Legora Reaches $100 Million ARR After Halting Sales to Rebuild System

Legora曾暂停销售以重构系统，现年度经常性收入达1亿美元 在企业级AI领域，行业专业知识和履历背景远不如纯粹的学习速度重要，因为直接深入用户并在内部构建专有评估体系，其发展速度超过了行业老牌企业和定制基础模型。 Max Junestrand回顾了Legora的发展历程：从最初被Y Combinator拒绝、在律所无窗会议室办公，到成长为增速最快的企业级AI软件公司之一。他详细阐述了一些反直觉的运营决策，例如在筹集3500万美元后冻结企业销售六个月以重构核心架构，以及在招聘时看重快速学习能力而非履历背景。在与YC合伙人Gustav Alströmer的炉边对谈中，Junestrand强调，深厚的领域专长不如强烈的学习意愿关键，并解释了Legora如何构建专有评估基准、重视对外传播，以及如何从被动提示词扩展到自主后台智能体。 自2024年10月全面推出（GA）以来，Legora在18个月内将ARR从100万美元扩大至1亿美元，覆盖了全球超过3%的法律行业人群。 (0:07) 尽管账面拥有3500万美元资金，该公司仍冻结了主动销售长达六个月，以从根本上重构其产品，确保为注重风险规避的企业律所提供零延迟的可靠性。 (13:17) Legora并未耗资数千万美元预训练专有基础模型，而是完全基于前沿基础模型构建，专注于企业工作流、合规性与分发渠道。 (29:14) Legora开发并开源了其专有基准Legora Bench，发现xAI的Grok在结合推理成本考量时，在法律推理能力方面名列前茅。 (34:09)

Legora co-founder Max Junestrand argues that rapid domain immersion and extreme execution velocity outweigh native industry credentials when building vertical enterprise AI. Scaling the legal assistant from $1 million to $100 million in ARR required embedding directly with lawyers, freezing sales to rebuild product reliability, and rejecting corporate modesty in favor of an intense performance culture. Overcoming Domain Deficits Through Extreme Immersion (3:16) Founding Legora without a legal background initially led to an outright Y Combinator rejection when the team could not identify specific legal customer segments. Rather than finding a legally credentialed co-founder, Junestrand and his team spent months cold-emailing practicing attorneys, offering to pay their standard hourly billing rates just to interview them over lunch. They subsequently convinced Nordic law firm Mannheimer Swartling to let them move into a windowless interior conference room, writing code side by side with active practitioners. Junestrand notes that while deep science verticals require academic specialization, enterprise software primarily demands an aggressive willingness to learn customer workflows firsthand. Because law firms operate in an environment where mistakes carry catastrophic professional penalties, lawyers hold near-zero tolerance for hallucinations. Embedding directly inside the firm allowed the founders to discover real operational friction points and build sufficient domain credibility to be accepted into YC two months after their initial rejection. Engineering Reliability and Freezing Commercial Growth (13:17) After securing a $9.51 million seed round led by Benchmark and a pre-empted Series A from Redpoint, Legora held $35 million in capital with a team of only ten employees. The company briefly earned more from bank interest than from operational software revenue. Facing fragmented feature sets born from internal democratic voting, Junestrand and his board made the counterintuitive decision to freeze the company's entire sales motion for six months. Junestrand emphasizes that enterprise software vendors get only one opportunity to make an impression with institutional law firms; system crashes or latency under load permanently terminate accounts. During the sales pause, the engineering team codified the Legora Product Manifesto in October 2024, overhauling the underlying infrastructure from scratch to handle shifting agentic frameworks. This architectural stabilization allowed the company to out-compete single-feature point solutions and scale smoothly from $1 million to $100 million in ARR. Model Neutrality, Proprietary Evals, and Proactive Agents (29:14) Rather than spending tens of millions of dollars fine-tuning proprietary base models—a capital-intensive path Junestrand observes failed at projects like BloombergGPT—Legora chose to build exclusively on general foundation models, betting that frontier labs would continuously outpace vertical-specific models. To handle diverging enterprise demands for high-reasoning accuracy versus low-cost processing, Legora developed its own testing benchmark, Legora Bench, employing in-house attorneys to continuously evaluate model performance on real-world legal edge cases. Junestrand projects that enterprise legal software is transitioning from reactive tools to proactive agents. Instead of waiting for prompt instructions from an attorney, background agents monitor enterprise data feeds, evaluate incoming agreements, execute straightforward contracts autonomously, and summarize complex M&A data rooms. Human intervention is reserved strictly for high-risk exceptions, multiplying individual attorney throughput. Building a High-Slope, Performance-Driven Culture (17:56) Scaling out of Sweden required actively dismantling *Jantelagen* (the Law of Jante), the Nordic cultural convention that discourages standing out or presuming personal superiority. To counter this, Legora instituted high-intensity operational values under the acronym LFG (Lean in, Fight for excellence, Grow together) alongside the Swedish athletic concept of *blodsmak* ("taste of blood"). Hiring explicitly prioritizes learning trajectory over corporate pedigree; for instance, Legora's top commercial performer is a 23-year-old college dropout who closed more than $10 million in ARR without prior sales experience. To preserve cultural alignment as headcount surpassed 750 employees, all international hires must complete onboarding in Stockholm. Operationally, Junestrand treats executive roles as requiring quarterly re-qualification, as running a business at $100 million ARR demands fundamentally different capabilities than managing early cohorts. This aggressive stance is balanced with high-leverage storytelling, demonstrated by unconventional brand campaigns featuring Hollywood actor Jude Law that broadened company visibility beyond traditional legal circles. With active users representing roughly 3% of the global legal profession, Junestrand estimates Legora has built only a fraction of its eventual software portfolio. The broader enterprise question remains whether vertical software providers can continuously re-qualify their architectures and leadership fast enough to stay ahead as foundational AI reasoning costs and capabilities shift beneath them.

[0:07] Max Junestrand: Legora is the agentic operating system for lawyers, and it handles complex legal work from start to finish so lawyers can achieve more than ever before. [0:07] If Legora went away tomorrow, I'd just go back to coaching high school basketball. [6:09] Max Junestrand: In law, you are not paid when things go right; you are punished when things go wrong. [7:22] Max Junestrand: The day after, we did not work. But the day after that, we got back to building, and two months later we got accepted under a new name and with a different platform. We had done our homework. We had done the work. [20:04] Max Junestrand: It's very hard to build one of the fastest growing enterprise companies in the world if you think that you aren't somebody [23:16] Max Junestrand: You shouldn't wait around for those big moments. You have to love the hustle. [26:55] Max Junestrand: You can just do things. Legora and myself is living proof of that. [29:14] Max Junestrand: We are like ChatGPT, but we are compliant in Europe. [37:29] Max Junestrand: The amount of things you learn is a function of the amount of discomfort that you are willing to endure. [45:04] Max Junestrand: I don't feel like I'm the boss. I feel like I'm a peer in the team that is running the company. And, you know, our CFO sometimes challenges our marketing team to be more ambitious... We're all just so forward-leaning and it's just so fun. [50:00] Max Junestrand: How quickly can we turn this around and delight the customer?

[0:07] This is a much warmer welcome than what [0:11] we had coming into Y Combinator the [0:14] first time. Legora's YC journey started [0:18] with a rejection, but more on that story [0:22] later. [0:26] In YC they teach you something and it's [0:28] called the two sentence description. It [0:32] is describe your company in two [0:34] sentences. [0:36] Legora is the agentic operating system [0:38] for lawyers and it handles complex legal [0:41] work from start to finish so lawyers can [0:44] achieve more than ever before. And over [0:47] 3% of all the world's lawyers are now [0:50] active users of Lora. You will learn [0:54] more about this when you get accepted [0:55] into YC. [0:58] When we started looking at the legal [1:00] industry, it felt like one of the [1:03] biggest opportunities to apply large [1:06] language models to solve real problems. [1:10] And the software that we were up against [1:12] looked like it had built in the '9s. [1:16] As software engineers, we have been [1:18] incredibly spoiled with great software [1:21] because we love to build for each other [1:24] and for ourselves. But in the legal [1:26] industry, this is what the software used [1:29] to look like and what we were up against [1:31] when we started out. [1:34] Now, Lor has grown into a global brand, [1:37] putting celebrities like Jude Law at the [1:40] New York Wall Street. And still I feel [1:44] like we are only getting started. My [1:47] most favorite Slack channel in Lagora is [1:50] customer love. It is where our customers [1:54] post things that they sort of truly feel [1:57] like they couldn't have achieved without [1:59] our system. And we mine many of our [2:01] sales calls for quotes like these. And I [2:03] wanted to share just a couple with you [2:05] because I think at the end of the day, [2:07] this is what it's all about. [2:09] A parent asked me at one of the [2:11] tournaments how I juggle work and all of [2:14] the golf travel. One word, Lagora. [2:18] Never has the path between an idea and [2:21] execute been so short. Never, man. If [2:25] Legora went away tomorrow, I just go [2:27] back to coaching high school basketball. [2:31] When Lora went into YC, we were [2:34] questioned about the idea of even [2:37] selling and working in the legal [2:39] industry. It is such a conservative [2:41] business that had never been sort of uh [2:46] lenient to software before. But from the [2:50] time that we went into GA of October [2:52] 2024 until the last end of quarter, [2:55] we've grown from 1 to 100 million in ARR [2:58] and starting from just three engineers [3:00] in Sweden to a company of over 750 [3:03] people all over the world. This is [3:06] awesome. And the fact that we've gotten [3:08] to do this starting from Europe, I think [3:10] just adds another flare in that hat. [3:16] But the Lora story almost didn't happen [3:22] and it actually didn't start with me. In [3:26] fact, it started almost like the great [3:28] beginning of a joke. And so a lawyer, a [3:33] physicist, an engineer, and a [3:35] psychologist back in 2020, [3:39] before GPT, before AI was cool, started [3:42] a company called Judelica. [3:45] And what they observed was that law [3:49] school students were going into their [3:51] internships basically to summarize court [3:54] cases. And with the early models from [3:57] Google called BERT, they started to [3:59] explore what this space could be and [4:02] what was possible to build. [4:06] My story started in the Swedish [4:08] archipelago [4:10] at an island which looks something like [4:12] this. and I met my two co-founders at a [4:16] volleyball game. [4:19] I think startup founder stories can come [4:22] from many different places and this [4:24] definitely wasn't one, you know, high [4:26] odds uh of a place to start. [4:30] But when I met August and SIGA, they [4:32] showed me a demo of the product they had [4:35] built. And our group chat actually [4:38] looked exactly like this. They gave me a [4:41] demo of the project and with GPT3.5 [4:46] they had built a very simple product [4:49] that could explain what a stock option [4:52] agreement really meant. And so we [4:55] started with that and I said this is a [4:58] super cool demo, a super cool project [5:00] and my co-founder Sega said I'm glad you [5:02] want to help out. And I ended up helping [5:04] out a little bit more than just that. [5:09] When GPT 3.5 came, that was the internet [5:13] moment of our generation. [5:16] I dropped out of college. I never [5:19] finished my master thesis because the [5:21] opportunity cost of not building had [5:24] become too large. But when we started [5:27] out, it wasn't obvious that we were [5:29] going to be a success. And so to learn [5:33] about what the legal industry was, what [5:36] it meant, how lawyers actually worked, [5:40] we came up with the ingenious plan of [5:43] cold emailing lawyers on their emails [5:46] that we could find on their websites and [5:48] writing to them on LinkedIn asking for a [5:51] lunch where we could learn more about [5:53] their different practice areas. And we [5:55] offered to pay their hourly fee to have [5:58] lunch with us. And many of them took us [6:01] up on that offer. But fortunately for [6:03] us, many of them didn't make us pay for [6:05] the hour and many of them even paid for [6:07] the lunch. [6:09] And once we had gotten, you know, our [6:12] feet under ourselves, we approached one [6:15] of the largest law firms in the Nordics [6:17] named Manheimr Swartling. But just two [6:21] years before that, their managing [6:23] partner had gone on TV and said that AI [6:27] back when they were starting to use it [6:29] was really more artificial than [6:32] intelligence. And so we had to work [6:37] really hard to change the perception of [6:41] what both technology and what AI could [6:44] be in the area of law. [6:47] In law, you are not paid when things go [6:51] right, you are punished when things go [6:53] wrong. And many of the initial AI [6:56] solutions in the space had required [6:59] immense amount of uh training. You had [7:02] to sort of bring a lot of uh use cases [7:04] and examples just to get very very [7:07] little results. And when GPT 3.5 came, [7:10] that really changed the story. And so [7:12] with Manheimmer and Swartling, we ended [7:15] up moving into uh into their offices and [7:18] moving and working really close with [7:19] them. [7:22] And as I said in the beginning, the um [7:24] the warm welcome here was quite [7:26] different from the first time we [7:29] interacted with YC. We applied in May of [7:33] 2023 [7:35] with uh with a promise of being able to [7:38] query all legal documents using large [7:41] language models. And when we came to our [7:44] interview, [7:45] one of the first questions that we got [7:48] asked was what type of lawyers are you [7:52] serving? And you know, not very [7:55] knowledgeable about the legal space yet. [7:58] Our response was, "What do you mean? Are [8:01] there different type of lawyers?" And [8:04] that did not land very well with the YC [8:06] partners at the time. And so I actually [8:09] remember very painfully um that Tom [8:12] Bloomfield started laughing in the [8:14] interview when we gave that response and [8:17] we knew that we were cooked. [8:21] But that also gave us the perseverance [8:24] and the [8:26] um the drive to continue. And so with a [8:30] lot of let down um the day after we did [8:33] not work but the day after that we got [8:35] back to building and two months later we [8:39] got accepted under a new name and with a [8:43] different platform. We had done our [8:45] homework. We had done the work and from [8:50] Sweden the idea of getting accepted into [8:54] Y Combinator felt like we were going to [8:58] Mount Olympus. We were going to dine [9:02] with the gods and we were going to, you [9:04] know, get to work with other startup [9:06] founders who came from all of the great [9:08] tech companies that we knew of. And once [9:11] we got to Y Combinator, we realized that [9:15] we actually knew a lot more than we [9:17] thought we did. And this is one of my [9:19] favorite pictures. And it is the [9:22] screenshot I took uh after we did the [9:25] video call after our second YC interview [9:29] when we knew that things had gone a lot [9:32] better than the first time. [9:36] So after we got accepted into Y [9:39] Combinator, we moved into this law firm [9:42] I mentioned earlier. Manheimer [9:44] Schwartley was the biggest law firm in [9:46] the Nordics, but they had very uh sort [9:50] of no real clue about how to work with [9:53] software and how to work with [9:55] technology. [9:56] And we ended up getting a conference [9:59] room with no real windows [10:02] where the AC would turn off at about [10:05] 5:00 p.m., but it was really close to [10:08] the Coke fridge. And so at about 6 p.m. [10:12] every day, one of the engineers would go [10:14] up, grab the door, and start waving it [10:16] back and forth to get more oxygen into [10:18] the room. And those conference chairs [10:22] still made an imprint on all of our [10:25] backs. I think 3 years later [10:28] and me in the car there with two more [10:31] screens is when we make the move from [10:33] the conf from the conference uh room of [10:35] a law firm to our first real office. [10:39] And I think one of the key takeaways [10:41] from this entire experience is that [10:43] first off, we quickly learned not to [10:46] take no for an answer. [10:49] But also the power of simply reaching [10:52] out and asking for help. I think one big [10:57] lesson from these first three years of [10:59] building Legora has been that it is very [11:01] powerful to be a person and to be a [11:04] company that others want to see succeed [11:07] and that others want to help. [11:11] And so we came to San Francisco, we got [11:16] a room uh at an Airbnb up in Bernal [11:18] Heights and we started building. [11:22] And [11:24] during YC, we grew from zero to a [11:28] million dollars in ARR. [11:30] And [11:32] we also learned the hard way that [11:34] working across San Francisco and Europe [11:37] took a big toll on the team. I couldn't [11:40] find a picture of this, but I bought a [11:43] influencer ring light that I put on my [11:46] laptop. And the reason I had that ring [11:49] light was because I was doing sales [11:51] calls between 1:00 a.m. in the morning [11:54] to 10:00 a.m. in the morning every [11:56] single day in San Francisco to Europe [11:59] because that was where our initial [12:01] initial customer base was. And after a [12:06] rocket trajectory during YC, which now [12:09] two years later is slow compared to what [12:12] some of the companies are doing, we set [12:15] off to fundra. And this was my first [12:18] time fundraising. [12:20] I had no idea what to expect, but we had [12:23] lined up 80 investor meetings in a week [12:26] and a half. And we had European [12:29] investors, American investors. And [12:34] the main takeaway from that experience [12:36] for me was that you need to become a [12:39] really good storyteller [12:42] in order to drive interest and to get [12:44] people on your team. [12:46] It was also a very different experience [12:49] talking to some of the European [12:50] investors versus some of the American [12:52] investors. [12:54] I feel like in the US we think more [12:56] about what can go right rather than what [13:00] could possibly go wrong. And we ended up [13:03] working with a venture firm called [13:05] Benchmark. This is a picture of Chaan, [13:08] our partner, um when we had signed the [13:11] term sheet. And I remember negotiating [13:13] that term sheet on that um on his desk [13:17] with a computer opened going on decimals [13:22] in the Excel to determine the exact [13:24] number of of shares and they ended up [13:27] investing $9.51 [13:30] million. [13:32] Now, following that investment, [13:37] Redpoint pre-letter series A just three [13:39] weeks later, and [13:42] we started to run into a challenge. [13:45] We had about $35 million on the bank. We [13:50] were 10 people in the team. [13:53] And there was actually a month following [13:55] that that we made more money from [13:58] interest rates on those 35 million than [14:02] we did from customers. [14:04] And that is not a very good sign when [14:06] you have become a bank. And going into [14:10] our first board meeting, we had Chathan [14:13] and we had Redpoint and it was me and my [14:15] co-founders. And we made a very hard [14:18] decision, which was to freeze our sales [14:22] motion. Because when you work with [14:25] lawyers, you only really get one chance [14:28] to get it right. [14:30] If you show up and the product doesn't [14:33] work or if the [14:37] um you know sort of lag in the system is [14:40] too high or if the system goes down when [14:43] there's too much traffic on your ashure [14:45] instance you are toast. And after that [14:50] six-month sales sales freeze was when we [14:52] really started ripping. And that's where [14:54] you see that curve from 1 million to 100 [14:57] million really starting. And what [15:00] happened at the start of that [15:03] was, you know, turning from a small [15:07] startup looking for PMF to really [15:11] establishing what our business and what [15:14] our product was going to be. [15:18] In the very very early days at Leia, so [15:21] even Pillegora, [15:24] the way that we would decide what to [15:25] build was by voting. We would make [15:29] democratic votes in the entire team. And [15:34] for anyone who has built software [15:35] before, you know that if you have too [15:37] many chefs in the kitchen, that [15:40] typically does not make one very good [15:42] dish. And that was exactly what was [15:44] going on. We were working on too many [15:47] features. [15:48] We had to rebuild everything for the [15:51] product launch that we were going to [15:52] have after the summer, after the sales [15:54] freeze. And we had to build a platform [15:58] where it was adaptable to foundations [16:01] that were constantly changing both with [16:04] the underlying models but also with the [16:06] frameworks that we were using to build [16:08] agentic workflows like lang chain etc. [16:11] And so in October of 2024, [16:14] we wrote down what we would call the [16:16] Leia product manifesto. This was [16:19] basically gathering everything we had [16:21] learned, putting it into one very simple [16:24] document that we shared with the entire [16:27] 25 uh strong person team back then and [16:30] that we would rally the troops around. [16:33] At the time, we were doing about $1.3 [16:36] million in annual reoccurring revenue, [16:39] and many of our competitors were doing [16:42] 10 times that [16:44] with products that only had one of these [16:47] features. And I also think that's a [16:50] function of having started in Sweden. [16:54] The market is not big enough to house a [16:57] real unicorn or a decorn or something [17:00] even bigger than that. And so one of the [17:02] earliest questions we got from YC was [17:04] when will you move to the United States? [17:07] When will you make the jump? And it was [17:10] off the back of this product refocus [17:13] that we [17:15] were able to not only compete out the [17:17] competition on product but build enough [17:20] momentum to make the jump over to the US [17:22] and start to compete for the biggest [17:24] logos in our markets. And I think as as [17:28] engineers and as tech people, we often [17:31] overemphasize the reliance on product [17:34] and on the tech. But if there's anything [17:37] I've learned in these past three years, [17:40] it's that it's all about the people. [17:44] And building product is one thing, but [17:46] building a company is a very different [17:49] thing. And we made many early mistakes [17:53] around this. [17:56] One of the hiring patterns that we had [17:58] to unlearn was looking for fancy logos [18:03] on resumeumés or what we like to call y [18:06] inter like y intercept. If you think [18:09] about somebody's skill curve, it might [18:12] start out really high, but if they don't [18:14] have a good trajectory upwards, they are [18:16] going to have a really hard time working [18:19] in a company that is scaling [18:21] exponentially. [18:23] And so instead, we started to look for [18:25] people like us who wanted to work [18:28] insanely hard with a very high growth [18:31] potential. And we doubled down on that [18:33] talent over and over again. [18:37] Our top seller today is 23 years old and [18:42] he sold over more than $10 million worth [18:45] of Legora. And he had no sales [18:48] background. He jumped out of university [18:52] and by continuously doubling down on [18:54] that type of talent, I think we have [18:56] built a very unique culture at Legora. [19:00] And one of those things has been about [19:02] bringing and building cultural fit. [19:07] Up until the point where we were around [19:09] 500 people, I interviewed every single [19:12] candidate outside of engineering. [19:15] Now I interview every director and [19:17] upwards. But what we have done is we [19:20] have set the cultural foundation in the [19:23] business where [19:26] it self- selects for itself. [19:29] We have three values at Lora. [19:32] Lean in, fight for excellence, and grow [19:35] together. And together they make LFG. [19:38] Let's go. Um [19:42] that also sends a signal to leaders and [19:45] other and others who join the company [19:47] who are coming from more perhaps [19:49] established backgrounds where joining a [19:52] company where the values um includes [19:56] profanity quickly sort of [20:00] tells them what this is about. [20:04] Another very complex cultural thing [20:07] starting a company in Scandinavia [20:10] is something called Yantan or the law of [20:15] yante. [20:16] This is um very foreign to our American [20:20] friends. [20:22] But in Swedish culture this basically [20:24] means that you should not think that you [20:28] are somebody or you are not more [20:31] important than anybody else. [20:34] and your ideas are not better than [20:37] anybody else's. [20:39] And I think a part of this [20:44] humility is good and it supports a [20:49] culture where the best ideas win and [20:53] where people earlier in their careers [20:56] feel empowered to speak up. [21:00] But it's very hard to build one of the [21:04] fastest growing enterprise companies in [21:06] the world if you think that you aren't [21:08] somebody or if you don't think that you [21:10] are better than anybody else. And so we [21:14] have had to really create a a nice [21:16] cocktail of US, European and Asian [21:21] culture all in one to create a global [21:24] company that is legor today. [21:27] And so I really think that building a [21:30] strong company culture is the best [21:33] guarantee to attract and keep the best [21:37] people. [21:40] Another way that company culture gets [21:43] expressed at Lora [21:45] is through articles like this one that [21:48] ended up in sifted. [21:51] For us [21:53] there is only winning. Everything else [21:57] is losing in a winner takes all market [22:00] or winner takes most market and the [22:03] company has to run [22:07] like there is no place for number two. [22:11] I think it actually would be fair to say [22:13] that when Legora was founded, it really [22:16] had no reason to exist. We were one of [22:19] many legal AI companies. [22:22] But [22:24] because of this mentality and because of [22:27] not feeling like we're safe even though [22:30] we have achieved extraordinary numbers [22:32] by any normal measures, we continue [22:35] acting like professional swimmers who [22:38] are looking down in our lane while the [22:41] competition is looking sideways at what [22:43] we're doing. And at Lora we have had to [22:46] build a culture to really support that [22:49] ambition. [22:51] So starting you know years behind some [22:54] of our biggest competition and [22:57] having to build this intensity in a [23:01] country where most people want to go for [23:04] fica at 300 p.m. which means having a [23:07] cinnamon bun and a coffee or you know go [23:10] home at 5. Um, one of the ways that that [23:14] has been done is [23:16] to [23:18] create cultural uh, values and cultural [23:23] moments that really distinguish us from [23:26] many others. And so in Swedish there is [23:29] a saying um, you can wake up with a [23:31] taste of blood because you're so excited [23:33] about something. and I was doing an [23:35] article in Swedish and I used that [23:37] saying and I didn't know that the uh [23:40] interviewer was going to release that [23:42] article in English and so when this came [23:44] out it got lots of fun reactions. Uh one [23:47] of them was is Legora made up of [23:49] vampires. Uh the second one was uh you [23:53] guys need to floss better. um many good [23:56] ideas, but this has now turned into the [23:59] word blue smok internally, which means [24:02] that and we just keep going at it. And I [24:04] think part of having our HQ in Stockholm [24:09] has worked to help us keep this. [24:13] Everybody globally from the company [24:15] makes their onboarding in Stockholm, [24:18] which is a lot nicer in summer than in [24:21] the winter. But that has created the [24:25] same vibe in any Legora office [24:28] regardless of where you walk in all over [24:30] the world. And I think one great example [24:33] of this was we had a a a very big law [24:37] firm in the US that our competition was [24:41] working with. But that law firm had a [24:45] specific problem around being able to [24:47] draft certain types of documents based [24:50] on incoming term sheets. And our [24:53] competition had promised that they were [24:55] going to solve that problem for them. [24:57] And they hadn't. And so one of our [25:00] US-based engineers, legal engineers, [25:02] lawyers, learned about this use case, [25:05] got on a flight to Stockholm, spent a [25:08] week with our engineers, solved the [25:10] problem in Sweden, and then went back to [25:12] the US and delivered it to the customer. [25:15] And I think that um Swedish base has [25:19] become one of Legora's real superpowers. [25:26] And so to round things off before we [25:28] jump into the Q&A, [25:30] I'm sharing this stage this weekend with [25:34] people who have built things at a scale [25:36] that I haven't yet gotten close to. And [25:41] so I will leave many of the big lessons [25:43] for them. But what I can tell you is [25:46] that three years in [25:50] is a little bit smaller than that. Most [25:52] of what actually mattered in the company [25:54] and in Lagora so far did not revolve [25:58] around signing a big client or closing a [26:03] new round of financing or open a new big [26:05] office in Manhattan. It happened in, you [26:09] know, a windowless room without much [26:11] oxygen or on a call on our way to the [26:15] airport, a Slack channel at 2:00 a.m. [26:18] trying to fix a bug before a big [26:20] presentation on Monday. And nobody puts [26:23] that in a pitch deck, but that is what [26:25] makes the company. And [26:29] you shouldn't wait around for those big [26:31] moments. You have to love the hustle. [26:33] And [26:37] I think if I can leave you with [26:38] anything, it's that even though we [26:41] started in a small law firm in Sweden, I [26:44] was born in the Swedish archipelago with [26:47] a class of 10 other people, I had I [26:50] thought I had a big plan for my life and [26:52] I I was going to uh go to Mckenzie after [26:55] school. But when GPT3.5 dropped in our [26:59] knees, we truly made that into our thing [27:02] and we captured what I believe to be our [27:05] generation's internet moment. And so you [27:07] can just do things. Um, Legora and [27:10] myself is living proof of that. [27:14] So with that, we're going to jump to Q&A [27:17] and I want to invite a very good friend, [27:19] our partner from Y Combinator, Gustav [27:22] Armstr, [27:28] [music] [27:33] [music] [27:34] we're really becoming professionals [27:36] [singing] at this now. This is not the [27:38] first time we're doing [music] this. [27:38] >> No, it's great. [27:40] >> It's good to see you. [27:41] >> You, too. It's good to see you. [27:43] >> Um, thanks for the talk. That was [27:45] awesome. [27:46] >> Really good. [27:47] >> Thank you. [27:47] >> And always thank you for coming back and [27:49] speaking to YC and to to founders here [27:52] at Startup School. [27:53] >> Of course. [27:54] >> To be back. [27:55] >> I want to ask the first question which [27:56] is um one that I've gotten from a lot of [27:58] founders who ask about you, which is you [28:01] guys are three founders or were three [28:03] founders who were not lawyers. [28:04] >> Yeah. [28:06] >> Uh and you started the best legal AI [28:08] company in the world. Does domain [28:10] expertise not matter that much anymore? [28:12] >> So there's actually a story of a Swedish [28:16] VC who was very bullish on doing the [28:18] preede of Leia. [28:20] >> Yeah. [28:20] >> But they ended up not doing it because [28:23] we did not have any lawyers in the team [28:25] and I met them at the airport. I met one [28:28] of the partners from there at the [28:29] airport quite recently and they were [28:31] very very bitter and Leia or Lora turned [28:35] out to be the lesson for them that [28:37] perhaps domain expertise is not what you [28:40] need to succeed. I think that you need a [28:42] willingness to learn about the market [28:45] which we did very very early by you know [28:49] exposing ourselves and spending lots of [28:52] time with lawyers like as much as we [28:53] possibly could. [28:54] >> Yeah. But uh I don't think you need [28:56] domain expertise. It probably you know [28:59] dependent on on the market that you're [29:00] going after. Like maybe if you're [29:02] building in quantum computing or in um [29:05] fusion that's a good idea but uh legal [29:07] was easy enough for us to pick up uh as [29:10] we went along. [29:12] >> Um [29:14] I spoke to some founders yesterday and [29:16] we spoke about sort of like the next [29:17] categories where the LMS are not [29:20] sufficiently good yet but they will be. [29:22] And it made me think of what you just [29:24] said when you were toying around with [29:25] Bard in the early days. Um, [29:29] what would you what gave you the [29:31] confidence to build a legal AI product [29:33] on a model that wasn't it wasn't good [29:35] enough? It was like good but not good [29:36] enough. And there's many other founders [29:38] probably in this audience who are [29:40] thinking about the same problem but in a [29:41] different domain maybe like I don't know [29:43] mechanical design or some some some [29:45] other part of um somewhere where the [29:48] models are just not good enough. [29:50] So I I think the sort of tribal [29:53] knowledge when we got started was that [29:55] uh you have to fine-tune a model. [29:57] >> Yeah. [29:57] >> Right. In 2023 2022 that was really like [30:00] the name of the game. I even remember I [30:02] think it was Bloomberg who spent you [30:05] know millions and millions and millions [30:07] of dollars building like a Bloomberg law [30:09] model. Our view was partly because we [30:13] didn't have enough money and partly [30:14] because we truly believe that to be [30:16] right um you know the models will keep [30:19] improving right like I'm sure Sam is [30:22] going to come up on stage and say that [30:23] later today. [30:24] >> Yes. [30:24] >> And so by just betting that the models [30:28] will keep improving and what you should [30:31] be focused on is how can you deliver [30:33] that the value that the models generate [30:35] to your market. That was the crux for [30:38] us. um even with the first generation of [30:40] of chat GPT that was unusable in law [30:45] partly because the model wasn't good [30:47] enough uh but partly because it actually [30:49] didn't support um you know private [30:52] conversations European data hosting I [30:56] remember the first sales I ever did we [30:59] basically said we are like chat GPT but [31:02] we are compliant in Europe like that was [31:05] the pitch h that was not a very strong [31:08] pitch Um, and you know, clearly ChachiP [31:10] would figure out how to do that later [31:11] on, but at the time it was good enough. [31:14] And so I think we recognized that you [31:16] don't have to build for the world over [31:19] here like you need to build for the [31:20] world today and maybe like one step [31:22] ahead. Yeah. [31:23] >> And it's kind of like this ladder where [31:27] uh you should maximize the opportunity [31:29] now. make sure that you get really good [31:31] relationships with your customers and so [31:33] that you can also have that conversation [31:35] with them because if everybody's betting [31:37] on this you know exponential uplift in [31:40] model capability uh the Lora platform [31:43] will continue delivering more and more [31:45] value [31:46] >> I think that we have built 1% of all the [31:49] software that we will build over the [31:51] company's history and [31:53] >> probably the the product will look very [31:56] different from what it what it does [31:58] today a few years down the line. [32:00] >> Yeah. Um, a lot of news on models this [32:03] week or last two weeks. [32:05] >> Um, [32:05] >> it's going to be like a great episode [32:07] for whoever makes the documentary. [32:08] >> I'm hoping that someone is making a [32:10] documentary right now about everything [32:12] that's going on. I'm not sure if someone [32:13] is, but I hope they are. Um, Open Weight [32:17] Small is uh very much um the topic right [32:21] now. And um when we spoke u on stage a [32:25] couple months ago, you talked about that [32:28] um you're not attached to one specific [32:30] model. You can work with many different [32:31] models. [32:32] >> Have um have you updated your view based [32:35] on this week? Um but like how do you [32:37] think about so like I'm sure there was a [32:39] moment like a year and a half ago when [32:41] it seemed like there was just one or two [32:42] models that were really sufficient to do [32:44] to power legor and now there seemed like [32:45] a lot more. [32:46] >> Yes. Which I think will be great. Yeah. [32:48] And and I mean you know clearly there [32:51] will be uh applications of different [32:54] models in different use cases and so if [32:57] you take customer support um you're [33:00] optimizing for you know speed of [33:02] turnaround um you know latency to the [33:04] customer and cost on you know solving a [33:08] particular ticket. Um I think Finn used [33:11] to charge like $1 per you know solved [33:14] customer ticket. In law, you actually [33:18] want the you know most amount of [33:20] intelligence quite often because the [33:23] fraction of token spend or software [33:26] spend compared to like you know human [33:29] expertise applied to the problem is [33:31] really tiny. [33:33] If you're solving you know complex [33:35] litigation, you're going to want to [33:37] throw the most brain that you can on [33:40] that problem. And so we have really sort [33:42] of two different camps. Like one group [33:45] of our users are asking for the ability [33:48] to loop really complex models to to work [33:53] on problems for a long period of time [33:54] like racking up big LLM costs and [33:57] another group of our users are saying [33:59] you know can we get access to open [34:00] source uh you know that is going to be [34:02] cheaper on our consumption and I think [34:04] the the answer is like somewhere in [34:06] between. Mhm. [34:07] >> Um [34:09] I think ultimately uh one of the core [34:11] IPs and muscles that I encourage as many [34:15] of you as possible to build is uh the [34:18] ability to evalu [34:21] use cases because that is the superpower [34:23] that then allows you to uh route things [34:26] effectively. Mhm. [34:28] >> We did a lot of that early on because we [34:31] hired a lot of lawyers into Lora [34:33] >> and we made part of their job customerf [34:35] facing but part of their job was also to [34:38] build out use cases that we would throw [34:40] our evals on and the the frontier of [34:44] like you know cost uh intelligence and [34:47] where that sort of uh frontier lies you [34:50] know yet to be seen. I think there's [34:53] also a compute problem where when we [34:56] lock in big contracts with the uh with [34:59] the labs and with our compute providers [35:01] that is actually a huge thing. We cannot [35:04] have our system go down when you know a [35:08] a number of percentage points of all the [35:09] lawyers in the world are working with it [35:12] and their clients rely on it. So [35:14] >> and and are there specific benchmarks [35:15] you look at for how um I'm sure you are [35:18] basically [35:18] >> well we we look at the Lora bench [35:22] >> so we actually announced and released [35:23] that on Friday last week and it has been [35:27] that has been work for for the past 3 [35:29] years and we've kept that uh internal up [35:32] until Friday and the reason for that was [35:34] it was basically open AI or or um or the [35:37] anthropic models and so it wasn't like [35:39] that interesting to publish benchmarks [35:41] on [35:42] >> and we we would only make those two [35:44] models available in our system and the [35:46] user could pick. Now the smoo board of [35:50] available options is a lot larger and [35:53] actually we found to our surprise that [35:55] SpaceX and Grock was one of the best [35:57] performant models especially given the [35:59] cost on our benchmarks [36:01] >> and we don't even have a a um have them [36:05] on our data processing agreement and so [36:08] we actually can't yet offer them to our [36:10] customers but now we will start to. And [36:12] do you see your customers having having [36:14] strong opinions about which models they [36:15] want to use or is it sort of like they [36:17] just want the best whatever is the best. [36:18] >> Some of them have they have a strong [36:20] especially like banks um you know big [36:23] law firms have a they don't want Chinese [36:27] models. [36:27] >> Yeah. Makes sense. [36:29] >> Um so [36:32] I want to go back to just a little bit [36:33] earlier like when you decided you want [36:35] to build a startup. A lot of people in [36:36] the audience here I I I've had two [36:38] dinners in the last two days with with [36:40] attendees at at Startup School. we [36:41] talked about when should you start a [36:43] startup? People are working at DeepMind, [36:45] people working at internships or jobs [36:47] right now. Often you're one or two years [36:48] into it. Um what would you say was the [36:53] trigger for you to say I'm just going to [36:54] do this and not do Mckenzie or the other [36:56] options that you had? [36:58] >> So um my my dad had two companies uh [37:01] during the com boom uh both of which [37:05] went to zero after raising too much [37:08] venture capital. Uh so maybe some [37:11] >> Has he been a good mentor for you? [37:13] >> Yes. Yeah, absolutely. Um and [37:17] >> how often do you speak about [37:18] >> Oh, we speak probably every day or every [37:20] other day. [37:21] >> Great. Yeah, about lots of stuff. Um [37:24] >> I used to now I'm like too tired to walk [37:27] home often. So I [laughter] would like [37:29] take the train or like take a cab, but I [37:31] used to walk home then I would call him [37:33] and that would be a lot of fun. And um [37:36] >> sometimes we call him senior adviser. Uh [37:38] we have actually one of our biggest [37:40] customer pitches. Um we we were [37:44] stretched for resources. So he came in [37:46] and did the PowerPoint for for that [37:48] client. [37:49] >> He knew everything at this point. [37:50] >> Yeah. Yeah. He knows most most of it. Um [37:54] so he had two companies in the both went [37:57] to zero and then he had a boat business [37:59] in the archipelago when I grew up. And [38:01] so, uh, I've always been surrounded by [38:04] entrepreneurship and, [38:07] um, I I always felt like I wanted to [38:10] prove myself in that arena. I'm very [38:13] competitive. I used to play video games. [38:14] I used to play Dota 2, um, like more [38:17] than I went to school. Um, and then I [38:20] think building a company is like the [38:22] optimal like it's the it's the biggest [38:24] arena that you can enter and you'll [38:26] either leave victorious or you'll leave [38:28] defeated or you'll be okay. And [38:32] I was very um decided in that [38:38] I want I I should probably learn more [38:41] about the world, which I thought [38:42] Mckenzie was going to be a great place [38:44] to start at, which I still think it is. [38:46] Um I think it really teaches you like [38:49] how to work. uh you learn work ethic [38:52] there but I did an internship and so I [38:54] think I I like got a bit of that even [38:56] during my studies [38:57] >> and then we had this like pivotal moment [38:59] with GPT and so we were talking about [39:02] this like you weren't at as excited [39:04] about investing a few years back like [39:07] prel because there wasn't a lot of [39:09] >> there wasn't a lot going on like like I [39:11] think the the last year or two before co [39:13] it was not a lot of new ideas [39:16] >> right and I remember sitting down preGPT [39:18] like here are all the startup ideas I [39:20] can think of, most of them were pretty [39:21] boring. I I don't think most of them [39:23] would have worked out. [39:24] >> Yep. Yep. [39:24] >> Um and then like small companies got a [39:28] really big uh advantage versus big [39:31] companies. [39:31] >> And I mean I think you saw this in in [39:33] just like the way that Microsoft would [39:35] build co-pilots. Like it it didn't work [39:37] for a really long time. Yeah. [39:39] >> And I remember when that came out and we [39:41] were like, "Oh my god, like we're so [39:43] screwed." Uh you know, every lawyer [39:45] already works in Word and Outlook and [39:47] now they're just going to use Copilot. [39:48] it turns out like it didn't work and the [39:52] um you know ability for us then to still [39:54] build a lot of value was huge. [39:57] I think now reflecting back on that you [40:00] learn so much by having to figure it out [40:06] and I I I really think that the amount [40:08] of things you learn is a function of the [40:11] amount of discomfort uh that you are [40:14] willing to endure. [40:15] >> Yeah. Um, when we started Legora, I was [40:17] like super introverted. I, you know, [40:20] really didn't feel comfortable talking [40:22] to customers. Um, I wanted to code as [40:24] much as possible. [40:26] You know, when you have to, uh, go [40:29] separate ways with an employee for the [40:30] first time, like all of this is so hard, [40:32] but you realize that startups is one of [40:35] the places where you can learn the most [40:38] in a short amount of time. I think, you [40:42] know, if I were to build a company again [40:45] or join a company, [40:49] I [40:50] still feel like I learned a lot working [40:53] at an other startups. I actually worked [40:55] at startups [40:57] >> before starting my own. [40:59] >> And [41:01] then I would say you you should think [41:03] really deeply about uh who you started [41:06] with [41:07] >> and the problem space. And I think we [41:10] picked a space more than we picked a [41:14] problem which I think is like two [41:15] different ways of going about it. It was [41:18] completely obvious that legal and AI was [41:21] going to be a thing. [41:22] >> Yeah. [41:23] >> How it was going to be a thing was like [41:25] very uncertain and so we just said let's [41:27] march in this general direction and then [41:30] we'll figure it out. That's one way of [41:32] doing I guess the other way of doing it [41:33] is like you find a specific problem, you [41:35] solve that very clearly, you generate [41:38] ROI and then you expand from that. [41:41] >> Yeah. Um you mentioned earlier, well I [41:44] know that you are extremely competitive. [41:46] Um how is that expressed inside Lora? [41:49] Like how does a Lora employee know that [41:51] Max is a very competitive C? [41:53] >> I think everybody everybody at Lori is [41:55] like that self selects. [41:56] >> How how does how's expressed? Is it like [41:58] a Friday meeting? Here's our goals and [42:00] like [42:00] >> Well, yes. So we do everybody knows [42:04] every every monthly goal. Uh I think [42:06] setting quarterly goals is like a little [42:08] bit tricky. [42:09] >> Is that how it was from the very [42:10] beginning with just three of you guys? [42:11] >> Um no then it was no goals. Then it was [42:14] just like every day you like try to [42:16] maximize the outcome. [42:17] >> Okay. [42:18] >> Um and then we got you know when you're [42:20] 750 people you need to plan a little bit [42:23] better and like have some better [42:24] internal communication. [42:26] >> Actually communication is like the thing [42:28] that really breaks when you scale. Yes. [42:30] Um I think a lot of people have talked [42:33] about that already. Um [42:35] but no we you really feel this uh like [42:40] drive from engineering that they want to [42:43] build the best product and the best [42:46] feature and if you compare you know our [42:49] tabular review versus you know the other [42:52] alternatives in the market like it's [42:54] really sort of you stack rank it. [42:56] >> Yeah. And [42:58] and we have lots of people on vacation [43:00] now in Stockholm because it's July. And [43:03] uh most of them have have most of the [43:05] engineers who are on vacation are like [43:07] more productive than they had have ever [43:09] been because they're like taking a bit [43:11] of time off, but then they also get to [43:12] code now on problems that they like want [43:15] to beat and they want to win. And [43:18] >> it's it's a team sport. And I think [43:21] >> you can be competitive but be it in a [43:23] team way. And so you really need to [43:26] solidify this like winning or losing [43:28] together strategy I think and when we [43:32] get a big win like that is really [43:34] celebrated and when we have a loss we [43:37] really like mourn it together. [43:39] >> Yeah. [43:39] >> And then we immediately make a plan of [43:41] like how can we flip it or like how can [43:44] we make a land and like turn it around [43:46] and I feel like people are in solution [43:49] mode rather than like blame mode. And I [43:53] think that has been a big um [43:57] you know momentum driver for us but also [43:59] we've been in like chasing right like [44:01] there were uh other legal AI companies [44:04] that were bigger than us when we started [44:07] and so we always felt at least in the [44:10] beginning like we were chasing [44:13] now we are much bigger and [44:17] >> let's say you were the biggest one at [44:18] some point very soon [44:19] >> yes [44:20] >> is is that going to impact your [44:21] motivation or your I think at that time [44:23] it so that's interesting. I I thought a [44:25] lot about like what happens at that [44:27] moment. [44:27] >> Yeah. [44:28] >> I think we need to [44:30] >> uh either pick out another enemy and go [44:34] like okay we have to be bigger than them [44:35] now or we have to like really find this [44:39] way of being better than ourselves [44:41] yesterday. And [44:44] I think that um having competitors is a [44:47] good thing. you should not be overly [44:49] focused on what they're doing, but it's [44:51] definitely like a carrot, I think. And [44:54] it's a it's a easy thing to like uh [44:56] rally around because what you don't want [44:59] is you don't want a lot of side [45:00] questing. [45:00] >> Yeah. [45:01] >> You want everybody focused on the same [45:02] thing. [45:04] >> So, uh one question I got from the [45:05] audience uh is how do you become more [45:08] ambitious? And when I ask this question, [45:11] I'm thinking of a company I'm working [45:12] with right now who's like, "Oh, we want [45:14] to do this country in Europe and then [45:15] we're going to do this other country in [45:16] Europe and then maybe next year we'll do [45:17] US." You guys did a lot of countries in [45:19] Europe just you're in the batch and then [45:21] like [45:21] >> and now we're at like 50 countries [45:23] >> and now you're everywhere and you scale [45:24] up really quickly. And is that ambition [45:26] or is that the product was working or is [45:28] >> No, it's definitely ambition. I mean I [45:30] think it's like [45:31] >> how does someone learn that or how does [45:33] someone become more ambitious? Is that [45:35] possible? [45:37] >> Yeah, I think it's possible. I actually [45:38] think the best way to become more [45:40] ambitious is to have a peer group of uh [45:44] other very ambitious people. So in [45:48] in like my final years of um [45:53] of like high school, [45:55] >> I was pretty lazy. I was playing a lot [45:58] of video games. I was um you know I had [46:03] a pretty easy time in school and I [46:05] didn't work that hard and I I didn't [46:09] know how big the world was like outside [46:12] of my sphere. [46:13] >> Mhm. [46:14] >> And the first two years in college was [46:18] the same. [46:18] >> Yeah. And then I made like a new friend [46:20] group of um very ambitious people and [46:27] that made it click for me. I went from [46:31] doing like zero internships to doing [46:34] eight jobs in one year. And I actually [46:38] think that's like one of the best parts [46:40] about YC. Um and you I think PJ talked [46:44] about that in one of the essays which is [46:46] like you come here and then you just [46:48] realize that like wow every peer like [46:50] everybody's really ambitious and so you [46:52] become more ambitious together. [46:53] >> Yeah. [46:54] >> And I think now we are like that inside [46:57] of Lora. [46:58] >> Yeah. [46:58] >> Because I I don't feel like I'm the like [47:01] boss. I feel like I'm a peer in the team [47:04] that is running the company. And [47:07] you know, our CFO sometimes, you know, [47:11] challenges our marketing team to be more [47:13] ambitious and you know, like a CFO maybe [47:17] typically wouldn't care about that. [47:18] >> Yeah. [47:19] >> But we're all just like so forward [47:21] leaning and it's just so fun. And then [47:24] you really get to assemble this like [47:26] group across the entire company. It's [47:27] not just at the top, right? It's like [47:29] everywhere. Yeah. and [47:31] and then you get to watch your creation [47:34] like spiral and become more and more and [47:36] more and more ambitious and that is [47:38] really fun. I also think you need to [47:39] have like big uh idols [47:42] >> and big companies ahead of you. I mean [47:43] in the beginning [47:44] >> who was who was your idol? Well, so we [47:46] would look at Spotify and CLA as [47:48] >> these are the biggest success tech [47:51] stories from Stockholm that have, you [47:53] know, ever happened. [47:54] >> And then we realized, wow, we have a [47:56] bigger market cap than CLA. [47:58] >> And that was like kind of a weird [48:00] moment. [48:01] >> Yeah. Yeah. [48:01] >> Um because [48:06] we feel like we have done zero to one. [48:08] >> Yeah. [48:10] >> And 1 to 10 and then 10 to 100 still [48:14] remains. that's in front of us and so [48:16] yeah [48:17] >> um would you say uh so so people in the [48:21] audience here are wondering what skills [48:24] are either more valuable to have now as [48:27] you're in school or you're graduating [48:28] from school um versus not more valuable [48:31] like basically is [48:34] is the skills that you need to start a [48:36] company or skills that you need to get [48:37] into like like basically your career [48:39] have they are they changing in a very [48:41] short amount of time now [48:42] >> I I think [48:45] storytelling is at least for the CEO. [48:49] >> Yeah. [48:51] >> I had completely underappreciated how [48:53] important and how good of a skill that [48:56] would be because what ends up happening [48:57] is you need to sell the company to uh [49:01] yourself. [49:02] >> Yeah. [49:03] >> Like working this hard if you don't [49:06] believe in your own story uh is really [49:08] hard. You need to sell employees. Um [49:12] like now we are hiring people you know [49:16] we are competing with the labs for [49:17] talent and you know we need to sell the [49:22] story of how working Lora at Lora is [49:24] going to be better than working at [49:25] Antropic. [49:27] You need to sell the story to investors. [49:29] You need to sell the story to customers. [49:31] And [49:33] I think that in college or in high [49:37] school or wherever that is, like [49:39] spending time to learn that is useful. I [49:43] don't know how to learn it. I actually [49:44] don't know how I learned it unless I [49:48] just had a lot of exposure. [49:49] >> Yeah. [49:49] >> Uh I always like liked stories. I like [49:51] video games. I like I enjoy a good [49:53] story. But I think that is like the most [49:55] important thing. [49:56] >> And and then maybe the other really tech [49:59] technical skills. [50:00] >> Okay. Um, well, I I'll I'll say one more [50:02] before we go to technical. [50:04] >> I used to be a really bad teammate, [50:07] >> uh, and a really bad collaborator. So, I [50:09] played volleyball for a long time [50:11] growing up. [50:13] >> And I would be the type of person on the [50:15] team who when my teammate missed the [50:18] serve, I would yell at them and go, "You [50:21] suck." Like, how could you miss [50:23] that serve? And [50:26] that was very that doesn't get you [50:29] invited to play the next game. [50:31] >> Yeah. [50:31] >> And [50:32] >> makes sense. [50:33] >> You know, it it it doesn't it doesn't [50:36] invite a followership and it doesn't [50:39] invite like a good vibe. [50:41] >> Yeah. [50:42] >> So in high school I like really [50:45] understood that and had to pivot that [50:47] way of of working. I think technical [50:49] skills. [50:50] I [50:53] think that [50:57] the sort of the velocity and and your [50:59] ability to iterate on customer feedback [51:01] still is probably the most important [51:03] skill to get started. Yeah. [51:05] >> Even now when we're bigger, very often [51:08] we will hear about a particular type of [51:10] problem a customer is having in a [51:13] customer call. I will dump that in the [51:16] product channel. and I'll go how quickly [51:19] can we turn this around and delight the [51:21] customer [51:22] >> and that's like a bit strange to do at a [51:24] you know thousand person company scale [51:26] but still how we do it [51:28] >> and you need to make sure that you don't [51:30] overly like zigzag and that you continue [51:33] building towards like the right vision [51:36] >> but I still think that the [51:38] uh your ability as a smaller company to [51:40] be fast versus your incumbent [51:42] competition is the superpower and when [51:45] you have a superpower you need to lean [51:47] into that and and and that's still true [51:49] for for me like I I should be doing the [51:51] things that I'm uniquely good at and [51:53] other people in the company do the [51:55] things that they're uniquely good at. [51:56] >> Yeah. [51:56] >> Um I think also [51:59] technical uh skills um [52:04] >> maybe they're different these days. [52:05] Technical means something else. [52:07] >> Yes. Well, but but it still means you [52:09] know being able to build systems at [52:11] scale, right? Like just because you can [52:13] produce a lot of code with AI [52:14] >> Yeah. is not a excuse for not being a a [52:20] very good software engineer. [52:21] >> Right. Right. [52:22] >> And I still think we need many good [52:24] software engineers. We are hiring a [52:27] hundred of them at Agora. [52:29] >> What's the most interesting problems [52:31] that people get to work on when they [52:33] join Agora? [52:35] Oh um [52:37] one of the most interesting problems [52:40] right now is this transition from [52:43] reactive agents to proactive agents. So [52:47] >> what's an example? [52:48] >> So for the last three years of working [52:50] with the product, you know, you will [52:51] give Lora a prompt or a set of [52:53] instructions and it's going to go off [52:55] and do that thing. Now we are basically [52:58] connecting Lora to different pieces of [53:00] context and when it gets a trigger it [53:02] will start to do something. Uh that [53:04] means that if the sales team gets a [53:06] contract it will get routed to a Lora [53:08] agent. The Lora agent will start working [53:10] on it in parallel. you know, if it needs [53:13] to escalate to a lawyer, it will. [53:14] Otherwise, it might just execute the [53:16] contract. [53:17] >> Or you connect the entire data room to a [53:19] Lorra agent and the agent automatically [53:21] starts um organizing that data room and [53:25] produce the due diligence report, right? [53:26] Like it's it's um having the agent do [53:29] things without you telling it uh what it [53:32] should do. [53:33] >> Yeah. And that for us is the unlock of [53:37] you know how does one lawyer produce the [53:39] outcomes of a team of 10 lawyers with [53:42] the help of Lora. [53:43] >> Yeah. [53:43] >> H I think that's very interesting. We [53:45] also have some very interesting problems [53:46] just from like a scale perspective right [53:49] like we're spending millions and [53:50] millions and millions of dollars on OCR [53:52] and like uh you know document parsing [53:55] and these are like real engineering [53:57] problems to go solve at a bigger scale. [54:00] >> Got it. Um and you grew from zero to 100 [54:04] millionaire in 18 months in April. Um [54:07] keep growing at a crazy rate. [54:09] >> Yes. [54:10] >> Your job has changed quite a lot in 18 [54:13] months. [54:13] >> Yes. [54:14] >> Um we talked earlier about you now [54:16] interviewing like exec teams like sort [54:19] of like we had this talk from the [54:21] founder of Dropbox at YCA like a couple [54:23] years ago. We talked about the mini [54:24] games that you play. You've done a lot [54:26] of mini games in 18 months and now [54:27] you're at the mini game of like building [54:28] out or hiring people to run your team. [54:30] How what's [54:31] >> Yeah, [54:32] >> that sounds hard and doesn't sound like [54:33] something you've done before. Like how [54:36] >> Well, it's uh definitely not something [54:38] I've done before, but [54:39] >> how do you learn something like that [54:40] that it seems extremely important to [54:42] hire the right executive? [54:45] I in a way I think [54:48] most of the work up until that point um [54:52] makes you qualified for that thing, [54:54] >> right? [54:55] >> Uh if you've [54:56] >> you get the next mini game by basically [54:57] seeing the last one. [54:58] >> Yes. But you need to make sure that you [54:59] learn uh because at part of every mini [55:03] game and I think you really need to um [55:06] put the company first um way ahead of [55:11] your own ego in a way. And you know I I [55:14] tell this to the executive team. I need [55:16] to re-qualify for the job as CEO of Lora [55:21] every quarter, right? It's a new [55:23] company, new challenges as do they, [55:25] right? Like running a sales team when [55:27] you are 1 million ARR is very different [55:30] from 150 million AR R. And I think for [55:34] myself, if you [55:37] like if you are um if you enjoy that, if [55:41] you enjoy learning new things, if you [55:42] enjoy being challenged, if you accept [55:45] the fact that now my most important job [55:49] is building out the executive team and [55:51] I'm going to go do that, you will [55:53] succeed at it. And then I think as [55:56] always, you try to build a good network [55:58] of people who have done it before. Um [56:01] the tough thing is there's very few [56:03] people who have done uh the company this [56:06] quickly as we are and you know even a [56:08] lot of the executives that we are now [56:09] bringing on they come from sauce and [56:12] they come from this you know world of [56:15] triple triple double double if you grow [56:18] 40% year on year you are doing fantastic [56:21] you're you know you're you're sort of on [56:22] track and [56:24] >> that's not enough anymore [56:26] >> it's not enough like we you know David [56:28] our CFO joined from H another YC company [56:31] very they've been very successful. [56:32] >> Yeah. [56:33] >> But they've grown from you know 30 [56:35] million to 300 million in in something [56:36] like four years and now we're doing it [56:38] in like one year and so uh from 200 [56:42] people to 1300 people and we're doing [56:44] over four years and we're doing it in [56:45] one year and so you need to just [56:46] compress timelines. I think you need to [56:49] be uh brutal at saying no to things and [56:52] you need to understand yourself where [56:54] you have the most competitive edge where [56:57] you can do the most. when I landed in [56:59] Stockholm uh in end of April, all of the [57:03] ads were basically Legora ads with Jude [57:06] Law. [57:07] >> Um and [57:10] I'm curious how you get Jude Law to be [57:13] the face of the company. Uh and he has [57:16] if you haven't watched seen the ads, I [57:17] recommend you go into YouTube and [57:18] watching the Leg Jude law ads. [57:20] >> Did you did you fall in love with law [57:22] again? [57:22] >> Yeah, I did. [57:23] >> Yeah, you did. Good. Um [57:25] >> what happened? How did how did you score [57:27] that? Okay. So, the the inside baseball [57:30] is [57:31] we had a meeting with one of our [57:34] marketing agencies and one of their [57:37] pitches was why don't we redo like very [57:40] famous um legal scenes like [57:44] >> um you can't handle the truth. Yeah. [57:47] Like like why don't we redo like famous [57:49] legal scenes with lawyers but like with [57:51] Legora? [57:51] >> Yeah. [57:52] >> And or but no, the pitch was this. It [57:54] was actually pretty funny. um all of [57:55] these movies about lawyers like what if [57:58] they just had Lora and you know the [58:00] movie would be over in like a minute and [58:02] so we got to this idea of like working [58:03] with actors and then somebody went well [58:06] legora is like AI powered law what if we [58:09] got dude law to be like AI powered and [58:13] that got us down the track of like huh [58:15] maybe we could get dude law um and [58:18] what's funny in Hollywood right now is [58:20] it's like pretty anti- AI with you know [58:23] writing script [58:25] you know, using AI in movies, etc. [58:28] >> And he thought we were completely [58:30] unserious when we reached out the first [58:31] time, like some Swedish AI company wants [58:34] to work with Jude Law to make billboards [58:36] and a campaign. But then we were so [58:38] persistent and I think um one [58:41] underappreciated form of negotiation is [58:45] nagging. It's one of the things my dad [58:47] taught me. And [58:49] we nagged and nagged and nagged and he [58:52] said yes. On one condition, he got to [58:55] pick his own script writer and [58:57] cinematographer. And who would have [58:59] known? He brings an SNL script director [59:02] and the cinematographer from Oppenheimer [59:05] [laughter] [59:06] to uh to like light the the scene and [59:09] like do the whole cinematography. And [59:12] they basically go into this the set. [59:14] Nobody from Lori is allowed in and you [59:17] know a few hours later like these [59:18] magical productions come out. [59:20] >> It's really good. [59:21] >> It's so funny. Um and what I think it [59:23] did for us was it put us in another [59:26] light where like even people outside of [59:28] law now know about Lorra and that has [59:31] been really exciting. [59:33] >> Awesome. Thank you so much for coming to [59:35] school. This is awesome. [59:36] >> Thank you Gustav for having me. [59:37] >> Thanks everyone.