Legal AI Startup Leya Raises $80M as AI Shifts Billable Hour Dynamics

法律AI初创公司Leya融资8000万美元，AI重构计费工时模式 这家YC校友企业阐述了在创始人缺乏行业背景的情况下，如何通过横向市场扩张与多模型基础设施打破保守律所的销售壁垒。 本次访谈回顾了法律AI初创公司Legora（亦称Leya）的迅速崛起。在入选Y Combinator 2024年冬季批次后的13个月内，该公司实现了巨大的规模扩张。通过打造具备表格化尽职调查网格的多模型工作空间，以及能够结合内部操作规范直接编辑文档的Microsoft Word加载项，该公司跨越了传统法律科技的障碍。尽管创始人缺乏法律专业背景，但他们凭借深入的用户研究、积极的欧洲多市场扩张战略，以及吸纳前创业者的扁平化组织架构取得了成功。讨论最后探讨了生成式AI如何推动律师事务所从低差异化的按小时计费工作，转向高价值的咨询业务。 针对高度谨慎的欧洲市场，Legora 严格确保本地托管与零数据留存合规，借助大语言模型处理非结构化文本的能力，将法律软件演进为全流程 AI 工作空间。 (2:22) Legora 打造了如同“律师版 Cursor”的 Word 插件与 Web 端多步工作流，并通过“表格化审查”（tabular review）网格将原本耗时数天的尽职调查任务压缩至几分钟。 (6:24) 客户拒绝继续为低差异化手工活支付高额时薪倒逼律所采纳 AI，Legora 借助“Playbooks”功能实现合同自动化标记，统一全机构的法律与合规标准。 (15:03) 创始人团队在无法律专业背景下通过深度访谈 100 位律师切入市场，在 YC 毕业后 13 个月内将团队从 10 人扩充至 100 人，并完成 8000 万美元 B 轮融资。 (30:20)

Max Junior Strand, co-founder and CEO of legal AI startup Leya (referred to early on as Legora), outlines how an outsider team scaled enterprise legal tooling to tens of thousands of daily users by embedding into existing attorney workflows and treating early technical constraints as competitive moats. From Fragmented Tools to Parallel Document Review (4:10) Before generative AI, legal tech was dominated by unintegrated point solutions for tasks such as redlining, translation, and basic research. While early BERT experiments struggled with non-English legal text, the advent of GPT-3.5 enabled developers to parse unstructured precedents. Strand points to tabular document review as his operational breakthrough: rather than evaluating files individually during due diligence, attorneys load hundreds of contracts into a grid where queries populate rows and columns with citations linked directly to source text. Running up to 100,000 queries in parallel required engineering systems that could parse legal cross-references and definitions without failing. Workflow Integrations and Playbook-Driven Automation (6:24) Because lawyers work primarily in Microsoft Word and PDFs, the company built an editing add-in alongside its web application, styling it as an assistant directly embedded in the Word sidebar. The platform integrates firm playbooks to automate contract markup; for instance, sales representatives or legal teams can check incoming NDAs or Master Services Agreements against internal rules and automatically insert approved fallback clauses. Attorneys have also adapted the system for real-time applications, such as a partner at Spanish firm Pérez-Llorca querying opposing evidence during a live hearing to catch spoken inaccuracies, and teams role-playing counterarguments prior to negotiations. Sales Dynamics and Enterprise Procurement Strategy (11:40) Historically, legal software companies struggled to sell to law firms and pivoted toward corporate legal departments. Strand argues this dynamic shifted as corporate clients pushed back against paying high hourly billing rates for routine tasks like document review. To counter partner skepticism and strict procurement hurdles, the startup initially sold to single partners to create internal success stories rather than attempting bottom-up seat sales. It satisfied European data governance rules by hosting all infrastructure in Europe, enforcing zero data retention, preventing client data from being used in training runs, and negotiating human-review exemptions with cloud vendors. Multi-Model Architecture and European Expansion (23:49) To manage inference costs and performance, the platform routes tasks through a multi-model infrastructure spanning AWS, Azure, GPT, Claude, Gemini, and Mistral. Classification models direct simple tasks to smaller models and reserve frontier models for complex reasoning. Strand chose to keep the company's base in Stockholm rather than moving to San Francisco following its Y Combinator batch, using the less crowded European market to establish a playbook across Finland, Norway, France, Germany, and the UK. After pausing sales for several months to re-engineer infrastructure to support onboarding 1,000 lawyers daily, the company raised an $80 million Series B led by Iconiq and General Catalyst to expand into the US. The Shifting Role of Legal Labor (35:06) Strand forecasts that over the next 5 to 10 years, attorneys will increasingly act as managers reviewing AI-generated drafts rather than executing manual text drafting from scratch. Because liability and final sign-off require human expertise, the startup exclusively targets legal practitioners rather than selling directly to consumer legal buyers. To maintain development speed against incumbents, the company prioritizes hiring generalists and former startup founders, noting that product roadmaps must adapt constantly because frontier AI labs are shifting from simple model vendors into full software platforms. While the platform has scaled rapidly across European and US enterprise firms, Strand acknowledges that rapidly evolving frontier models make traditional multi-quarter product roadmaps difficult to maintain, requiring legal tech startups to avoid platform lock-in as AI labs expand directly into core enterprise software environments.

Legal AI初创公司Leya（原注中亦称Legora）联合创始人兼CEO Max Junior Strand阐述了非法律背景团队如何通过解决欧洲合规痛点、打造嵌入既有工作流的表格审查与Word插件，以及以律所合伙人模式切入企业级市场，最终完成8000万美元B轮融资并实现跨国扩张。 技术演进与表格化尽调创新 (4:10) 在生成式AI爆发前，法律科技多为模板生成、特定翻译与基础检索等单一工具。创始人团队早期曾尝试Google BERT模型，但其在瑞典语表现极差；直至GPT-3.5开放开发者权限，团队才构建出能解析非结构化法律文本的初期产品。随后，团队通过RAG架构接入瑞典法规库，向北欧头部律所证明了法律问答能力。Max指出，其核心突破在于将传统逐篇审阅的尽调工作转变为“表格化审查”（Tabular Review）系统：用户将百份协议置于行，将关键审查问题设为列（如是否含知识产权保护条款），系统能够并发运行高达10万次查询，并在数分钟内给出附带源文档精确引用的结果。 产品形态与多模型分发架构 (6:24) Leya将产品拆分为Web应用和被称为“律师界Cursor”的Microsoft Word插件。Web端已从基础问答升级为可执行多步工作流的自主智能体，借助模型上下文协议（MCP）适应知识产权、重组与争议解决等专业方向；Word插件则在紧凑的侧边栏内直接编辑文档，支持依据内部谈判策略清单（Playbook）自动置换退让条款。在技术底层，Leya将基础设施托管于企业客户普遍使用的Microsoft Azure，并逐步过渡至混合多模型架构，综合调用AWS、Claude、Gemini、GPT及Mistral。为控制运算成本并维持利润率，系统使用分类模型对用户查询进行分流：简单任务采用低成本小模型，复杂任务才路由至大模型。 欧洲严格合规与律所销售破局 (2:22) 打入对风险极度敏感的欧洲法律市场，必须在创立初期即落实严苛的数据合规标准。Leya确保数据全量驻留欧洲本土，不将客户数据用于模型训练，严格执行零数据保留策略，并争取到了AWS和Azure等云服务商的人工审核豁免。面对律所传统按小时计费与采购壁垒，Leya采取了与律所建立长期利益同盟的反常规销售策略。随着企业客户拒绝为低门槛重复工作支付高昂账单，律所不得不采用AI来保持竞标优势。在实操案例中，西班牙Pérez-Llorca律所合伙人在庭审中实时利用系统核对证据与对方辩词；北欧大型银行则将文档规则审查（Playbook）由法务部门进一步横向拓展至合规、风控乃至销售团队。 组织扩张与以斯德哥尔摩为跳板 (30:20) 团队没有法律教育背景，早期通过在LinkedIn上约访100位律师共进午餐来摸清细分领域的运作差异。从Y Combinator毕业后，公司规模从10人迅速扩充至约100人，并在获得Iconiq和General Catalyst领投、YC、Benchmark等参投的8000万美元B轮融资后，暂停对外销售4至5个月以专注于提升系统可靠性，达到单日承载1000名律师入驻的工程指标。Leya选择先扎根斯德哥尔摩，横向打通芬兰、丹麦、挪威、西班牙、法国、德国及英国等多元化欧洲司法区，建立起标准化的跨国扩张模型后，再进军纽约设立办公室并签下美国律所。 Max Strand认为，未来5至10年律师的工作重心将从手动起草转向审核AI产出并担任管理角色。为此，垂直领域AI初创团队应避免与底层大模型实验室正面对抗或过度绑定单一供应商，而应聚焦于沉淀深度的垂直工作流与专有格式化能力；同时，面对每周快速迭代的基础模型，长期产品路线规划的难度也在显著上升。

[0:00] Max Junestrand: aged five years in the last one [2:22] Max Junestrand: horrendously bad in Swedish [4:10] Managing Partner: more artificial than intelligent [4:10] Max Junestrand: you kind of see it on his eyes, like it's the aha moment [6:24] Max Junestrand: Cursor for lawyers [15:03] Pérez-Llorca Partner: another piece of armor [23:49] Max Junestrand: sometimes you don't need a bazooka when you just need a water gun [17:17] somebody people want to help [20:31] rate of change [27:51] Max Junestrand: The International [30:20] drag [35:06] just do things [38:34] Max Junestrand: If something breaks, we get a phone call immediately [38:34] small fish in a small pond [38:34] crocodile or a shark [42:01] Max Junestrand: boats, so that when the tide rises, everything gets better

[0:00] AI is continuously developing super [0:02] super quickly and that means we need to [0:04] do the same. We're finding that as we go [0:06] deeper and deeper and deeper in the [0:08] entire legal software stack, we're also [0:11] seeing that the line between software [0:15] and service is blurring. I think that's [0:18] been one of our strengths as a company [0:20] to say we don't know exactly where the [0:23] future's going, but neither do you. So, [0:25] let's work together to make sure that [0:27] we're both winners in whatever happens. [0:33] [Music] [0:34] Today, I'm joined by Max Junior Strand. [0:36] He's the CEO and co-founder of Legora. [0:38] Legora was in winter 24 and they are the [0:41] leading AI workspace helping lawyers and [0:43] legal professionals do their work. [0:44] Welcome, Max. [0:45] >> Hey, thanks Gustav. [0:46] >> It's been 13 months since you did the [0:48] batch. It's been a really busy year for [0:49] you. [0:50] >> It has. Uh, it feels like it was a [0:52] really long time ago. I feel like I've [0:53] aged five years in the last one. [0:55] >> For those who don't know, tell us about [0:56] Legora. Yeah. [0:57] >> What are you guys building? [0:58] >> At Lagora, we're building the AI powered [1:00] workspace for lawyers. We're essentially [1:02] transforming the way that they complete [1:04] their work. Everything from reviewing, [1:06] drafting, researching. Essentially, [1:08] within legal, you've had this incredibly [1:10] fragmented software space where there [1:12] was a lot of point solutions and AI was [1:15] never good enough to actually work with [1:16] unstructured text, precedent, legal [1:19] documents. And when GPT 3.5 got out, [1:22] that just completely changed the game. [1:24] So we were quick to, you know, build a P [1:27] and then now we've scaled that all the [1:28] way to an enterpriseg grade system [1:30] serving tens of thousands of lawyers [1:32] daily. [1:33] >> Those point solutions were basically [1:35] workflow tools. So what were they [1:36] before? Because it's been a history of [1:38] of a legal technology industry that [1:41] existed before. This is not started [1:43] right now. [1:43] >> No, I mean legal tech has been a [1:44] category for a long time, but um it was [1:47] really unsexy for a long time, I think. [1:49] And you'd essentially have a broad range [1:51] of point solutions. Everything from [1:52] templating tools where you would sort of [1:54] codify a contract to special translation [1:57] tools or redline tools or research [1:59] tools. And all of them work with text [2:02] somehow. And generative AI came into the [2:05] game and just kind of threw up [2:06] everything off the off the table. And [2:08] then when it landed, you very clearly [2:11] saw how you could solve a lot for a lot [2:13] of these use cases with the same [2:14] underlying tech. So chatp came uh maybe [2:17] 8 months prior to you guys starting this [2:19] company. Describe that moment. Was that [2:20] an important moment for the company's [2:21] founding? [2:22] >> We were playing around in AI and legal [2:24] way before Chat GPT and we were using [2:26] these early models from BERT coming from [2:28] Google. They were decent in English but [2:31] they were just horrendously bad in [2:32] Swedish and you know the first [2:34] observation that kind of sparked the [2:36] founding of the company was one of the [2:38] co-founders friend who was a lawyer [2:40] spent four months during a summer just [2:42] summarizing court cases for a big law [2:43] firm. We basically saw that GP 3.5 was [2:46] released to developers started building [2:48] I think the first thing that we built [2:49] was a stock option reader that would [2:53] explain how a stock option contract [2:55] worked [2:56] >> useful [2:56] >> right you know as startup founders with [2:58] no legal background that was seemed [3:00] reasonable and then very quickly the the [3:03] sort of focus changed to how do we build [3:06] this more wall-to-wall or or endto-end [3:08] system that every legal professional [3:10] wants to work with on a daily basis and [3:12] the first product was really quite [3:14] simple. Uh especially building for [3:16] Europe, you got to go through a lot of [3:17] hassle to kind of conform with all the [3:19] data processing requirements. So you [3:21] know all data hosted within Europe, [3:23] nothing for training, no retention, [3:26] exemption from human review when you [3:27] look at the way Ashure and AWS is [3:29] structured. And we we kind of jumped [3:31] through all those hoops and just built a [3:33] system that was compliant for law firms [3:36] to work with. And then very quickly as [3:38] the general sort of AI platforms [3:40] continued to develop with chat GBT with [3:42] cloud with Gemini the requirements for [3:45] what we had to build to be much much [3:47] better you know continuously increased [3:49] >> in some industries or some some [3:51] categories like coding or law for [3:53] example it seems like the models are [3:56] just magical like like they do things [3:58] that the people that were in those [3:59] industries before could not even imagine [4:01] be possible. Could you describe sort of [4:03] like the first time you um used Lora um [4:07] to do something that was magical for a [4:09] customer and and how they experienced [4:10] it? [4:10] >> Yes, I think the first time was when we [4:12] deployed Legora into the biggest law or [4:14] largest law firm in the Nordics, man. Um [4:17] they their managing partner had a famous [4:19] saying in the newspaper that AI was more [4:21] artificial than intelligent which was [4:24] back from the early models. Yeah. I mean [4:27] a lot of firms burnt themselves buying [4:29] expensive tools that didn't solve [4:31] anything. Mhm. [4:32] >> And I came into that meeting, you know, [4:34] I bring out my laptop and I just ask [4:35] him, you know, put in a query. [4:37] >> And he puts in this uh legal research [4:40] query and we've tied Legora to Swedish [4:42] legislation with a rag system. Yeah. And [4:44] it answers perfectly [4:46] >> and you know, you kind of see it on his [4:47] eyes like it's the aha moment and now [4:51] when we're [4:51] >> Is that your aha moment as well? [4:53] >> No, I think my personal aha moment was [4:56] just using chat GPT generally, right? [4:58] Like it was amazing. It felt complete [5:00] sci-fi that you could talk with the [5:01] computer and it talked back and you know [5:03] as an entrepreneur you you you kind of [5:06] quickly you know from that you [5:08] understand that all right we can apply [5:09] it in this space in this way and in that [5:11] space in this other way and I think for [5:12] legal specific the chat experience I [5:15] think was always cool but when we took [5:19] the the same models and and sort of [5:21] applied them differently um one of the [5:23] first use cases we did was due diligence [5:25] where you have hundreds or or you know a [5:27] lot of documents that you want to view [5:29] and instead of going through them one by [5:31] one by one, we just made this large grid [5:34] where essentially every document [5:36] represented a row [5:37] >> and then you could put your queries in [5:39] the columns, [5:40] >> right? And as you then put in, you know, [5:42] 100 employment agreements and you ask uh [5:45] does does all of them include an IP [5:47] clause [5:48] >> where the company protects its [5:50] intellectual property and it just starts [5:52] to rattle and it goes yes yes yes no no [5:54] no yes yes yes and it always links backs [5:57] to the citation. and you realize like [5:59] holy this is transformational. It's [6:02] taking tasks which used to be you know [6:05] days or hours and it's turning them into [6:08] minutes. [6:08] >> By the time this is is live uh you will [6:11] have announced that you have raised uh [6:13] series B. How much did you raise? [6:15] >> We raised $80 million led by iconic and [6:18] general catalyst and you know grateful [6:20] for YC's continued participation as well [6:22] as benchmark and red points. [6:24] >> What is the software like? So, so as a [6:26] lawyer using Legora, what does my [6:28] day-to-day look like? [6:29] >> So, it's really broken up into two [6:31] pieces. The first one is the web [6:32] application and the second one is our [6:34] word addin. So, we integrate directly [6:36] into Microsoft Word. Right? [6:37] >> So, if we start with the web [6:38] application, the first thing that we had [6:40] was just a simple chat chat over your [6:42] own documents and files. This has [6:44] quickly developed into its own agent [6:46] that's able to use a lot of the other [6:47] endpoints in the app and also external [6:50] tools to solve more complex sort of [6:52] step-by-step workflows. Mhm. [6:54] >> So you could imagine saying um hey I [6:57] want to write a memo and the first step [6:59] of the memo is to go out and do some [7:01] research. The second step is to take all [7:03] that research and conform it into the [7:05] standard language of the firm and the [7:07] third step is to write the report and [7:09] then output is a report. [7:11] >> And does it do all that? [7:12] >> It does all of that. Wow. [7:13] >> Right. Um, and I think we can talk more [7:15] about it later, but MCP and the way that [7:18] you can scale the tool usage of these [7:19] agents is something that I'm super, you [7:21] know, keen on and and that we're leaning [7:23] very heavily into because a lot of firms [7:25] have different needs in terms of how [7:27] they want to adopt the tools to solve [7:29] for their specific workflows. And it's [7:31] different if you work in intellectual [7:33] property or if you work in restructuring [7:35] or if you work in corporate or if you [7:36] work in uh disputes. The second piece [7:38] outside of the chat is uh well the grid [7:41] that I talked about before. We call it [7:42] tabular review. [7:43] >> Y [7:44] >> um it's essentially input any number of [7:46] files and then input any number of [7:47] queries and we sort of cross run that [7:49] across each other. And the big [7:51] innovation there does not really come [7:53] from you know how do you prompt and and [7:55] work with the model but it's how do you [7:57] make this run at scale [7:58] >> right? [7:58] >> You know how do you run 100,000 queries [8:01] in parallel at the same time and make [8:02] sure nothing breaks all the citations [8:05] are correct. There's a lot of chunking [8:07] sort of rag searching within the [8:10] individual documents because sometimes [8:11] they're very very long. Y [8:12] >> and with legal docs there are certain [8:15] intricacies where you need to always [8:17] include things like the definitions y [8:19] >> and there might be cross references [8:20] within each clause to each other. So [8:23] taking all of that into consideration [8:24] that kind of serves the grid. Looking at [8:27] the word add in I think you could phrase [8:28] it as cursor for lawyers. [8:29] >> Lawyers basically use word that's that's [8:32] a like a known fact for a long time. [8:34] >> Yeah. I mean they draft and they review [8:36] contracts in word or PDF form and [8:38] >> what we really wanted to do is similar [8:40] to cursor how do we bring generative AI [8:43] into the you know existing work [8:45] environment of a legal professional and [8:46] that means integrating in word [8:48] >> now the difference is you can't fork [8:50] word and you can't take up all the real [8:52] estate you want you're basically [8:54] conformed to this sort of right hand [8:56] column [8:57] >> and then you got to get really creative [8:59] it's basically like designing a mobile [9:01] app almost because that's all the real [9:02] estate you get And the first thing that [9:05] we built there was just, you know, how [9:06] do we integrate an assistant or a chat [9:08] that's able to not only read the [9:10] document, but also create edits. So you [9:12] might say, I want you to um, you know, [9:15] renegotiate this MSA for the buyer. Y [9:17] >> and do that using this internal [9:19] checklist that I have or this internal [9:21] sort of playbook or precedent. And now [9:23] we've scaled that to not only work in a [9:25] chatby chat basis, but also more [9:27] extensive workflows. So you can say, [9:29] here's a contract. I want you to take my [9:31] playbook that consists of 20 different [9:33] steps and make sure we negotiate from [9:35] the starting positions and have, you [9:37] know, different fallbacks included. [9:39] >> Do you have a specific example of [9:40] something that was impossible a couple [9:43] years ago for a lawyer? Like literally, [9:44] you couldn't do it and now you can do [9:45] it. [9:46] >> Yeah, I mean I think well there's a lot [9:48] of it, right? The early ML models were [9:51] really bad at legal language and what [9:53] they were really bad at was when the [9:55] language looked different in across [9:58] documents, right? You could train a [9:59] system to find let's say a change of [10:01] control clause if it looked the same way [10:04] across all the documents. Yeah. [10:05] >> But it was really frankly bad at [10:09] >> finding the meaning of a change of [10:11] control if the class didn't look that [10:13] way. [10:13] >> Mhm. [10:14] >> And so what the LLMs have allowed us to [10:15] do is to just take tasks where [10:17] especially on like large contracting and [10:19] large document extraction. So how do we [10:21] pull the insights from this? Another one [10:23] is just you know redlinining. So [10:25] redlinining files within word against a [10:27] president or playbook completely [10:29] impossible. Or take deep research across [10:31] you know hundreds or thousands of [10:33] judgments where you need to conform not [10:35] only the judgments but also pull in [10:37] things like legislation and regulation [10:39] all into the same place that since the [10:41] cost of intelligence is going down it [10:43] also increases the amount of queries we [10:45] can do. Right? So, one pretty cool thing [10:47] is, you know, embedding making one [10:49] search against your own documents and [10:51] files. Yeah. Making another one on the [10:52] web and making another one against um [10:55] court cases and judgments and and [10:57] legislation and then combining all of it [10:59] to create effectively like a memo that [11:02] >> maybe they they couldn't afford to do in [11:03] the past. They just like just didn't do [11:05] it. [11:05] >> No. And similarly with with due [11:07] diligence when if you go way back, it [11:10] used to be a physical data room. That's [11:12] why it's called a room, [11:13] >> right? You used to go into the room, you [11:15] had all the documents and all the [11:17] contracts and then you'd sit down and [11:18] read through all of them. Yeah. [11:19] >> And you had to mark them with a pen. So [11:21] making and doing a due diligence on a [11:23] company was really expensive. And now [11:25] it's becoming almost a commodity where [11:28] you're expected to do it, but [11:30] >> clients are also not really that excited [11:33] to pay for very simple contract review [11:36] when they know that AI can do, you know, [11:39] 99% of it. [11:40] >> Wow. Yeah. [11:41] >> So, in the time that I've been at, we [11:43] have funded some legal software [11:45] companies, but the hardest challenge for [11:48] all of them was selling to law firms [11:51] >> and selling to legal like most of them [11:52] would end up selling to companies cuz [11:54] law firms were just like not possible to [11:56] sell to. That radically changed um just [11:59] like two years ago. [12:00] >> Yeah. [12:01] >> Can you tell us sort of like what do you [12:03] think changed and how do you do it when [12:05] you go and sell to uh one of the major [12:06] law firms in the world? So for everybody [12:08] listening, this was also one of the [12:10] questions that I remember you pushing [12:12] really hard on during the interview. [12:14] >> And I think we were quite contrarian to [12:16] say, you know, no, it's different this [12:18] time. Trust us. [12:19] >> Y [12:20] >> I'm glad we were right. [12:22] >> I think the way that we approached the [12:24] problem was always with this idea of we [12:28] win if you win. So let's align our [12:30] incentives with saying as a law firm, [12:34] this technology is revolutionizing. [12:37] you're you're going to need to adopt it [12:38] in some sense, shape or form and we want [12:41] to be that long-term partner [12:42] >> and somehow they know that. [12:43] >> Well, so what what happens is um a lot [12:46] of legal work is low differentiation. [12:49] You know, if you're doing a DD from, you [12:51] know, law firm X or law firm Y kind of [12:53] getting the same deal [12:54] >> and so when you have this perfect [12:56] equilibrium of services and somebody [12:59] disrupts that by taking a new approach, [13:02] >> clients are quick to switch. [13:04] >> Yeah. I mean clients are under price [13:06] pressure. They want to be effective. [13:08] Legal fees are very high. And so if this [13:11] equilibrium breaks, you are almost [13:13] forced to adopt it. And you you are [13:16] incentivized. [13:17] >> It's kind of the same as when lawyers [13:18] adopted computers, [13:19] >> right? If you're billing by the hour, [13:21] you could say, well, let's have a person [13:23] walk to the library, you know, find the [13:25] right book, you know, find the right [13:27] cases or the right president and use [13:29] that for whatever work we do. [13:31] >> Or you press C++ F, [13:33] >> right? There's always this dilemma of [13:35] you want to serve your client in the [13:37] best way possible because that drives [13:38] you more revenue over time. Yeah. And [13:40] for a lot of a lot of the firms that we [13:43] work with their you know brand [13:46] reputation trust as always putting the [13:48] client first is what matters the most. [13:51] And so a lot of the firms also want to [13:54] be uh leaders here. [13:55] >> Yeah. You know, some of them want to be [13:57] fast second movers, but many want to be [13:59] first movers because they're [14:01] understanding if you have this perfect [14:03] equilibrium and you take a simple type [14:05] of work that gets disrupted, you should [14:08] get more market share by moving down [14:11] quicker. [14:12] >> Mhm. [14:13] >> But then it's not a race to the bottom, [14:14] right? [14:15] >> It's a question of okay, if we take [14:17] >> every country has a ranking of law firms [14:18] basically, [14:19] >> right? And and it's also not a race to [14:20] the bottom in terms of pricing because [14:22] if you pull down let's say the cost of a [14:24] due diligence, you free up more time to [14:27] spend with a board on advising them on, [14:30] you know, a really complex merger or a [14:32] really complex acquisition. And so what [14:34] typically ends up happening is you're [14:35] under time pressure. You could do more [14:37] work, but you just have all this stuff [14:40] that needs to get done. And that's what [14:42] AI is really good at. But it's also [14:44] serving lawyers in very creative ways. I [14:47] mean, we've had use cases where, you [14:49] know, we get a call from somebody and [14:50] they say, "I played a role role playing [14:52] game with Lora, you know, trying to win [14:55] this argument and I'm asking it to act [14:58] as the other party, right?" Wow. [15:00] >> There was this amazing uh situation that [15:03] one of the Spanish um partners at firm [15:06] called Perorka had where he went into [15:08] court. M [15:09] >> he had put all the evidence and all the [15:11] documents from the opposing party in [15:13] Lora and he was actively querying it [15:16] during the hearing and during you know [15:18] at the time when the other attorney was [15:20] speaking because then he could [15:21] immediately interrupt if he found [15:23] something that was uh that was wrong. [15:25] >> Wow. [15:25] >> And he phrased it very nicely. He said [15:28] when when he goes into the battlefield [15:30] having Legora is like having another [15:32] piece of armor. And I thought that was [15:34] that was very poetically. [15:35] >> Could you use Legora to do negotiation [15:37] on your behalf? Yeah. So the way that we [15:40] built that is I think the LLMs by [15:42] themselves are not good enough for that [15:43] yet and we can talk about that but it's [15:46] it's interesting to to build these [15:48] products knowing that the models will [15:50] get better. Yeah. [15:51] >> And where do you stop? [15:52] >> Yeah. [15:52] >> Right. On every feature but so so that [15:54] feature in the Gore is called playbooks. [15:56] >> A playbook is essentially a collection [15:59] of rules where you either approve or [16:01] disapprove something. So you might say [16:04] for the way that you would sign NDAs [16:06] here at YC, you always want the [16:08] definition within a confidentiality [16:10] agreement to look a certain way. So you [16:12] provide the rule, you provide some [16:13] example language and then you say, "All [16:15] right, if the opposing party will not [16:17] accept this definition, we have some [16:19] fallbacks." So fallback one and fall [16:21] back two. And you just open a document [16:23] in Agora, you open the playbook, and you [16:25] press play. And it goes through every [16:28] rule and runs it against the contract [16:30] and it marks it up. [16:31] >> Yeah. So if it does not conform with [16:34] your playbook, it gives you the [16:35] suggested language so that it will and [16:38] the really cool thing about this is it [16:39] scales outside of just legal [16:42] departments. So um at Lora every sales [16:46] rep is using Lora to negotiate NDAs [16:49] before sending it to our legal team. And [16:52] we just started working with this very [16:53] large bank in the Nordics and it's very [16:55] quickly moved from, you know, the legal [16:57] team to compliance to risk [17:00] >> and now to sales. [17:01] >> Wow. [17:02] >> Because everybody can leverage the [17:04] system. And the cool thing about it is [17:05] it's not only faster and more accurate, [17:07] but you agree on a standard [17:10] >> because the legal team then creates the [17:11] playbook and that becomes the standard [17:13] that everybody uses. So it actually [17:14] increases quality and consistency over [17:17] time. [17:17] >> YC's next batch is now taking [17:20] applications. Got a startup in you? [17:22] Apply at y combinator.com/apply. [17:25] It's never too early and filling out the [17:27] app will level up your idea. Okay, back [17:30] to the video. None of you guys when you [17:32] started were lawyers. [17:34] >> Oh, [17:34] >> so you still are building one of the [17:36] largest or fastest growing legal AI [17:38] company in the world. How do you do [17:39] that? [17:40] >> I think at this point I've become a [17:41] hobby lawyer. Uh but how we approached [17:44] it was being incredibly humble. Humble [17:47] for the fact that we did not know the [17:49] industry. We were quick to create [17:51] relationships with our early partners [17:54] where feedback was, you know, happening [17:56] daily. Yeah. [17:57] >> And I think that's been one of our [17:59] strengths as a company to say, we don't [18:02] know exactly where the future's going, [18:04] but neither do you. So, let's work [18:06] together to make sure that we're both [18:08] winners in, you know, whatever happens. [18:10] And I think now we of course have the [18:12] privilege of having hired a ton of [18:14] lawyers into the team that work directly [18:15] with the product teams and directly with [18:17] the customers. Especially in an industry [18:19] that is now going through such big [18:21] change. It was useful to come in with [18:24] more naivveness if you will saying [18:28] why does it work this way? You know it [18:30] could work this way instead. [18:31] >> Let's say you're a founder watching this [18:33] right now. You're like I want to build [18:34] AI software for logistics or for [18:36] insurance or finance. Is your advice [18:38] basically you don't need any expertise? [18:40] How do you how do you how would you [18:41] >> how do you learn about the things you [18:42] need to learn though? [18:43] >> I think my advice is, you know, learn [18:44] about them, right? Like we we went into [18:47] this and the first thing I did was I [18:49] interviewed 100 lawyers. I had this good [18:51] hack on LinkedIn. I I texted them asking [18:54] if we could have lunch and I would pay [18:55] their hourly rate and I I could def you [18:58] know definitely not afford it and none [19:01] of them would, you know, impose that. [19:02] They would just say, "Oh, that's [19:04] amazing. Like I'll have the lunch with [19:05] you anyways." One of the attributes that [19:08] have been very helpful in in my career [19:10] has been that I'm I'm I'm somebody [19:13] people want to help. [19:15] >> Um I think that's a very underrated [19:18] >> skill. I think there there are things [19:19] you can do to be more like that. Um you [19:22] can be a bit uh fearless in in your [19:26] approach to people and you can also be [19:27] very veryful and grateful and [19:30] appreciative of the work that other [19:32] people help you with. If we hadn't done [19:34] that, we would not be where we are [19:36] today. [19:37] >> And then how do you conduct a lunch with [19:38] a lawyer when you're starting a startup? [19:40] You know not much about law. [19:42] >> So you'd sit down like this. Uh you'd go [19:44] to somewhere decently nice because again [19:46] they make a lot of money. And it took me [19:48] some time to even understand that the [19:50] way that departments work are [19:52] fundamentally different. Like a [19:53] transactional lawyer works nothing the [19:56] way a lawyer within the corporate [19:57] department works. You just ask them a [19:59] ton of questions. And I think also [20:01] giving them something back. So, you [20:04] know, I'd reach out, they see my, you [20:06] know, tech background and you try to be, [20:09] you know, give them nuggets of, oh, [20:12] that's really cool. Have what do you [20:13] think about this like you give them [20:15] ideas, you make them engaged in wanting [20:18] to give you advice and Yeah. [20:21] >> And people generally feel good giving [20:22] giving founders advice of course like [20:24] it's like something that [20:25] >> you should take advantage of. [20:27] >> Yeah. and something that I'm you know [20:28] really happy to do now from the position [20:30] where we're at. [20:31] >> There are some large companies in in [20:33] legal technology. Um are you going up [20:35] against all of them or what how do you [20:37] think about the existing market of legal [20:38] tech? [20:38] >> Right. So there's been a lot of sort of [20:41] large M&A machines and incumbents in [20:43] this space for a long time. They're not [20:46] very popular with the end users. [20:49] >> Um I think they have very kind of [20:52] far faring roots. um there's some [20:55] advantages and you know data modes and [20:58] so on that that come into play but [21:00] effectively what AI has done is really [21:02] changed the game in terms of how quickly [21:04] you can ship something and it's created [21:06] a new category so a lot of again these [21:09] existing point solutions were in maybe [21:12] suites of of these M&A machines and now [21:15] a lot of it is becoming irrelevant very [21:18] quickly and the cost of building [21:20] software is also going down very very [21:22] rapidly So our ability to out ship or [21:26] you know uh outd deliver these teams of [21:29] you know thousands of engineers with [21:31] just 30 [21:33] >> is insane [21:34] >> and so [21:35] >> we have instead [21:38] managed to build a company with I think [21:40] at the at the time of recording it's [21:42] about 100 where like our velocity is way [21:45] higher than companies you know 100 times [21:48] our size. I think that's interesting in [21:50] and of itself in terms of how we built [21:51] the company over the last year because [21:53] when we came out of YC we were roughly [21:55] 10 people and now we're 100 and that [21:57] means we've onboarded on average like [21:59] two people a week and hiring correctly [22:02] it's really hard like it's a skill you [22:04] need to learn [22:05] >> and hiring for velocity hiring for you [22:08] know entrepreneurship and ownership of [22:10] different products and things but also [22:12] scale because the company is growing [22:14] exponentially so you need your your [22:16] teammates to scale exponentially as [22:18] well. If people scale linearly, at some [22:21] point it's a really large delta and then [22:23] you know things aren't working out [22:24] anymore. [22:25] >> Do these big companies have lock in like [22:27] the big legal tech companies? [22:29] >> So these big companies have a couple of [22:30] advantages but I think the the [22:33] disadvantages outweigh uh the advantages [22:36] almost 10 to1. There were very large [22:39] data advantages and you know being like [22:43] an incumbent where you lock in a large [22:44] contract. [22:45] >> Yeah. But I think the buyers have also [22:48] changed aptitude here. So we're not [22:51] seeing anybody want to lock in a [22:52] five-year contract, [22:53] >> right? [22:53] >> Because the world is moving so fast. [22:56] >> Um so we instead see them, you know, [22:58] doing one-year contract. [23:00] >> It sounds like a good motivation for [23:01] companies moving faster. [23:02] >> It is. Yes. And and but even law firms, [23:04] right? I mean, they don't want to be [23:06] locked in with a vendor. So they're [23:07] doing one or two year contracts. And you [23:10] know, as we see them now coming up in a [23:13] lot of places, they're also looking [23:14] outside of their existing alternatives. [23:16] So you might have made a bet back in [23:18] 2023 or 2024 when it was experimentation [23:22] days, but now you're looking at what are [23:23] we going to deploy, you know, more long [23:26] term and there what I'm seeing is yes, [23:29] people look at the technology, but even [23:31] more so they're they're zooming out and [23:33] they're looking at your rate of change. [23:34] >> They want to work with the partner [23:36] that's going to get them from point A to [23:37] point B. And they can be different [23:39] things. It might be we want to be AI [23:41] first and drive our top line or we want [23:44] to drive profitability and you know [23:46] streamline our operations. It can be you [23:48] know very different motivations. [23:49] >> Right. [23:49] >> How does your tech stack look like [23:51] what's under the hood [23:51] >> internally? Yeah. [23:52] >> Yeah. So building our infrastructure I [23:54] think from the beginning it was pretty [23:55] clear that we wanted to be on Azure just [23:57] because it was the same that our [23:58] customers were were on and in the [24:01] beginning I think OpenAI and and GPT was [24:04] really the only model that you could [24:05] serve via Azure. Now we have much more [24:08] um you know options available to us. So [24:10] we we use AWS and claude and Gemini and [24:12] GPT and Mistrol kind of interchangeably. [24:15] The biggest thing there has been how do [24:17] we build everything in such a way where [24:19] we can hot swap the models whenever we [24:22] want and also build it in such a way [24:25] that the models become better everything [24:27] improves. And now we've also looked into [24:29] classification models where you know if [24:31] you do a simple query will serve you a [24:33] simple model. If you do a complex query [24:35] we'll serve you a complex model and [24:37] that's just because that's just you know [24:39] to keep the margins down but also you [24:41] know sometimes you don't need a bazooka [24:43] when you just need a you know water gun. [24:46] >> So who is the buyer? Uh my understanding [24:49] is that law firms have but maybe you [24:51] could explain to me like there's a bunch [24:52] of partners and there's other people [24:53] there too right. How is a law firm or a [24:56] legal team at a company generally [24:58] constructed and who are there and who [24:59] buys it and who uses software? [25:01] >> It changes a bit depending on size. [25:03] >> Uh so if you start with the with the [25:05] biggest firms of course you have the [25:06] partner group that kind of runs things [25:08] but you very often have an innovation [25:10] department which sometimes have more or [25:12] less influence. Um if it's a very strong [25:15] innovation department, they make their [25:16] own choices. They procure software and [25:18] they're responsible for the entire [25:20] innovation agenda. I've frankly got the [25:22] most energy out of working with the [25:24] innovation folks who are really smart [25:28] about these things. Uh because there's a [25:29] lot of people that just want to kind of [25:31] check the AI box and then others who [25:33] really want to push things forward. And [25:34] the interesting dilemma there is, you [25:36] know, they're basically driving [25:38] efficiency across the across the stack [25:40] or across the firm, but they're not the [25:42] users themselves, right? However, you [25:45] might often have innovation [25:46] practitioners that work in the M&A group [25:49] or the disputes group or arbitration and [25:52] then they will work with those teams to [25:54] drive an upskill. So they will have a [25:56] very kind like process-minded way of [25:58] working and then they might use Lora to [26:01] build use cases for the end users [26:04] because when you work in a big law firm [26:06] you need to hit your building targets. [26:08] Yeah, [26:08] >> they work a lot. Like we grind as [26:10] startup folks, but [26:12] >> lawyers [26:12] >> lawyers grind as well. And if you know [26:15] that there's a way to solve something [26:16] and it's going to take 6 hours for you [26:18] to do that [26:19] >> and you know a way how to do it in 6 [26:21] hours, you might not take the chance of [26:24] exploring a way how you could [26:26] potentially solve it quicker or you know [26:28] with a with a with a higher quality. [26:30] You'll just conform to the way you're [26:31] used to working. [26:32] >> Yeah. So innovation teams have a huge um [26:36] opportunity and frankly you know mission [26:39] to drive that across the firm and if you [26:41] go down a bit so you have sort of [26:43] mid-size firms more often than not you [26:45] might not have an innovation department [26:46] and so it's the partners who are making [26:48] the move or or the decision and [26:51] >> what I found is [26:53] >> it's hard to get the entire partnership [26:56] to buy in. go deeper on this point [26:58] because I know a lot of founders is [26:59] asking me how do I sell to like a [27:01] financial firm or law firm or something [27:03] like that and it seems like this is the [27:05] the tricky part is like you have to [27:07] convince everybody. [27:07] >> You have to convince everybody or you [27:10] start smaller. [27:11] >> Okay. [27:11] >> You say let's work with this partner and [27:13] their team and make them rock stars. [27:15] >> Mhm. [27:16] >> And then everybody else looks at them [27:18] saying what's that guy doing? [27:20] >> Right. [27:20] >> That looks awesome. We also want in. And [27:23] then you expand. But the key here is to [27:25] sell sell sort of like not top down but [27:27] sell to the senior people first. [27:29] >> Right? So there's it's impossible to do [27:31] a bottom up motion in our industry [27:34] because you don't uh procure software [27:38] individually. You take it through [27:39] procurement and you take it through it. [27:41] >> Yeah. [27:41] >> And there's a lot of security checks. [27:43] There's a lot of uh data privacy checks [27:46] that you need to go through in order to [27:48] actually you know serve client data in [27:50] your systems. [27:51] >> You were 23 when you co-ounded Lora. [27:53] Yeah. Um, [27:54] >> by then you've already done a lot. You [27:56] had some stints at other YC companies [27:57] like multiple different ones. [27:58] >> Yeah. [27:59] >> What was your background before you [28:00] started this company? [28:01] >> When I was 18 and it was time to apply [28:04] to college, um, I actually had two [28:06] options. I was either going to go down [28:08] the route of becoming a professional [28:10] Dota Dota 2 player or go to college. [28:12] >> I knew this [28:13] >> and my thinking at the time was okay, [28:16] what's the best case scenario in each of [28:18] the outcomes? So best case scenario in [28:20] Dota would be to win the international [28:22] the biggest tournament in the world you [28:24] make $10 million that would be amazing. [28:28] Uh but then I was thinking what happens [28:30] then [28:31] >> you know it kind of feels like then life [28:33] stops. [28:33] >> Yeah. [28:34] >> And the best case scenario with going to [28:35] college was basically this what I'm [28:37] doing now. So I decided to go to [28:39] college. Um and when you apply to [28:42] college in Sweden um you go to one [28:44] school to do one program. So, the [28:46] engineering university is completely [28:47] separate from the business university, [28:49] which I think is really weird. Like, we [28:51] don't mix at all, which is bad. But [28:53] there was a hack so that you could you [28:55] could make an an admission to one of the [28:57] schools and then kind of pull the [29:00] admission to make another one [29:02] >> or pull your application to make another [29:03] one and then call them and say that you [29:06] messed it up and you wanted to get, you [29:07] know, reapplied. So, I I ended up uh [29:10] making it so that I could go to both [29:12] universities in parallel. It was a [29:14] really good timing during COVID to do [29:15] that because that means when you have [29:17] two lectures at the same time, you can [29:19] just have two laptops. [29:21] >> Record one. Yeah. [29:21] >> Yeah. And there were multiple times [29:23] where I had like exams at the same time [29:26] with both universities and you would you [29:28] kind of sit with one camera over here [29:30] and one camera over here pretending that [29:32] you were just doing one of the exams. [29:33] And so like one or two years into it, I [29:35] was working as a programmer. I was [29:37] building statistical models for esports [29:40] betting and that was really fun. But I [29:43] think I I also wanted to kind of see [29:45] what the business side looked like. So I [29:47] had the the the privilege of working at [29:50] a company called Norhen. It's like Y but [29:52] for impact and it's based in Stockholm. [29:54] And I think I got um a lot of exposure [29:57] to other entrepreneurs. And [29:59] >> what struck me then was one a few of [30:01] them were not super ambitious to build [30:04] companies that we're doing now, but they [30:06] sort of had this like fiveyear plan to [30:08] conquer Nordics. Yeah. So I think [30:10] immediately like I had a different take [30:11] on it and then did a short stint at [30:16] McKenzie um and worked at BAMLO and just [30:19] one week at Depict. [30:20] >> Depict was one of those companies uh is [30:23] is one of those companies that was an [30:24] incredible talent magnet. Yeah. [30:26] >> Like some incredible people have come [30:27] out of depict like an Lovable was one of [30:29] the founders. Um but there's a bunch of [30:30] others. You're starting the Gora even [30:32] though you spend a week there. But is it [30:33] it's like kind of cool how you have [30:35] these magnets that that that spun off to [30:37] a bunch of other cool companies. No, [30:38] they're amazing and u you know we're all [30:41] good friends in Stockholm. It's a small [30:42] ecosystem and it's really fun to kind of [30:46] cheer on each other as well. [30:48] >> And YC ended in uh April uh last year. [30:50] Can you walk us through sort of like the [30:52] company growth and your personal [30:53] development in this time like you were [30:54] 10, now you're 100. Um what happened? [30:56] >> We grew really fast and we were also [30:58] feeling the [31:00] >> the the drag. [31:01] >> Yeah. [31:02] >> You know like we we took the product to [31:04] market and you know we would sell it in [31:06] a demo. Mhm. [31:08] >> And when law firms start to buy things [31:11] after one demo, [31:12] >> you're doing something right. And so the [31:15] rationale was like, we should be doing [31:17] more of this and we want to do it [31:18] everywhere all at once. And this is also [31:20] a space where it's kind of obvious that [31:23] legal and LLM is a good fit. And so [31:26] there were a lot of other companies in [31:27] the industry. I like to say like there [31:29] there were so many legal AI assistants [31:31] and now it just feels like many of them [31:32] have kind of fallen off and they're [31:34] emerging a couple of winners. With that [31:36] rationale, we wanted also to get [31:37] American capital uh on the in the [31:40] company because we wanted to be able to [31:41] make the move from Stockholm to the US [31:44] when the time was right. After we raised [31:46] the money during our first board [31:48] meeting, we sat down and I remembered [31:50] the look on on some of our board members [31:53] faces when I basically said, "We're not [31:55] going to sell for the next four to five [31:56] months." And the reason for that was [32:00] when we got the chance to onboard a [32:03] client, it took a lot of work. took a [32:05] lot of work to get them to a level of [32:07] understanding of what they could [32:08] accomplish in the platform. And also the [32:11] first experience of a of a legal [32:13] professional logging in is the one [32:15] chance you have. If you mess that up, [32:17] they're not coming back. [32:18] >> And we had a couple of of of situations [32:20] where we'd onboarded a lot of people and [32:22] we had done, you know, some misses and [32:25] we didn't want to ruin that. So we [32:29] worked really hard on reliability, [32:30] scalability, got the system to a place [32:33] where we could comfortably onboard a [32:34] thousand lawyers a day. And once we had [32:36] that, we kind of let it rip. And that's [32:39] also when we really started to hire. So [32:42] we were maybe 25 in the beginning of [32:45] October. And just 6 months later, we're [32:47] now 100. [32:49] >> So what we did was we said, "Okay, we're [32:52] now going to scale across every market [32:53] in Europe and we're going to start [32:55] scaling towards the US." and our initial [32:57] conversations in the US some time [32:59] because we were a small Swedish startup. [33:02] Um so I made you know multiple trips [33:04] back and forth to New York and now we [33:07] open up hubs both in New York, London, [33:09] Stockholm and also have people locally [33:11] in Spain, France and Germany. [33:13] >> So we've really gone at it and just said [33:15] hey we want to do everything everywhere [33:17] all at once and let's do it now. [33:20] >> And for you personally and Sig like what [33:22] was your experience in that? I think the [33:25] biggest takeaway and learning is going [33:28] from being an IC into delegating. Yeah. [33:31] >> And that move, you know, you know how to [33:35] do something, but you that's not going [33:37] to scale. So, you need to teach somebody [33:38] else to do it. And you need to hire [33:40] people who are way better than you [33:43] >> on a lot of different topics. [33:45] >> So, one of the early sort of hires that [33:47] we made were actually another YC founder [33:49] and we've ended up Jake. Yeah. And we [33:53] actually we've scaled the team with a [33:54] lot of entrepreneurs. And that's not [33:58] only like the skills we're looking for, [33:59] but it's also like the way that we built [34:01] the company because we're effectively [34:03] running multiple companies within the [34:05] company. [34:05] >> It's sort of like a secret playbook that [34:07] a lot of YC companies, some of the best [34:08] ones, are all following is that the [34:10] first people you want to hire all former [34:11] founders. And it's kind of actually an [34:13] advice I got from Paul Graham back in [34:15] the days is that um sometimes you think [34:17] you're a founder that I work in this [34:18] company for 3 years didn't go well. Am I [34:20] less attractive in the job market? Like [34:22] if you're here or if you're in a startup [34:24] center, you're actually more attractive [34:26] in the job market because people [34:27] actually want to work with people like [34:28] you. [34:29] >> Yeah. And we want to hire them. So um [34:31] it's been amazing. And also the agency [34:35] and the the attitude to problem solving. [34:38] That's kind of what you're looking for. [34:39] And then sometimes you need to hire for [34:41] scale, right? Like now we have a [34:43] significant sales team and you need [34:45] somebody who's seen the 100, you know, [34:48] 10 million to 500 million because that's [34:50] the journey that we're on. [34:51] >> And my learning from M&B, which probably [34:52] I'm sure applies to you, is the culture [34:54] in the beginning is the people that you [34:55] hire, [34:55] >> of course. And when we've now scaled the [34:58] hubs, um, we always send a person from [35:00] Stockholm with them. It's the best [35:02] people from the Stockholm office that [35:03] then travels and set up the new hubs. [35:06] >> You seem like the kind of person who [35:07] embodied the attributes, you can just do [35:08] things. So can you tell me how that is [35:11] reflected in your company? [35:12] >> You can't just do things. And when we [35:15] started building this company, we didn't [35:17] know anything about law, [35:18] >> right? [35:18] >> Um I think that was pretty apparent in [35:20] our first interview and we you know made [35:23] the right moves from then from then to [35:25] the second one where we showed that we [35:27] could do it. [35:28] >> You apply for two different batches. [35:29] >> Yeah. The first one didn't go as well. [35:30] And so about this attribute, it's [35:32] something I look for in others as well. [35:34] Um during a lot of the interviews I do I [35:36] ask I often ask the question you know [35:38] what have you done outside of your role [35:40] for the company and here I'm looking for [35:44] creativity ability to spot problems and [35:47] solve them [35:48] >> and to take responsibility for more [35:50] things than just the stuff that you're [35:51] doing. Right. Right. And I think in [35:53] terms of starting companies and you know [35:56] building the future because frankly we [35:58] need to reimagine a lot of like the [36:01] stuff that we're doing we don't want [36:03] people who are bogged down by your boss [36:06] telling you to do something right we [36:08] have a very sort of flat organization [36:12] where let's say our marketing team we [36:15] want generalists who are using AI to do [36:18] 10x more work than they could have done [36:21] in the past And where you might have [36:23] needed a 30 person marketing team, you [36:25] now need five. [36:27] >> And you want those five people then to [36:29] be complete, you know, yes. And to go [36:33] out, you know, above and beyond. And [36:35] that that characteristic, I think, is [36:37] increasingly important as well in an in [36:39] an age where if you're really ambitious, [36:42] you can get a lot of leverage out of [36:44] tools. [36:45] >> Absolutely. So, if we fast forward like [36:47] five or 10 years, uh how does the [36:49] day-to-day job of a lawyer look like? [36:51] That's interesting. We think about that [36:53] a lot, right? Um I I'm kind of viewing [36:56] it as you're more and more entering a [36:59] work space of reviewing work than [37:02] actually doing it [37:03] >> and you are managing the expectations [37:05] from your clients and the expectations [37:07] and the work from your AI agents, right? [37:10] You're you're effectively instructing [37:11] them. you're watching them go out and do [37:13] work and you're making sure that [37:15] everything they're doing is not only [37:17] correct and sort of at your standard but [37:19] you're also managing how that work gets [37:21] delivered to the client because I think [37:23] you know you you will always want [37:25] somebody who knows their stuff. [37:26] >> Yes. [37:27] >> On this and there's a big reason for why [37:29] we're working with lawyers and not with [37:32] the people who might you know use the [37:34] legal services because a lawyer is [37:36] needed and and necessary to deliver the [37:38] end product. But looking 5 10 years [37:40] ahead in in these days is also [37:42] >> it's hard. [37:43] >> It's hard, right? Um if I knew where the [37:45] AI models would be 10 years from now. [37:46] >> Yeah. We're looking weeks ahead now [37:47] right now. [37:48] >> Yeah. It's and and that's funny just [37:50] with our product road map. Um I tried to [37:52] do them kind of like many quarters [37:54] ahead. [37:55] >> Yeah. [37:56] >> Yeah. It's hard. [37:57] >> It's really hard. [37:58] >> Do you think uh that the large AI labs [38:01] are going to try to attempt at doing [38:03] law? Maybe not law specifically, but I I [38:05] do feel like they're more and more [38:07] becoming platform companies rather than [38:09] model providers. I mean, Google is [38:11] building Google Workspace with Gemini. [38:14] Anthropic is running very hard on the [38:16] MCP idea of building kind of a universal [38:19] entry point into a lot of applications. [38:21] >> I think the the expectations on [38:23] companies like us are pretty clear. you [38:26] know, whatever comes out of a model lab [38:28] is kind of expected and then [38:32] >> everything else we're adding on top is [38:33] kind of like icing on the cake. [38:34] >> How does it feel to product market fit? [38:37] >> I think the feeling is best summarized [38:38] by almost like this drag feeling or kind [38:41] of infinite [38:42] >> you're being pulled into the market, [38:43] >> right? um it like it literally feels [38:45] like we have infinite demand and I think [38:47] that's it's coming from a point of the [38:50] product is working and it's moved from [38:53] being in this experimental AI bucket [38:56] into we are relying on this for core [39:00] work that we are delivering right now if [39:03] something breaks you know immediately we [39:06] get a phone call saying hey we can't do [39:08] this like what's going on right and we [39:10] fix it it's basically been this point of [39:12] you start out. You hope that what you're [39:15] doing is the right thing and you try to [39:17] get early partners excited about what [39:18] you're doing. And in the beginning, to [39:20] be, you know, really frank, a lot of [39:22] people got on with us because they [39:23] wanted to be on the journey and they [39:25] took a bet. [39:25] >> Yeah. [39:26] >> And I am so thankful and happy that they [39:29] did that because now we've taken them [39:31] from from point A to point B and we're [39:34] continuously scaling from here. [39:36] >> So we tell companies to move to San [39:37] Francisco [39:38] >> generally. [39:39] >> Uh you decided to not take that advice. [39:41] Can you just like tell us a bit about [39:42] the thinking here and and maybe like if [39:44] you have some pros and cons? [39:46] >> Yeah. [39:46] >> About not being here? [39:47] >> Yeah. The reason why we stayed in [39:49] Stockholm was um we needed a market to [39:53] grow in. And if you go to the US, it's [39:56] not only more competitive, but I think [39:58] it kind of pushes you into becoming a [40:01] more narrow company. You start building [40:03] really horizontal and then you realize, [40:05] wait a minute, we're really good at [40:07] this. So you start to scale it in other [40:09] markets and you quickly notice ah we're [40:12] the best in Finland too and we're the [40:14] best in Denmark and we're the best in [40:15] Norway and then you scale to Spain, [40:17] France and Germany, London and then the [40:20] states and at that point we had always [40:22] you know we had already done [40:24] >> 15 new market entries. The algorithm or [40:27] the the method was already kind of [40:29] established. Of course, the US is a [40:31] bigger undertaking, but we had also then [40:34] grown from this small fish in a small [40:37] pond to crocodile or a shark in in the [40:40] bigger pond now. [40:41] >> So, you you've raised $80 million like [40:44] in midmay. You open an office in New [40:46] York. Uh you launched with one of the [40:48] most famous law firms here in the US. [40:50] >> Um it seems like you're trying to [40:51] position yourself as the category leader [40:53] of AI law in the world. [40:54] >> Yeah, I mean 100% and I think in many [40:56] aspects we're already there. It's for me [41:00] more of a question around ambition and [41:02] what's next. It's very easy to say, hey, [41:05] we see this problem. Let's go solve it. [41:07] And then you get satisfied. [41:09] >> But it feels to me like every time we [41:11] solve a problem, a new one emerges. And [41:14] we're finding that as we go deeper and [41:16] deeper and deeper in the entire legal [41:19] software stack, we're also seeing that [41:22] the line between software and service is [41:27] blurring. AI is continuously developing [41:30] super super quickly and that means we [41:32] need to do the same. And so in my mind, [41:35] the category leader in the space does [41:38] not only build software, they serve as [41:40] the strategic partner to these large [41:43] firms and they make them win in this [41:45] transition because it's a very large [41:48] transition. And that's also why we've [41:51] basically scaled headcount as as quickly [41:53] as we've could whilst maintaining kind [41:56] of culture, urgency, and velocity. So, a [41:59] lot of founders that I I meet are asking [42:01] me questions about how you build a [42:02] vertical AI company that seems like the [42:03] kind of companies people are building [42:04] now. Do you have any general advice you [42:06] want to give to those founders who are [42:07] just starting out? [42:08] >> The first kind of obvious tip is don't [42:09] get locked in with a provider and don't [42:12] compete with the AI labs. [42:14] >> The AI labs ship, right? And so does [42:17] companies like Perplexity and and [42:19] others. And so, I think you want to be [42:20] really clear and honest to yourself [42:24] where you're adding value and where [42:26] you're adding long-term moat. And this [42:28] is something that we've thought a lot [42:29] about at Leguire, like how do we build [42:30] things as boats so that when the tide [42:33] rises, just everything gets better. [42:35] >> If you're just starting out, you got to [42:37] realize that you do not have the [42:38] capacity to outperform any of those [42:40] companies. [42:41] >> You kind have to find a nar a narrow [42:43] category to do it where you know the [42:45] models won't get to or [42:46] >> either that or finding out a way to [42:48] leverage the models very creatively. I [42:51] mean in in a way that others haven't [42:52] done it. I think take AI scribing is a [42:55] good one. Like typical AI scribing is [42:58] hard to do [42:59] >> and you need to embed a lot of like [43:00] custom prompts and ways to get it right [43:04] so that it uses the right medical [43:05] language [43:06] >> which is very similar to law. Like you [43:08] needed to write clauses in a way that a [43:10] lawyer would write a clause, [43:12] >> not just what the model spits out as the [43:14] most probable answer. [43:15] >> If I'm watching this video and I'm like [43:17] I'm thinking about applying for a job at [43:18] Agora. [43:19] >> Yeah. [43:19] >> Uh tell me about what I should expect [43:21] either from the application process or [43:23] from working there. The things that we [43:25] look for are ambition and the [43:28] willingness to say we got this huge [43:30] problem. There's this huge mountain. How [43:32] do we climb it and we're also very [43:35] upfront with candidates that this is not [43:37] a 9 to5 and we're not the traditional [43:40] Swedish working environment? We have the [43:42] good stuff. We have the Fica, but uh we [43:45] we have a lot more hunger and you know, [43:49] frankly, a lot higher expectations and [43:53] we we want that not only for ourselves [43:55] but for each other because we want to [43:57] grow as people and we want to grow as [43:59] entrepreneurs and as a company as [44:01] leaders and I think there just looking [44:04] at like our our application process the [44:07] biggest thing we do is a lot of cases [44:09] right if you want to come in and work in [44:12] our go to market team, you need to come [44:13] and pitch us our product. [44:15] >> Yeah. [44:15] >> And you need to do a really strong [44:17] pitch. And you know, if you take the [44:19] engineering team, we basically ask you [44:21] to build a P of Legora [44:24] >> and you know, we want you to work with [44:27] AI generated code, but we also want you [44:29] to be able to explain it. Yeah. Right. [44:30] And to design systems that scale. And I [44:34] think [44:35] >> Stockholm is a small ecosystem. And so [44:37] it's also quite easy to make references [44:40] and see who's actually good and who's [44:41] who's been in a company and made them a [44:44] success. You know, not only was there [44:47] the right time exactly and another [44:49] really big piece is we're hiring all [44:51] over Europe. So um we've had people move [44:54] from Madrid, from Amsterdam, from [44:56] Germany, from Paris all the way to [44:59] Stockholm. uh we tend to not onboard [45:01] them in November when it gets nasty, but [45:03] I feel like we've started to build this [45:05] sort of AI hub together with many other [45:08] companies that is not only like super [45:10] fun, but also, you know, great companies [45:13] come out of it. [45:13] >> Thank you so much for coming back to IC. [45:15] >> Thanks, Gustau. [45:20] [Music]