(upbeat music) Traylon X, a conversation with the co-founders of Harvey, one of the most closely watched companies in Legal AI. Gabriel Pereira and Winston Weinberg started Harvey in 2022 when there were roommates sharing an apartment in San Francisco. Gabe had been working on AI research at Meta and Google while Winston was a first-year litigation associate at Old Melvany and Myers. Today, they still share that same apartment but Harvey has grown into a global company serving more than 1,000 customers in achieving a valuation of $8 billion. In this conversation, we go back to Harvey's earliest days when Gabe and Winston were reaching out to thousands of lawyers on LinkedIn trying to get anyone to take a look at their product. We talk about the pivotal moment when they got early access to GPT-4 and how that transformed what they could build. When we discussed the breakthrough that came when Alan and Overy became their first major client and announced it was deploying Harvey across its entire global firm. Let's also look ahead. We talk about Harvey's vision for becoming the operating system for legal work, their strategy for building agentic AI that can handle complex workflows and their focus on moving beyond one-off tasks to client-matter-centric work. We even get into how two young founders with no amount of experience have learned to scale a company that now employs hundreds of people or the 20% of whom are lawyers. I'm Bob Ambrogie and this is law next, the podcast that features the innovators and entrepreneurs who are driving what's next in law. Before we get to that, please take this moment to hear from the sponsors who so generously support this podcast. Legal Week is back this March. Join thousands of legal leaders at its new home, the North Javits Center in New York City, March 9th through 12th. Discover why it's called the premier legal event for information, education, and networking. Gain actionable insights to help you restructure, rebuild, and reinvigorate your firm or legal department. Don't wait, secure your 2026 pass now at legalweekshow.com. Running an immigration law firm requires cutting edge technology that keeps pace with your practice. Bollylaw delivers with over 50 predefined workflows, instant form updates, and multi-lingual client management. Access 150 USCIS, DOJ, and EOIR forms, all updated within an hour of release. While streamlining your entire case process from intake to completion, transform your immigration practice with intelligent, comprehensive case management software. Visit lollilaw.com, that's L-O-L-L-Y-L-A-W.com. Now let's get to that conversation with Gabe and Winston. Gabe and Winston, welcome to law next. - Thanks for having us. - Nice to see you guys, and looking forward to our conversation today. A lot I want to cover, but I have to start with this question because two weeks ago I saw the two of you featured in the New York Times as two of the new billionaires of the AI boom. And since I may never again have the chance to talk to a billionaire, I have to ask, what's it like? - It's literally like kind of a ridiculous article. I mean, Gabe and I literally live in the same apartment as we did when we started the startup. We have a third roommate who also works at the company. I don't think it has really changed much at all. I think we've both helped our parents out in some instances, but in terms of our own personal lives, to be honest, we work the same amount of hours. Our apartment, I think, has the same furniture Gabe, basically, as it did in the beginning, which by the way is no furniture. - My mattress is still on the floor. - Yeah, we had one of our friends who works at the big labs come over kind of recently and he was just like appalled at basically, just like it's all boxes and everything like that. So I don't know, I mean, these things are all like on paper so it's not, it doesn't really make sense. - I get it. - It was just that article struck me as kind of funny. And I was wondering whether you're still sharing an apartment because I knew way back when I talked to you, when you guys were just getting started with this, you were roommates at that point, but yeah. - Good. At some point maybe you'll be able to afford your own place. - Working on it. - All right. - Well, from billionaires to beginnings, I want to kind of go back a little bit to some of the early days and ask a couple of questions about that. It's interesting, I was actually looking back at my own notes. I'm an inveterate note taker and I first gave, I first talked to you back in June of 2022 when you were really early stages, I think of thinking about building a product and what it was you wanted to build, and maybe wrong, but I kind of got the sense at that point that you were kind of more focused on thinking about almost a consumer facing kind of a product. We talked a lot about issues around unauthorized practice of law and how that could impact development. That was June 2022 by November 2022. As the next time I talk to you at that point, you guys had a product, you were talking to law firms. I think you, GPT 3.5 came out at that point. I think you are already having access to four, to early access to four because you're a relationship with OpenAI. So I'm kind of curious, how did the product evolve to that concept that you were building in November from how did you're thinking evolve around what it was you wanted to build and how you wanted to use this technology? Gabe, maybe that's good. - I was gonna say I would love to see those notes from our first conversation. I remember that in full detail. I would say the biggest thing that changes just the capability of the models. So when we first started, Winston had been doing litigation on Malvani and so we had a good sense of how good the models need to be, to be useful in big law. And when we first started the company, they just weren't there. And then I think we were also just interested in consumer access to justice, things like that. And then through the relationship with OpenAI, I got early access to GP4. And we actually, from there, the product you saw, we went and showed that to law firms, startup lawyers, in-house lawyers, kind of all the different lawyers we could get in front of. And when we actually found the most traction was big law firms, which I thought was new. - Who was pretty interesting, but that market pool, I think, kind of put us in this direction. - Were you showing it to other sectors of the legal market? And that's where you got the interest or that's where you just kind of started showing it? - Yeah, so we showed it to in-house teams. So for example, one of the early demos we did was to John LeBarr, who's now our general counsel, who's in-house at Snowflake, and he was super excited about it. - He's like our seven tires, something like that. And I remember our VCs were just like, why are you hiring a general counsel, is there seven tire? Like, you usually do that when you're at like 300, 400 people, or something. - Yeah, was he the one, you told me the story in November 2022, and I've heard you repeated since then, of showing it to a general counsel who said, this is crazy, my lawyer would have charged me 10,000 or there's sort of 20,000, but there's over here what the number was, was that that person? - Yeah, I mean, the use case was, this is public now, and was then Snowflake was kind of opening offices in Indonesia, and he needed a memo on the data localization laws. And so it was just draft me a memo on this topic, and he had already done that. And so he could really verify the outputs of the model, and I think even then he was already pretty impressed. Obviously, like things missing. But that was, yeah, that was one of the early, like, like, old moments for sure. - Yeah, as I said, I think 3.5 was out then, but you had early access to 4.0. As you were building this product, were you, was that eye-opening when you got your hands on 4.0 and started to work with that? I mean, how did that kind of change the, thinking about what you were building in the direction of what you were building? - Yeah, I mean, I still remember Winston going in his room for like 14 hours when we first got it, and kind of just mimicking a bunch of the legal work he had done at his firm, and I think I had intuition from watching him do it, and then as I was building the product, I was using these models to code, and so by analogy, you could kind of see how much better the reasoning was. But yeah, we basically just like ran back and tried everything that I couldn't get three and 3.5 to do, and all of a sudden, you know, really the main difference was just the reasoning jump, right? So it's like I was giving it the same data, and data is in like the prompt, right? et cetera, just like a bunch of context in the prompt. But the main difference was just like, oh, it could actually like reason through that data and come to a correct conclusion, whereas the earlier models, they could really only do that if you had something that was like really, really well defined rule based wise. So like when we started in the beginning, a lot of what we did is landlord tenant, like that's what we were looking at, and the reason why is there's just like, hey, there's so much public data on this, but be it's like safety security deposits in California. Right, there's like a certain amount of rules, every single fact pattern falls into like one category of those rules, and it was really, it was easier to do. What I couldn't get the model to do is like, here's what a bunch of cases say about, you know, like section 11 claims and things like that and securities law, and it was just absolutely couldn't do it. And that's where like the reasoning jump was just like astronomical. So yeah, probably the most pivotal, pivotal, that's a tough word to say. The most pivotal point in your early development, or at least from a public facing point of view, was came in February 2023 when Alan and Overy announced that they were going to be deploying you throughout their firm of 3,500 lawyers then at the time and that they had been testing the product for like a number of months at that point. How did that relationship ever come about in the first place? How did that start? - Yeah, so we got intro basically too. So it was interesting, you know, 'cause we obviously did that like wait list in the beginning, right? And I think a lot of folks don't realize that the reason that wait list happened is what happened was, we reached out to like basically everyone on LinkedIn, like every single firm. Like I think Gabe and I reached out to like about tens of thousands, maybe? I mean, we would limit it. Yeah, we get blocked basically every day of like you reached out to too many people and then we'd have make like another account or we'd like have a friend's account or like something like that. Honestly, like very few people responded like almost none and none of the big law firms responded. And then we had happened to do this with A&O Sherman and then we also met someone at Stanford Business School who had been at A&O Sherman and we basically were able to do like a live in person demo with him, right? And he actually emailed David Wakeling and said you gotta like take a chance on these guys and like do a demo, right? And so we did that demo and then we started meeting more and more people at A&O Sherman and then all of a sudden after the A&O Sherman press happened a lot of those law firms reached back out, right? And they're interested. We were four people at the time and the fourth person had joined literally as the like the day before the press announcement had came out with A&O Sherman. So we were four people, three of whom had been there for, you know, Gabe and I had been there for at least six months or something like that and the other person been there for a month. And then all of a sudden all the law firms are reaching out. And that's like the position we were in. We had basically just done a rollout to like a 4,000 person org with four people. And so it was like a very kind of crazy time for us because we had so much demand and we quite literally couldn't respond to enough folks. - Yeah, so how did you manage that? - We did a wait list, which in hindsight I think there could have been better ways to do this. Like if I could have done it over, I think what I probably would have done is like just been like more public about this. Like we both would have just been like public and been like, "Hey, we're a small team." Like introducing Harvey to the world basically, right? Instead of the way that we were introduced was through a press announcement of A&O Sherman adopting it instead of us coming out and being like, "Hey, this is our team." Like meter team, this is our resources. This is how we did the onboarding. And you know, this is where we're at. We're hiring as fast as we can and we'll bring more people on after that but we'll do some public demos or things like that. - Yeah, I'm curious just, this was so early. I mean, you guys were obviously super early in bringing a product out. This was before case text that unveiled co-counsel before the whole Thomson Reuters acquisition and all that. This was really early. From working with Alan and over at that point and kind of doing the testing with them and then ultimately going forward on deploying it, what did you kind of learn about your product that maybe you didn't know? What did they want out of it that you hadn't thought about? Maybe. - Yeah, give you one go first. 'Cause the list is long. (laughing) - I was gonna say like, I think, and people still experiencing this, there's just a huge gap from like an imprinted demo to like something that worked in production for high stakes legal work. And I think you just very quickly found the rough edges of these models and just the sheer amount of product that you needed to build around them of, how do we connect this to our data? It's missing case law. I want to save documents. I want to share this with teammates. We just got, I mean, just from Alan and over, like, you know, a three-year product group map of. We want to be able to do all of these things and then also all of the security, data privacy, ethical walls. And so I think it was just kind of like, it was super exciting of getting this to a bunch of lawyers and then at the same time, so much feedback and so much like product intuition from kind of all of those users. - Yeah. And the scaling is really different. And I feel like that's something we've, maybe like, I feel like we learn a scaling issue on the product side of like every three months, basically. Where, like, for example, we, you know, for a while, it's just the assistant product, right? Which uses a lot of tokens, but not like a crazy amount. And then we created a vault, which was like, you know, at the time, I think it was, what was our minimum of the maximum of the time, like a thousand documents, I think, like when we first released it, something like that. And it's like, now all of a sudden, you're doing multiple model calls on a thousand documents across, like, simultaneous. And it's like the architecture you need to set that up. And when people actually start using it and pressing on it, it's just completely different. And so I feel like what was nice about working with Aino Sherman in the beginning is we worked with a scaled customer immediately, right? Whereas if we would have started smaller, we would have ran into those issues like later on. And now we're doing, like, we're going the other way, where like our mid market org is like growing really, really fast and we're selling more and more to smaller law firms and smaller corporates. And like, it's actually for us, I'm on the product side, it's actually easier to go down, right? Just because like you don't have to deal with the same like architecture and security concerns. What's a smaller law firm when you say a smaller law? Even like, I mean, we're selling to firms that have like 10 attorneys and stuff like that. We aren't doing solo shops yet, just because we don't. We haven't set up the infrastructure for self-serve. I were eventually going to do that, but we haven't done it yet. Yeah, I'm sorry, I cut you off. I was just going to say it was also a really interesting experience in the early days where you think of kind of the Googles of the Microsofts of having kind of unlimited cloud capacity and also just constraints of, we don't have enough model capacity, we don't have enough embedding capacity and so kind of like building that alongside them. And like even now as we're scaling, like these are actually a lot of the problems we're solving of, we want models in different regions. They haven't stood up regions. And so kind of like figuring out how to solve that globally, I think has just consistently been an interesting problem. The data processing still a problem. Right back to John Lovar's question, or you know, like a bunch of these regions have data processing laws, right? And especially for financial data and financial data, if you have data processing laws for financial data, you have data processing laws now for law firms because a lot of what they deal with is financial data, right? And so we had to basically set up, as your instances, basically across, you know, or in 59 countries right now, and almost every country wanted their own instance, right? To make sure the data doesn't go back to the US and it stays in their country. So I think a lot of that was difficult at the beginning. - Yeah. What kind, as firms are reaching out to you, you started working with Alan and over, what kinds of response did you get from, kind of the rank and file lawyers in the firm? I mean, I could imagine some of the leadership was eager to embrace this technology, but we all know how lawyers are about new technology and how a lot of them have been about generative AI in particular. Was that an obstacle for you that you had to challenge for you? - Yeah, I actually think it was a little bit the reverse, to be honest, where there were a lot of, you would get like a bunch, like you kind of have, and it wasn't practice based, by the way. Like you'd think it would just be like the tech groups or something like that, that they're more interested in it. It wasn't practice based as more like individuals, but you'd have individual attorneys who were really bullish on it and really creative and like coming up with their own solutions and doing these crazy complex prompts. I mean, there were some attorneys in the beginning that like they were showing us what they were doing and it was like six paragraph prompts. And it was like just like crazy, and this is like 2023. Like this is like really complex, interesting creative things. It was actually harder to get leadership at law firms to be bought into these things. And I'd say it started with individual lawyers than it went to innovation teams and they were the ones driving it. I'd actually say that law firm leadership has just started really kind of pushing this. In 2025, I think that was the first time that for like overall market wise, that leadership started really like taking this step forward. - Yeah, I think one thing really early on that at least was very surprising to me is having senior partners talk to us about this and be like, this is really useful to me. Not, oh, this is useful for my associates to use. Like I'm using this technology and I think that was one of the things that at least gave me a ton of conviction of if we can get the product to a point where not just the people that are super deep into kind of early days of language models, but everyone could get to that experience and you kind of have the ceiling of some of the best senior partners using this. That's when it was like really eye-opening. - Yeah, I'm gonna fast forward a little bit. Obviously, a lot's happened over the last couple of years with Harvey and the product. I'm curious, not exactly how the product has evolved over the last couple years, but how you're thinking about the product has evolved. How your approach to it has evolved, as you have seen greater and greater adoption and gain greater and greater traction. Where are you now on that? - Yeah, I mean, I can start giving if you wanna add to it. One of the main switches that I feel like we have started doing is building things for scale. So like something that I think is interesting is like building something that demos super well versus something that works at scale. And I think that has changed a little bit part of our product direction and things like that. Like I think in the beginning we used to do a lot of things that were, I mean, I'll admit it, like flashy and stuff like that. But at the end of the day, like if you had 50,000, 100,000, couple hundred thousand users using it, it actually like wouldn't work that well. And so I think we've thought, we've been a little bit more thoughtful about our product roadmap than we used to be, where we think about like, okay, if this is a new thing that we wanna build, we wanna build shared spaces, right? And make it so that it can scale. Maybe shared spaces might be the best example. You have to think about the architecture of that in a world where eventually you have seven parties working in the same shared space, right? Like you have to make sure that you're building for that version of it. And that might mean that there are certain changes that you make on the architecture and upfront investments that make you like slow down in some instances, but they're really helpful because then it makes it so the clients are much more willing to actually adopt these tools, or you can have like multiple clients in the same one and things like that. So at least like, this is super high level and instead of like feature by feature, I think a lot of what we thought about is like, how do you drive innovation as much as possible, but also make sure that whatever you're building is actually gonna get adopted, right? 'Cause I think that we have a, there's a lot of market, like, oh, there's a new shiny thing, et cetera. And we definitely believe in that. Like memory is a really awesome direction and I'm really excited about that. But that is gonna take tons of soak time to get memory right, right? Like a lot of soak time. And so we're thinking about like all the architecture permissioning, all the different types of memory and all of those things up front, which makes our product roadmap like you have to actually think like a year ahead, two years ahead, et cetera. And I think related to Winston's point for me, it feels like the biggest shift in thinking is when we started building the product, it was very much, how do we build the best co-pilot for an individual lawyer? And I think this is still a big part of the roadmap, right? You want the assistant to work well, you want to connect it to all your data, a case law word, email all of these things. And there's still a ton of work to do there and make that better. But our general sense is like next year, the big blocker is going to be, how do you deploy and cover this technology at scale for these very large law firms or even small law firms and enterprises? And to give one, I think concrete example, PWC was our second customer. And a lot of what we built for them doesn't actually show up in the product, but you need to be able to deploy at that scale. And so like under the hood, they have 50 different workspaces because they need every country to be isolated. They need a global admin layer so that admins can, with eyes off, without seeing the actual queries of all the different countries, aggregate that data and look at it. They want their compute in different regions. And it's a lot of this infrastructure work of how do you make that configurable where we can do that for a PWC. We can do it for these large global law firms, but it also works for these 10 person law firms. And then this also works for enterprises. And then to Winston's point, like this has built a lot of the foundation where now you want to do collaboration. And you have a lot of these pieces built. But I think as we get bigger and bigger, we've just realized how much of building these enterprise products is most of it you don't actually see when you interact with the product. It's solving kind of these problems so that the admins, the CIOs, can confidently deploy this at scale with all of the ethical walls, permissions, data is in place. And I think that's become increasingly important as these systems are becoming more business critical. - Yeah, it slows you down, but then it speeds you up. So it's like long horizon agents, pulling data from multiple client matters simultaneously. That would be awesome, really cool. And work on incredibly cool document drafting processes, analysis, checking what's market, all of those things. If you don't have ethical walls in place, and you try to do that, it's gonna be an absolute nightmare. But if you get like a very strong foundation for ethical walls and permissioning, et cetera, now all of a sudden you can build kind of like whatever you want and release that. And there's the trust that it doesn't matter because anything that you build in this like ecosystem of these walls will respect that security, will respect that permissioning. And so a lot of this year is gonna be figuring out those so that when really interesting things come out from us like memory, personal memory, institutional memory, client memory, all of those things, you can just turn it on and you're good to go, right? And you don't have to worry about like different configurations for different problems. - Yeah, and one thing to add to this, I think the other thing, I think we've done a good job of this but it's become increasingly important. It's how do you also connect with that ecosystem? 'Cause I think when you start building any product, you kind of just think about the user living in my product. And very quickly when you deploy in law firms at enterprise, you realize the tax stats at both of these size of the market are super complex. And so a lot of the value of the product is not just, this is a good product by itself. It's this product integrates and fits well within the rest of my products. There's one thing that's super important for us is we don't want to rebuild data rooms, DMSs, ethical walls, all these things. But law firms are going to need all of these products to connect with all of these systems to be able to get the value out of Gen.A.I. as it continues to approve. As you heard Gabe describe, Harvey went through a fundamental shift in how it thinks about its role in the legal technology ecosystem. In the early days, it was about building the best AI assistant for an individual lawyer. But as Harvey has scaled to serve thousands of lawyers across major law firms and across enterprises such as PWC, the challenge has become something much more complex. How to build infrastructure that can support 50 different workspaces maintain ethical walls across global operations and integrates seamlessly with the dozens of other systems that law firms already use. And as you heard Winston say, getting that foundation right actually enables Harvey to move faster later. Because it can build new features like memory and long horizon agents on top of a secure permission-based architecture that everybody already trusts. We'll continue the conversation with Harvey's founders in just a moment, but first, please take this opportunity to learn about the sponsors who so generously support this podcast. Deadlines don't care how many documents you need to review. BriefPoints Autodoc keeps you ahead of them. Autodoc automatically generates responses to your opposing counsel's requests for documents, complete with tailored objections, baited production packages, and substantive responses, citing the responsive documents by baits numbers. All done at the rate of two to 10 seconds per request. No setup required. Stress less about discovery deadlines. Head to briefpoint.ai and take the pain out of your practice. Think about all the time intensive tasks slowing down your immigration practice, hunting for updated forms, translating documents, chasing payments, and managing workflows. Quality law's immigration case management software automates it all. With government forums updated with an hour of release, more than 50 customizable workflows, and native payment processing in over 20 languages, see the difference at lallelaw.com. That's L-O-L-L-Y-L-A-W.com. (upbeat music) Welcome back to my conversation with Gabriel Pereira and Winston Weinberg, the co-founders of Legal AI Company Harvey. Before the break, we were talking about Harvey's evolution from its early days as an emerging technology to where it is today. Now we'll focus more on where Harvey is today and where it's headed over the coming years. By any measure, 2025 was an extraordinary year for Harvey. It doubled its valuation to $8 billion, raised $760 million in funding, and again, that's just in 2025 on top of what it had already raised, surpassed 1,000 customers, and reached $190 million in annual recurring revenue. But perhaps more significant than the numbers is what Winston gave described as a fundamental shift in the market. The year when other vendors finally started partnering with Harvey rather than viewing it as a competitor. Let's get back to that conversation. 2025 was an amazing year for Harvey, obviously. You ended the year with, you doubled your valuation over the course of the year to $8 billion valuation, you raised some $760 million, I think, just this year, just in 2025. You've now reportedly passed, that was in customers, $190 million in annual recurring revenue. And you put out, I was just reading the annual year and review that you folks recently put out. And it started with the Senate saying 2025 was the year Harvey moved from emerging technology to essential infrastructure. Is that what you mean by that game what you're just kind of talking about about the need to sort of be this infrastructure for organization? Yeah, I mean, go for a game. I was going to say, we see that starting to happen, where luckily our product doesn't go down a lot, but when it does, we get emails from customers saying, hey, we're working on this litigation or this diligence, we need this and able to do this work. And I think enterprises are going to start relying more on these Gen.A.I. systems because they're so powerful, but you need kind of that confidence of these systems. Like, I think we've moved past the let's wait and see, let's figure out if this technology is a real thing and we're very much now in the territory of, okay, this is going to become mission-critical infrastructure and we need to start thinking about how to deploy that scale, govern it, make sure it's reliable. Sorry, I got what I said. Oh, I was going to say, especially with client-facing. Like, right, that's going to be a whole new level of this. It's one thing, if it goes down and you're working on, you're doing a diligence and you're using vault to do it, et cetera, that's awful and that's terrible if the product goes down. Imagine this is also client-facing the clients in that vault trying to do something in a coast down. So I think there's maybe two reasons for looking at that way. One is everything that you need to do on the back end for security, reliability, scale, infrastructure, et cetera. And that's a very different type of engineer, too. A very large portion of our company is infrastructure engineers way larger than most AI companies to make sure that that works. The second one is, no one would partner with us until 2025. I'm serious, right? And like, you do mean other vendors like your Lexus Nexus in that, right? Yeah, exactly. And I think a lot of what happened in 2025 is we ended up kind of being able to partner with all of these other vendors, right? And hopefully the reason why that happened is one, a lot of large user base, right? So the users are asking for it. I think the second realized thing is they realized that we aren't rebuilding a DMS. Like we aren't rebuilding XYZ, right? Like we're trying to, you know, I think like all companies have like a one area where they have like a right, or they think they have a right to win, right? And our right to win isn't rebuilding like legacy legal tech companies, right? And so I think that started like a snowball of more folks actually not thinking of us as like a direct competitor, but as something that's like part of their ecosystem that the users use, right? And I'd say like almost if anything, one of the biggest things that changed last year was how many folks partnered with us. And we have a lot more that we're going to announce soon. And I think that's a net good, like net positive for the industry. And one thing we want to do in 2026 is actually do the other way around too. So there are some incredible companies that are in other verticals that I'm probably not supposed to say this. And if my head of sales was listening, he'd get pissed, but hopefully you won't hear this podcast. There are companies in other verticals of hopefully he doesn't listen to it. I don't know, we'll see what happens. There are companies in every single vertical and in every legal vertical. And some of them have like much better products than Harvey does for that particular vertical, right? - Yeah, yeah. - Like there's like some in like certain like patent areas or other areas. And it's just like their output is like better and the product's better than ours. For that vertical, right? For that specific use case. And we're not trying to build it all. And so what we'd like to do more of this year is actually like have them integrate with our platform too, right? And figure out how to, you know, if you're asking a certain query that might work on a different platform, can Harvey identify that query and route that query to the product that's best for it, right? In the same way that right now, one of the things that we're doing as a revamp to our product is we're starting to route everything. So like the new version of Harvey is basically gonna be you start typing. And it's gonna route you to different parts of our product or different documents you've uploaded in the past or things like that. It's gonna be proactive in coming up with like what part of the product should you use for a certain thing? You're gonna imagine a world where it like also routes you to maybe a different product too, right? If it's like better for that particular use case and your firm's already bought that. So I don't know, I think of like partnerships as one of, if not the core like thing that we're trying to do. - Yeah, and one thing we heard a bunch, especially at the end of last year from CIOs was kind of exactly what Winston was talking about if we have 100 legal tech solutions. This one's great for chronologies. This one's great for patent drafting, but our lawyers can't find it, right? It's somewhere it's come direct that they don't know about and they just forget to use it. Wouldn't it be great if when you ask the question, Harvey, it could redirect you to that product and you go use that product and it kind of helps you remind that. And so a lot of what we're building like Winston mentioned is how do you build kind of MCPs in both direction or just write PIs that let firms customize more of this so they can surface all these great tools to the users? - So I get all that. I have to say the partnership with Lexus Nexus certainly felt like competitors partnering with each other. I mean, Lexus Nexus, the lot of overlap in terms of what Lexus Nexus has been building with its protege and some of its other tools. Why did the two companies want to do that? What's the benefit of that partnership? - Yeah, I mean, I think the reality is like, Lexus has been collecting data for decades, right? And I think the reality is like, even if today we started to go out and try to collect that data, A, it'd be difficult to get and there is data, like we do do a lot of data collection ourselves as well. It'd be a difficult to get and B, I think there's a trust here element, right? And like Lexus has an incredible brand and people really trust their product and they trust that if you do an exhaustive search in Lexus, that's an exhaustive search and so much, you know, as a litigator and it's like so much of litigation is find every single case that has ever mentioned or refer to this other case like all of that, right? Like the worst thing you could possibly do is not mention a case in a brief and it's pertinent and material, et cetera. And the other side mentions it and you have to explain to the partner why you didn't mention that case, right? And so I think for us, you know, we hope and think that we have a right to win on kind of like UX, UI, AI core engineering systems, things like that, we are in a data company. And so it felt like an actual partnership for us and I think that's why it happened. - You had talked, when I talked to you guys in 2024, you were talking about possibly building up a collection of case law all around the country, all around the world. I mean, and trying to put together some of that data, what happened with that plan? - We still do that and we still like collect our own data. I think that like the end of the day is like, we'd rather take like a partnership strategy and we've done a bunch of these two, these data partnerships like across the world, right? Like Lexus and TR are the main ones in the US, but in a lot of other countries, there's local players that are really, really big, right? And we've done a lot of those partnerships too. So we think of it more as like, if we can partner a great and we'll partner, if no one wants to partner with us, then we'll have to go out and go get the data ourselves is like basically how we do that. - Yeah, makes sense. One of the other changes for Harvey in 2025 was I think your decision to kind of more deliberately or aggressively expand into the corporate market as opposed to just law firms. I think Walmart is one of your customers now. Why did you do that and why, what do you have to offer that market? - Yeah, so a lot of it actually, the first ones were brought to us by law firms. So the law firms like brought us to their clients and said, "Hey, we're using Harvey." And like, these are some really cool ways that your legal department could use it as well. I think there's a lot of value. It definitely is, we actually have our product org now split in the middle, where we have like a law firm product org and an in-house product org. So we are building both. There's a lot of similarities though, right? Like building a really good word plugin and things like that, like that's gonna work for everyone. A lot of the like vault use cases was just analyzing hundreds of thousands of documents that's gonna work for a law firm. Also some folks do that in-house, legal research, et cetera. So there really is like a lot of overlap, but we are starting to build out specific features for the in-house group that might be different, right? So like law firms don't do contracting that often, right? So like, there's certain things there that you need. Another thing too is they're just, there actually are different security needs, especially we work with a lot of banks now, and they have like different security needs sometimes than the law firms. And so it kind of was like the process all over again of becoming enterprise grade, again, for the big banks. I don't know if you wanna talk a little bit about that. - Yeah, it was literally like the process all over again. - Yeah, no, I was laughing 'cause I was thinking about like one challenge that I think law firms are struggling with is outside council guidelines. So they get all of these conflicting rules from their clients that say use Gen-A-I, don't use Gen-A-I, if you do use Gen-A-I, use it in this way, and they need to just keep track of all of these. And I was laughing because we had one conversation where we talked with a law firm and they said, "Hey, our client said that we can't use Gen-A-I in this way." And so I talked with the client and they said, or a law firm said we can't use it. And I was like, "But it's your client data." And they were just like, "Yeah, we have no idea." And so people are still figuring it out, but I think one of the interesting opportunities of collaboration or building for both is how do we work with both of these parties? - Yeah. - What is the correct, secure way to leverage this technology on these client matters or other sensitive areas where they're working together? And so that, to me, has been one part of a really interesting conversation. - Yeah, there's so much cool work to be done there. There's so many law firm in-house pairs where the law firm basically works as the in-house is DMS. Which is really interesting. Like the law firm basically stores all the documents of their in-house team. And so something we're doing with shared spaces is like, because you have all of that client data, is there stuff that you could do to make it so you can support that client better, right? Like is there way to like give them more insights into how you work? Is there more insight, like predictive analysis on things like that, like risk profiles? And so I think one thing of shared spaces that will get really interesting is, how do you make it so that these law firms that have all of this kind of like client data and really have like really close relationships with them? How do you make it so that that relationship is actually even stickier? And that's one of the main things that I think law firms like about shared spaces is they can do that. The second thing they like is a lot of them are using it to get new business. So folks are basically tech firms will use it to get private equity clients. And it's like, hey, we're gonna do the work differently and this is our like value prop, right? So I think both of those are interesting, but it's interesting to work with both sides of it to see what they want and don't want. - Yeah, wanna ask a little bit about the competitive landscape. But obviously that you have competitors out there, you have the sort of traditional companies that have been there for a long time as other startups, but I think one of the interesting questions is to what extent the kind of the foundation models themselves are potentially becoming competitors for you? I mean, obviously you started with a strong relationship with OpenAI, you've since gone to a multi-model structure. There's a lot of talk about, you know, why don't the Pioneer models just kind of jump in and do what Harvey or LaGoura or Thompson Reuters or anybody else is doing? What's your moat there? Well, how do you kind of protect yourself against that kind of competition? Or what do you think the likelihood is of that kind of competition? - Yeah, Ella gave go, but the first thing I wanna say before this is we think about it a lot and even if we're wrong, great. We prepared, we over-prepared, great. Like I'm fine with if we end up being wrong and they don't end up going after legal and it made us go faster and create a more innovative product, faster. I don't see the downside, but thinking-- - We're around the kitchen at home. - Yeah, I don't know. - I would say definitely our thinking has changed a lot here. I think there's one article where the thing I think quote is like the model is the product and I think this was true in the very early days. And then I think the thing we've learned is when you go work with a Walmart with a Goldman Sachs, a lot of the problems that you are solving that are valuable enterprise problems are not intelligence problems. And so we have no doubt that the foundation models are going to the same way they've done clot for financial services, all these things do the same at legal, but I think they will do it the way they did clot for financial services. Right, the capabilities of the model better in all these specific domains. But I think the best framing I've found for this is like the problem we are solving is different. And so I think a lot of the most stems from the fact that you are just not solving the same problem. Like the problem we are solving is not make the model smarter at legal. Like that's part of it. But the real problem we're solving is how do we make law firms more profitable, right? And if you go talk with any law firm today and you ask them, hey, if Chachabitiy or Harvey or any of these products was 20% smarter, are you 20% more profit? And the answer's no, right? And eventually the thing that is going to be very valuable in these vertical specific platforms is you can drive those business outcomes. And same for enterprises, right? If you go to Walmart and they keep scaling and they say, hey, the CEO comes to the GC and says, I need the same or higher quality legal work, but I need you to cut the cost of that by 30%. If the GC comes back to the CEO and says, I gave everyone Chachabitiy. And like they're not able to show that. Like I think you're going to run into this same issue as the model is being 20% smarter, doesn't solve this problem because so much of the problem is the coordination, the governments changing the transformation, how you structure these organizations. And so that to me is like the very big opportunity that I think Harvey has of how do we help every law firm become AI first and how do you help every in-house team become AI first? And I think so much of that problem, like model intelligence is a key piece of this, but 99% of it is also all of the enterprise product. And then something that I actually didn't realize at the start, so much of it is also your GTM org. It is training like these law firms, enterprise organizations on how to do this transformation. Right, like these industries are just way too complex to just be like, here's a product, go figure it out. - Chachabitiy, go figure, change your entire business model and everyone adopted instantly. - Yeah, and a lot of these, like we don't know the answer. Like I spend a bunch of talking with partners, managing partners, CIOs of how should we think about structuring our practice areas? Should we hire more or less in this practice area? Should we do fixed fee for this part of work? Like this to me is like the big value of like what we want to help the industry solve. And I think solving that problem leads to kind of a very different solution than what the horizontal providers are solving. And you need both, right? Like we can build this solution on top of the horizontal platforms. - Yeah, I'm watching the clock and I know I've got so many more questions I want to ask, but we're not going to get to them all. But what I do want to ask, came up in the in the red at AMIA that you did, which is relating to making your product available in the access to justice realm. I know you have the partnership with the courts in Singapore. I'm heading off in two weeks to the Legal Services Corporation's technology conference where I know there's any number of legal aid offices in the world who would drool over getting access to something like Harvey and would make such a dramatic difference. What are your plans around expanding and access to the technology you've built? - Yeah, we've done a bunch in California, but it's mostly through like my old ties at O'Meldeny and kind of like different aid societies that they work with. We're actually starting on looking at how we want to spend this up at like a much larger scale. It'll also be like free, it's the same. Like it's not expensive to deploy these systems. It's expensive to deploy them in some ways. It's not as expensive to deploy them if you can kind of do like here's like the base platform and here's legal research and all of these things on top. And the reality too is like the very expensive part of the platforms is like analyze a million documents which just like happens less in a lot of those use cases. It just does. I think the bigger thing is you can't just do the product side of it. It was what Gabe was talking about too. It's like process helping and implementation and like change management. And so in order to do it well, the thing you need to do is set up like a team to do it. And they need to be experts at that and actually go in and help too. Because otherwise what you're doing is just giving them a tool and being like, I hope this helps, right? Which I don't think is good enough. I think you have to do what Gabe was talking about as well with implementation. Gabe, I don't know if you wanted to add that. - I was like agree with Winston. It's just something that I think one of the things that always motivates me about this company is the opportunity to kind of use the core business to fund things like this and so something both that should happen. - All right. Well from a connected people at Legal Services Corporation, let me know I'll do that. - That would be great. That would be great. The government's much harder to work with in some of these instances. - Oh, they're rapidly. - We've definitely. - They're actively. - We've tried hard to switch to this area. So, you know, as I said, your 2025 report said the year ended with you becoming essential infrastructure at law firm. What's the 2026 report going to say? What's the year going to look like at the end of next year? - Yeah, it's a good question. I'll let Gabe go to. One of the things I'm really excited about is memory-based systems. And when I say memory-based systems, basically like, how do you make it so Harvey has all the specific context for your legal work and the legal work that's going on today instead of how does it have the specific context for legal work overall? - Would you kind of be developing this? - Exactly, right. So you can think of like there's two ways to improve the system. One is make it better at legal work. Just like, and every single vertical, every single use case. The other way is make it better at my legal work. And so I think like that's the thing I'm really excited about. And super briefly, like the four tiers I see here, like personal memory, which is like lawyer memory, matter specific memory. So everything that we've done on this particular matter, right? And then I see basically institutional memory, which is like all the things that a law firm has done over time. And then I see the client institutional relationship, right? So how do you think about like what has been the memory over this type of client? And then how do you do M&A, you know, LBOs over time? And like how do you do those things? So I think like, I mean, this is like a five year roadmap. I think to get this right across all of these different areas. But that's something I'm really excited about. And I don't know if you use it in chat, UBT, but even just in your personal life, like it just makes the product a hundred times better. Yeah. Yeah, and maybe it's on, I would say on the product side for me, two of the big pushes, or how do we make the product more client-matter-centric? But I think right now most of these products, you're still using as kind of one-off assistance. And I think with things like cloud work, the coding products, you've seen this become more of a workspace. And for years, what does that mean? It's when you log into Harvey. You see, here's the five client matters I'm working on. Here's all the context relevant to those client matters. Here's how it's evolved over time. And I can use this technology alongside my team to kind of move that client matter along and collaborate with your clients, hopefully all in Harvey, while also applying ethical walls to all of this. It's connected to veteran, all of these other systems. I think to me, that's kind of the really excited product, exciting product vision. But like when, since it will take a while to build this, but I think we're moving in that direction. And then I think on the business side, really starting to show some of these initial ROI results, I think, at the time, at the period level. So really being able to say, with this law firm, they use Harvey, they transform their fund formation practice, their M&A practice, their IP litigation practice. They did processes in this new way because of the platform. They charged potentially in a different way. And here was the outcome. That to me, those would be, I think, both of the stretch goal, but I think that, like some of the big things on the product side, we're really pushing for it. One last question, and I'll let you go. But I'm really curious how the two of you have kind of grown into this role as the company has grown. I mean, Winston, you're a first year law firm associate. Gabe, I don't think you would ever run a company for you. You'd work at Google and work at Meta. How have you learned how to do what you're doing? Gabe, you're on the first step. How would I want to give you your stress that I don't see right? I mean, just failing again and again and again and again. No, I'm serious. It's like, one thing, one of the hardest things about hiring lawyers, and about over 20% of our company is lawyers. As a lawyer, you kind of have this expectation-- I mean, even before you go to law school, and you take your GPA, LSAT, all of those things-- that everything needs to be perfect. You get in trouble. I remember one of my bosses, who she was fantastic. But I remember she got really mad at me once, because they used a hyphen instead of an amdash. I'm sure, in a sense. But Shatchy B.T. would have helped me with-- Just a hyphen, I can relate to that. Right. But my point is, you kind of learn that perfection really matters. The problem with that is, if you're moving at a really fast pace, it's impossible to be perfect. You are just going to make so many mistakes. And so I feel like the main way that you learn how to do this is you just have to have a super high-paying tolerance for making mistakes. And when you make a mistake, you just got to admit that you made a mistake. And say, I messed that up. And hopefully, I'll do 50% better next time. And then you go to the next day. And hopefully, you do 50% better. It's not an amazing answer, but for lawyers specifically, it's such a different mindset of, yeah, not everything is going to be perfect. Some things are going to lose. And how do you handle that? And I think you have to grow a tolerance for that really, really fast. Otherwise, you don't get better fast enough. And you just kind of collapse at a certain scale. Yeah. Gabe, what about you? I was going to say probably for me, the biggest learning or changes most of my career before this was research, where you kind of sit by yourself, read a bunch of research papers, and you're just kind of trying to find what's the right answer, what direction is this going. And I think that works for a surprisingly long time at a company where when we were pretty small, it's like, I could write a lot of the code. I could typically like guess the right answer. And then you just hit a size where it's so complex that you just can't hold everything in memory. You can't have this context. Unless you're Winston, who can kind of somehow do it with all the people in the company. And so that transition of realizing, I don't need to have the answer. And in fact, my job is never to have the answer and so much of the role at this size is how do you build the right organization? How do you get the right people? How do you start just thinking about like listening to everyone on the team and not just the team like customers? I think this was also a big learning where I think when you're first doing a startup, you're like, oh, I know the answer. And I think everyone has listened to your customers, but I think everyone secretly wants to be more of like the Steve Jobs. And we get into these domains where they're so complex, I'm not a lawyer. I think like so much of it is how do you figure out how to listen to everyone and kind of synthesize these signals. And so I think it's been a really fun process of realizing this new way to like work and scale where now we have so many engineers, product people, lawyers, and you can't see what everyone's doing. You don't have contacts, but if you set things up right, then you're pleasantly surprised where six months later, you check in, someone's built this really cool demo, someone's solved the scaling problem that you weren't aware of. And so I think seeing that organization grow, I think that's been like a really cool experience. Sounds good. - Well, Gabe and Winston, thank you very much for taking the time to come on the podcast and talk. Appreciate it. - Thanks so much. - God, we always time it like that. (upbeat music) - Thanks for joining us for today's conversation with the co-founders of Harvey. I hope you enjoyed it. If you'd like to share your own thoughts or comments, please do so by messaging me on LinkedIn or other social media or email me directly at
[email protected]. If you're a fan of Lawnext, please leave us a review wherever you get your podcasts. Lawnext's a production of Lawnext Media. I'm your host Bob Ambrogie. Hope you'll join us again next time for another episode of Lawnext. (upbeat music)
PodcastTranscript AI