Swedish AI legaltech founder: Most big law firms “not like to leverage Chinese models”

Hello and welcome to the Tech EU podcast. My name is John Reynolds, the host. This week we have another top-notch guest. We are joined by Max Junestrand, the 25-year-old co-founder of Swedish legal tech startup Lagora, one of the hottest startups not only in Scandinavia but in Europe at the moment. So thanks a million for joining us, Max. You have realized you're super busy and we're grateful for you to take time out to speak to us. Just first up, just give us an overview of Lagora and your own background, please. Yeah, thank you so much, John. Really excited to be here and might I add that we've just expanded to the US and so I think we've already made a large footprint in New York and are quickly expanding from there as well. So super quickly about Lagora, we provide lawyers with top-notch legal AI tools and it's all about how do we empower legal professionals to do more with their time? I mean, from the moment AI sort of came out in the form that we know it today with the large language models back in 2022, I think many looked to the legal space as one where it had the most to gain from this, from frankly, revolutionizing technology. And my own background started out in engineering. Before that, I was a professional video game player, so competitive from a young age and spent a short time at McKinsey in a couple of other startups backed by Y Combinator, a really short time in VC that, you know, it's always helpful to know what the other side is thinking when you go into those types of conversations. But frankly, for the last couple of years, I haven't thought a lot about other things than legal tech. Okay, no, that's a really good overview. So I looked at the press coverage of Lagora. So you obviously did this, you announced the $80 million fundraise last month. And some of the coverage referred to you as a red hot startup. And then in my intro, I called you one of the hottest startups in Europe at the moment. I mean, am I overhyping you? Or I mean, do you feel as though that there's kind of more eyes on Lagora than ever before? Do you feel not to big yourself up too much, but do you feel as though you're a hot startup right now? And if so, how does that kind of manifest manifest itself? Can you talk about that? Well, I mean, it's hard to sort of clap yourself on the shoulder too much. But I do think we're doing extraordinarily well. And the reason I say that is not only that we're experiencing sort of infinite customer demand, right? Like there's more firms and legal teams to serve than it feels like we have time to. So we're just scaling as quickly as we possibly can. And at the same time, I mean, we're receiving hundreds of applicants a day who want to come and join this rocket ship. We quickly analyzed some of those statistics. And I think we figured out that we had a much, much lower acceptance rate than Harvard on folks applying for a job here. And so, of course, I feel like we're doing incredibly well. But in many ways, I feel like we're also just getting started. We've done zero to one. Now it's 10, sort of one to 10, and then 10 to 100. So there's 100 applicants a day. Are you personally getting back? I guess someone else is getting back to that. I presume you're not getting back to them all personally, are you? Oh, well, I actually interview every single candidate who joins Ligora. But we, of course, have a fantastic recruiting team who sort of, yeah. Okay. No, that's the top of the funnel. Yeah. So just before we dig more deeper into Ligora, I just wanted to, obviously, you're an AI company. I just wanted to ask you about a couple of kind of big AI news stories recently, which are kind of longstanding AI stories. So the CEO of Anthropic, Dario Amadeo, recently, I don't know if you saw this, but it's got a lot of press coverage. You gave an interview with Axios. And he said that AI could wipe out half of all entry white collar jobs and spike unemployment to 10 and 20% in the next one to five years. And he said that AI companies and government need to stop sugarcoating what's coming. And kind of, you know, bluntly saying it's a possible mass elimination of jobs across technology. And I guess some of the criticism has been that he's kind of overhyped it because he's in his best interest. I mean, do you think that's true or do you think he's kind of overreacting the situation? I think the difficult part is, I feel like AI is in one part completely overhyped and in another part completely underhyped, if that makes sense. It kind of exists in this Schrodinger's equation, where on the one hand side, it can do incredible things. It can write an application almost from scratch. It can produce a contract that is extremely good. It can review hundreds or thousands of agreements, I mean, much, much faster and at a similar accuracy rate to a human. But at the same time, it falls short on very simple tasks from time to time. And so, you know, of course, it's, you know, maybe in his interest to sort of maybe overpromise and, you know, they're fighting a big back battle with open AI. But I think when I look at the application application layer and how these tools are fundamentally being adopted on a daily basis, I think last year, it was very much about, okay, we see efficiency gains, we see we're experimenting. But now I think the conversation has slightly shifted to also just how do we produce higher quality work? And how do we produce more work and less sort of more sophisticated work? Because at the end of the day, at least in legal, there's some fundamental sort of simple tasks, but the more complex one still requires a lot of human judgment and they require a lot of context around the legal system and the interplay between, you know, that contract is referencing another contract and how could that play out in a potential future scenario? And for all those things, I mean, AI just isn't there yet. Okay, no, that's great. And it might be slightly unfair, because I realize you don't work for one of these large language model companies, but you are kind of kind of deeply involved in this area. And there's also a lot to talk about kind of superintelligence and AGI, but there seems to be, from what you're saying there, you're talking about still having humans in the loop and there's basic things that it can't do. So that this superintelligence would be, in your opinion, you know, a considerable amount of time away. Well, I don't know if it's a considerable time away in the sense, but it depends a lot on how you sort of define superintelligence. I think if you define it as something as, can this system make money on its own, I think we'll be there quite shortly, frankly, like, we're already writing something like 70% of all our code with AI at Legora today. All of our marketing material is being reviewed by AI. We're using AI in our own legal team, and actually in our sales team, right, like they are reviewing contracts using Legora much, much faster and puts less pressure on our legal team to sort of, you know, turn those things around quicker. And so I do really think that professionals who start leaning in and who takes time to not only understand how this technology is going to impact their day to day, but also kind of understands like, okay, what's happening with the broader capabilities of these systems, they will do much, much better in the future to come. And we're also seeing a big difference between lawyers and associates who are coming out of college and university today, and they're much better prepared for that future because they've worked with AI themselves in the last years, and in their own sort of, you know, day to day lives, like they're much more tech savvy. Okay, no, that's a great answer. But I mean, you are, I mean, at the start, you mentioned you're getting hundreds of applicants for jobs. I mean, I presumably you ask, you do still need human recruits, but perhaps, well, I mean, there's lots of talk, isn't there, about kind of people recruiting less, but you are still openly recruiting people. We're recruiting a lot. And of course, you might use systems to do better, better initial screenings, or, you know, automated outreach, or, you know, like top of funnel activity. But at the end of the day, I mean, joining a company is also a very large decision. And it's a very human decision. Do you, like, do you get a good vibe of the people that you work with? Like, that frankly, is my biggest, I think, job contribution to building Legora. It's, who do we bring in? And who's going to take this company after we've done the zero to one, and make the one to 100, right? And I think building that team is something when I, when we started the company, I maybe didn't think enough about. But now I just see that as my most crucial job. Yeah, I think I think I listened to a podcast when you talked about one of the great upsides of being the boss of your own company is kind of picking the people you work with. And perhaps you hadn't had that in, was it a university, you talked about it? But I mean, that's a good segue. So I mean, it's amazing that you only launched in 2023. But I think in when I listened, you're one of these people who's always wanted to launch a start at them. So I think I, I did, right? And the way that I see it, it's, it's about competing on the largest scene in the world. And we are just fundamentally a group of extremely ambitious people. And for us to start in a small sort of back office here in Stockholm. And now to, you know, build this company on a global stage. That has been the most rewarding thing that I've ever spent my time on. Okay, and you're three co founders. So let me get the pronunciation right. Cigar Labor and August, I can't pronounce you so now, S.C.S. is it. So just talk about how, how, how, how, how, how the three of you, how do three of you met to be great? Yeah, I think me and August actually met over a volleyball game here in Stockholm. And I think immediately hit it off around a, a mutual appreciation for machine learning technology and, you know, all things tech. And I was introduced to Cigar, because August and Cigar were sharing a flat back in college. And, you know, when they started fiddling around with this idea, the, the sort of sophistication of the early models, like Bert just wasn't there to, you know, solve interesting problems. And so it was a lot of bashing our own heads against the walls, you know, before we got access to systems like GPT 3.5, that really just changed the game in terms of what you could accomplish. Okay. And I think so you went through, you went through YC and I think I've been doing my research. I heard you tell us quite an amusing anecdote about how was there an English interviewer there who laughed at your kind of, or you were at the speed on all the legal knowledge. Is that right? Well, I think when we did the first interview, we have just been running at this for, for a week or two. And, you know, as with, with a very common YC advice is launch quickly. And so I think we, we tried to launch quickly and then quickly figured out that there were some more learnings that we had to take with us before we, you know, finalize what, what we were going to build and how we were going to approach the market. So we quickly figured that out over the next month or so. And then we reapplied very early for the winter 24 batch in in July, got our first interview in August and then got accepted, which also meant that we had a lot of time before the YC batch kicked off. So we were one of the companies, I think, coming in with the most traction into the batch already, which was a lot of fun, but also a slightly different experience than I think many of the other companies had when they were still figuring out what to work on and so on. Okay, so just, let's just, just so we've got our heads clear. So your platform is obviously built on top of large language models. And so the product is, is effectively like you're researching, reviewing and drafting contracts, is that right? And I think you've hunted it. So you've got, just tell us, you're working with, I've got figures here, lawyers across 250 firms and legal teams in 20 markets. Can you give a bit more detail, maybe? Yeah, for sure. So, I mean, as you think about the sort of tasks that we can look to solve, it's not all about contracts, but if you talk about reviewing, for instance, there's, there's, you know, tasks like due diligence, when you have a lot of contracts that you need to review, and you need to find errors or risks, and then you need to package them and make the client aware of, of the underlying issues in the sort of collection of a company that they're buying. But you could also just say, Hey, we need to review a lot of contracts that come in on a daily basis because our sales team is growing so quickly. On the drafting piece, I mean, most of the times you have sort of existing precedent. So existing previous good examples that you can leverage to do drafting at a much faster pace and at a much better quality, frankly. And so it's a lot about how do you put LLMs on top of some of the existing structures and systems that also exist in these organizations. And when you talk about research, that's a sort of separate thing to, to contracting where you say, okay, we've got all this case law, we've got all this legislation, now we need to interpret it, and we need to synthesize it in a way so that it makes sense for the question or the problem that we're trying to solve for. And so you can leverage LLMs in many different ways across all these tasks. And what we quickly find, found when we arrived in the legal industry is that there's just a lot of software that isn't, you know, very nice to put it nicely, right. And so it's not very inspiring, it's not very exciting. And since LLMs are now powering many of these systems, it also makes a lot of sense to bring them all together to create a much more unified experience. Yeah, that's really well explained. So which models are you built on top of? And how easy is it to switch between models? Oh, extremely easy. So from a sort of product perspective, I've said since day one that we should be completely model agnostic in terms of what we work with. So working with the GPT models, the cloud models, and the Gemini models, but also open source. You never know what's going to be the best model tomorrow. And so you need to be ready to make a 180 to upgrade the systems. So from an architectural perspective, that creates a lot of exciting challenges where you for one need to be model agnostic in terms of which one you work with, but you also need to build a very good framework for how you test the quality of new models. So how well will they perform on the tasks that we solve? So we've invested a very significant amount of time, energy and money into building large scale evals, which basically means input and output pairs, where we can use both human judgments and LLM as a judge methods for determining how good the system is on, you know, different tasks. What about Chinese models? What about those? Oh, most big law firms that we work with would not like to leverage Chinese models. What confidentiality to do with? Well, they want to be able to very clearly explain to their clients, which in turn might be governments, large financial institutions, and these types of organizations I think are not yet like AI is moving really fast, and they're not really used to moving at that pace. And so to throw a Chinese model into the mix just gets a bit tricky. Okay, no, that's interesting. So obviously, given your lack of legal knowledge, how difficult was it to gain the trust of legal firms initially? And how much of a, I mean, you've got some of these blue chip legal clients, how much is a kind of word of mouth between them? If you do a good job with one, does it kind of, you know, is it kind of, is it quite a clubbable industry where they will tell other legal companies about you? Also, I think I very, I was very quick to create a strong baseline for myself. And at this point, I could probably call myself a hobby lawyer, if nothing else. And when it comes to, you know, building trust and getting to work with these very, you know, frankly, you know, high stakes organizations where there's a lot on the line, that takes time to build up, right? It's a continuous relationship where you show that we're going to make these promises, we're going to deliver on these promises. And at the end of the day, we're going to help you win. And so for sure, it's the case of if you do a good job with at one place that that can quickly spread. But more importantly, I think it's how do you meet these organizations where they are in their journey of adopting AI and rethinking their processes? And how do you best support them on the journey that is to come? Because I don't think it's the case of you buy solution X, and you're done, and you just check the, check the box on AI. This is a technology that is ever evolving. And so the way that it impacts the, the way that these businesses work, both in terms of the products that you deliver, the way that you price, the way that you structure your teams, that is ever changing. And so we, you know, fundamentally serve as a really trusted partner and a strategic partner to the clients and organizations that we work with. Okay, just just just on the models too, just one thing which I because today today I write about fintech and there's a lot of fintechs which are kind of deploying AI powered applications in kind of customer facing roles. The thing that I can't really for my own ignorance, if these models still hallucinate and still make errors, I mean, how can, I mean, is that not a problem in some of the, in how it's been deployed in Lugora's case in legal companies? I mean, other examples where errors are still made. So hallucinations, I think, were a big problem back in 2023 when we started out. And now there's, you know, there's a lot of techniques that makes it possible to reduce hallucinations. Both the models have, you know, become significantly better. I think people have become better users where they are providing more context to the models. So they do a better job. And, you know, techniques around guard railing and basing the or grounding the the answers in, you know, context that that is provided via documents or things like this have improved drastically. So I think, you know, hallucinations in themselves is frankly like a non-existent problem anymore. And maybe it comes up from time to time, but, but, but not really. I think it's more around now, how do you make sure that the answers that you're providing are fully, fully exhaustive and fully, fully exhaustive. And also, you know, everything that it provides is spot on because you don't want to overflow the context that you don't want to provide too much and everything that you provide has to be exactly the thing that the user wants. And so I think some of the biggest challenges now around how do you build large scale AI systems are not within the singular kind of input output, but also how do you chain together a genetic system where a genetic systems where you're using multiple tools and input of one step becomes the or the output of one step becomes the input of the next step. And that just vastly expands the universe of the type of things that you can do. And in those places, the evals and how you test and structure these use cases is kind of your your secret sauce in many ways. So the answers are exhaustive enough at the moment, that's like a work in progress then? Oh, of course. I mean, everything is a work in progress, right? I in many ways still think we've we've just scratched the surface of what's possible to do and accomplish with these models. And there's a lot of software still yet to be built. Right. Okay, Max. So I looked at some figures for your annual recurring revenues. And I've got, I think, 72. Well, I'll put it bluntly. Have you hit $100 million in ARR? Oh, we're not at 100 million yet. But I'm confident that we'll get there. Okay, but can you give me a figure about what where you are at the moment? Yeah, we're on our way to 100. Okay. So and you mentioned so you've got so when did you you're in you so you've got an office in New York and have you know? Yeah, at Union Square. Okay, so you've got offices in Sweden, New York and where else? In London. Okay, so I mean, and how do you I mean, I mean, that's pretty impressive, isn't it? I mean, and what's the headcount at the moment, too? Oh, I believe I saw the last number in Slack and it was at 104. Okay. So I mean, and how I mean, what will it be? I mean, that's, I mean, as soon as you're only launched in 2023, I guess that's enough offices for the time being that there won't be any plans immediate plans to have offices elsewhere. Well, I mean, we're always, you know, opportunistically looking. I believe that we're starting to run out of space in our Stockholm office soon. We thought we got a really huge space when we first arrived. But now with the amount of meetings and things going on, we might need to look for an upgrade. And how much you personally, how much of your time are you spending in the US in terms of, you know, the percentage of your, I mean, is that what you're looking for for new customers to? Oh, I mean, we're working very globally, right? And so just personally, it's a lot of travel these days. But yeah, New York, London, San Francisco, you at Stockholm. I was recently in in a pack as well. It's, it's really a global phenomenon in terms of, you know, how we're waking up to this new future, where we're all going to be working with AI in one shape or another. Okay. And what about, I mean, it's quite yours is quite a competitive area. You've got the likes of luminance, luminance is it? I mean, how do you, how are you kind of standing out from the competition? Right? Well, I think the biggest thing is the space is evolving very, very quickly. And I think we've not only proven but keep innovating in the space, and we keep pushing the boundaries of what's possible. And I think as we look to see how sort of how is the next generation of tools going to look like, we're building already for that future. And so, you know, a couple of examples would be when you think about building agentic systems, maybe don't just allow them to do kind of super standard forms in the sense of, you know, here's step one, here's step two, here's step three, but really allow the agents to work with different tools via, for instance, MCP, connect that to our customers databases, connect that to the web, connect that to other tools that they've that they're already leveraging, and build for where the puck is going and not where it is today. Okay. And finally, we're in June now, aren't we? I mean, you've kind of hinted at a lot of trouble. So just give us a kind of an overview of what you've got planned for the rest of the year then. Oh, for the rest of the year. I barely know what's happening for the rest of June. What have you got for June then? Yeah, I'm off to San Francisco next week. Then I believe I'm due for Finland, and then Amsterdam, and then London, back to Stockholm. And then I'm spending all of July in New York. Do you enjoy the traveling? Not as much as I enjoy actually doing the work. Right, okay. Okay, Max, I'm really that's it. I'm really appreciative of you. I realize you're really busy. I wish you all the best. It sounds like a really exciting starter. So that's the TechieU podcast.