on the product side, one of the first areas we used AI is when you're importing or exporting content, there's a lot of fuzzy logic that's needed. So particularly on importing, right? You could get an Excel sheet or CSV, it could label first name, name, ah, all these different things. And a lot of times you need a human to say, oh, this is a first name, this is a last name, this is a telephone number. Well, AI can figure that stuff out pretty easily. It's been capable of doing that for some time. So one of the very first product implementations we had was fuzzy matching of imported data using AI. But now AI does literally everything. You can forward an email to our system about a deal, for example, it will figure out what that deal is, put it in the deal pipeline. It will even rate the deal to see if it's a good fit. It will write a synopsis of the deal, take any files for the deal like investment agreements and, and decks and the like and put them in the right folders. Another thing that we do that's pretty interesting is that we have a plugin for Chrome You can go to any page on the internet and hit the Chrome plugin and it will read the page, figure out the content. And if it's a deal, it will say, hey, do you want to import this as a deal into your deal pipeline? If it's about an LP, it's like, do you want to import this to the LP pipeline? So we're using AI internally. We're using it for marketing. and we're using it for product at scale.

No, that's great. And I, I'm a heavy user of some of those products with the Chrome.

Yeah, go, go. Yeah, some of them, right?

Yeah, yeah, yeah, yeah. The Chrome, the Chrome plugin is amazing. So it definitely saves a lot of time. So it's a great example of, of a, of a product that kind of you put in front of your, or your customers that it definitely, it's a game changer, right?

So let me— Ignacio, can I just— I want to cut to the— let me drive for a second here. Look, if you're thinking of building something in 2025, you have to be thinking AI first. So let me give you an example. We're the market leader in AI. I was working on AI earlier today. I'm working with AI all day long. To do my job. We work with AI in the product, in the marketing, again, in the company, everywhere. Okay? We're— there is someone who's going to disrupt us with AI, 100%, because AI is moving very quickly. And we're also moving very quickly, but we're, you know, we have a lot of legacy stuff that's not as cutting edge as the newest cutting edge AI stuff. So we're at risk of being disrupted. So I actually have an AI tiger team, and their job is to disrupt us from the inside before someone outside disrupts us. And so we're making tools to disrupt ourselves all day long. And so let me tell you, if you're an entrepreneur, you're thinking to launch a company, I'm gonna tell you how to think about AI in 2025. So first and foremost, you can prototype any idea that you have using tools like Replit right now in minutes. Okay? I built some of the early stuff on Replit, and it was a terrible, horrible— I'm not saying it's a perfect process right now, But compared to when I started using Replit to today, building something on Replit is wicked easy. You're like, I want to do a thing that does this, this, and this. Any idea that you have, you can probably build with AI using a tool like Replit right now. And maybe it's not the final tool, but it's a prototype. And then you can start getting users to use the prototype and give you feedback if what you're building is any good, right? And a lot of the, frankly, you can use Replit, which is an AI application and builder to build an application with AI, right? right? And then get users to use your application with AI to see if your application has any real traction with consumers, right? And so you can prototype very complex business ideas very easily using a tool like Replit to see if there's market demand And then if there is market demand, you, you could probably use the tools to finish it or build the team out, etc. So I think entrepreneurship in 2025 is AI first for a number of different reasons because you wanna prototype what you're going to build very quickly using AI and get customers using it. The tool itself may use AI to deliver the value to the consumer. In fact, I would say that if you're prototyping a tool with AI, that tool better be using AI as well. Because everything that you can do today without AI, you could probably do better with AI. And then last but not least, you've gotta get people using the tools. And you gotta do outreach to them, outreach on social, outreach on, with videos, and you can make all the marketing. I would say one of the areas where AI excels today is in marketing. You can pretty much have AI produce marketing for you that will require minimal or no editing whatsoever and be rather good. Probably better than you could pay someone to do at an affordable rate as a starting entrepreneur. So you should use AI to make the product. You should use AI in the product, and you should use AI to market the product.

No, I totally agree with that, Leo. And can you tell more about— I mean, there's a lot of talk about the data that you have to have in order to make this AI defensible, right? I mean, you— so is there an aspect of what you guys have done that you really kind of have a strategy behind it that in order to create data and have a kind of unique set of data that no one else has? So how do you guys think about data in collaboration with kind of this technology, these tools?

Yeah. So look, there's no one answer to that question, right? So you can create great AI tools that don't really require data. For example, one of the first comprehensive tools that I built was a something called Dumroll. And it was just for fun over the holidays. I built a multi— massive multiplayer text-based game, and different players could create different worlds, and the AI acted like a dungeon master. So you could say, I want a wizardry world, or I want a dungeon world, or I want a fantasy world. And then the AI would be the dungeon master, right? And you would play the game with your friends and you could chat with your friends, all the stuff. It actually was pretty, pretty interesting. And they're like, you know, the AI is entertaining. I'm not really worried about being disrupted, right? Because if you get critical mass of usage, you have a unique idea, it's like kind of nichey. And there's no real data strategy there. The, if anything, it's like get players to create interesting games that get other players to play them. And, you know, AI is, makes it entertaining, right? Then there are other business types where, you know, data, there's training data and then there's actual data to operate the AI on. And, and, and let me explain how I look at those things differently. So AI is very smart, right? Because it's trained on tons of information on the internet, tons of information in video, etc. However, it, it, it's like it knows so much that in a weird way, it's like the jack of all trades and master of none, in my experience, right? So you hear some of these AI companies say like, oh, AI now is better than your average lawyer. Well, you know, you're not going to trust your average lawyer with a billion-dollar transaction. You're going to trust a great lawyer with a billion-dollar transaction. And AI is nowhere near the caliber of a great lawyer. And it's a long way away from being that. Right. I know everyone says AGI is coming and stuff. But even AGI, which let's hope is benevolent and doing great things in the world, there's still this level of mastery that you have to get to, to be the best of a certain field. And I can assure you that even with AGI, it will, will probably not be the master of everything that humans do. So what, what I've found is you can get AI to start kind of giving answers as though it were a master if you can provide it with a corpus of very focused training material and you use what is traditionally referred to as a RAG, which is basically a system to access pieces of the corpus of training data to answer a specific question. So, and there you can really start upleveling AI in areas where the jack of all trades master of none AI is not really great at like deep complex legal concepts. But if you give it a corpus of great training of deep legal concepts, it can actually start to replicate the work of a very high-end lawyer, but it needs, on top of this generalist model, it needs that focused data set. And so that's definitely an area where entrepreneurs can differentiate today by collecting a corpus of highly specialized training data to make generalist AIs into really experts on a subject matter. And that's gonna be very difficult to replace for the foreseeable future because as we keep upping the generalist quality, it's still a generalist, right? It's not a specialist and it needs that specialist training data to become a specialist. And that's today done with a RAG. So that's very differentiated. The other area of content is there is proprietary content, right, that you can have. Maybe you're a movie studio and you have your collection of movies. And, I'll give an example, like, you're Disney and you own the Star Wars franchise and people wanna make their own Star Wars movies, right? So they're like, I want a movie where Darth Vader beats Luke and Luke did it, right? Well, you're gonna, like in the case of Disney, they own Star Wars. So in order for you to make your personal Star Wars movie, you might need to access the proprietary data of Disney to make that happen. Similarly, you may create like in our case, our fund managers are storing all of their data on fund performance, deal reviews, etc., in their respective parts of our application. And then they can use their fund data with AI to do specific things. And this is their proprietary data, their proprietary analysis of deals, their proprietary performance. And they can use that with AI to get to find new deals that look like the deals that did best for them to manage their operations in the way they like their operations managed. So you'll have proprietary datasets that can be used by AI either to make consumer services like the Disney example or just to make, your life better. You'll use your proprietary dataset to improve your operations or whatever. So that would be a third area of opportunity.

Yeah, and I was asking that, Ariel, because in this world in which you can basically create these applications with Replit or similar tools very quickly, I mean, just having the technology by itself is not going to get you anywhere, right? So And we always have talked about the speed and execution being kind of the biggest kind of defense that you have in entrepreneurship. So nowadays it's even more important, right? So will you think that, I mean, execution and speed is all still kind of what is gonna ensure your success eventually?

Well, sort of. I mean, look, so yes, you have to move fast, guys. There's no question about it. If you're not moving fast, I mean, I've got a team. I am the market leader and I have a team to disrupt me because I realize that someone else with less overhead, with less people, with less to worry about can disrupt me, right? So, you're in a world where moving fast is now table stakes. If you're not moving fast, you yourself will be disrupted by someone who is moving fast, and then they will be disrupted by someone who's moving faster, and that pattern will continue. Now, what I'll say though is that, and then you asked about execution, so I actually think, um, It's not so much about execution anymore, though. It's, we're in this transition period. It's really about mastery because there is going to be very soon a billion-dollar founder who has no team. It's all AI. That's inevitable. That's gonna happen. It's gonna happen soon. The technology's reaching the point where that's going to happen any minute, right? When I say any minute, I don't mean like tomorrow. I mean any minute in the scheme of the proverbial reference, probably in the next 3 to 18 months, right? It really depends. And then the question is, who is going to be this first billion-dollar solopreneur? And mainly using AI, right? And it's gonna be the person that masters AI, right? So I take the mastery of AI very seriously, and I have machines that run it here locally. I run it in the cloud. I use the APIs. I build apps myself because if I am not a master of this coming tool suite that is highly disruptive, then I'm guaranteed that someone else will be the master and disrupt me. So I'd rather be in the disruptor versus the disruptee position now. So speed matters, right? And I would say execution also matters, but really more than execution, it's this mastery. Mastering the craft of using AI to accomplish your goals and ambitions will put you in a much better place than any peers that haven't mastered it who are really ripe for disruption.

No, amazing. So just a couple of more questions, um, before we wrap up. One is, um, you talk about this tiger team that you have, but when we talk about early-stage founders, they're not going to probably have the luxury to have a team to disrupt themselves. So what would you recommend to them?

Yeah, when you're a one or two-person team You are the disruptor, not the disruptee. I've got a 50-person team. I am the disruptee. That's so I need to have a tiger team, including myself, that's out there saying like, so I talk with managers, they're like, oh, I use this system to do this. I'm like, shit, we didn't build that system. Someone else did. Because you can build things so fast. So here's what I would say to if you're a 1 or 2 person founding team, uh, the main metric of your success is the speed at which you're able to create tests to learn your market and then get to the next test, right? So you're, you're really optimizing for the speed of learning to master the market you're trying to go after. So you have to look at everything you release as sort of a test. You have to try and get usage. You have to examine the usage, learn from it, launch a new task. And if that takes you 2 or 3 weeks to do, you're probably moving too slowly. You want to be doing tests and learning from the tests in days. I would say 2 to 5 days is pretty fast. 1 to 2 days is really fast.

Yeah, so it's a very interesting world we're living in there. So You kind of touch upon the question that I have last, but is with technology moving and advances moving this fast, right? I mean, how small teams can keep up with all the, with all the advances and everything that is going on is, and doing everything else that they need to do.

But there's only one thing they need to do, which is master the tools.

Yeah.

To get the speed of releasing product to learn lessons so that they can then do another test and learn more lessons. So look, you're gonna eventually need to do accounting, hiring, all this type of stuff, right? At some point. But when you're starting out, the main thing you need to be doing is releasing product and learning from those releases to release better product and learn from those releases and release better product until you have this really zealous customer base that's like, I love it, I want it, you know, make it happen. So you want to be releasing As I said, 1 to 2 days is very fast. 2 to 5 days is fast, and anything past that is okay, but not great. And, um, as a small team, you're in a great position to do that because you don't have anything else to do. Like, what else do you have to do? Like, cook dinner maybe? And like, hang out with your family? Like, and then work on this stuff until you get it right. And once you get it right, then you're in a pole position to go out and think about things like fundraising or hiring or whatever else may be next on the agenda. Now, look, there are some areas where you're probably going to need to have money to even be able to do any of this. And, but I think you could probably prototype a V1 enough so that you can raise money to get the infrastructure in place to get on that fast learning curve.

Thank you so much. This has been great. I really appreciate the time. And, and, yeah, thank you for, for giving your insights and advice to, to founders.

I use AI every day. Hopefully it's helpful, guys. Good luck. Thank you.