Legora CEO Max Junestrand：单日新增700万美元ARR｜Harvey vs Legora：法律AI是否赢家通吃？
- Max Junestrand 对核心市场的判断是：法律AI“完全是赢家通吃——就像所有SaaS一样，第一名拿走90%，第二到第十名瓜分剩下的10%”。 他用一张Bloomberg信息图为Legora对抗Harvey的势头背书：除Microsoft Copilot外，Legora已是英国前200大律所部署最多的生成式AI工具；客户数一年内从50家增至750家，员工数从30人增至300人——“谁先到并不重要，重要的是谁最好”。
- 标题里的企业模型份额信号是：Legora在2023年及2024年大部分时间里只用OpenAI，如今则“相当铁杆 Anthropic”。 转换发生在Sonnet 3或3.5前后。他对未来24个月的排名是：Claude或Gemini第一，OpenAI第三（“不，我们完全不会把Grok放进来”），理由是“Anthropic越来越偏企业，OpenAI越来越偏B2C”。但忠诚有前提：“我们会非常花心——这句话可以截出来。”
- 谈到能力，他认为Opus 4.5在编码方面已经“应该直接把它定义为AGI，然后专注于优化成本”。 GPT-3.5“像是在管理一个不太聪明的员工”；Opus 4.5“就像一位副总裁，你告诉它：这是我想要的事，去执行，然后它就做了”。他认为Harvey早期的错误，是在2023年投入精力微调模型——“我们应该造船，潮水上涨时，所有产品都会一起变好”——因为80%的价值来自应用层“构建普通软件”。
- “Harvey赢了美国、Legora赢了欧洲”这句话里，我认为有一句是真的。 Legora年初在美国一个人也没有，如今已有50人；美国已经是其收入最大的单一市场，预计第一季度末将超过整个欧洲——此前它采用的滩头阵地策略，是在开设第一家美国办公室前，先从欧洲签下2家AmLaw 200律所（White & Case和Goodwin Procter）。
- 本期最亮眼的增长数据是：Legora在2025年12月单日新增700万美元ARR，超过2023年和2024年新增ARR之和。 在连续6个季度每季翻倍后，年底达到200这个目标“确定，绝对确定”。但他坦承，所有客户仍在“把这当成一次延长版试点，以及押注AI的期权——就像看涨期权”：合同期限为1至3年，没有5年合同；拿它与Harvey声称的98%留存率和178% NRR相比还为时过早——“要到2026年，才能知道真实数字会是多少”。
- 他罕见地坦言，按席位收费“对买方来说最优，但我不认为这对我们最优”，而且3年后的席位定价“绝对不会”继续存在。 重度用户会推高不可持续的LLM成本。当前毛利率“还行……但不是SaaS毛利率”，上限仍然很高，因为AI工作的定价基准是“我会花多少钱请一位律师真正出去完成这项工作”，而不是其他软件的价格。眼下是“圈地时间”，还不是优化利润率的时候。
- 对终端市场的判断是，律所将进入由PE推动的整合期：“我不认为还会有AmLaw 200，我认为会变成AmLaw 20，或者可能是AmLaw 12。” 大型律所和小型律所会胜出，中型律所受到挤压，初级律师数量减少，而计时收费的变化速度会远慢于预期。Harry反驳称，劳动力替代将在12至24个月内反映到统计数据中；Max部分让步：“从总量层面看，是的，可能如此。”
1. 法律AI是赢家通吃——“没有第二名”
- Max对这一赛道的框架是：“这完全是赢家通吃——就像所有SaaS一样。第一名拿走90%，第二到第十名瓜分剩下的10%。” 现实层面的后果是：“你得拼命跑，必须赢。没有第二名……只有胜利，其他一切都是失败。”
- 这不是Uber对Lyft：网约车“没有产品差异化，产品基本一样”。但在法律AI里，产品差异化非常重要，可构建的空间是“这个从未真正被建立起来的法律科技宇宙”——生成式AI之前的法律科技创始人，往往在营收只有几百万美元时就接受改变人生的收购报价，“没有独角兽结局”。
- 对Harvey品牌领先的反驳，是一张Bloomberg信息图：除Microsoft Copilot外，Legora已是英国前200大律所部署最多的生成式AI工具，Harvey排名第二——“谁先到并不重要，重要的是谁最好。AltaVista还是最大浏览器时，Google并不是你首先想到的名字。”
- 势头的证明是：员工数在整整12个月内从30人增至300人，客户数从50家增至750家；这些客户来自一场场比选，律所“基本上是在玩VC”——它们买的不是一个工具，而是“今天就能得到的结果……以及一支AI赋能的法律团队将会是什么样子的愿景”。
2. 从只用OpenAI到“相当铁杆 Anthropic”
- Legora在2023年及2024年大部分时间里只运行OpenAI，如今则“主要使用Anthropic”，转换发生在“Sonnet 3……或者3.5”前后。背后的逻辑是，模型开始要求不同的提示词，于是“选一个好模型，然后押注到底，围绕它构建整个应用”。
- 他看到的结构性分化——Harry也以“100%”表示认同——是：“Anthropic越来越偏企业，OpenAI越来越偏B2C”；“而我们是企业级系统，所以应该更多受益于它们的模型”。至于OpenAI转向允许用户微调模型，这条路径“到目前为止，我没有太多理由相信”。
- 他对未来24个月的模型排名是：Claude或Gemini第一，关键取决于上下文窗口是否成为决定性因素；但目前还不是，因为Legora“已经围绕如何处理上下文窗口不足，搭建了大量架构”。接下来是OpenAI。至于Grok：“不，我们完全不会把Grok放进来。”
- 但他没有感情用事：“我们会非常花心——这句话可以截出来。”客户把Legora托付为自己的AI伙伴，所以“如果Gemini更好，我们会立即切换……前提是评测结果更好”；同时，用户可以固定具体模型，以支持确定性工作流。
3. 应用层胜过微调——Opus 4.5“就像一位副总裁”
- 被问到Harvey做错了什么，Max点名的是2023年微调模型：通用模型进步太快，“我们应该造船——潮水上涨时，所有产品都会一起变好”。他承认这个判断部分是被现实逼出来的——3名工程师和5万美元天使资金，不可能花300万美元做微调——“但结果证明它是对的”，而且“我怀疑今天这会成为决定胜负的差异”。
- 他的价值金字塔以模型为底座，上面是Legora对模型的“法律解释”；但“80%的价值来自构建普通软件”，也就是企业级基础设施。顶层则是客户在做差异化的事：律所正在“内部用vibe coding写工具”，并搭建MCP服务器，因为律所过去靠专业能力、招聘和费率竞争，而如今“越来越是在技术上竞争”。
- 模型是否正在触顶？“不，Opus 4.5很棒……它在编码领域已经到了这样的程度：你应该直接把它定义为AGI，然后专注于优化成本。”对比之下，GPT-3.5“像是在管理一个不太聪明的员工”；而“Opus 4.5就像一位副总裁。你告诉它：这是我想要的事，去执行。然后它就做了。”
- 他正把Claude Code和新“co-work”工具中的范式搬进法律行业：给Legora智能体接入整个生态的工具和客户MCP服务器，让它规划并执行——“基本上就像合伙人与资深律师一起工作”，成为“律所层级中一个全天候可用的新底层”。合伙人已经会把同一项任务同时交给律师和Legora，“而且很多时候质量相当不错——这会带来实质性影响”。对于Jason Lemkin关于今年将出现全天候推理的判断，他说：“Legora里还没有任何任务需要运行12小时”，但睡前启动任务的世界“肯定会到来”。
4. “Harvey赢了美国，Legora赢了欧洲”——“有一句是真的”
- 对于美国VC圈的共识，Max给出的反事实是：Legora年初在美国没有任何一名员工，如今已有50人；本周将在曼哈顿开设办公室，预计容纳150人；今年还将开设另外3家美国办公室。美国已经是其收入最大的单一市场，预计第一季度末将超过整个欧洲的收入，包括北欧——在那里“我们基本上和所有大律所都有合作”。
- 考虑到欧洲市场曾有项目多次失败（“我想Klarna可能试了几次才真正成功”），他的进入策略是先从欧洲签下并服务2家AmLaw 200律所。他通过演示、飞行出差和“始终具有竞争性的试点”，拿下了华尔街白鞋律所White & Case，以及“都排在前20名”的Goodwin Procter；在此之前没有大举投入美国市场营销或B2C内容。
- 他提醒创始人注意一个结构性优势：美国的终止期为2周，而瑞典大约是3个月。当公司每季度翻倍时，“我知道需要某个人的那一刻，如果他要等一个季度，我们已经是另一家公司了”。
- 至于“美国人更努力”的说法，他认为“有点扯淡”——Legora的美国文化是从欧洲移植过去的。不过他承认，纽约“永远在线，永远醒着，这就是我的节奏”，而斯德哥尔摩“有点困”。去年他有200天在路上，实际上“住在飞机上”。
5. 单日新增700万美元ARR——以及拒绝出售公司的6个月
- 头条数字是：2025年12月某一天，Legora单日新增700万美元ARR，“超过我们2023年和2024年新增ARR之和”；此前它已连续6个季度“每个季度都翻倍”。年底能否达到200？“确定，绝对确定。否则……你们可以羞辱我。” 他真正的目标是成为默认选择：“如果你是一名律师，要做严肃的法律工作，你就会用Legora。它就像Figma。”
- 对于是否能达到Harvey声称的98%客户留存率和178% NRR，他回答：“这两个数字，都可以。”但他认为，在增长规模远超2024年续约基数的情况下引用NRR并不公平：“要到2026年，才能知道真实数字会是多少。”
- 关于客户黏性的坦诚保留是：每个买家“仍然把这当成一次延长版试点，以及押注AI的期权——就像看涨期权”。合同期限为1至3年，而不是5年——“很多律所已经存在了200年……两年只是眨眼的功夫。”
- 他最有魄力的决定，发生在第一次董事会会议上：当时公司只有12名员工，刚从Benchmark融资1000万美元，并在一个月后以1.5亿美元估值从Redpoint融资2500万美元；他宣布Legora接下来6个月完全不出售公司，先重建基础设施，因为“我们在这些律师身上只有一次机会……如果产品不能运行，他们不会再回来”。到2024年10月1日，公司已经可以每天为1000名律师完成入职；“如果我们继续向前硬推，只会把所有客户都流失掉。”
6. 按席位收费不对——他直说
- 他罕见地坦言自己的收入模式存在问题：按席位收费“对买方来说最优，但我不认为这对我们最优……我其实不认为这是正确的定价模式”。单个用户可能产生高额LLM成本，最终变成“按用户计算不可持续”。3年后的席位定价？“绝对不会。” 转向按用量收费，要等“客户准备好按用量购买”，而不是等Legora准备好。
- 与此同时，用量扩张“对留存极其有用，但对收入优化没用”——更多使用“实际上会贵很多”。被追问毛利率时，他说：“我们的毛利率还行……但不是SaaS毛利率。”不过他相信最终能达到理想水平，因为定价上限并不由软件可比公司决定：“你的定价基准是，我会花多少钱请一位律师真正出去完成这项工作。”
- 为什么现在不效仿Lovable，通过模型路由优化毛利？他们确实做了一些，但“我不认为现在是优化模型利润率的时候”。Harry的说法被他当场接过来：“你们现在处于圈地时间。”
7. 产品纪律：删掉代码、押注3件事、卖铲子
- 他承认有两个错误：2023年夏天的v1版本只有点击式使用场景，没有智能体或聊天功能，“显然走错了方向”，YC录取后所有代码都被删掉；2024年则出现产品失控，12名工程师在Max外出销售期间做了“6或7个不同的东西”，最后变成“弗兰肯斯坦怪物”。修正方案是2024年10月至11月的“Leya产品宣言”（当时Legora仍叫Leya）：砍掉5到6个产品，坚定押注智能体、助手、表格审阅和Word插件，把它们做成“钱能买到的最佳组合”。
- 对于竞争对手抄袭——有一家竞争者甚至上线了一个字面上就叫“表格审阅”的功能——他认为这没关系，因为在竞争性试点里，“用户会开始把它们拆开比较。那时你就能看到，一个产品是劳斯莱斯，另一个可能只是沃尔沃。”（Harry补充说，Harvey内部据称称Max为“他们的首席产品官”。）
- 对于Crosby式的AI原生律所，他认为这不是赢家策略：低复杂度切入点（NDA、MSA）很快会拥挤，大律所本来就免费做NDA，以赢得“昂贵的私募股权业务”，而且“只要AI能完成一项任务，它就会完成那项任务”。他更愿意成为“卖铲子给全世界那些才华横溢的律师的人”。Solve这样的垂直专业公司可以赢——Harry称其为面向专利律师的AI公司，对应一个4850亿美元的市场——也可以成为中央枢纽的节点，由Legora发起调用：“我们为什么不直接联系Solve Intelligence，让它来帮我们写这份专利？”
8. AmLaw 200变成AmLaw 12——初级律师减少，但计时收费仍会保留
- 结构性判断是：律所将进入“一个相当显著的整合期”，而且私募股权如今也想参与进来——“我不认为还会有AmLaw 200。我认为会变成AmLaw 20，或者可能是AmLaw 12。”但不会变成四大，因为监管和时间仍是约束。
- 机制在于：基础的并购业务缺乏差异化，均衡价格假设为10万英镑。第一家借助AI、以更低价格提供同等质量，或者更高速度的律所，会“打破均衡——所有人都必须跟进”。市场将变成争夺份额的游戏：大律所凭借护城河、品牌和数据胜出，小律所凭借私人关系也能做得很好，而中型律所受到挤压——“把AI加入其中，竞争只会更加激烈”。
- 初级律师和实习生会减少，但律所整体规模会变大。他已经看到律所在营收增长的同时不再补充空缺职位，“所以利润更高”，并流向合伙人。他用工程师作比喻：“我们雇佣了更多工程师，尽管他们写的代码更多。” Harry对此反驳称，劳动力替代将超过预期，并在12至24个月内反映到劳动力数据中。Max部分让步：这正是“AI在我们的垂直领域足够擅长、能够非常确定地完成端到端任务”的时间窗口；单家律所可以做大蛋糕，但“从总量层面看，是的，可能如此”。
- 计时收费不会消失。“计费方式的变化会比你我想象的慢得多”，因为客户自己就要求看到明细；固定费用会按业务领域逐步增加，同时在其上方保留“一种模糊的计时收费”。
9. 传教士、雇佣兵，以及如何选择合作伙伴
- 当前最大的挑战不是竞争，而是在2个季度内把员工数从300人增加到600人，同时保持“让我们走到今天的雄心、诚信、团队协作和纯粹的韧劲”。他仍然亲自面试每一个人——“我想培养传教士，而不是雇佣兵”——并有意制造竞争，甚至细化到微观层面：除夕签单、圣诞热红酒派对上实时展示销售看板。他自己的变化是：过去拿到McKinsey offer时，只是“买了一包花生庆祝”；今年Legora学会了庆祝，因为感受到胜利，才会感受到失败，而“你必须看清世界本来的样子”。
- 团队结构本身就是一种判断：工程团队全部位于斯德哥尔摩，按pod组织；工程、产品和设计人员中有10%是YC创始人——“有一位YC创始人在负责产品的一部分，他们会为自己的事情拼命推进”。
- 他身体力行的YC建议是：“把生意做好，融资自然会很容易。” Benchmark给出的价格“是所有种子基金里最低的”，另一家机构则提供了高得多的抢先投资条款；但他为了选择那位合伙人，接受了折价（可能是Benchmark的Chetan）：“我大概在两周内见了80位合伙人。没有人比他更了解我的领域……他已经让3家公司上市。” 下一家公司只能带一位投资人？“Chetan。很简单。” 他收到过的最佳建议，来自YC以及可能是Sana Labs的Joel：“选Benchmark的支票，别选其他任何人的。”
- 他希望自己早知道的是，全情投入的心理成本：两年半里什么都不想，只想着这一件事，“就像在马拉松中跑一场超长冲刺。我很享受它。”但如果一开始就知道这一点，“我可能会更擅长处理那些无法投入同等精力的人生部分”。
- 他与Spotify的“可能是Daniel Ek”见面时非常紧张。成长过程中，他从“可能是Niklas、Daniel和Sebastian”以及他们打造的成功瑞典科技公司中获得灵感；如今则钦佩接替Daniel的Alex和Gustav。他没有每天崇拜的偶像，而是从优秀范例中汲取灵感，然后朝着它们全速奔跑。
It doesn’t really matter who was first; it matters who’s best.
Last week we had Harvey on the show. This week, we have their biggest competitor, Legora. Joining me is Max Junestrand, co-founder and CEO of Legora.
In a single day in 2025, we added $7 million in ARR in 24 hours. That was more than what we did in 2023 and 2024 combined.
There’s this perception that Harvey has won the US and that you have won Europe.
I think one of those statements is true. It’s totally winner-takes-all. You know this: number one will grab 90%, and numbers 2 through 10 will share the remaining 10%.
What that means for us is that you’ve got to run like hell. You’ve got to win. There’s no number two. There is only being number one. There’s only winning, and everything else is losing.
Ready to go. Dude, we did our last show, and I have to admit, I was so surprised—not that this sounds awfully rude, but it’s the end of the day—by how well it did, specifically in this incredible founder community, where I got pinged by 300 or 400 founders, which is more than normal, actually.
That sounds like a big number.
Yeah, it is pretty solid. Mostly, it’s just VCs. Thank you so much for agreeing to do a second show with me.
Always. I’m happy to be back.
Before I grill the shit out of you, 60 seconds: what does Legora do? Just to set the scene for people who don’t know.
Legora is the platform where legal work happens. I think that’s a better pitch than the one I had last time.
What started to happen more and more is that AI is doing more and more parts of legal work, and this has to happen on a centralized platform. We started out with simple assistant-based use cases, but this has grown tremendously, and it’s solving different types of tasks for different types of lawyers.
If you’re a transactional lawyer and, as part of a due diligence process, you need to review the data room and find the red flags in that data, Legora can do it. If you’re a litigator and you’re preparing a brief and drafting that in Word, Legora can help you do it right.
More and more of these tasks are being bundled into the platform, and what we’re seeing is that a bigger and bigger part of a lawyer’s day is being spent on Legora, which is amazing.
That’s my favorite data point. Is that the number one metric you use in terms of product metrics?
Yes. Time spent on the platform, the number of messages, the number of queries, and the number of actions taken—I think those are the best KPIs.
1. Why Does Everyone Think Harvey When They Hear Legal AI?
When people think AI law, there are a ton of players around the space and across the verticals, but there’s you and there’s Harvey. If we’re blunt, Harvey is the first name that comes up.
When you think about that, why is that?
I don’t necessarily think that’s the case anymore. The reason I say that is that I just saw this report. I was on Bloomberg a couple of weeks back, and they had a big infographic that said the most deployed generative AI tool in the top 200 law firms in the UK, outside of Microsoft Copilot, is Legora. Number two was Harvey.
We are moving, and the category is moving, at such a rapid pace that it doesn’t really matter who was first. It matters who’s best, and it matters which clients are coming back and want to do more work with you.
What often happens is that firms will throw many vendors into a bake-off because they’re in this luxury position. Basically, they’re playing VC. They get to bring in all these different vendors and say, “We’re going to do a bake-off.” In the bake-off, it’s up to the vendor to display why it should be their partner of choice.
I say “partner of choice” because I don’t think that these law firms or big in-house legal teams are buying just a solution. They’re buying an outcome today, but they’re also buying an outcome tomorrow, and they’re buying into a vision of what an AI-enabled legal team can look like.
When they look across the board and see all these different companies, they make qualified bets. More and more, we’re seeing those firms make that bet on Legora.
This year alone, I went into our HR system before I came in here, and we went from 30 to 300 people in 12 months exactly in headcount. This time last year, we were working with roughly 50 clients. Now we’re with 750.
A name might be associated with a category. I’m pretty sure Google was not the name you thought of back when AltaVista was the biggest search engine, but now it’s synonymous with searching on the web.
I want to unpack a couple of elements. You said that partnership is very central to how they think and how they choose.
I had Matt Fitzgerald, the founder or CEO of Invisible, which is kind of like a Mercor or Turing competitor, and he said that it is essentially impossible to sell into enterprise without an FTE model.
Mm-hmm.
Do you agree with that? Are you seeing that?
We have a very big team of legal engineers who are ex-practicing lawyers from the top-tier firms, but they’re not fully seconded. They’re forward-deployed, in the sense that their main job is to make you successful.
An example would be a big firm that just went with Legora, White & Case. White & Case now has the challenge of adopting AI across their entire firm. It’s a big firm, and it’s such an enormous change-management undertaking to equip all the lawyers across all the different practice areas, across all the offices, and across all the skill levels—from associate and senior associate to partner—with AI proficiency.
We need to make them successful because if they’re not successful on Legora a year from now or 2 years from now, it’s going to be a really sad conversation. We invest a ton of upfront manual labor, time, and effort in doing the implementation and activation right.
I do think that’s necessary for enterprises where you’re changing the way they work. If you just think about a process—for example, if you’re working in legal and you’re working with AI contracting—you basically have a contract lifecycle management system. You send a document somewhere, generate some redlines, and then put it back. That’s pretty easy. You don’t need a forward-deployed engineering or legal-engineering model for that. You just deploy the stuff, and then you’re done.
In our case, I actually like to think of it as analogous to the way that accountants had to learn Excel or architects had to learn CAD. Before, you would actually go out to the site, draw the building, go back to your office, and do all the math.
Now, you just get a picture of the site, throw it up in CAD, put in the blueprints for the building that you want, and the system generates it. Then you look at the math behind it, and you bring taste and design.
When architects were learning CAD, I think it was an enormous change-management effort. All the old architects would say, “Harry, are you going to use that computer system to do the hard work for you?” You would say, “Yeah, I’m super savvy, and I’m going to get to spend more time doing the design or having creative ideas about how to solve my client’s architectural problems.”
I think that’s synonymous with what’s happening today.
You said that the value is not in the first-mover advantage, and that the value can actually come from being second. What did Harvey not do well that you learned from, as specifically as possible?
Some of the things that we observed were that they spent a lot of effort on fine-tuning models, and that always seemed to me, at least back in 2023, like a waste of time.
The general models were improving at such a fast rate that it felt like we should be building boats, and then, when the tide rises, all of our products just get better.
Frankly, my initial team was 3 people. We were 3 engineers.
That thesis plays into the team structure, right?
We had $50,000 in angel funding, so it wasn’t really a question of spending $3 million fine-tuning a model. It happened to be right as well.
We always believed that the majority of the value in our category would come from the application layer.
So you think work being done to fine-tune models does not give you an inherent advantage?
It didn’t do that back in 2023 as a starting strategy. I doubt that’s going to be the difference-maker today.
Do you think models are plateauing in performance today?
No. I think Claude Opus 4.5 is awesome, and it continues getting better.
How much better is it?
It depends on the task. In coding, I think it’s gotten to the point where you should just coin it AGI and focus on optimizing the cost, pretty much.
What makes you say that? I’m sorry, I’m naive.
I think it’s the understanding of my intent and its ability to execute on my intent, given the tools that it has available today. It’s so good.
You don’t need to sit with GPT-3.5 and feel like you’re talking to a lobotomized system. If you go back, it was crazy bad, and this was only 2 years ago. You would have to give it so many instructions over and over and over again. It felt like managing an employee who wasn’t very intelligent.
Whereas Opus 4.5 is like a VP. You give it, “Here’s the thing I want. Go execute,” and it just does it. It’s amazing.
Has Anthropic won the Claude Code game? Does anyone on your team use Cursor?
That’s a good question.
We're in both.
You are?
We have both. Yeah. I'm actually not sure what the full team split is, but I know that we're using both very much. But we are pretty die-hard Anthropic right now at the company in terms of the models that we deploy in our system.
Wow.
So initially, we were only OpenAI. So, 2023 and most of 2024 were only OpenAI, and now we're majority using Anthropic.
What changed? Just 4.5?
No, not 4.5. This was prior to that. I think it was maybe Claude 3 Sonnet or Claude 3.5 Sonnet that we made the switch. But I think what also happened was you had to start prompting the models quite differently, and then there's a question of where we actually want to put our effort: just pick a good model, double down on it, and build all the application around it.
I feel like our job, frankly, if you think about the pyramid of value, is that you sort of have the underlying models and the frameworks, and then Legora's responsibility is to build the legal interpretation of those models. So how do we make them the most useful in a legal setting? A lot of that comes from the models, but 80% comes from building normal software—enterprise-grade software around the models.
So, all the scaffolding, all the ways that the users can actually interact with the system. And then, at the very, very top of that pyramid, we need to enable our partners and our clients to do differentiated things.
Our clients—the law firms and the legal teams we work with—are very ambitious. They're vibe coding tools internally, they're developing MCP servers, and they've understood that if you compete as a law firm, you used to compete on expertise, on marketing, on your ability to recruit, and on your hourly rates. But increasingly, you're competing on tech. Tech is the main lever for our clients to differentiate from their competition.
And so they will use a platform like Legora to get everybody up to speed. Then they will develop things internally to try and get a little bit of a head start against some of their peers.
Unpacking so many things, just doing it chronologically there: how promiscuous do you think you'll be with model usage over time? You mentioned you're pretty much exclusively Anthropic now. When you look at the next 12 to 36 months, will you switch between them as they improve in model efficiency, or do you think there'll be a continuing loyalty towards Anthropic?
We will be very promiscuous, and that's a clip.
That's a real clip. That's the intro right there.
Well, I think it's our responsibility to be that because our clients have entrusted us to be their AI partner and to deliver them the outcomes that they need, based on everything that you can do with AI. We need to deliver them the best possible thing at the best possible price, using all the tools available to us.
So, if Gemini is better, we will switch immediately. If OpenAI is better, we will switch immediately. If a new model comes out that's better, we will switch immediately, provided that the eval is better.
2. 24 Months: Which Foundation Models Will Win?
But then, in specific workflows, you could also let the users pick, right? So let's say they have a very deterministic thing that they want to run, and they always want to run it on this specific model to not break the system. Then you could also allow that.
Can you help me, in 24 months, rank the model landscape for me?
I'll bet, based on what I've seen in the last 3 to 6 months, that for our type of work, it will either be Claude or Gemini that is the top model, and it will be dependent on whether context window is a very important factor or not.
So far, it is not, because we've built so much architecture around handling a lack of context window that we still prefer the Claude models, or the Anthropic models. And it seems to me like OpenAI is going down the “let users fine-tune models” type of journey a bit more, which so far I don't have a lot of reason to believe in.
Okay. So, we're going for Anthropic, Gemini, and then OpenAI.
I think. I also think there's a difference between solving enterprise needs and solving B2C needs. To me, there's a split happening—or a perceived split, at least, from my vantage point—which is that Anthropic is going more enterprise and OpenAI is going more B2C.
And we're not going to throw Grok in there at all.
No, we're not going to throw Grok in there at all.
Okay. He said it. Don't kill me; it was not me.
100%.
Yeah, and we're an enterprise-class type of system, thus we should benefit more from their models.
And Jason Lemkin, a dear friend of mine—you said you know the shows with Jason—said on a show recently that we're going to see inference running 24/7 for a portion of the knowledge-worker economy, and how that really is going to be the defining theme of the year. Do you agree with that? And if so, what ramifications do you think that has?
So, basically, you're continuously running tasks at all times?
Continuously, 24 hours a day. Exactly. When lawyers leave the office, they're going to still have inference running for the projects that they have.
For what it's worth, I don't think that we're there yet, when we have tasks that take that much time to run. We don't have any task in Legora that would take 12 hours to run yet.
But when you can put the models in loops and they get better and better and better the more loops it takes, then you can for sure allow that. I do think that we're going to move into a world where we start a lot of things as we go to bed, and we wake up in the morning and it's done. For sure, I think I already started doing that with Deep Research when that came out for the first time, and it would take 25 to 30 minutes to run.
It was magical. It was so cool. But you so quickly get used to our new shiny toys.
What do you think we don't talk about enough, or don't see in the model environment today, that more people should see or talk about?
One of the things that's very impressive with Claude Code—and do they call the new thing Cowork?
Yeah, Cowork, I think.
Cowork, I think, is maybe not so much about that tool itself, but about the paradigm that it has shown is useful and probably the right direction to go in. The more we were working with Claude Code and Cursor in our engineering team, the more we thought, “Hey, let's apply the same principles to the way that Legora works.”
Basically, having the Legora agent access all the other tools available in our ecosystem, as well as any MCP servers that the client brings, and then basically letting it roam. You just give it this overarching task, it gives you back its plan, and then you say, “That looks awesome. Go execute.”
It's pretty much the way that a partner would work with a senior associate, and a senior associate would work with an associate. I think this is adding another layer in that hierarchy, basically, with AI at the bottom, but then that's available to everyone 24/7.
One of the themes that I've noticed is that partners at these firms that we work with are starting to think that the technology is so good that, as they give a task to their team member, they will simultaneously give that task to Legora. Very often, the quality that they get back is pretty good, and that has real implications.
We're going to go to the structure of law firms in the future. I just want to unpack another thing that you said earlier, which I don't want to forget: you said you're spending more and more time in the US.
There's this kind of perception when I speak to especially US VCs that Harvey has won the US and that you have won Europe.
I think one of those statements is true.
One part of that statement. My worry with you is your lack of confidence, my friend. Why is that not true? Because they seem to have the Magic Circle in the US.
Well, that's not true. When we started the year, we were zero boots on the ground in the US. Now we are 50 people. We're opening up our new office in Manhattan this week. It's going to be 150 people.
We've got 3 more offices in the US opening this year, and the US has, by revenue, become our biggest market.
Wow. Revenue-wise, you have more?
It's the biggest country by revenue.
Wow. Do you have more in the US than you do in Europe?
In total? No, we have so much in the Nordics, actually, because we basically work with all the big firms. But it will be, I think, by the end of Q1.
Super interesting. So, by the end of Q1, you'll have more in the US than you will in Europe.
Yeah. And the cool thing, right, is that if you look at the Am Law 200—and, by the way, many of these clients that we're working with in the US, it's not the mom-and-pop shops, right? We work enterprise.
When we came to the US, we had a strategy, which was: there are so many Nordic or European companies that have launched in the US and failed very close to home. I think Klarna tried to launch in the US a couple of times before it really worked.
And so I had this heuristic, which was, if we can sign and serve 2 of the Am Law 200 law firms from Europe, we are ready to open in the US. So we clearly got White & Case, a white-shoe Wall Street firm, and Goodwin Procter, one of the best VC firms in the world. I think both are in the top 20 law firms in the US.
We were able to work with both of them and give them confidence that we could support them better than anybody else.
3. Lessons Scaling from Europe into the US
And that gave me the confidence to go to the US and hire a team. The awesome thing about building a team in the US is that it takes 2 weeks for people to leave. We can talk about some of the differences between the US and Europe, but I think the termination period in the US versus, let’s say, Sweden is actually one of the structural benefits of having a big office in the US.
So how does it compare—2 weeks to leave in the US versus 3 months?
Everybody has 3 months, basically.
And for you as a founder, that is a night-and-day difference in terms of ramp.
Well, we’ve doubled in size every quarter, and the minute I know that I need somebody, if they wait a quarter, we’re a different company, right? It’s wild. For one, I need to try and predict our headcount plan much more diligently in Europe than I do in the US, because there it’s like, bam—really awesome people can just turn up in 2 weeks.
So your biggest advice to founders on scaling in the US without committing large resources would be that you can do a presale and test it from Europe.
Yes, for sure. I think we could, so why can’t you? We’re very enterprise, so that might be different, right? We did not need to invest a ton in marketing or B2C content in the US. I could just get on demos and get on a few flights, demo the product, run a few pilots—always competitive pilots—and then, again, on the partnership level, show that we were willing to work with these firms at their ambition level, because it’s very high.
The firms that we work with are not treating AI as a check-the-box exercise. It’s not, “Oh, let’s buy this thing, let’s roll it out, and we’re done.” We want to be the firm that dominates our market because we understand AI and technology better than any other firm.
Totally get that. Going back to the US and the expansion there, do you regret waiting as long as you did?
No. Did I tell you that I took the decision not to sell the product for 6 months?
No.
To give you some context, we were a YC company. We raised $10 million from Benchmark. A month later, we raised another $25 million from Redpoint, and we had our first board meeting. We were about 12 people at the time. The first board meeting was Benchmark, Redpoint, and the 3 founders, and we sat down. I told them that we were not going to sell at all for the next 6 months.
Redpoint looked at me, and they were, I think, a little nervous that they had met me for basically 1 hour and 45 minutes and given me a price, and I was showing up and saying, “We’re not going to sell.”
What price was the Redpoint round?
$150 million.
Huh. And yeah, that’s interesting. Do you regret taking—not saying Redpoint, but doing that round?
No.
Because that’s a lot of dilution.
Yeah.
$25 million at $150 million, when you didn’t need the money 2 months after Benchmark.
Yeah. But you couldn’t know that you didn’t need the money. If I remember correctly, one of our competitors did another round quite quickly after. So I think it was good to solidify that there was interest in Legora’s stock.
Now, at this point, I’m just archiving emails because I’m getting too much inbound. But back then—
Which is why I send WhatsApps.
Yeah. But back then, that’s a good thing. That’s a good thing. I turned around and said, “Hey, we’re not going to sell. We need to get to the point because we only have 1 shot.” We only have 1 shot with these lawyers because they are very impatient. If it doesn’t work, they’re not going to come back. That’s why activation and getting to that time to value is so important in the product.
To go back, we took 6 months, calculated the time, and said we were not going to sell. We had to solve our infrastructure, our reliability, and the scalability of the product, and we needed to rebuild and refactor a lot of it because it had just been quickly put together.
What we told all the clients was—we were lucky, it was summer—“It’s summer in Europe, so we’re not working, and we’re going to wait to onboard you after summer.” There was so much demand that we were going to have to do it in October, because September was completely full and we couldn’t onboard more clients.
But on October 1, 2024, we were ready to onboard 1,000 lawyers a day comfortably on the product. We got to that point, and then we started to do RFPs. I’m very proud that I had the guts to tell the investors that was the right plan, because I think if we had continued to push, we would have just churned everything.
When you look back at the timing of the US expansion, do you not think you could have gone sooner and not ceded so much ground?
Well, for what it’s worth, I don’t think we conceded a lot of ground, and we are winning back a lot of ground, if you put it that way.
So customers aren’t loyal?
Well, for what it’s worth, no. I think everybody is still treating this as an extended pilot and an option on AI. It’s like a call option, right? They’re not doing 5-year contracts. They’re doing 1- to 3-year contracts. In law firm time, that’s a blink.
Many of these firms have been around for 200 years, so 2 years might be half of our lifetime, but for them, it’s a short time.
When we look at the numbers, bluntly, Harvey’s retention is 98% logo retention and 178% net revenue retention. Do you have as good numbers?
For both those numbers, yes. But on NRR, I don’t think that’s a fair number for me to comment on, because so much of our growth is not about renewing contracts from 2024, right? We added $7 million in ARR in a single day in December 2025—1 day, 24 hours—and that was more than what we did in 2023 and 2024 combined.
NRR and logo retention—it’s up to 2026 to determine where those real numbers will be. For what it’s worth, the ability of these products to go quite broad will be very interesting. To some extent, there are initial use cases that you can solve with AI that we target, and then the more time we spend with our clients, the more problems and opportunities we see.
4. Why Seat Models Are Not Dead in SaaS?
What’s happening to the products is that they’re growing quite a lot. I think this will be a suite, a platform kind of play, that just becomes more and more of the central system where they do their work.
From an NRR standpoint, do you charge on a per-seat basis, on a per-task basis, or on a volume-per-task basis?
We charge on a per-seat basis.
Is that optimal?
I think that’s optimal for the buyer. I don’t think that’s optimal for us.
And is that not solving for a historical norm, not a future optimization?
It is. I actually don’t think it’s the right pricing model. I think it should be consumption-based.
Yeah.
Because you can have individual users racking up such big LLM costs that it basically becomes unsustainable on a per-user basis. The reason why we have that is that you need to make it easy for the buyer, right? If they don’t know how to manage a consumption-based pricing model, you can’t have it.
I think that will pivot, but I’m unsure exactly what that timing is. I think the timing is more around when the clients are ready versus when we are ready. Task expansion is incredibly useful for retention, not for revenue optimization.
In other words, the more they do, the more likely they are to retain, but it doesn’t actually help your dollars.
Right. But actually, it costs a lot more. That’s a bad thing: the more they use the product, the more it costs.
You have good margins.
We have okay margins.
I respect the honesty of that answer.
Yeah. It’s not SaaS margins, and I think it will take time to get there.
Will it get there, do you think?
Yes, I do. I actually think it might, to some extent. Yes, I think it will get there. Not only do I think it will get there, but I think your ability to price versus traditional SaaS products will be insanely high.
Why?
Because you used to use a lot of these products to do very narrow parts of the work, and they were all disconnected. To give you an insight into the life of a lawyer, you have one product over here that you use to compare 2 contracts. You have another product over here that you use to extract relevant data from contracts. You have another product over here where you go and look up legislation. You have another product over here where you go and look up case law—all these different things.
As the human, you had to sit there, comb through all these different systems, and aggregate the stuff yourself. But now, similar to Claude Opus 4.5, you’re just going to send the task to Legora. It can be pretty arbitrary, and then you let it figure it out. It just goes and does all the work—or at least a big portion of the work—in a much, much, much, much shorter time at a very high-quality level.
When you do that, you’re not being priced against the other SaaS products. You’re being priced against, “What would I pay a lawyer to go out and actually do this work?”
In 3 years’ time, will you still have seat-based pricing?
Absolutely not.
When does that change?
When our clients are ready to buy on consumption.
Why do you sound so confident that will be within 3 years, respectfully? You know, Cursor’s consumption—
A lot of other enterprise tools are for consumption. I just think legal takes a little bit more time. But within 3 years—well, look, you have to understand my vantage point: 3 years is longer than I’ve been CEO at Legora. So 3 years for me is a very long time going forward.
On the margin-optimization side, is it a little bit like what we’re seeing with Lovable, where they’re able to do model selection dependent on the task and optimize margin because of that?
You can do that, of course, which we do to some extent, but I also don’t think we’re in the model-margin-optimization time yet. You know what I mean?
You’re in the land-grab time.
Yes, that’s the right way to phrase it.
5. How to Use Competition To Drive a Fire in Your Team?
Yeah. In the land-grab time, what is the biggest challenge that you face?
The biggest challenge that we have right now is growing from 30 to 300 and then doubling again in the next 2 quarters, from 300 to 600, while maintaining the ambition, integrity, teamwork, and raw grit that got us here when you double the team.
I think we just hired 2 new people in the US, and they were really surprised by how late everybody was working. They were like, “At the other place I was at, which was another legal tech provider, everybody left at 6, and we have dinner in the office at 8.”
When your entire team globally operates at that level and at that pace, with that goal in mind, that’s awesome, but I care a lot about maintaining that.
How do you maintain that?
Well, I still interview everyone, so I ask quite brutal questions about why they would take a hard job. You could go work somewhere else. I try to create missionaries, not mercenaries, and I think we’ve successfully done that.
I also think that you get pulled in. When you see everybody else doing it, you’re just like, “Okay, of course I’m going to do it.” Momentum breeds momentum.
We were signing deals on New Year’s Eve. We had a big Christmas dinner, and whilst we were having mulled wine—we had the big sales dashboard in front of us at the wine thing—everybody kept looking at it because everybody wants momentum. Everybody wants to win.
When you join a company and you feel like a winner, I think you get burned out doing work where you don’t feel like you’re winning.
Do you think competition is helpful in creating that vibe?
100%. 100%. Of course.
Do you light the tinder, so to speak, and fuel the fire?
Oh yes, of course. I think I’m quite good at it, actually. Competition can be played at a macro level, where you think us versus them, but you can also do it at a lower level. Our marketing team wants to beat the other marketing team, or our engineers want to build a faster document-upload time than the other team.
You compete on all these micro levels and you celebrate them like crazy, right? What I’ve learned this year is that I actually used to be quite bad at celebrating.
I remember when I was in business school, my dream job was to go to McKinsey because I thought that’s where all the amazing people went. I found out maybe that that was not the case. But when I got the call and got the job, I was in the grocery store, and I celebrated by buying a bag of peanuts.
Wow.
Yeah, that was a bit crazy. So I was really bad at celebrating. Really bad.
Explains why you’re so thin.
But this year we’ve learned to celebrate, and it’s amazing. You celebrate the wins really hard because then you also really feel the losses. I think it’s easy to get blindsided if you have momentum and success. You need to see the world for what it is.
I heard from some of your investors that people internally at Harvey call you their CPO, speaking of comparisons of teams.
Well, I think you’ll have to ask them. That’s funny.
Have they ripped your product?
I think we take a lot of pride in developing our product as fast and as well as we can. There are 2 main parts to our product development. One of them is improving the parts that we have, and the other one is making new, qualified bets. I think we’ve had a history of making bold and correct bets.
There are many legal tech products that, on the surface, look pretty similar. There’s even another product where they ripped our name, and it’s our Tabular Review. It’s just called Tabular Review in their product, which is totally fine.
But what happens when you then go into these competitive pilots and the user starts to rip them apart? That’s where you see that one product is a Rolls-Royce, another product is maybe a Volvo, and the other product is perhaps a cheaper version.
What product decision did you make that, with the benefit of hindsight, was a mistake? And what did you learn?
The first version of the Legora product, back in the summer of 2023, was completely the wrong direction. We built it centered around a couple of core use cases, and we did not have an agent or a chat that could operate over those tasks. It was a click-and-point use case—clearly the wrong direction.
After we got accepted into Y Combinator, we deleted all of that code. I think we made some good product decisions by very early on recognizing that LangChain was too bad at the time, so we built our own agent architecture. We did that very early, which I’m very proud of, and that was the right direction to continue in.
6. Is Legal AI a Winner-Take-All Market? How Does It End?
You still have that today?
No, it’s been completely rebuilt many times. I last committed code in October 2020.
But I’m intrigued as to how you think about that. We’re seeing more and more companies like Deel or—not Replit—Revolut build their own complete vertical software.
Yeah, I think we’re not the size of Revolut or Deel, right? So it probably doesn’t make sense for us to do that, and LangChain and a lot of the surrounding tools have gotten a lot better. I think we’re in the job of—and our engineering team is in the job of—picking the best third-party things.
Yeah.
The other product decision we made that was wrong was that we were doing too many things. In 2024, we were 12 engineers, and we were trying to build 6 or 7 different things at the same time. That was creating a lot of confusion because I was very involved in the product decisions at the time, and I was basically out doing go-to-market.
So we’d make a lot of product decisions without me in the loop, and then it kind of looked like a Frankenstein monster. There’s actually a pretty funny doc from October or November 2024 called the Leya product manifesto.
Gosh, I remember.
Yeah, we were Leya then. We were still called Leya.
Yeah, I remember.
It basically outlined that we were going to do 3 things, but we were going to do those 3 things so well that our suite was the best stack money could buy. It was our agent, our assistant, our Tabular Review, and our Word add-in.
We were competing with local products for these different things, right? The Word add-in was competing with a bunch of other legal tech companies that were only focused on the Word add-in. Our Tabular Review was competing at the time with Hebbia and companies that were only focused on Tabular Review, or the matrix, I think they call it.
Then we had our agent, but we said that if we have all of these 3 things and combine them in a very user-friendly way, that suite is going to be better than buying all of these 3 things separately. That’s kind of the platform play, or the suite play.
That was totally the right move. So we removed 5 or 6 other things that we were building, just deleted the code, and hard-committed on these things.
7. Why OpenAI is Toast? Switching to Anthropic!
Can I ask, when you look at the landscape, when you think about bundling versus unbundling, and how you thought about that, when you look at the landscape today, how does that landscape look in 3 to 5 years? Is this a winner-take-all? Actually, a much better way to ask that is: is this an Uber and a Lyft, or is this Google Cloud, AWS, and Azure?
The reason why I don’t think it’s an Uber and Lyft is because, in Uber and Lyft, there was no product differentiation. The products were pretty much the same.
Yeah, but it was very hard to build something different. I mean, Uber, for what it’s worth, the product looks the same today, basically, right? With the addition of bells and whistles, but it’s the same fundamental thing.
The difference to our story is that product differentiation really matters, and the amount of things that you can go and build is so vast. It’s like this universe of legal technology that just has never been built.
One of my theories is that maybe in legal tech, before generative AI, there weren’t that many exciting things to build, and it was really hard to scale a good company. As soon as you got to single-digit millions in revenue, you would get an acquisition offer, which would be life-changing money for the founders, but no unicorn outcomes.
So the product strategy will impact the trajectory of all of the businesses in our vertical tremendously. I think it’s totally winner-take-all, like all SaaS. I mean, you know this: number 1 will grab 90%, and numbers 2 to 10 will share the remaining 10%.
And so I think what that means for us is you’ve got to run like hell. You’ve got to win. There’s no number 2. There is only being number 1. There’s only winning, and everything else is losing, and the type of people who think like that are the people who work at Legora.
One segment that I do find interesting, or 2 segments, is verticalization. We’re investors in Solve Intelligence—AI for patent lawyers. Love them. They’ve done amazing.
You have verticalization in that sort of way, and then you also have verticalization in the, “We’re going to own the whole vertical and do like a Crosby. We’re going to be your law firm and use ours.” How do you feel about those 2?
Yeah. I don’t think that owning the entire service layer and software layer is a winning strategy—basically building an AI-native law firm. The reason is, I think there are so many talented lawyers who are very good at utilizing software, and I would not want to compete for them.
I think it’s easy to get into that space and try and solve—maybe this is unfair—lower-complexity tasks. I mean, you’re starting out with nondisclosure agreements, master service agreements, and I think you can get there pretty quickly, but then that will be a very crowded space in and of itself.
I would much rather be the shovel seller to all the world’s amazingly talented lawyers who want to turn their firms into software-powered entities. I don’t think that there will be a ton of margin or profit in the low-complexity work, because I think as soon as AI can do a task, it will do that task. The only question is, where does it get transacted?
The big law firms already do NDAs for free for their clients because they’ve got to win the expensive private equity work.
And so it’s like, are you going to show up as Crosby or somebody else and go, “Hey, pay me $50 an NDA”? I’m not sure.
Okay, so we think that’s not a good strategy. What about the verticalization?
Yeah, I think you can get to a lot of value more quickly by verticalizing. I’m not sure what the TAM looks like within each of those verticals.
If you look at patents, it’s a $485 billion market.
Yes, it’s a huge market. So maybe you go win patents, and that’s amazing. Or patents become part of a broader thing. Unclear.
I think there will be many winners in different layers of the ecosystem. I also think that there’s one part which is kind of the central hub that will—let’s say somebody goes to Legora and they want to write a patent. Why don’t we just ping Solve Intelligence and say, “Hey, Solve Intelligence, come write us this patent,” and then it comes back and does it? Or maybe that’s what Copilot wants to do, right?
I think there are a lot of Venn diagrams and a lot of overlap. Right now, I think it’s more about execution than about the underlying market or being smart about the markets.
You have hired now 150 people in the US. 50 or 150?
No, we’re at about 50.
50 in the US now. Okay.
I think we’ll be at 150 before summer.
Do you think the canonical wisdom that the US works harder, they’re harder-driving, they’re more transactional but bullish, and they’re in the office late, while we in Europe just like to chill the fuck out, is fair, having hired 50?
I don’t think that’s fair. I think we were very good at seeding the Legora culture in the US, and a couple of our best people from Europe went to the US and have spent a lot of time there.
So actually, the idea that, “Hey, we’re all such hustlers. We’re so good in the US”—it’s a bit of bullshit. I think it’s a little bullshit. But for what it’s worth, the culture that we have in our US office now is—I almost want to spend more time in the US just to be there because it’s electric.
New York is on fire, and I have a hard time spending a lot of time in New York and then coming back to Stockholm, because New York is always up. It’s always awake. It’s my tempo. Whereas Stockholm, when you walk out on the street—not when you’re in the office—it’s a little sleepy.
Why are you not living in New York or living in the US?
I’m de facto living on a plane. I had 200 travel days last year in total, and last year I spent a meaningful amount of time on product. We have all engineering in Stockholm.
We’re 10% YC founders in our engineering, product, and design team, which I think is a high number. 2 of our batchmates have actually joined the company.
Wow.
I think what’s cool about it is we’re structured in a way where my management style is very—there’s a lot of delegating. It’s not very micromanage-y. It’s very much, “Here’s the thing. Go run with it.”
I think founders do very well with that. I also think it basically gives all the different components of our platform—which is separated into pods, with one team working on this piece of the product and one team working on that piece of the product—a YC founder running part of the product. Then they just run like hell on their thing, and they’re competitive about their part being better than all the other parts, or rather, their equivalent in other products.
I also think YC has a way of attracting very ambitious people who want to win.
Before we discuss the future of law firms and then do a quick fire, I have to ask you: I’m going to ask Harvey on the show, which, you know—
Great.
—about their revenues. I have to ask you for yours. Where are you guys at?
I’ll tell you a little bit later in the quarter, but in December we added $7 million in a day, and we’ve basically doubled every single quarter for the last 6 quarters.
Will you be at $200 million by the end of the year?
Definitely. Otherwise, I’ll come back and stretch you. You can shame me.
What’s the stretch goal?
You’ll have to ask Patrick, our CRO.
If you were at $300 million, would you be happy?
Well, I don’t view the number in isolation. I’ll be happy by the end of the year if we deliver on all the promises we made to clients.
We don’t even need these competitive pilots, right? It’s just a no. It’s like, “Oh, if you’re a lawyer and you do serious legal work, you’re on Legora.” It’s like Figma: if you’re a designer that makes money, you’re on Figma. I want that to be the truth.
If we do that, I’m sure $300 million is the number, or even more, right?
8. The Future of Law Firms: Do Juniors Get Fired?
Okay. In terms of the structure of law firms—dude, I’m dating a lawyer. She, by the way, and they thank you. They thank you for the work that they now no longer have to do because you do it.
But you will need far fewer trainees. You will need far fewer juniors. Do you agree? Will you all need far fewer juniors, and what is the structure of a law firm in the future?
So I think law firms will go through a quite significant consolidation period, because so far there haven’t been a lot of incentives to consolidate law firms. But now, actually, private equity wants to get in on the action and wants to fund different law firms that want to become AI-powered, again coming back to the idea of: do you want to own the whole stack, or do you just want to work with the best service layer?
And so you’re seeing a roll-up play where you integrate and juice up the—
I don’t think there’s going to be an Am Law 200. I think it’s going to be an Am Law 20, or maybe Am Law 12. I don’t think it’ll be a Big Four because of regulation and it just takes time, but it will definitely consolidate.
At the end of the day, Legora and I care about working with the winners of that space because, as we talked about, I think the technology lever will be one of, if not the most important, levers to utilize in competing against other firms in that environment.
It looks different for different practice areas and in different calibers. But let’s take a bread-and-butter M&A transaction. In a bread-and-butter M&A transaction, the legal work that the law firms do is pretty much undifferentiated. You get pretty much the same thing depending on which firm you go to, if you just look at the diligence and the work.
It’s an equilibrium where the price gets offered at—let’s say it’s £100,000. The minute that one of the firms playing in this game is able to offer that at a lower price but with the same quality, or maybe higher speed or something like that—some attractive aspect of running that deal—let’s say they’re offering it at £80K, you kind of break the equilibrium.
Everybody has to move to that equilibrium. It’s kind of a market-share game. It’s like, who can run the fastest on technology and all the other aspects of what it takes to run a law firm? Who can win most of the market and then hold that market?
Then I think you’re able to deliver additional services that go outside of what the very, very, very competitive things are.
Will we have fewer junior lawyers and trainees? Will law firms be smaller?
Well, I actually think law firms will be bigger, because they will consolidate. But I don’t think that you will need the same number of lawyers running a transaction as you have today.
If you have a physical data room, that’s why it’s called a data room. You used to have to send people there. They would open up all the boxes, read everything, and make all the commentary, right? And then it became a virtual data room.
So then you have to assume that there’s going to be more transactions in the future.
Yes.
Yes. So more transactions, probably fewer people running those transactions, but every person can run more transactions because of technology.
So, just to really be clear, you do not think there will be fewer trainees and junior lawyers?
I do think that there will probably be fewer junior lawyers and trainees, because I think you just won't need as many people to execute the work that the firm has.
Also, most law firms are partnerships, correct?
Right.
And so partnerships accrue profit.
But I'm already seeing patterns of this where firms that we work with will have somebody leave, they will not fill the vacancy, but they're doing more revenue than they did last year, and so it's a higher profit. I think it might also create an opportunity for—I think big law will do really well because they have a lot of moats, a lot of brand, and a lot of data. I think small law can do really well because it's still operating on a very, very personal basis. I think where it might get difficult is midlaw, where it's very competitive. You're competing hard on price. You throw AI into the mix, and it will make it even more competitive.
But it's like, for engineers—in engineering, let's just take another example. We're hiring more engineers, although they're writing more code, because there's more stuff to do. So the thing for the law firms is that they need to increase the size of the pie.
So you think we'll have more engineers in 2 years' time?
I think so, actually, because I think there'll be—you can just do more stuff. You'll have more people writing code and doing things, or starting new things and projects, in 2 years than you have today.
I think you're being quite optimistic, which is really nice. But I think we're going to see more labor displacement than I think people expect, and I think we're going to see that in the next 12 to 24 months. Jason Lemkin, a young friend, said this is the year that we're going to see AI displace large amounts of knowledge work, and that's going to show up in labor figures. Do you think that's too soon?
Well, I think it's within that time frame that AI gets pretty good at completing end-to-end tasks very deterministically within our vertical. So unless they can find something else to do within the firm or grow the pie, then, yeah. But here's the cool thing, right? If you use technology to complete more of the work, then you need to think about how you win more work from the other firms.
But I'm talking about an individual firm.
If you're talking on the total level, yeah, probably.
Do you worry about the demonization of you as a technology leader displacing labor?
No, that's not something that I worry about. It's something I should worry about in time.
Can I ask you about timesheets? Timesheets are how lawyers spend a lot of their time. Every 2 days they have to do their timesheets. Did you know this?
No.
Yeah, it's wild.
Fucking wild.
Adrien, who's our VP of Product, his YC company did AI timekeeping.
Fucking wild.
Are billable hours over?
9. How We Raised $200M and 3 Rounds with No Deck
No, I don't think they are. I think billing will move much slower than both you and I would think, because very often it's actually demanded by the clients. They want to get a breakdown of everything that the lawyers actually do. It will start to be replaced by fixed fees in different practice areas and for different types of tasks, but you will still have a sort of ephemeral billable hour above that.
Dude, I'm going to ask you a quickfire. Does that sound okay?
3 rounds, no deck.
What's the biggest advice on fundraising?
The advice that I got in YC was, “Build a good business and it's very easy to fundraise.” That's what I've lived by. I don't think I'm a master fundraiser by any means.
With respect, I'm going to push you. Sorry. You got Benchmark early, which then led to Redpoint. Do you think Benchmark is just a massive signal, which led to a quick successive round that you wouldn't have had otherwise?
No. Redpoint was not the fund that first wanted to preempt us. There was another firm that wanted to preempt us for the Series A round, and they gave a very competitive term sheet for the Series A at a much higher price than Benchmark. Benchmark was, I think, the lowest price out of any seed fund.
But you placed so much value on them that you took the discount.
Yeah. I placed value on Chetan.
Chetan, not Benchmark.
Yes.
Interesting.
So I picked one partner. I heard Benchmark was good. I can do that research. But I met, within 2 weeks, probably 80 partners. Nobody knew my space better than him. I thought it was great. He's taken 3 companies public, and that's what I wanted to do. I thought I'd be much better off working with that guy.
10. Quick-Fire Round: Best Advice, Closest Mentor, Biggest Mindset Shift
Unthinkable reality: You start another company, but you can only bring 1 investor with you.
Yeah.
Who do you bring?
Chetan.
Really?
Oh, yeah. That's easy.
Okay. You can delete 1 investor from your cap table. Who do you delete?
Well, I think I'll delete YC.
Do you regret doing YC?
No, I love YC. But all the other investors are on my board, so it would feel very rude saying somebody else.
That'd be a fucking awkward board meeting.
No, no, but, yeah, I think that's unfair. They're extremely helpful still. YC rocks. I still speak with Gustaf every quarter. But all the others are either board observers or on the board of directors, so I can't comment on that.
What's your most unpopular belief about where AI is heading?
I really do believe in the platformization. I think there are way too many point solutions that will not survive a winter or a bubble, whatever you might call it, and they would deliver more value as part of a broader ecosystem.
What do you know now that you wish you'd known at the start of Leya?
How the intensity of doing and thinking about nothing else would impact your own psyche and personality. For the last 2.5 years, I've basically done nothing else than think about Leya, Legora, or the business.
When I was in college or prior to that, I was doing so many different things. It was nice to do many different things, and you got to have many different contexts and many different types of activities. When you got tired of one thing, you could go and do another thing. This is not that, right? It is like running a mega-sprint as part of a marathon, and I love it.
But I think if I could go back and also have that expectation going in, I think it would have made it—I think I would have been better at handling other disappointments, or parts of my life that I couldn't focus as much on.
I think our job is much easier as venture investors than we give credit for. When you find obsessed founders that are just quite unhinged and psychopathic, I put you in that category, to be honest. I put myself in it too, to be honest. But when you find them, it's quite obvious.
When I sit down with Alan Chang at Fuse Energy, I said, “Do you angel invest?” And he goes, “Are you fucking stupid?” I said, “Potentially. My mother thinks so, but tell me why.” And he's like, “To angel invest, I'd have to either sell Revolut shares or Fuse Energy shares. That would both be a terrible fucking decision. So no, I don't angel invest.”
No, no, because you can't focus on doing that. You can't make good decisions. You can invest $10K in a friend's company because it's nice, but you can't do it seriously.
Penultimate one: Which founder do you most respect, admire, and look up to?
I'm pretty bad at having idols. I wish I was better at it. I was super nervous the first time I met Daniel Ek from Spotify.
Oh, wow.
I remember—I thought that was the coolest thing ever in 2024, and he had reached out to us and was like, “Hey, love what you guys are doing.”
He's amazing.
And I think, when I grew up looking at Niklas, Daniel, and Sebastian and seeing these Swedish tech companies succeed, that was phenomenal. That was a huge inspiration. And now, having met Alex and Gustav, who are taking over from Daniel, they're also fabulous.
But I don't go around on a daily basis thinking, “Oh, I look up to this person. I want to be like them.” I try to draw inspiration from where I see good, and then I run at it.
Final one. What's the best advice you've ever been given?
The best advice I had ever been given was probably from YC and Joel at Sana Labs, which was to take the check with Benchmark over anyone else.
I love that, dude. Thank you so much for doing this. It's so lovely to do it in person. I really appreciate the friendship. Thank you for not letting me on the cap table. I think about it every day.
It's fine. I've got over it, you prick.
But you're a hero, my friend.
Thank you so much, Harry. Let's go.