Eve AI Automates Plaintiff Law to Boost Attorney Capacity Fivefold

Eve AI助原告律所实现自动化，将律师承案能力提升五倍 通过将人工法律工作流转化为自动化算力，AI将法律软件转变为一种影子劳动力市场，使小型原告律所能够以盈利方式扩大案件规模，并承接以往不具备经济可行性的诉求。 Jay Madheswaran与Alex Rampell探讨了为何与依赖计费工时的企业辩护业务相比，实行风险代理收费的原告法律业务尤其适合生成式AI的颠覆。通过将客户接待、判例检索和病历收集等耗时任务自动化，法律AI平台Eve使独立执业律师和精品律所能够将单名律师的承案能力从100起提升至500起，并承接以往因高昂间接成本而不得不放弃的诉求。对话涉及了从按席位收费的SaaS向按结果收费的算力定价的经济转型、原告律所在面对机构保险公司时遭遇的信息不对称，以及AI智能体将如何重塑谈判与争议解决流程。他们最终认为，技术杠杆将使小型律所能够扩展为估值达数十亿美元、具备软件特性的企业，同时让执业律师重新专注于法庭辩护。 与依赖计费工时的企业大型律所不同，原告律所采用风险代理收费模式，因此采用AI更像是一个创收扩张引擎，而非单纯的成本削减工具。 (0:51) AI自动化已使较早采用该技术的人身伤害律所将律师的承案能力提升了五倍，从每名律师管理100起案件跃升至500起。 (13:57) 前沿基础模型在原告法领域面临结构性壁垒，因为超过99%的争议解决发生在审前，通过公开网络数据集中不存在的非正式和解谈判完成。 (20:25) 传统的风险代理收费模式面临高昂的人工间接成本，其中约50%的费用用于人工，30%用于营销，导致律所拒绝高达96%的劳动与就业潜在案件线索。 (3:50) Eve通过部署外呼语音智能体调取病历，实现了大体量诉讼工作流的自动化，其中包括每起人身伤害案件所需的约10,000次离散微交互。 (47:18)

## Background 本期内容由风险投资机构 a16z 于 2026 年 10 月 30 日在 YouTube 频道发布。视频围绕法律科技公司 Eve 展开对谈，探讨当 AI 将法律实操成本大幅压低后，如何让原本因无利可图而被律所放弃的大规模平民维权案件重新具备经济可行性，进而推动小型律所向数十亿美元体量扩张。 ## TL;DR a16z 合伙人 Alex Rampell 与 Eve 创始人兼 CEO Jay Madheswaran 深入剖析了原告律师事务所独特的商业模式，指出法律 AI 在该领域的价值不在于裁员降本，而在于打破劳动力瓶颈以实现收入爆发。通过端到端的 AI 智能体工作流，律所单名律师管理的客户量从 100 起跃升至 500 起，不仅使 96% 因核算成本过高而被拒的微小劳动与侵权案件重获代理可能，更推动法律服务从传统的“按工时计费”向“交付结果计费”范式转移。对谈还描绘了未来原告与被告双方皆由 AI 智能体展开高频交涉的对抗博弈生态，以及律师向数据驱动决策者转型的趋势。 ## Atmosphere & style 整场对谈兼具风险投资人敏锐的商业洞察与一线创业者的实战兴奋感，语调轻松明快且极富思辨性。Alex Rampell 善于用咖啡馆偶遇出庭律师、自己怒用 ChatGPT 起诉富国银行等生活化切片抛出尖锐问题，而 Jay Madheswaran 则以技术理性的视角层层拆解底层逻辑，穿插“爱因斯坦空降律所”与“感恩节开示突袭”等风趣案例。两人的互动火花四溅，既有对“大卫战胜歌利亚”式法律公平的理想主义共鸣，又充满对软件重塑古老行业底层生产关系的智识推演。 ## Bulletin - [07:00](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=420s) 传统原告律所受制于法务助理等人工成本上限，全美劳动与雇佣维权咨询中有高达96%因律所预期收益低于5,000美元保本线而被拒之门外。 - [13:57](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=837s) 洛杉矶某标杆律所接入垂直法律AI平台Eve后，借助端到端智能体流转，将单名律师可同时管辖的客户体量从约100人提升至500人。 - [15:53](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=953s) 法律科技的本质是一个隐藏的“影子劳动力市场”，原告律所不按小时计费，AI通过供给廉价Token劳动力直接解锁了原本因算不过账而被放弃的千亿美元级潜在索赔标的。 - [48:15](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=2895s) Eve推出Auditor等夜间全量扫描工具，并部署自动化Voice AI接管单个案件周期中多达10,000次的琐碎交互，甚至出现了双方机构AI在电话线上相互试探防线的情景。 ## Key takeaways - [00:00](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=0s) AI在原告法律领域的经济逻辑是“营收扩张”而非“裁员降本”，旨在帮助律所吞吐原本无力承接的海量线索。 - [00:50](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=50s) 原告法领域普遍采用胜诉分成的风险代理机制，这天然构成了普通人抗衡巨头机构的制度杠杆。 - [03:50](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=230s) 小额纠纷面临严重的制度性市场失灵，单案预期提成若低于5,000美元门槛，律所接案即意味着商业亏损。 - [07:00](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=420s) 劳动雇佣法由于需要在前期耗费大量工时排查法条与诉由，导致96%的求助电话被拒之门外，AI的介入逆转了这一经济门槛。 - [09:17](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=557s) 长周期智能体使端到端处理耗时半年至一年的复杂案件成为可能，大幅摊薄了维权前期的边际成本。 - [10:02](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=602s) 原告律所长期受困于劳动力供给短缺的动态瓶颈，前端巨额营销带来的潜在案源流失率高达90%。 - [13:57](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=837s) 律师的角色正在从繁琐的案头文书撰写者，跃迁为类似于“数据科学家”和系统流程设计师的操盘人。 - [15:53](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=953s) 原告法律处于高度对抗的信任边界两端，天然缺乏协同办公需求，使得该领域的软件市场本质上是一个数字化的“影子劳动力市场”。 - [20:25](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=1225s) 通用大模型的“原生智能”在原告法领域难以直接发挥效用，因为超过99%的真实和解数据与战术博弈完全不公开。 - [24:47](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=1487s) Auditor等产品利用夜间自动化拉网式巡检，挖掘卷宗中因人类精力受限而遗漏的关键信息，以此提升获赔率。 - [27:37](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=1657s) 当博弈双方均配备AI后，常规文书的往复拉锯将演变为智能体对战智能体，人类律师得以回归真正的法庭审判与高维策略主张。 - [33:01](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=1981s) 诉讼门槛骤降可能引发司法体系交通大堵塞，未来将催生由AI主导的自动化争议解决（ADR）与非诉分流体系。 - [35:15](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=2115s) AI赋能使小型原告律所摆脱作坊式运营，有望通过极度纵向的专业化数据积累与规模化获客，成长为数十亿美元营收的软件化企业。 - [39:16](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=2356s) SaaS按人头售卖的席位模式正在被瓦解，原告法的风险代理机制成为整个商业世界中最超前跑通“基于结果定价”的范式。 - [48:15](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=2895s) 单个诉讼案在全生命周期内可能涉及上万次往复沟通，Voice AI在调取医疗记录等低效流程中逐步实现全自动化。 - [49:37](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=2977s) 在面对被告律所掐点发起的节假日恶意证据开示突袭时，原告方借助专属智能体在数小时内自动化反击，彻底扭转了资源攻防态势。 ## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=0s) 原告律师业务的劳动瓶颈与 Eve 的切入点 在传统观念中，原告律师事务所（plaintiff law firms）的运营极其轻便，过去往往只需要一部手机和一个邮箱地址就能开张接案。传统法律软件之所以未能带来根本性变革，是因为它们从未真正解决这个行业最核心的痛点——律师与法务助理的人工劳动瓶颈。大部分人提及人工智能时，下意识的反应往往是企业将借此大举裁员、削减开支，但在原告法律领域，技术带来的其实是截然相反的扩张效应：律师借助 AI 工具并不是为了裁减团队，而是为了让律所能够承接成倍增长的案件体量，从而创造远高于以往的营业收入。 由于普通民众根本不可能为了一起充满变数的官司，以每小时 500 到 600 美元的无上限支票持续支付律师费，原告法律从诞生的第一天起就演化出了风险代理模式（contingency fee）。Eve 致力于通过纯粹的 AI 能力端到端（end-to-end）接管原告案件的繁杂流程，将服务贯穿从当事人打入的第一个咨询电话直至达成最终和解。这种效率上的解放已经带来了惊人的现实跨越，例如洛杉矶的一家律所利用该系统，直接将单个律师能够同时承载管理的客户体量从 100 人跃升至 500 人。 面对外界关于为何垂直类法律 AI 会优于通用“原生智能”（raw intelligence）的疑问，原告法律复杂的实操环境给出了答案：即使今天将阿尔伯特·爱因斯坦直接空降到一家律所的办公室里，在相当长的一段时间内他也无法产生任何实际的生产力。法律工作充斥着海量的特定程序规范、证据整理逻辑与本地化操作惯例，而这正是垂直自动化系统真正发力的地方。 - 传统原告律所依赖极其简陋的工具（手机与邮箱），受困于人力劳动上限。 - 市场误区认为 AI 仅用于“裁员降本”，但在原告端其实际表现为“营收爆发”。 - 洛杉矶标杆律所案例：单名律师管辖的活跃客户量从约 100 人激增至 500 人。 --- ### [00:50](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=50s) 硅谷视野之外的“平民法律”与大卫对决歌利亚 在硅谷的主流叙事中，人们谈到法律行业时，脑海中浮现的往往是企业服务与白鞋所（white shoe law firms）。那些毕业于哈佛法学院的精英律师出入 Wilson Sonsini 等顶尖律所，按每小时 2,000 美元的标准向大型商业机构开具账单。然而，美国法律体系中存在着另一个体量更为庞大、分布更为广泛的板块——原告法律。每一位驾车行驶在加利福尼亚州 101 高速公路上的司机，都能看到路边密密麻麻的原告律所广告牌；在公众知名度与实际盈利能力上，像专注于人身伤害与平民诉讼的巨头律所 Morgan & Morgan，其利润表现甚至远超声名显赫的 Wilson Sonsini。 Eve 的创始人兼首席执行官 Jay Madheswaran 指出，Eve 的核心定位是“面向普通人的法律 AI”（legal AI for real people）。它不服务于拥有雄厚资本的大企业法务部或既得利益机构，而是专注于向普通民众提供法律救济。原告法律在本质上是一场资源极度不对称的较量：在诉讼过程中，投入的法律研究与证据搜集工作越充分，事实陈述就越严密，争取到的和解或判决结果就越好。然而，普通消费者无力承受按小时计费的沉重负担，原告律师只能采用“不赢不收费”的风险代理机制，充当普通人抗衡庞大机构的制度杠杆。Eve 的目标正是将这一杠杆最大化。正如 Jay Madheswaran 所总结的那样： > “What Eve is aimed to do is really handle that whole thing end to end, and of how do you represent a client all the way from the very first call to getting them a settlement. How do you get that end to end thing purely with AI, and give that to the lawyers to then amplify the outcomes they can actually provide. And that's what Eve is doing, and we're really arming, you know, think about giving David a slingshot to take down Goliath. That's what we're doing.” ```mermaid flowchart LR accTitle: 从初次来电到案件和解的端到端自动化 accDescr: 展示 Eve 如何在原告法律全流程中替代繁重的人力劳动 subgraph Intake["客户接入阶段"] A["潜在客户初次来电"] --> B["Eve AI 自动接听与意向过滤"] end subgraph LegalWork["实质法律工作"] B --> C["自动匹配诉由与法条检索"] C --> D["起草初步索赔函件 (Demand Letter)"] end subgraph Settlement["律师复核与救济"] D --> E["执业律师审核介入"] E --> F["达成案件和解 (Settlement)"] end ``` --- ### [03:50](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=230s) 经济学供需失衡与小额案件的“市场失灵” 从入门经济学（Econ 101）的供求曲线视角来审视法律服务，能够清晰揭示现存法律市场的制度性失灵。当索赔金额极高时，供给端的律师有极其充沛的动力介入；但从需求端来看，随着潜在法律开销的攀升，普通大众对法律服务的有效需求被迅速压抑。假设一个普通人的实际损失只有 1,000 美元，他显然不可能去雇佣每小时收费 1,200 美元的商业律师；但在现有的风险代理体制下，他同样会被绝大多数律所拒之门外。 即使原告律师愿意按通行的三分之一提成收费，1,000 美元标的对应的 333 美元收益也根本无法覆盖律所支付给法务助理（paralegal）与基础文书调研的人工成本。这种机制导致律师为了维持生计，被迫放弃成百上千个前来求助的小额真实案件，将全部精力集中在极少数可能产生巨额赔偿的大案上。许多当事人的实际处境甚至演变成一种荒谬的困境：对律师而言，即使案件打赢了也是亏损的，因为办理案件的垫付时间与沉没成本远远高出最终能分得的法定赔偿。 - 价格与供给正相关：高标的案件吸引过量法律精力，低标的民事纠纷无人问津。 - 律所底线法则：在当前市场环境下，若单起案件预期无法为律所带来至少 5,000 美元的律师费收入，全美几乎没有原告律师愿意建档承接。 - 结构性悖论：即便胜诉概率极高，案件持有成本（carrying cost）一旦超过预期获益分成，律所接案即意味着商业亏损。 --- ### [07:00](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=420s) 劳动雇佣案件 96% 拒签率与 AI 带来的经济学逆转 在现实生活中，这种由于法律劳动成本高昂导致的司法正义缺位随处可见。例如发生了一起轻微交通事故，只要车损赔偿低于车辆保险免赔额（deductible）——即使这笔 3,000 到 4,000 美元的自付费用对一个工薪阶层家庭来说可能直接关系到下个月生计——全美也几乎没有任何律师愿意接手。在劳动与雇佣法（labor and employment）领域，这种矛盾尤为尖锐。例如服务业员工长期遭遇职场霸凌或客户骚扰，雇主本可通过完善管理规章加以制止，但受害员工却很难在司法体系中获得立足之地。 出现这一现象的根源在于前期评估成本过高：由于美国大部分州实行自由雇佣制（at-will employment），原告方必须将遭遇严密对应到受法律保护的特定诉由（causes of action）上。律师首先要花费 30 到 45 分钟与求助者通电话梳理案情，随后至少还要投入整整 10 个小时去调取判例法、核实证据细节、反复追问当事人，以确定其是否符合立案资质。在耗费了十几个小时的昂贵工时之后，律师往往无奈地发现这起纠纷的提成根本达不到 5,000 美元的门槛。正因如此，全美劳动与雇佣维权领域的来电有高达 96% 最终直接被律所拒之门外，普通人除了咽下委屈外别无选择。 Eve 改变了这一经济学底色。当求助电话接入时，AI 能够自主接听并进行深度对话，依托其学习的律所内部决策标准与法律规则，实时研判该案件是否具有可诉性。在瞬间完成以往需要 10 小时人工才能理清的基础法条匹配后，系统能直接自动化生成初步诉求函（demand letter）或前期文书雏形，让律师能够零边际成本地快速推进案件。这种技术手段彻底打破了法律服务的人力劳动瓶颈，使原本在经济账上完全不可行的大规模平民维权案件重新具备了立案价值。 - 雇佣法现实门槛：需耗费 30-45 分钟接听初访电话，再叠加至少 10 小时人工研究案情判例。 - 行业惊人数据：全美劳动与雇佣类别的法律维权咨询中，因人工核算成本过高而遭到拒绝的比例高达 96%。 - 自动化效率重构：Eve 将长达 10 小时的前期尽职与诉由排查压缩为即时完成，并自动生成首期索赔文书，使原本亏损的小微案件在商业上重获可行性。 ### [09:17](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=557s) 突破漫长案件周期的瓶颈与普及法律代理 当前基于大模型的法律 AI 仍处于极其早期的探索阶段，但 Eve 展现出的核心能力在于能够跨越极长的时间跨度，端到端地处理法律诉讼结果。原告律师事务所面对的案件往往需要历时六个月甚至一年以上才能推进到诉讼阶段，整个流程中包含数以千计的往复沟通与操作。随着模型精度的持续提高，处理这种超长周期的任务成为可能，从而将极其复杂繁琐的法律工作分摊降低为近乎恒定的边际成本。这种技术突破最直接的社会效益，就是让成千上万原本因为请不起律师或案件标的太小而无法获得代理的普通人，重新享受到法律维权的权利。 正如 Jay Madheswaran 所强调的那样：“So, now all these people without representation get representation.” 如果每个人都在生活中自省一下，假如向律师寻求帮助没有任何门槛与摩擦成本，自己一生中会咨询多少次法律援助？这个潜在的需求数字远超公众日常的想象。 - 案件推进跨度通常在 6 个月到 1 年以上，中间涉及上千次往返交互与操作。 - AI 准确度的提升使得超长周期案件的端到端自动化成为可能，大幅分摊了案件前期的边际成本。 - 降低法律咨询与代理门槛，释放了社会中长期被压抑的海量法律求助需求。 --- ### [10:02](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=602s) 收入扩张而非成本削减：原告律所的真实商业逻辑 社会大众对 AI 的第一反应往往停留在“裁员”和“削减成本”的刻板印象上，但 Alex Rampell 分享了一个生动的亲身经历：大约九个月前，他在加州洛斯阿尔托斯（Los Altos）的一家 Blue Bottle 咖啡馆向朋友解释 Eve 的模式，强调 Eve 的核心逻辑根本不是帮律所省钱，而是让律所能够接下更多案子、实现收入扩张。话音刚落，旁边一位牵着大狗、看起来颇具威严的男士径直走过来，让原本有点害怕大狗的 Alex 以为对方要发难，结果对方开口道：“I’m a trial attorney. I want to try it.” 这一插曲恰恰印证了出庭律师对于扩大业务吞吐量的极度渴望。 在全美的主要交通走廊上，无论是加州的 101 高速还是纵贯东海岸的 I-95，道路两旁密密麻麻树立着出庭律师和原告律所的巨幅广告牌。律所投入巨额营销预算就是为了吸引潜在客户来电，但由于内部人力匮乏，他们不得不拒绝掉多达 90% 的线索。Jay Madheswaran 指出，大众开车经过广告牌时，只看到律师光鲜的面孔，却忽略了每通咨询电话背后所需要的庞大人工工作量。原告律所的经济模型非常纯粹——代表的委托人越多，赚的钱就越多；限制他们的从来不是市场前端的客户需求，而是后端的劳动力供给。 ```mermaid flowchart TD accTitle: 原告律所线索处理瓶颈与 AI 介入 accDescr: 展示了传统模式下由于人工瓶颈导致的大量线索流失，以及 AI 介入后实现端到端吞吐量扩张的过程。 subgraph FrontEnd["前端获客"] A["巨额广告投放 (高速路牌/营销)"] --> B["海量客户致电咨询"] end subgraph Bottleneck["传统人力瓶颈 (放弃90%线索)"] B --> C{"人工供给不足"} C -->|"案件标的低/人力太贵"| D["拒绝代理 (放弃海量潜在营收)"] C -->|"有限人力受理"| E["个案繁重沟通与推进"] end subgraph AISolution["Eve 平台赋能"] B --> F["AI Agent 自动化接待与初筛"] F --> G["消除内部流转损耗与信息孤岛"] G --> H["律师仅在核心节点介入裁决"] H --> I["规模化实现收入扩张"] end ``` 这种劳动力短缺使律所陷入了不断转移的运营瓶颈。单打独斗的小律所往往只有一名出庭律师配一个案件经理，电话打进来先记在纸上，律师只能在碎片时间狼狈应对；当律所尝试规范化、外包呼叫中心作为接线团队后，一旦广告效果太好、咨询量暴增，呼叫中心又会迅速被击穿。律师们上的是法学院而不是商学院，他们的初衷是帮助普通人维权，却被困在小微企业的琐碎运营与流程泥潭中。Eve 的作用正是将这些流程收拢整合，托高服务底线，让律师把精力集中在真正需要专业判断的节点上。 - 诉讼律师的核心诉求是扩大案源与收入，而不是削减内部开支。 - 传统原告律所因为处理成本过高，不得不拒掉将近 90% 的进线索赔机会。 - 律所的产能瓶颈呈现动态转移：从单人兼顾接待的混乱状态，到接听团队被过载营销线索击垮。 --- ### [13:57](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=837s) 从人均 100 起到 500 起：重塑人效与律师角色 为了说明这一工具带来的剧烈变革，Alex Rampell 询问是否有客户实现了类似“生发广告前后对比”那样立竿见影的飞跃。Jay Madheswaran 举了一个洛杉矶真实律所的案例：该律所全面接入 Eve 后，单个律师平均管理的委托人案件量直接从 100 起暴增到 500 起。这一巨大增幅的背后，是律所将 Eve 的自动化能力应用到了极致——从最早期的电话接入、自动资格审查，到消除律师与案件经理之间来回传话的沟通损耗，全部由系统内的各类 Agent 无缝承接。 这种提效不仅停留在收案阶段，更一直贯穿至关键的和解谈判环节。在律师与调解员展开谈判之前，系统已经自动生成了完整的摘要简报，不仅精准提炼出本案应聚焦的核心争执点，还基于该律所过去处理类似案件的大量历史数据，推算出应当争取的目标和解金额区间，并附带曾经行之有效的具体谈判策略。 > “That is I think the future of law. Right? Where you as a lawyer are really going to look more like a almost a data scientist and engineer engineering something to be better and better.” —— Jay Madheswaran 顶尖律所的律师们正在经历这一关键的认知转变：他们不再被案头文书所困，而是演化为不断调试、优化法律交付系统的“数据科学家”与系统工程师，借由技术杠杆推动案情稳健向前。 - 洛杉矶标杆律所实现人均承办案件量从 100 起提升至 500 起，效率提升达 5 倍。 - 调解前置准备全面自动化：系统自动生成案件简报、定位争点，并根据历史相似判例输出目标和解额与谈判策略。 - 律师的职业角色发生深刻跃迁，从劳动密集型文书撰写者转变为类似“数据科学家”与系统设计者的操盘人。 --- ### [15:53](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=953s) 隐形劳动力市场：原告律所软件的真实体量 Alex Rampell 深入剖析了法律科技长期以来被风险投资人严重低估的本质原因。在传统的风险投资视角中，原告律所的软件市场显得极其狭小。对于 Wilson Sonsini 这样按小时计费的大型商业律所而言，最核心的工具是工时追踪软件；但在风险代理（Contingency Fee）模式的原告律所中，律师根本不向客户按小时收费，他们不需要复杂的计费系统，甚至连日常使用的核心生产力工具也仅仅是微软的 Microsoft Word。如果投资人只看律所购买微软办公套件的开销，会得出整个原告法软件市场每年只有几十亿美元的错误结论。 然而，这一行业每年撬动的实际案件结算金额高达数千亿美元。律所之所以无法交付更多结果，核心卡点在于缺乏极其廉价的基层劳动力——如果律所能以每小时 12 美分（即调用 AI Token 的成本）招到合法律务助理，他们完全能消化掉涌入的所有 5,000 个销售线索，但现实物理世界中根本不存在这样的劳动力市场。Alex Rampell 指出，AI 法律软件的本质根本不是工具，而是一个庞大无比的“影子劳动力市场”（Shadow Labor Market），其真实市场空间远超过去任何人的想象。 > “So, it turns out the software market is kind of like this shadow labor market, and it’s like massively bigger than anybody would have expected.” —— Alex Rampell Jay Madheswaran 进一步揭示了原告法领域在软件采用上的特殊生态。历史上，一个刚从法学院毕业的原告律师甚至只需要一部手机和一个电子邮箱就能开业，许多独立执业者出于成本考虑连 Microsoft Word 都不愿订阅。更底层的逻辑在于诉讼工作与商业合同审查截然不同：合同审查需要双方反复批注与修订（Redlining），具备强烈的协同属性；而原告诉讼处于高度对抗的信任边界（Trust Boundary）两端，原告律师与辩护律师之间天然敌对，根本不存在跨团队协作修改文档的需求，这导致了该领域在 AI 到来前几乎处于软件荒漠状态。 | 维度 | 传统商业律所 (如 Wilson Sonsini) | 传统原告律所 (风险代理模式) | AI 赋能下的原告律所 (Eve) | |:--- |:--- |:--- |:--- | | **计费与营收模式** | 严格按小时计费 (Billable Hour) | 风险代理分成 (胜诉抽成，不按小时计费) | 风险代理模式，依靠极高案件周转量实现收入倍增 | | **核心软件堆栈** | 工时追踪软件、电子审阅 (e-Discovery)、Word | 智能手机、电子邮箱、最基础的 Word（甚至不用） | 全生命周期法律 Agent 平台 (Eve) | | **协作特征** | 双方频繁红线批注 (Redlining)、多方合同协同 | 对抗性强，处于信任边界两端，零跨机构文档协同 | 内部高度自动化协同，AI 辅助梳理争议焦点与对策 | | **市场规模认知** | 传统垂直 SaaS 视角（受限于员工人头订阅费） | 极低（IT 支出仅限基础办公工具） | **影子劳动力市场**：吞吐原本无法承担的千亿级法律索赔标的 | - 风险代理律所由于缺乏按小时计费的需求，长期依靠手机和邮箱运作，传统 IT 软件开销极低。 - 诉讼业务属于对抗性领域，缺乏商业合同审阅所需的协作红线批注需求，因此传统文档协作软件渗透率极低。 - 法律 AI 软件的商业本质是向律所供给每小时成本仅几美分的“数字化劳动力”，打破了实体劳动力短缺对业务扩张的硬性约束。 ### [18:24](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=1104s) 原告律师律所的劳动力瓶颈与法律软件的预算错配 以往的软件从未真正解决过原告律所的核心难题——劳动力限制。因为原告律所掌握着劳动力且善于管理人力，在他们的视角里，任何问题都可以通过堆人手来解决，律所也一直沿用这种模式发展壮大。这也导致法律软件行业出现严重的资源倾斜：绝大多数软件公司（包括 Harvey 等）主要面向辩方或大型辩护机构，因为那里的预算充沛且付费爽快。 相比之下，原告律所通常是锱铢必较的小微企业。如果一款软件无法从第一天起就为他们带来实打实的业务成果，他们绝不会续约，甚至会毫不客气地找供应商算账。然而，原告律师的底层诉求极为明确：代表更多当事人并打赢官司。如今的软件迎来了本质上的跃迁——一旦某项手动工作（包括处理复杂数据和转化信息）被完成过一次，AI 就能将其固化为标准化流程。这种过去依赖“人员运营”（people operations）却因每个案件细节不同而屡屡失败的模式，如今已被 AI 攻克，单个优秀律师得以打破认知负荷上限，同时统筹多达 500 起案件事实迥异的复杂纠纷。 - 辩方律所拥有充裕的软件预算，是大多数法律 AI 初创团队的首选切入点。 - 原告律所对软件价值极度敏感，要求立竿见影的投资回报。 - 传统原告律所受制于招聘门槛，无法通过单纯堆砌人力来规模化处理高度非标的案件。 ### [20:25](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=1225s) 为什么垂直领域的 Eve 能胜过大模型“通用智能” 面对众多应用层创业者担忧自己被研发底层基础模型的大厂沦为“路杀”（roadkill）的疑虑，Jay Madheswaran 指出，Eve 的核心优势并非对抗大模型，而是深度拥抱并押注底层模型会愈发强大。他打了个生动的比方：哪怕今天拥有 1000 个爱因斯坦来处理诉讼，把他们直接扔进一家律所，他们在很长一段时间内也毫无生产力可言。通用大模型在原告法律领域的短板在于缺少未公开的私有数据与实操语境：原告案件中仅有不到 1% 的案件会最终判决并形成公开的判例意见，绝大部分都在发送索赔函前或立案前就达成了私下和解。 Alex Rampell 举例说明这种博弈的现实场景：比如遭遇车祸且自身无责的当事人，即便车子报废、人身受损，保险公司的首要原则永远是“拒绝、拒绝再拒绝”，试图用 1 美元打发受害者。逼迫当事人聘请律师后，保险公司可能会悄悄赔付 29,000 美元达成和解，这种实际赔付数据极具价值，却无法从公开互联网中抓取。保险巨头（如 Geico）同时与上万家律所打交道，掌握着类似 Colossus 这种计算“断一根拇指值多少钱”的定价软件，原告律所面对的是完全信息不对称的对手。即便去 Google 搜索“如何针对 State Farm 撰写一份顶级软组织损伤索赔函”，也搜不到任何现成结果。 > “The problem is if you drop Albert Einstein into a law firm today, he will not be productive for quite some time.” —— Jay Madheswaran ```mermaid flowchart TD accTitle: 传统原告律所与 Eve 赋能模式的对比 accDescr: 对比传统原告律所在信息孤岛下的低效博弈与 Eve 结合非公开和解数据后的高效决策链条 subgraph Defense ["辩方机构（保险公司）"] Colossus["内部理赔软件（如 Colossus）"] DenyRules["策略：无差别拒赔与压价"] GlobalData["掌握数十万案件的赔付底牌"] end subgraph EvePlatform ["Eve 赋能的原告律所"] Intake["案件线索智能初筛 (Intake)"] Auditor["Auditor 系统全量夜间巡检"] KnowledgeBase["隐性经验与战术沉淀"] SmartDemand["定制化索赔函与谈判策略"] end Colossus --> DenyRules DenyRules -->|"信息不对称施压"| Intake GlobalData -.->|"壁垒：非公开数据"| KnowledgeBase Intake --> Auditor Auditor --> KnowledgeBase KnowledgeBase --> SmartDemand SmartDemand -->|"高胜率对弈"| Colossus ``` - 原告法领域 99% 以上的案件数据无法通过公开网络抓取，基本在判决前私下和解。 - 保险公司使用专门软件（如 Colossus）推算理赔金额下限，并对普通当事人采取惯性拒赔策略。 - 地方司法管辖区、法院体系及特定辩方律师的偏好差异极大，构成了高度复杂的上下文要素。 ### [24:47](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=1487s) 律师数据隐私、经验共享与 Auditor 产品的夜间巡检 在跨地域律所的数据壁垒与知识共享问题上，Eve 严格遵守行业既有的规则与红线。由于“律师-委托人特权”（attorney-client privilege）的存在，律所对案件事实与当事人隐私要求极高，Eve 明确保证绝不对客户的私有案件数据进行模型训练。与此同时，律师行业本身有通过参加法律继续教育培训（CLE）和线下研讨来交流破解大型保险公司抗辩策略的传统。Eve 内部聘请了专业律师，旨在将这些共享战术与成功经验沉淀为系统能力。 Eve 还开发了一款名为 Auditor 的产品。该产品在夜间自动对在办案件的每一页文档进行拉网式全量扫描，模拟顶尖律师的审查水准来寻找漏洞与遗漏。在现实中，受制于人工精力和遗忘曲线，绝大部分案件卷宗中都存在被律师或助理遗漏的关键信息。Auditor 的全量审计能力消除了因疏忽导致的失误，直接提升了案件的获赔率与和解金额。随着对抗机制演进，双方都将拥有 AI 超能力，法律博弈正在演变为系统级能力的碰撞。 - 客户专属案件数据受“律师-委托人特权”保护，绝不用于公用模型训练。 - Auditor 能够在夜间逐页分析卷宗，挖掘人工极易忽略的细微证据与关键线索。 - 平台通过量化案件胜率指标来检验软件对律所的实际增益，并以此沉淀针对常见抗辩的应对范式。 ### [27:37](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=1657s) 当博弈双方都拥有 AI：从例行拉锯到“智能体对战智能体” Alex Rampell 抛出了一个博弈论视角的比喻：如果你想预订一家极难预约的热门餐厅，而对方依靠人工接听电话，如今你每小时只需花上一分钱就能发起类似 DDoS 的呼叫轰炸；但最终对方也会意识到自己被攻击，从而升级防御。在法律领域也是如此，以前像 Geico 这类大保险公司可以对受害人轻描淡写地表示“你的损失只有 2,000 美元”，即便受害人实际损失了 2,500 美元，由于涉案金额太小、诉讼成本过高，保险公司根本无需担心被起诉；但现在 AI 赋能了那些原本处于弱势的“大卫”，让他们能够奋起反击“歌利亚”。那么，当 Geico、State Farm 这些制度化巨头同样引入 AI，长周期的智能体（long horizon agents）介入后，对抗过程会演变成什么样？ Jay Madheswaran 认为，无论未来 AI 在案件处理中承担了 99% 还是 99.9% 的工作量，个人依然需要律师作为代理人，最核心的关键始终是那最后一点点人类的温度与判断。保险公司等防御方一直以来的强项在于更雄厚的资金、更充足的资源，以及最重要的武器——时间；普通人拖不起诉讼，而大机构可以通过标准化和流程拖垮对方。但未来，AI 将打破这种壁垒，所有常规的文书拉锯都将自动化： > “很多案件都将拥有充足的人手支持，而所有常规的往复交涉，我认为都将通过智能体与智能体之间的对话来完成。”——Jay Madheswaran 最终的分水岭在于人类律师对当事人的深入理解、对具体事实的洞察，以及对当地司法环境和对手律所的深谙度。 ```mermaid flowchart LR accTitle: AI时代法律博弈演进流程 accDescr: 展示了AI如何处理常规对抗，并将人类律师解放至核心出庭辩护。 subgraph S1["常规博弈层 (自动化)"] A["原告 AI Agent"] <-->|"海量常规文书 / 往复交涉"| B["保险/被告 AI Agent"] end subgraph S2["核心价值层 (人类驱动)"] C["原告律师 (深入洞察案件事实与当事人)"] --> D["法庭实地辩护与审判 (Trial Advocacy)"] end S1 -.->|"沉淀底层数据并释放精力"| S2 ``` ### [31:04](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=1864s) 重返真正的主张：从被迫和解到法庭审判与数据决策 当大量繁琐的案头工作被剥离后，律师的实际执业形态正在发生深刻转变。过去，对于绝大多数中小型律所而言，“走上法庭审判（go to trial）”几乎是不可承受之重——一次审判可能需要全所律师停下手头所有事情，通宵达旦地准备材料、挑选陪审团，因此绝大多数案件最终都会退化为妥协性的和解（settlement motion）。 如今在 Eve 的辅助下，许多客户反而主动选择把更多时间花在真正的出庭辩护（trial work）上。这使律师能够回归纯粹的主张本身：当当事人退无可退、身处绝境时，律师应该如何真正在法庭上为他们争取核心权益。与此同时，这也将律师推向了类似“数据科学家”的定位： - **诉讼投资（Litigation Finance）的启发**：以往只有资金雄厚的诉讼金融机构会利用庞大的法官倾向、历史判决数据来决定是否资助某起诉讼； - **打破数据孤岛**：由于大部分和解金额完全保密、缺乏公共数据库，普通律师以往只能凭感觉评估报价； - **量化最优决策**：拥有海量底层数据后，律师面对 Geico 提出的 12,000 美元和解金时，能够精准研判是应该反击争取 15,000 美元，还是直接推向审判博弈 100,000 美元。 > “他们正在利用软件的力量去完成以前从未想过要开展的工作，以此来改善案件的最终结果。”——Jay Madheswaran ### [33:01](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=1981s) 司法系统的未来：机器调解与自动争议解决（ADR） 谈及整个司法体系的瓶颈，Alex Rampell 分享了自己的亲身经历：他曾因某件事与富国银行（Wells Fargo）纠缠了 4 个小时未果，愤怒之下打开 ChatGPT（Codex）输入“起诉富国银行”；仅仅 4 个小时后，系统不仅起草好了全部起诉材料，甚至在线找到了送达传票的人（process server），正式完成了对富国银行的起诉，只是开庭日期被排到了次年二月。这种门槛的骤降，类似于专利法领域中写专利变得极其容易、但美国专利商标局（USPTO）却成了严重堵塞的交通瓶颈一样，全社会的法庭体系将不堪重负。 Jay Madheswaran 预测，未来并非所有纠纷都能挤进法庭系统，法律界必然会分流： - **既有非诉机制的延展**：如同退伍军人事务部（VA）纠纷由专门委员会处理、劳动与雇佣纠纷先走平等就业机会委员会（EEOC），或是企业内部广泛采用的调解与仲裁； - **自动化争议解决（ADR）的崛起**：未来中立的仲裁机构或调解方将全面引入 AI 工具，基于历史案件事实与判例数据快速厘清因果脉络、剔除人为偏见，并在极短时间内出具调解结论。 > “最终必然要建立起一整套分流体系，让绝大多数争议根本不需要涌向法庭系统。”——Jay Madheswaran ### [35:15](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=2115s) 从家庭小作坊到十亿美元巨头：律所的软件化革命 随着生产力飞跃，Alex Rampell 提出了一个关于律所商业形态的根本疑问：如果一个律师过去只能处理 100 起案件，现在能处理 500 起，未来可能处理 5,000 起，为什么一家律所还要局限在纽泽西某个街角的小办公室？目前全美原告律师界虽然有像 Morgan & Morgan 这样年营收数十亿美元的庞然大物，但绝大多数仍是极其分散的“夫妻店（mom-and-pop）”和生活方式型企业（lifestyle business）。在 AI 的杠杆放大下，一家小型律所是否有可能成长为规模达十亿美元的巨头？ Jay Madheswaran 给出了肯定的判断。他指出，律所形态正迎来不可逆的重塑： > “即将发生的转变在于，大量律所的运作模式将变得越来越像软件公司。”——Jay Madheswaran 律所将摆脱以手工计时和线性雇人为基础的作坊模式，转而依靠高密度的数据资产、自动化智能体管线和规模化获客引擎，以接近科技公司的边际成本和扩张节奏实现爆发式增长。 ### [36:36](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=2196s) 律所经济学与纵向专业化的双向重构 Jay Madheswaran 剖析了传统律所过去 30 年被软件革命边缘化的底层经济原因。以一起经典的 25,000 美元和解案为例：律师通常抽取三分之一（约 8,000 美元）的胜诉提成，其中约 50% 耗费在人工成本上，30% 用于获客与市场营销，剩下极微薄的利润。在这种成本结构下，律所几乎无法通过技术再投资来提升对下一个客户的服务质量，唯一的规模化手段就是不断高薪雇佣更多律师。但随着 AI 平台的引入，律所开始能够沉淀属于自己的私有数据集与案例库，从而在细分法域中取得远超同业的结果。 这种变化推动了进一步的垂直专业化。历史上，法律行业曾经历过从“处理酒驾、知识产权到雇佣纠纷的全能街角律师”到“大类专业化律师”的演进；而如今，AI 正在将专业分工推向极其微观的颗粒度——例如专门钻研“特定事实组合下的软组织损伤诉讼”或“产科婴儿缺陷案”。对此类精准案例的定向投放不仅极其高效，获客成本也远低于像 Morgan & Morgan 这种依靠在全美铺天盖地打广告牌的传统大所。Alex Rampell 将其概括为：“先在本地把某一细分领域做到极致，然后凭借对该领域极高的胜率把握，将业务扩张至全美。” 与此同时，行业另一端也在发生剧烈的规模整合。以巨型律所 Morgan & Morgan 为代表的机构正在大举并购其他律所。以往此类收购往往受阻于漫长而痛苦的组织再造和流程整合，但现在通过 Eve 这样的系统，律所可以屏蔽人员组织层面的巨大变动与摩擦，直接围绕客户的核心诉求——“法律结果”——迅速完成资产与流程的统一，大幅缩短并购获得正向回报的时间。 ```mermaid flowchart TD accTitle: 传统原告律所与 AI 赋能律所的经济结构对比 accDescr: 对比传统原告律所依赖人力资本导致利润微薄与 AI 律所通过数据沉淀实现超细分专业化和规模化的差异 subgraph Traditional ["传统律所经济模型 ($25,000 和解金)"] Settlement["和解金: $25,000"] --> AttorneyFee["律师提成 (1/3): 约 $8,000"] AttorneyFee --> Labor["人力支出: 50% (~$4,000)"] AttorneyFee --> Marketing["获客营销: 30% (~$2,400)"] AttorneyFee --> Profit["微薄留存利润: 20% (~$1,600)"] Labor -.->|"阻碍再投资，只能靠加人扩张"| ScalabilityCeiling["扩张天花板"] end subgraph AIEra ["AI 驱动的专业化与规模效应"] EveData["Eve 私有案例数据库沉淀"] --> DeepSpecialization["超垂直领域专业化 (如特定软组织损伤)"] DeepSpecialization --> LowerCAC["超精准定向广告 (更低获客成本)"] DeepSpecialization --> HigherWinRate["更高胜率与赔偿额"] EveData --> FastMA["无摩擦并购整合 (平滑组织管理剧变)"] end ``` ### [39:16](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=2356s) 从“按工时计费”向“交付结果计费”的范式转移 Alex Rampell 指出，整个软件行业目前正在经历向“基于结果定价”（outcome-oriented pricing）的深刻变革。早期的企业级 SaaS 依赖按人头（per-seat per-month）收费模式，被他戏称为软件销售的“中杯、大杯、超大杯（Tall, Grande, Venti）模式”——打开任何 SaaS 网站，几乎都是千篇一律的 79 美元、99 美元或 129 美元月费标签。以 Zendesk 为例，其被 Permira 和 Hellman & Friedman 私有化前 ARR 已达 20 亿美元，但随着 AI 取代人工座席，企业所需的席位数量甚至可能归零，这迫使整个行业转而追求按结果交付并按结果计费。 与之形成鲜明对比的是，原告人身伤害法律从诞生之日起就是百分之百的“按结果收费”（纯胜诉提成制）。Alex 认为，包括移民法甚至辩护律师等其他法律门类，从商业逻辑上也早该转向结果导向定价，但阻碍这种演进的核心障碍是“数据不对称”与“工作量不可预测性”。Alex 分享了自己聘请 McDermott Will & Emery 的信托遗产律师的经历：其费率极其昂贵，但很难按结果定价，因为“我猜我的交付结果就是我的死亡，那我可不希望它早点来”。在传统公司法中，律师按小时计费并非故意敲诈，而是因为客户脑海中的意图与实际口头表达之间存在巨大鸿沟，案情随时可能剧烈膨胀；这也导致律所经常把早期融资法务作为微利的“引流亏损产品”（loss leaders），再试图通过后期的并购（M&A）或 IPO 大单将利润赚回。 Jay 认同原告法业务是法律甚至整个商业世界中最超前跑通结果定价的领域。为了实现这一点，Eve 在过去一年多里深入开发了所谓的“长任务代理”（long-horizon agents），使得 AI 能独立推进复杂流程直至交付结果。Jay 强调，法律真正走向民主化的关键，在于能否从一组输入事实中精确测算出达成特定结果所需的算力规模。他直言： > “最终，我认为法律民主化所必需的是：从一组既定事实出发，你能够清楚知道自己需要投入多少算力资源才能推导出一个确定的法律结果。” 对于企业客户在法律上开具“空白支票”的现状，Jay 指出传统按小时收费的本质是“律所自身也不知道其工作实际的期望价值（expected value）究竟是多少”。 - **SaaS 的定价断崖**：按座席收费的“Tall, Grande, Venti”模式正被 AI 瓦解，软件业被迫从“卖工具人头”转向“卖业务结果”。 - **公司法中的失衡博弈**：比如企业进行 C 轮融资法务，若律所按 25,000 美元包干计费，极可能因遭遇突发复杂条款而耗费 4,000 小时律师工时，面临灾难性亏损，这也是按小时计费模式在商事领域长期存续的主因。 ### [44:03](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=2643s) Eve 的商业模式进化与算力循环 当谈及 Eve 自身的商业模式时，Jay 坦言公司在初创时也曾掉入思维定式，采用了按席位卖软件的“Tall, Grande, Venti”传统订阅套路。但深入法律实务一线后，团队迅速发现原告律所的核心痛点只有两个：如何达成更好的法律结果，以及如何在不线性扩张人力的情况下服务更多委托人。因此在一年半前——这在迭代迅猛的“AI 时代”已相当于上百个狗年——Eve 便坚决转向了基于成果价值的定价路径。 这一转变契合了原告律所的利润分配模型：原本占胜诉酬金 50% 的繁重人工劳动支出被大幅缩减。律所不再需要把大量资金消耗在初级律师的机械审查与文书撰写上，而是将这笔省下来的开销反哺于 AI 算力——在 Eve 上消耗更多 Token，完成更深度的调查、案情拆解与文本推进，从而驱动更稳健的赔偿和解。这种由算力直接转化为律所营收的正向商业循环，让律所摆脱了对工时堆砌的依赖。Jay 总结道，虽然按小时收费的机制可能不会彻底消亡，但在大势所趋的效率红利面前，它必将不可逆转地走向式微。 ### [45:23](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=2723s) 摆脱按席位收费：交付端到端的法律结果 Jay Madheswaran 首先深入探讨了法律科技的商业模式。他指出，对于某些特定的法律细分领域，按席位收费（seat-based model）可能是最合适的方式，因为案件往往需要耗费数年才能看到结果，很难在早期量化其价值。然而，在如今的大量诉讼场景中，案件的走向和结果其实是可以预期的。律所理应能够把工作成本摊销并直接兑现价值，而这正是 Eve 的核心押注方向。 Eve 从第一天起就致力于端到端地交付最终的法律结果，将操作体验简化到极致。正如 Jay Madheswaran 所形容的：“Here's this client I got in. Give it to Eve. Eve will pull you in when you're needed and you're good to go.”（律师只需说：‘这是我接到的客户，交给 Eve 吧。Eve 会在需要你时叫你介入，其他时候你尽管放手。’）这种产品哲学直接颠覆了传统法律软件单纯作为辅助工具的逻辑。 ```mermaid flowchart LR accTitle: Eve端到端案件处理流程 accDescr: 展示从客户录入到律师仅在关键节点介入的端到端自动化流程 subgraph INTAKE ["案件接入"] A["录入新客户案件"] --> B["交由 Eve 接管"] end subgraph AUTOMATION ["自主流转与协作"] B --> C["自主调用底层模型<br/>(Claude 等)"] C --> D["推动多环节任务落地"] end subgraph RESOLUTION ["案件终局"] D --> E{"是否需要人类决策?"} E -- "是" --> F["提醒律师介入处理"] E -- "否" --> G["达成端到端案件结果"] F --> G end ``` ### [46:08](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=2768s) 智能体工作流与工程瓶颈 Alex Rampell 随后指出，原告律师本质上受限于供给端而非需求端：由于获取客户成本（CAC）与生命周期价值（LTV）算不过来账，律师不得不放弃海量的潜在客户。Alex 进而追问，Eve 自身高速增长的限速器（rate limiter）是什么，需要哪些投入才能进一步爆发？ Jay Madheswaran 回应称，市面上公开发布的法律基准测试（包括与他们紧密合作的 Anthropic 和 OpenAI 所关注的指标）大多还停留在单一孤立任务的表现上，比如测试模型完成某项单独任务是否合格。但真实世界正在迈向智能体（agentic）协作，而 Eve 从一开始就只在乎智能体带来的最终端到端结果。虽然目前技术和流程中仍需一定的人工干预，但每个月这些干预都在被大幅压缩。Eve 正在把每个案件类型当作一门精密科学来攻克，致力于构建一个能够消化海量数据、调用诸如 Claude 等工具并推动结果的“微型律师大脑”。 - **研发瓶颈**：当前 Eve 增长最大的限制因素是工程师团队规模，面对无限的研发待办事项，需要足够的技术力量来实现全自动化。 - **市场空间**：只要能彻底攻克机动车交通事故（moving vehicle accident）这一个细分场景的端到端自动化，就能为全球释放数千亿美元的商业价值。 ### [48:15](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=2895s) 一个案件里的 10,000 次交互与 AI 对话 AI 当 Alex Rampell 询问是否已经实现了从接听电话、入职客户到从 GEICO 拿到赔偿支票的“完全自主闭环”时，Jay Madheswaran 坦言目前还没有彻底做到，但这正是他们急需优秀工程师的原因。 Jay 揭示了一个令人震惊的现实：单个案件在生命周期中可能涉及多达 10,000 次来回沟通。例如在人身伤害（PI）案件中，调取客户过往医疗记录的环节通常需要长达两周时间，工作人员必须反复向保险公司和医疗机构拨打电话。Eve 开发了具备自主语音能力的对外呼叫智能体（Voice AI）来执行这些电话联络，现场甚至已经出现了极其荒诞且滑稽的一幕：Eve 的 AI 智能体在电话线上与对方机构的 AI 智能体斗智斗勇，相互试探并寻找绕过对方防线的破局之法。 ```mermaid flowchart TD accTitle: 医疗记录调取环节的智能体对抗 accDescr: 展示Eve语音智能体与保险机构系统沟通并推进案件的环节 A["客户提交人身伤害线索"] --> B["Eve 语音 AI 呼出电话"] subgraph CALL_FLOW ["智能体交互战场"] B --> C["医疗机构 / 保险公司 AI 接线员"] C <-->|"智能体对抗与突破规则"| B end CALL_FLOW --> D["成功调取医疗记录<br/>(以往需耗时两周)"] D --> E["生成法律文书并向被告发起交涉"] ``` ### [49:37](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=2977s) 感恩节证据开示突袭与攻防战逆转 Alex Rampell 接着探讨了被告方（如保险巨头 GEICO）面对这种生产力跃迁时的反应，好奇他们数十年来固化的理赔与和解规则是否会因对手成本骤降而改变。Jay Madheswaran 指出，辩护律师（defense side）大部分依然沿用传统的计费工时模式，在节奏上完全跟不上原告律师如今的技术跃迁，双方形成了巨大的成本与定价不对称。 Jay 分享了一家合作律所的经典攻防故事：过去，被告律所经常故意把证据开示（discovery）请求的截止日期拖到感恩节或重大节假日前一天，趁原告律师准备休假时搞突袭，企图用堆积如山的工作量毁掉对方的假期，迫使对方在仓促中出错或交出低质量答复。面对这种卑劣的惯用套路，该律所直接利用 Eve 构建了一个专属智能体。当防守方再次掐点发起假期突袭时，Eve 智能体在当天几小时内便自动化完成了全套专业答复并甩了回去，直接打得对方措手不及。Jay 评价这一戏剧性转变时表示：“And and like we're seeing there's so many variants of that that is happening now which is causing defense to realize, wait, is this this is what happens when the other side has resources? Right. Oh crap. And and that's that's what's fun to see.”（辩护方终于意识到：等等，原来当对手拥有充沛资源时会变成这样？天哪。看他们吃瘪真的很有趣。） 在对谈结尾，Alex 对单个案件涉及 10,000 次交互的惊人细节感慨万千，并对能作为合伙人支持 Eve 推广至未来 10,000 家律所感到兴奋。Jay 总结道，Eve 如今已协助服务了数十万名客户，但未来仍有数以亿计的当事人需要法律援助，他呼吁更多优秀的工程人才加入 Eve，共同用 AI 改变普通人的法律处境。 ## Notable quotes > [35:15](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=2115s) 即将发生的转变在于，大量律所的运作模式将变得越来越像软件公司。 — Jay Madheswaran > [15:53](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=953s) So, it turns out the software market is kind of like this shadow labor market, and it’s like massively bigger than anybody would have expected. — Alex Rampell > [13:57](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=837s) That is I think the future of law. Right? Where you as a lawyer are really going to look more like a almost a data scientist and engineer engineering something to be better and better. — Jay Madheswaran > [27:37](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=1657s) 很多案件都将拥有充足的人手支持，而所有常规的往复交涉，我认为都将通过智能体与智能体之间的对话来完成。 — Jay Madheswaran > [39:16](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=2356s) 最终，我认为法律民主化所必需的是：从一组既定事实出发，你能够清楚知道自己需要投入多少算力资源才能推导出一个确定的法律结果。 — Jay Madheswaran > [20:25](https://www.youtube.com/watch?v=iRMmPjunm3Q&t=1225s) The problem is if you drop Albert Einstein into a law firm today, he will not be productive for quite some time. — Jay Madheswaran ## Mentioned Eve, Jay Madheswaran, Alex Rampell, a16z, Wilson Sonsini, Morgan & Morgan, Harvey, GEICO, State Farm, Colossus, Auditor, Wells Fargo, ChatGPT, Codex, USPTO, VA, EEOC, Zendesk, Permira, Hellman & Friedman, McDermott Will & Emery, Anthropic, OpenAI, Claude, Microsoft Word ## Video description What happens when AI makes legal work cheap enough to take cases lawyers couldn't afford to touch before? Eve CEO Jay Madheswaran argues that AI won't just cut costs for law firms. It could expand access to legal representation by making an entirely new class of cases economically viable. Jay joins a16z General Partner Alex Rampell to explain how Eve is automating plaintiff law from the first client call through settlement, and why some firms are already using it to increase the number of clients each attorney can manage from roughly 100 to 500. They also discuss how AI could shift lawyers toward higher-value advocacy, why agents may eventually negotiate with other agents, and what happens to the legal system when bringing a case becomes dramatically cheaper. Timestamps: 0:00 Intro 0:51 Plaintiff Law and the David-vs-Goliath Pitch 3:50 Why You Can't Get a Lawyer for a Small Case 8:42 Revenue Expansion, Not Cost Cutting 13:57 From 100 to 500 Clients Per Attorney 15:53 Legal Software Is Secretly a Labor Market 20:25 Why Eve Beats "Raw Intelligence" 27:00 When Both Sides Get AI Superpowers 29:13 What's Left for Human Lawyers 33:01 Suing Wells Fargo With ChatGPT 35:15 Can a Small Firm Hit a Billion in Revenue? 39:16 The Death of the Billable Hour 46:08 The 10,000 Touches in a Single Case 49:37 The Thanksgiving Discovery Trick Resources: Follow Jay Madheswaran on LinkedIn: https://www.linkedin.com/in/jayanth1/ Follow Alex Rampell on X: https://x.com/arampell Stay Updated: If you enjoyed this episode, be sure to like, subscribe, and share with your friends! Find a16z on X: https://twitter.com/a16z Find a16z on LinkedIn: https://www.linkedin.com/company/a16z Listen to the a16z Show on Spotify: https://open.spotify.com/show/5bC65RDvs3oxnLyqqvkUYX Listen to the a16z Show on Apple Podcasts: https://podcasts.apple.com/us/podcast/a16z-podcast/id842818711 Follow our host: https://x.com/eriktorenberg Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details please see http://a16z.com/disclosures. [Source](https://www.youtube.com/watch?v=iRMmPjunm3Q)

[35:15] Jay Madheswaran: 即将发生的转变在于，大量律所的运作模式将变得越来越像软件公司。 [15:53] Alex Rampell: So, it turns out the software market is kind of like this shadow labor market, and it’s like massively bigger than anybody would have expected. [13:57] Jay Madheswaran: That is I think the future of law. Right? Where you as a lawyer are really going to look more like a almost a data scientist and engineer engineering something to be better and better. [27:37] Jay Madheswaran: 很多案件都将拥有充足的人手支持，而所有常规的往复交涉，我认为都将通过智能体与智能体之间的对话来完成。 [39:16] Jay Madheswaran: 最终，我认为法律民主化所必需的是：从一组既定事实出发，你能够清楚知道自己需要投入多少算力资源才能推导出一个确定的法律结果。 [20:25] Jay Madheswaran: The problem is if you drop Albert Einstein into a law firm today, he will not be productive for quite some time.

[35:15] Jay Madheswaran: I think the transformation that's going to happen is a lot of these law firms are going to look more like software companies. [15:53] Alex Rampell: So, it turns out the software market is kind of like this shadow labor market, and it's like massively bigger than anybody would have expected. [20:25] Jay Madheswaran: The problem is if you drop Albert Einstein into a law firm today, he will not be productive for quite some time.

[0:00] Historically, plaintiff law really [0:02] didn't need more than a cell phone, [0:03] email address. That was kind of it. No [0:05] software really solved their core [0:07] problem, which is labor constraint. [0:09] >> Most people think of AI is like, "Oh, [0:11] you're going to fire everybody and cut [0:12] costs." And it's like, "No, actually, [0:14] like if I use Eve, I make a lot more [0:16] money for my firm." [0:17] >> People can't pay $500, $600 an hour on a [0:20] blank check for however long [music] the [0:22] case might take. What Eve has aimed to [0:24] do is really handle that whole thing end [0:26] to end. One firm in LA went from [0:28] managing about a hundred clients per [0:31] attorney to five. Wow. That is a type of [0:33] thing that is possible. [0:34] >> Why is Eve better than just kind of like [0:37] raw intelligence? [0:39] >> drop Albert Einstein into a law firm [0:41] today, he will not be productive for [0:42] quite some time. That is a general [0:43] problem that Eve is solving. And why [0:45] won't that person be productive? Well, [0:47] it's due to a number of things. One is [0:50] >> Hi everyone. I'm Alex Rampell, partner [0:53] at Andreessen Horowitz, and I'm very [0:54] happy to be here with Jay from Eve, uh [0:57] co-founder and CEO. [0:58] Um and everybody knows a lawyer, right? [1:02] Everybody knows law. [1:04] But most people, I think especially in [1:06] Silicon Valley, when they hear law and [1:08] legal, they kind of think about [1:09] corporate law. Like really big, you [1:11] know, supposedly white white shoe, I [1:12] guess is the term, law firms, where you [1:14] go to Harvard Law School and you bill [1:15] $2,000 an hour. But there's like another [1:18] side of law, which actually is bigger, [1:21] more distributed, and I guess we would [1:23] call that plaintiff law. Um so like [1:25] every time you drive on 101, you see all [1:27] these billboards. Um and maybe kind of [1:30] talk about like what is maybe start off [1:34] like what is Eve? [1:36] Um what form of law and legal AI do you [1:39] focus on? And maybe a little bit about [1:41] the industry in general, because I feel [1:43] like most Americans are familiar with [1:45] it, but they don't necessarily realize, [1:47] I mean especially for people I think [1:49] here in Silicon Valley, they've heard of [1:50] Wilson Sonsini, but they haven't heard [1:52] of Morgan&Morgan, right? And Morgan& [1:54] Morgan is probably a lot more profitable [1:56] than Wilson Sonsini. But like what what [1:58] is plaintiff law and also kind of maybe [2:00] explain Eve in your own words. [2:02] >> Absolutely. One, thank you for having me [2:03] here, Alex. Always fun to talk about [2:06] Eve. Um [2:08] what Eve is is is legal AI for real [2:10] people, right? Like people like you, me, [2:13] not institutions, not in-house legal, [2:15] not the people that already have all the [2:17] resources. But, how do we actually [2:20] provide legal services to people um that [2:23] otherwise can't afford it. [2:25] And this is how the plaintiff law [2:27] industry really came about. Because what [2:29] ended up happening was you talked about [2:31] how Wilson Sonsini might charge $2,000 [2:34] an hour, and that's that's really the [2:35] business model. Lawyers'times are [2:37] expensive. And you're trying to point [2:39] lawyers towards more and more expensive [2:41] time. [2:42] The same is true for most lawyers. They [2:44] they have a lot of bills to pay, and [2:46] they need Today, legal work, the more [2:48] work you put behind it, [2:50] you can represent your facts better. You [2:52] can get a better outcome. That is the [2:54] state of world we live in today. [2:57] To account for this, [2:58] lawyers understand people can't pay [3:01] $500, $600 an hour on a blank check for [3:04] however long the case might take. So, [3:06] the plaintiff industry came about uh on [3:09] working on contingency law. So, what [3:11] this means is they only get paid when [3:14] they win the case for the client. And [3:16] this is what's allowed them to even have [3:19] any sort of leveling playing field with [3:21] people versus institutions. [3:23] And what Eve is aimed to do is really [3:26] handle that whole thing end to end, and [3:28] of how do you represent a client all the [3:30] way from the very first call to getting [3:33] them a settlement. How do you get that [3:34] end to end thing [3:36] purely with AI, and give that to the [3:39] lawyers to then amplify the outcomes [3:41] they can actually provide. [3:42] And that's what Eve is doing, and we're [3:45] really arming, you know, think about [3:47] giving David a slingshot to take down [3:48] Goliath. That's what we're doing. [3:50] >> So, what I find fascinating about this [3:52] is uh you really I I kind of think about [3:56] Econ 101, where you have a supply curve [3:58] that goes up into the right, and what [3:59] does that mean? Um just remember there's [4:02] price and quantity. So, here's here's [4:04] quantity and here here's price. So, as [4:06] the price goes up, right? Then the [4:09] quantity supplied goes up. [4:11] >> Mhm. [4:11] >> Right? So, if I can make $100,000 by [4:14] doing something, wow, I'm totally in. [4:16] But, the demand curve is the opposite [4:17] direction. That's right. So, you know, [4:20] price and quantity go this direction. [4:22] So, the demand, the number of the units [4:25] that I I want, um actually I should say [4:27] this this this is quantity, this is [4:28] price here. So, you know, price goes [4:30] down, quantity demanded goes up [4:32] considerably. And if I think about [4:33] applying this to law, like what if I [4:36] have $1,000 of damages? [4:39] Right? Um, [4:40] I can call a lawyer [4:42] even for contingency, right? Um, [4:45] they might not want to take my case [4:47] because if they take a third of the [4:49] outcome, it's like they want a bigger [4:51] case and it's going to require a certain [4:53] amount of work. And AI really it it kind [4:56] of liberates these market failures to a [4:58] certain extent. Because, um, you [5:01] mentioned kind of giving David a bigger [5:03] slingshot. Like a lot of times like [5:04] David just Well, number one, if I have [5:06] $1,000 of damages, I certainly can't pay [5:08] $1,200 an hour for a lawyer. So, like [5:10] I'm out for that. Well, enter [5:12] contingency law. The problem is that [5:15] like the settlement night might not be [5:17] big enough. So, like the lawyer might [5:19] want to serve me, but like he or she has [5:21] paralegals that they have to pay and so [5:23] on and so forth. But, if the software [5:25] can do the work, now the market [5:27] opportunity is so much bigger. Right? [5:29] You can actually deliver justice for a [5:31] lot of people that otherwise they [5:33] weren't priced out of it because how do [5:35] you how are you priced out of [5:35] contingency? But, like it's not really [5:38] in the interest of a lawyer. They they [5:40] want to serve, you know, 1,000 customers [5:42] that they get, but they can't. They have [5:44] to focus on the biggest ones to the [5:46] detriment of the smaller ones because [5:47] again, it's kind of to the right of that [5:49] equilibrium point. Um, I don't I don't [5:51] know if you agree with that or how you [5:52] think about that. [5:53] >> That is exactly right. I think that is [5:55] the actual market problem we have today, [5:58] which is you know, if if you go back to [6:00] first principles, if if you're a a you [6:03] deserve a great lawyer [6:04] to represent your case. And usually only [6:07] need a lawyer when your back is against [6:09] the wall, right? The best of us, the [6:10] worst of us comes out this time, and we [6:13] don't have the skill sets to represent [6:14] ourselves. That's why this whole [6:16] profession exists. That's why the legal [6:18] system is about third of our our [6:20] government. It's for this reason. Yeah, [6:22] um [6:23] And today, what's happening is because [6:27] lawyers still need to get paid, it's [6:28] still bottlenecked on legal work. [6:31] Today, if a lawyer doesn't make at least [6:33] 5,000 fees off a case, they do not take [6:36] it on. Yeah. That is the market dynamics [6:38] today. And even at 5,000, they're losing [6:41] money on a lot of cases. They're really [6:43] banking on few cases paying out very [6:45] large. They're they're not even [6:46] guaranteed to win, period. [6:47] >> That's So, they already take that That's [6:49] the contingency aspect. [6:50] >> That's exactly right. [6:51] >> But, it's kind of like, even if I win, I [6:52] lose because my costs of carrying the [6:55] case are going to be higher than the [6:57] actual potential outcome, even if I win. [6:59] >> That's exactly right. I mean, I think [7:00] about like most people here have been in [7:03] an car accident, fender bender of some [7:05] sort. If it's below the deductible, no [7:08] lawyer in America will touch that case, [7:10] right? It's just not economically worth [7:11] it. Even if the deductible's 3,000, [7:13] 4,000 dollars, to an individual, that is [7:16] a difference between a paycheck being [7:18] gone versus not, right? And there's no [7:21] representation they can get for that, [7:23] right? And the the same thing. And And [7:25] the reason is actually from their side, [7:28] we have to understand from their side, [7:29] too, right? So, let let's say you've [7:31] been getting harassed at work, and your [7:33] employer could stop it. Like in the [7:34] hospitality industry, let's say, you [7:36] could, you know, tell your customers, [7:38] "Hey, don't harass my employees." And [7:39] change policies to make that happen. [7:41] Today, that doesn't happen because when [7:45] that person gets calls, the lawyer on [7:47] the other side has to pick up a call, [7:49] field that call that's about 30 to 45 [7:52] minutes of work and then after that, go [7:54] do legal research; because in employment [7:56] in America, pretty much everything's at [7:58] will, and you have to tie a problem to, [8:02] uh, [8:03] you know, like a protected protected [8:05] clause of some kind. And it's called [8:06] causes of action. So, you have to tie to [8:08] a cause of action [8:09] and that work requires them to go spend [8:12] 10 hours. Yeah. To go research your case [8:15] law, research the facts, go back and [8:17] forth and ask them more follow-up [8:19] questions to see if they qualify. [8:20] And while they spent all that time, they [8:22] might still realize, "Hey, this doesn't [8:25] even make me $5,000." [8:26] >> Right. [8:27] >> So, [8:28] that is why 96%of cases on the labor and [8:31] employment side, for example, gets [8:33] rejected. [8:35] >> Right. So, you can never find anybody to [8:36] represent you. [8:37] >> to come out of pocket and many cases [8:39] people don't have money to come out of [8:40] pocket. [8:41] >> That's That's exactly right. That's [8:42] exactly right. So, you just kind of take [8:44] it unless something really egregious [8:46] happens. [8:47] Um and that is the current state and [8:49] it's all bottlenecked on legal labor. [8:51] And this is what Eve is meant to do is [8:54] to How do you now take that 10 hours is [8:56] now immediate with Eve. A call comes in, [8:59] Eve actually answers the call and [9:02] automatically uses its insights on the [9:03] law firm to know whether or not there's [9:05] a case the lawyer has to take on or not [9:08] and then hit the ground running with a [9:10] demand or the initial preliminary work [9:12] done. [9:13] Right? And that completely changes the [9:15] economics of that type of case and we're [9:17] still at the very beginning stages. You [9:19] know, like what's What's happening is [9:21] Eve is getting more and more capable of [9:23] handling outcomes [9:25] end to end over a very long horizon. [9:27] These cases take 6 months, a year to [9:29] take to launch and through them probably [9:32] a thousand actions back and forths that [9:34] are happening. [9:35] And as accuracy of Eve is getting higher [9:37] and higher, this long horizon becomes [9:39] possible. So, the more this becomes [9:41] possible, even more complicated work [9:43] gets amortized down to constant case. [9:46] So, now all these people without [9:47] representation [9:48] get representation. Yeah. Um and if you [9:51] really ask yourself like [9:52] how many times in your life [9:54] if [9:55] asking for legal help wasn't a friction [9:57] point, would you have asked it? [9:59] That number is way higher than we think. [10:01] >> Yeah. [10:02] So, it's a funny story, this is true [10:03] story. I was in Blue Bottle, a little [10:06] coffee shop in Los Altos, and I was [10:07] explaining Eve to somebody probably 9 [10:09] months ago. [10:10] And I was explaining it because most [10:12] people think of AI is like, "Oh, you're [10:14] going to fire everybody and cut costs [10:16] and it's all about cost savings." And [10:17] it's like, "No, actually like if I use [10:19] Eve, I make a lot more money for my [10:21] firm." And it's not because I'm cutting [10:23] costs. It's like I can take on more [10:25] cases. It's actually revenue expansion. [10:27] So I'm explaining this [10:28] as as an outsider investor can so not as [10:31] well as you probably would. And at the [10:32] end of this, I mean we were talking [10:33] about many topics, but Eve came up. And [10:35] this guy comes up to me, looks a little [10:36] angry, has a big dog, and I'm kind of [10:38] afraid of big dogs. He's like, "What was [10:41] that company you were talking about?" I [10:42] was like, "Uh-oh, is he going to get [10:43] mad?" He's like, "I'm a trial attorney. [10:45] I want to try it." Oh, great. So I got [10:47] you a new customer there. Thank you. But [10:48] the whole point is that you expand the [10:51] revenue opportunity. But maybe kind of [10:53] walk through like what do the dynamics [10:55] because everybody is driven on 101 or [10:56] I-95. I-95 is the corridor that goes [10:59] from Florida all the way up to to Maine [11:00] or something, you know, up and down the [11:02] Eastern Seaboard. And it's like I've [11:04] driven almost all of that and it's like [11:06] I feel like half of the ads are for [11:08] trial attorneys, [11:09] plaintiff law. [11:10] So they're obviously advertising there [11:12] because you want they want to get [11:13] called. It's like, "Hey, I I I have this [11:15] accident or I have this issue." Um but [11:18] what do the economics look like? So like [11:20] I I spend all of This is why again I [11:21] think Eve is so fascinating because I [11:24] have to turn down 90%of the leads that [11:26] I'm getting, I assume. Right? Or like [11:28] what what do the numbers look like? I [11:29] spend money on a billboard, I get lots [11:31] of inbound. How many of those turn into [11:33] cases? How many of those cases turn into [11:35] revenue? Maybe kind of like a before and [11:36] after because you are you're not a cost [11:39] savings play for lawyers, you're a [11:41] revenue expansion play for them. Uh [11:43] that's that's from the legal side. But [11:45] what what what do the economics look [11:46] like for a law firm? That That's exactly [11:49] right. When most people drive by, they [11:50] see the face and they forget about the [11:54] people behind them and the amount of [11:55] work every single call actually takes. [11:58] That's That's what happens today. That's [11:59] what Eve is there to help with. And the [12:02] math on the law firm side is very [12:04] simple. The more people you represent, [12:06] typically speaking, the more money you [12:08] make. But because you can take on more [12:10] people and you [12:11] honestly are denying and not taking on [12:14] many cases because they're too expensive [12:15] to take on, [12:17] the demand from their perspective is not [12:19] really constrained. It's really the [12:21] supply of labor. Like, do I have enough [12:24] people uh inside to actually service [12:27] that problem? [12:28] And to make things more complicated, [12:30] different firms here are in different [12:32] realms of problems where the bottleneck [12:34] keeps shifting. Right? So, like a small [12:36] firm might just have one attorney that [12:39] is has a case manager or some sort of [12:42] uh office manager that feels the initial [12:44] call and just notes it down and then the [12:47] lawyer then [12:48] responds back in their free time. Right. [12:50] They're kind of running chaotically, [12:51] running around. And then there's a next [12:53] level, which is they find a way to [12:55] systematize it. You know, they have a an [12:57] intake team of call center like a call [12:59] center somewhere that intakes it. And [13:02] now that becomes a new bottleneck. So, [13:03] now when you increase your marketing [13:05] spend, let's say you looked particularly [13:07] attractive in that billboard and a lot [13:09] of people are calling in, [13:10] uh suddenly now that team is [13:12] overwhelmed. And you're limited again by [13:14] how many people you can represent. [13:16] >> Right. [13:17] >> And this is the daily challenge these [13:20] small businesses are dealing with. It's [13:21] they're small businesses, they're not [13:23] massive enterprises. Uh they want to [13:25] represent individuals and everyone here [13:28] went to law school, not business school. [13:30] They want to be there for the people. [13:32] Um so, we're basically consolidating all [13:35] of that into the Eve platform such that [13:38] they're getting pulled in only for what [13:40] they need to. [13:41] Right? Like their lawyer insights, their [13:43] particular insights on how do they take [13:46] this case, Eve keeps raising the floor, [13:48] how do I take it from this floor to the [13:50] next level? And that is [13:52] where I believe really all lawyers [13:54] should spend time, talking to their [13:55] people and helping get client outcomes. [13:57] >> Well, this is kind of the the market [13:59] failure point to a certain extent, which [14:00] is I could serve all 5,000 leads that [14:03] came in if only I could hire a paralegal [14:07] for 12 cents per hour, [14:09] right? Or whatever the token cost of it, [14:11] but it's like I can't. Mhm. So, I would [14:13] just there's no labor market that could [14:16] actually function. But with AI now I [14:19] can. [14:21] I mean, do you have anecdotes of just, [14:23] like, kind of before and after? It's [14:24] just like, you know, like the Hair Club [14:25] for Men it's like, you know, "Here I am [14:26] before, here after; I have a nice head [14:28] of hair." But, like what, what's that [14:30] for kind of your best customers that you [14:33] work with? I imagine the revenue [14:34] transformation is pretty significant. [14:36] >> Yeah, one one firm in LA went from [14:38] managing about a hundred clients per [14:41] attorney to 500. [14:43] >> Wow. [14:44] >> Right. And that is that is a type of [14:46] thing that is possible. And they went to [14:48] to their credit they worked really hard [14:50] to achieve this. They utilized Eve up to [14:53] the max [14:54] of all the way from intake take fielding [14:56] the call doing qualification [14:58] automatically removing all the relay [15:01] problem that they have of passing things [15:03] from one attorney to another right to a [15:05] case manager. All that is now handled [15:07] within agents in Eve. [15:09] And all the way to settlement where it's [15:11] actually [15:12] briefs before they go to a mediator [15:14] discussion that actually tells you [15:16] here's what you need to focus on. Here's [15:18] what the settlement outcome should look [15:19] like for cases like this that we worked [15:22] on. And here are the tactics used before [15:24] that got you here. Yeah. And that is [15:26] what the future looks like. It's truly [15:29] moving the case forward continuously and [15:32] you're helping making it do a better job [15:34] than it could do by itself. That is I [15:36] think the future of law. Right? Where [15:39] you as a lawyer are really going to look [15:41] more like a almost a data scientist and [15:45] engineer engineering something to be [15:47] better and better. And that is a shift [15:50] the top law firms on Eve mentally kind [15:52] of go through. Yeah. [15:54] >> Well, I kind of think of it if you if [15:57] you look at [15:59] uh your average law firm [16:02] where they don't really know AI and and [16:05] actually like this is kind of an [16:06] interesting kind of take on on the world [16:08] of of AI enabled applications in general [16:11] like what was the software stack for law [16:14] before? Well, if you're a big firm like [16:17] Wilson Sonsini probably the most [16:18] important thing is like your your time [16:20] tracker, right? Because you're billing [16:21] by the hour. If you're in contingency [16:24] law, it's like, well, your time matters, [16:26] but you're not billing your customers on [16:28] that. You probably need to have like a, [16:29] you know, each case, here's how much [16:31] money I'm spending on the case, but like [16:33] I imagine your primary tool is Microsoft [16:35] Word. And there wasn't This is how like [16:38] VCs would have like gotten this industry [16:39] wrong before, which is like, oh, well, [16:42] the software market for plaintiff law is [16:44] small. They all buy Microsoft Office. [16:47] Yeah, Microsoft Office is kind of [16:49] expensive, so therefore like the [16:50] software spend is like in the tens of [16:52] billions of dollars a year or something [16:54] you know, seemingly comically small. But [16:56] wait a minute, the outcomes that they [16:58] deliver is in the hundreds of billions [17:00] of dollars a year. Um, they can't [17:03] deliver more outcomes because they're [17:05] really constrained on labor. [17:07] Um, and they can't hire labor for 12 [17:10] cents an hour because it doesn't exist. [17:12] So, it turns out the software market is [17:14] kind of like this shadow labor market, [17:16] and it's like massively bigger than [17:18] anybody would have expected. Um, and and [17:21] like it's always just a weird thing [17:23] because it's like you can't say the [17:24] software market is small because you're [17:26] a you're a legal software company. [17:28] But people know Harvey or they know like [17:30] corporate laws or like you know, [17:32] e-discovery tools that have been around. [17:35] But I imagine like what was the main [17:37] software stack before Eve entered? What [17:39] was it just Microsoft Word or like what [17:41] what what was a in order for me to go [17:43] into business? I graduated from law [17:44] school. I want to be in this particular [17:47] you know, subsector of the field. What [17:49] was my software stack? [17:51] >> You know, this is the unfortunate thing [17:52] about plaintiff law and institutions is [17:56] historically [17:58] plaintiff law really didn't need more [17:59] than a cell phone, email address. [18:02] Honestly, even Microsoft Word, some [18:03] people on the solos would not buy it for [18:06] a cost reason or other reasons. No, it's [18:08] it's because you if you think about it, [18:10] you actually don't do that much [18:11] redlining. It's a trust boundary [18:13] situation. Right? So, you're you're one [18:16] side is kind of opposition to the other [18:18] side, so you're not really collaborating [18:20] in [18:21] >> In same way that you are when you're [18:22] doing contract review or or something [18:24] along those lines. So, that was kind of [18:26] it. And no software really solved their [18:30] core problem. Which is exactly what you [18:32] mentioned. Their labor constraint. That [18:34] is the core of how they do anything at [18:37] all. And because they have labor and [18:39] they're good at managing their labor, [18:41] they found a way to solve in every [18:42] problem to them look like solve with [18:43] labor. [18:44] And that's how they [18:45] they've grown up. [18:47] And as a result of that [18:49] what ended up happening was a lot of [18:50] resources on software went towards [18:53] defense institutions. Because that's [18:55] where the budgets were. For Harvey, you [18:58] know, really any company honestly, legal [19:00] software company today vast majority of [19:02] them are on the defense side. Because [19:04] that's where the budgets are. [19:05] >> Right. [19:06] >> Right. And that's what's easy to build [19:07] for, easy to get money for. And on the [19:10] on the plaintiff side [19:12] these are small business that are [19:13] watching their pocketbooks every month. [19:16] Right? If you're not driving outcomes [19:18] for them from day one, you're out. [19:21] You're not getting renewed. They're [19:23] they're going to curse you out, you [19:24] know, like any any anything they can do, [19:26] they're going to come after you. Right? [19:28] And that's what's made things so [19:29] difficult. But what we realized is, well [19:33] their core problem is we know what their [19:34] incentives are. They want to represent [19:36] more people. They want to win. [19:38] And we can do that now. [19:40] You know, there's Once the beauty of [19:42] software today is once you do a task [19:45] manually, even tasks that include [19:48] crunching new piece of data and [19:49] transforming it that's kind of what AI [19:51] is really good at. You can now encode it [19:53] as a process. [19:55] That is a fundamentally new shift. That [19:57] has never happened before. That's what [19:58] they were trying to do with people [20:00] operations. [20:01] But never worked because every single [20:03] case still [20:04] different enough that you can't do it. [20:06] It's still bottlenecked on hiring smart [20:08] enough people to be able to handle it. [20:10] And now one smart person can actually [20:13] handle [20:14] like we talked about 500 different cases [20:16] of very different facts, very different [20:19] situations. Um and how to you keep that [20:21] in your brain, Like it's hard, right? [20:23] But now you don't have to. [20:25] And I guess one of the things that a lot [20:27] of founders are thinking about right now [20:29] at the application layer is like uh-oh, [20:31] maybe I'm roadkill from the big guys, [20:34] right? Like the labs that produce the [20:36] intelligence, maybe they're going to go [20:39] tackle MySpace next and I'm I'm toast. [20:41] There are many ways of defending against [20:43] that, but defending kind of sells it [20:44] short because if you're able like if you [20:47] have a network effect or if you have [20:49] data that actually nobody else has, you [20:52] can deliver better outcomes. So, you [20:54] know, who cares if everybody uses Chat [20:56] GPT or or Claude behind the scenes or on [20:58] their own, but maybe talk about like I'm [21:00] sure you've thought about this a lot. [21:02] You know, why is Eve better than just [21:05] kind of like raw intelligence? And how [21:08] does it get better because it does [21:10] uh the more practitioners that use the [21:12] product? [21:12] >> Yeah, we're we're betting on we're [21:14] betting on these models getting better [21:17] and better. [21:18] Um and our value is exactly that. You [21:20] know, if you have 1,000 Albert Einsteins [21:23] to work with to do litigation or or you [21:26] know, the best attorney ever to to do [21:28] every single task, we would take it. [21:30] The problem is if you drop Albert [21:32] Einstein into a law firm today, he will [21:34] not be productive for quite some time. [21:37] Um and that is a general problem that [21:39] Eve is solving, right? And why won't [21:41] that person be productive? Well, it's [21:42] due to a number of things. I think one [21:44] is in plaintiff law, the vast majority [21:46] of data is actually non-scrapable. About [21:49] less than a percent of cases actually [21:52] settle and get turned into case [21:53] opinions. [21:54] >> Right. [21:54] >> Vast majority settle pre-demand or [21:57] pre-litigation. And some even before [21:59] that, right? Where it's actually just [22:01] negotiation of hey, this is how I was [22:03] wronged, please fix it. [22:04] >> And to make that concrete, it's like I'm [22:05] in a car accident, I'm not at fault. [22:08] Um you know, the adversary, if you will, [22:10] like the other side's insurance company [22:12] is like, "Oh, you weren't that hurt. [22:13] We'll give you a dollar." Right? They [22:15] always try to like deny, deny, deny. [22:17] Like rule number one of of an insurance [22:19] carrier is deny, rule number two is [22:20] deny. You know, it just goes on. Those [22:21] are all the rules of insurance carriers. [22:23] So, now I hire a lawyer um and to [22:27] actually get what I'm owed because I [22:28] actually am injured and my car is [22:29] actually totaled and they just don't [22:31] want to pay me, um, you get a [22:32] settlement. It doesn't go to case. It [22:34] doesn't go to actual trial, but you know [22:37] >> That's right. [22:38] >> right, that Geico actually did pay [22:40] $29,000 and that's very, very useful [22:43] information, I assume. [22:45] >> That's right, cuz insurance is the one [22:46] that's dealing with 100,000 different [22:48] law firms. [22:48] >> have the shared community idea. They [22:50] have They know all the cases, but [22:52] because it's shared across all the [22:53] different local firms, they don't know [22:55] exactly Geico's every single move. [22:58] >> That's exactly right. And they, you [22:59] know, they even have ominous-sounding [23:00] software like Colossus, which tracks, [23:03] you know, they they're they try to [23:05] basically say, "A broken thumb, this is [23:07] how much we think it's worth." [23:09] Which is a hard thing to do for in their [23:12] defense, like it's a hard thing to do, [23:13] but they are doing that, but also [23:15] saying, "This is how much we can get [23:16] away with giving them away for." And [23:19] irrespective of that individual's actual [23:22] need [23:23] of of their particular situations. [23:24] That's what the legal system is meant to [23:26] fix. That's That's what it's there for. [23:28] >> Right. Um [23:29] >> And what now is happening is, and this [23:31] is this is the interesting bit and the [23:33] challenge is if you go and Google, "How [23:35] do you write an amazing demand letter [23:38] for against State Farm for soft tissue [23:40] injury cases?" Like that today exists [23:42] nowhere. Even for a new [23:44] paralegal that is joining a plaintiff [23:46] firm, this is something they have to [23:47] learn. [23:48] And on top of that, it gets more [23:50] complicated because when you're in local [23:53] jurisdiction, the local en-environment, [23:55] like the court systems, uh, your [23:57] opposing counsel, they all behave a [23:59] little differently and that context [24:01] matters, too. [24:02] So what Eva's doing is it's turning this [24:05] his-historically incredibly difficult to [24:08] utilize multi-attribute, [24:10] multi-dimensional information that is [24:12] basically "What is the best thing I can [24:14] do at this particular moment to improve [24:17] the outcome of the case?" [24:18] >> Yeah. [24:18] >> And that same question is repeated [24:21] thousands of times during the case. And [24:24] a human brain is just not capable of [24:26] kind of keeping all of that in mind. Um [24:29] and but AI is right? And And that is the [24:32] type of data we help build for law firms [24:35] and they're seeing the immediate value [24:37] from it. And the first place where they [24:38] see value is, of course, intake. Who do [24:41] I even take on? Who am I good at [24:43] representing? And maybe start there and [24:45] keep getting better. [24:46] >> Right. [24:47] And I imagine um [24:49] because these aren't giant firms, [24:53] right? Like if I'm a lawyer in New [24:55] Jersey with a four-person, four-attorney [24:57] practice and uh over there there's [25:00] another lawyer in Arizona with a [25:02] seven-attorney practice, but their [25:04] counterparties tend to be State Farm and [25:06] Geico and and and whatnot. Uh at the [25:09] limit, it's actually fine to share [25:11] information. I guess how do you think [25:12] about data? Because I mean this is [25:15] always a this is always, you know, hard [25:16] for companies when you're getting off [25:18] the ground. It's like, "No, no, this is [25:19] my data." Um but there's also there's a [25:21] give to get, I assume. It's kind of [25:24] like, you know, I work with a a dental [25:26] AI company and it's like, "Well, the [25:27] dentist in Iowa can't serve patients in [25:29] Florida, but they want to learn from the [25:32] collective actions of each other." Um [25:34] but I guess how do you think about you [25:36] know, just data sec-sequestration and [25:39] and whatnot? Um does that ever become an [25:42] issue or how do you think about using [25:43] it? [25:44] >> We We've largely followed the rules of [25:46] engagement that already exist in this [25:47] industry, but we've amplified it with [25:49] software. That's kind of our stance, [25:51] right? So, the rules rules of engagement [25:53] is for their own cases [25:55] uh they want to be very private. Like [25:57] that is important by law, it's [25:58] attorney-client privilege. And we don't [26:00] train on that data ever. Yeah. Right. Uh [26:04] however, what attorneys also do is to [26:06] share this knowledge, they actually [26:08] travel, they attend CLE, and they share [26:10] tactics each other on how they [26:12] went up against State Farm's X and Y [26:14] defense [26:16] um because State Farm repeats that [26:18] defense across the board, right? And uh [26:21] not picking on State Farm, anyone is [26:22] fair game here, but that is that is what [26:25] they're already doing. So, we hire own [26:27] attorneys and track success rates of [26:28] cases for a platform and we want to help [26:31] if you're using Eve, you should have [26:33] more success. That's what we track. [26:35] Um and we have this product called [26:36] Auditor, which actually analyzes every [26:39] single case [26:41] through every single page overnight and [26:43] looks for [26:44] uh situations that a perfect attorney [26:48] would catch every time. [26:49] And [26:50] across the board, unfortunately, vast [26:53] majority of cases have almost every [26:54] single case has missed information. [26:57] Right? And now that's no longer a thing. [26:58] Yeah. [26:59] >> And maybe talk about the future of where [27:01] where things go because um on the one [27:04] hand, [27:05] I can imagine a world where it's like [27:08] one of these days Geico's going to get a [27:09] little bit smarter [27:11] um and they're probably going to like [27:13] you see they're deny, deny, deny plan. [27:15] That's how they got smart before, right? [27:17] Because they realized it is an [27:18] adversarial system like, you know, I [27:21] lost my house in a fire and I'm going to [27:23] claim I had a Picasso in there and then [27:25] they're going to think that I'm claiming [27:26] that I have a Picasso. They're just [27:27] going to deny the claim. Deny, deny. [27:29] Like so each party is kind of fighting [27:30] each other, but they're going to get [27:31] superpowers. I mean, you see this [27:33] playing out a little bit with Muse and [27:34] all the consumer agents where I want a [27:37] reservation at this restaurant that's [27:38] impossible to get and they have a human [27:41] answering the phone. I can DDoS them now [27:44] that I have to pay one penny per hour or [27:46] something and then eventually they're [27:48] going to realize that they're getting [27:49] DDoS'd and they're going to prevent [27:51] that. So, you know, I guess on the one [27:52] hand, like how do you think about, you [27:55] know, your your adversaries here? Not [27:57] not your adversaries, but like your [27:58] clients' adversaries. How are they [28:01] adapting to a world where it's like, you [28:03] know, it was great for Geico where they [28:06] could say, "Hey, you have $2,000 of [28:08] damages." The consumer is like, "Yeah, I [28:09] did have maybe $2,500 of damages." But [28:12] it's like they never really had to worry [28:13] about that before because of the [28:15] quantum. And now, unfortunately for [28:18] them, but fortunately for the Davids of [28:19] the world, like people that are actually [28:21] aggrieved and and have been victims, now [28:23] they can fight back and you're powering [28:25] that. So, that's kind of one thing, but [28:27] also you mentioned long horizon agents. [28:29] Um because, you know, what does the [28:31] practice of law look like when a lot of [28:34] the stuff can actually be done by AI? [28:36] You have you I think you you framed it [28:37] as like the uh the attorneys are almost [28:39] like the data scientists of the future, [28:42] but you know, maybe talk about like what [28:43] does the adversary or the adversarial [28:46] process look like? You know, what, what, [28:47] what are the the State Farms and Geicos [28:49] using, uh, potentially as a result of [28:52] you entering the scene? But also like [28:54] how, big big, question, but how does law [28:57] evolve and emerge just given that, you [29:00] know, you're kind of manipulating words [29:02] and finding facts and doing all this [29:04] kind of stuff where, like, the humans [29:05] are not as important? Um they obviously [29:08] have an important role of shepherding [29:09] the case to completion, but like what [29:11] does the AI do versus not do? [29:13] >> Yeah, I mean, ultimately my my framework [29:16] for thinking about this is ultimately [29:18] when you have an individual, [29:20] they're going to need representation. [29:21] Whether that representation, like who [29:23] does the work on the representation, you [29:25] can assume it's [29:26] 99%AI, maybe 99.9%AI. There's always [29:30] going to be that little bit [29:32] that is the human touch, right? And we [29:37] do believe that people are going to make [29:39] a positive difference to the outcomes of [29:41] the of the case. [29:43] And that is exactly what you have, [29:44] right? So, like what do you have on the [29:47] on the defense side or really [29:48] institution side is what's always [29:50] happened, which is they have more [29:51] resources, they have more money, [29:53] uh they have more time, which is [29:55] actually another weapon they have, which [29:56] is quite powerful. And individuals are [29:59] in in debt to pay their fees, [30:02] um and that is a very powerful tool they [30:04] have. [30:05] And [30:06] uh that is that is a strength they're [30:07] working with, and they're trying to rely [30:10] on you know, centralization and [30:11] standardization. On the defense side, [30:14] you have a attorney that is fighting [30:16] your case [30:17] and pushing them to get to litigation. [30:20] And I think that is going to be what [30:21] happens. I think [30:23] a lot more cases are going to have peep, [30:25] and all the routine back and forths, I [30:27] think it's going to happen through [30:28] agents talking to agents. I think [30:30] that's, you know, it's crazy future [30:32] world, but I think that's where we're [30:34] headed to. And the difference maker is: [30:36] what does the attorney understand about [30:39] that client, that set of facts, um, and [30:43] their local environment of who they're [30:45] going up against locally in that law, in [30:47] in that defense firm. [30:49] And they're using that insight to make a [30:51] difference for their client. [30:54] And I don't think that ever goes away. [30:55] Um it's it's in fact something we're [30:58] seeing already where many customers on [31:00] Eve are actually choosing to spend more [31:02] time on their trial work. [31:04] >> Yeah, because it actually enables that. [31:05] Previously, going to trial took down [31:08] your entire firm. [31:08] >> Right. Suddenly, you're just spending [31:10] all your time doing trial trial prep, [31:12] interviewing juries. You want things to [31:14] settle. [31:14] >> Exactly. It's more of a settlement [31:16] motion. [31:16] >> That's exactly right. And instead, I [31:18] think things are going more into real [31:20] advocacy. How do I actually make [31:22] our point known? How do I You know, my [31:24] client is against [31:25] back of the wall. [31:26] They're doing bad things, they're doing [31:28] good things. How do I represent them in [31:31] what actually matters for them? [31:33] >> I imagine this is kind of the data [31:35] science side of what you're doing as [31:36] well with this giant data set that's [31:38] probably never been accumulated before. [31:40] Because as you mentioned all these [31:41] things, it's so bespoke, there's no like [31:43] communal repository for like all of the [31:45] settlement amounts. But like I've seen [31:47] this with litigation finance a little [31:49] bit where [31:50] if I'm a lawyer uh if I'm a trial [31:52] attorney and I'm not getting paid by the [31:54] hour by you, my client, well, where do I [31:57] pay myself and my staff and my office [32:00] rent and everything else? Well, there [32:01] are companies that will show up and say, [32:03] "I will fund this case." [32:05] >> That's right. [32:06] >> Right? And I'm funding this case cuz I'm [32:08] counting on you winning it. And like [32:10] they'll do all sorts of clever things [32:11] like here's the judge and here's this. [32:13] Like they're using data. So, outside of [32:14] I I think those guys have been very data [32:16] science forward. But historically, like [32:18] this data just like never really lived [32:20] anywhere. And now that it does, you can [32:22] make smarter decisions as an attorney. [32:24] I'm like, do I take it to trial or do I [32:26] accept the settlement or do I push back? [32:28] Oh, Geico offered me $12,000. Should I [32:31] push back and fight for 15? Should I you [32:34] know, take it to trial and try to win [32:36] 100? Like how do I make those decisions? [32:38] >> That's exactly right. [32:39] >> That's which is why we can help them [32:41] make the best decision every single case [32:43] and know when an optimal outcome has [32:45] been achieved. Yeah. And it's still I [32:47] would say it's still the attorney's [32:48] responsibility to push it even further, [32:50] right? And I think that is what I meant [32:51] by data science-like work because [32:53] they're using the power of software to [32:56] do work you never even thought to do [32:58] to improve the outcome. [33:01] >> And I mean, where do you think This is a [33:02] very broad question, of course, but like [33:03] where does the legal system go? [33:06] >> I was having some some trouble with [33:07] Wells Fargo. [33:09] And I'm like, you know what? I've tried [33:10] like for 4 hours to do this, I'm just [33:12] going to sue them. So, I went to Codex, [33:14] you know, Chad GPT was like, "Sue Wells [33:16] Fargo." 4 hours later, it had sued Wells [33:19] Fargo. It actually found a process [33:20] server. [33:21] Um I got the notice of sir like we're [33:23] going to court, but you know, the date [33:25] is in February. Like another kind of [33:27] similar vein is happening with patent [33:28] law where before it's so much work to go [33:31] file a patent, now it's less work to [33:33] file a patent, but you still have the [33:35] USPTO, which is the bottleneck. So, I [33:37] guess I mean, if you were to [33:38] prognosticate, like where does the legal [33:41] industry go now that AI is becoming [33:43] table stakes and you're delivering your [33:45] your part of that table, obviously for [33:47] plaintiff law? But if you think about [33:50] the entire practice of law in general, [33:52] like how does it change if you were to [33:54] fast forward 5 years? What does it look [33:55] like? What's similar? What's different? [33:58] >> Most likely, I think it's inevitable [34:00] that you're going to have to create [34:02] systems where not everything goes to [34:04] court system. [34:06] And this already is a framework that is [34:09] used in law, right? Like I mean, for [34:11] some types of Veterans Affairs cases [34:13] that goes through a different committee. [34:15] For even labor and employment, that goes [34:17] through EEOC in a lot of cases before it [34:19] goes through. [34:20] Uh and you know, employers have figured [34:23] out different ways to mediate and have [34:26] arbitration that doesn't necessarily go [34:28] through the court system. [34:29] >> Right. [34:29] >> And this is this is the framework that [34:32] actually naturally evolves and I expect [34:34] that to happen more and more. And I [34:36] think the tweak that's going to happen [34:37] is: if for some pockets of it, we're [34:39] probably going to go through some [34:41] version of like ADR (automated dispute [34:43] resolution), [34:45] where [34:46] AI-enabled mediators or central parties [34:49] that have a neutral view of both sides [34:52] are going to use AI to really speed up [34:55] how they come up with judgment. And I [34:57] don't think the judgment will really be [34:59] replaced, but what they will get is a [35:02] set of facts that shows historically [35:05] here's how this happened, here's how [35:06] this happened, and it's up to them to [35:08] remove the biases that might exist. And [35:10] I think we're just going to spend more [35:11] and more time on that side of things. [35:15] >> And do you think the small law firms get [35:17] bigger? Because that would be the other [35:18] thing. It's like if I could only handle [35:20] 100 cases per attorney, now I can handle [35:22] 500, but in the future maybe 5,000. You [35:26] know, why do I limit my office to like a [35:28] very very small New Jersey location? You [35:31] know, may-maybe I should go more broad. [35:34] Um and you're going to have kind of [35:37] larger, almost like corporate-like law [35:39] entities. And like I I think the biggest [35:41] trial attorney firm is Morgan&Morgan. Is [35:43] that right? Um and they have billions of [35:45] dollars in revenue, but your average [35:47] firm is is quite small. But you can [35:50] imagine powered by Eve, and with all the [35:52] changes that we're talking about, it's [35:54] like the small firm could actually [35:55] become a billion-dollar revenue [35:57] business. Um it's kind of like venture [35:59] outcomes versus kind of like but what [36:01] you would call lifestyle business. You [36:03] were a VC once upon a time, you know [36:04] that term. Uh some somewhat pejorative [36:06] term for like you make profits. Um but [36:08] [laughter] how do you think like is is [36:11] that a reasonable take that like some of [36:13] these small guys can actually I mean [36:14] obviously you can't go public as a as a [36:16] law firm for now. Um maybe in Arizona or [36:18] something you can, but um [36:20] uh how do you think there's a chance for [36:22] some of these local mom-and-pop uh [36:24] stores, if you will, to become very very [36:26] large institutions? [36:27] >> I think the transformation that's going [36:29] to happen is a lot of these law firms [36:32] are going to look more like software [36:34] companies. [36:35] I think they're going to finally [36:36] participate in the software revolution [36:38] that's been happening for the last 30 [36:40] years they've been left out of, right? [36:42] Um and if you if you kind of deep dive, [36:44] you'll understand why. Let's say you [36:46] have a $25,000 settlement. The attorney [36:50] gets maybe a third of that. Uh, let's [36:51] call it 8,000 for simplicity. Uh, and [36:54] you know, like of that 8,000, about half [36:57] of it goes towards labor spend. 30%goes [37:00] to marketing, and the rest maybe is [37:02] profit. Right? Like that that's kind of [37:04] the business as it exists today. [37:06] Because of that business it is so [37:09] difficult to reinvest back into [37:11] servicing your next client better. [37:13] Right? Like the only way you can really [37:14] do that is by hiring a better attorney [37:16] today. [37:16] >> Right. [37:17] >> What changes is now they can actually [37:20] accrue a database and data set that is [37:22] unique to them that gets them results [37:25] that is better than their competitors [37:27] for a subset of law. [37:29] Right? And this is kind of [37:30] specialization. And historically this [37:33] industry already went through that. You [37:35] know, like 30 years ago there was a [37:38] uh, you know, your neighborhood lawyer [37:39] that handled everything, you know, from [37:42] >> Right. [37:42] >> from employment cases to IP to what have [37:45] you. [37:45] >> DUI, yes. [37:45] >> Exactly. Exactly. And and now what we [37:48] have are specializations of practice [37:49] areas. You have someone that's really [37:51] good at IP, someone that's good at [37:52] employment, and now it's getting even [37:54] more specific, right? Like someone [37:56] that's good at, let's say, nursing birth [37:57] defects or or something along those [37:59] lines. And I kind of anticipate that to [38:02] go even another level where now someone [38:05] could get really good at soft tissue [38:07] injury cases that involve X and Y fact [38:09] patterns. And it advertising for those [38:13] cases might be a lot more cost effective [38:15] than some other larger firm like Morgan [38:17] &Morgan [38:18] >> Right. [38:18] >> just showing their billboard everywhere. [38:20] >> So, get good at something locally and [38:22] then expand [38:23] >> That's right. [38:23] >> across the entire country because you [38:25] have such high confidence around that [38:26] one field. [38:27] >> That's exactly right. And we're so I [38:29] think we're going to see that type of [38:30] specialization. And I think we're also [38:32] going to see the opposite where some of [38:34] the large [38:35] we're also working with some on the [38:36] scale of Morgan&Morgan. Like some of [38:38] them actually are coming together. So, [38:40] those ones that got as [38:42] >> big [38:42] >> are actually going around buying other [38:43] law firms because they I with Eve they [38:46] can very quickly consolidate what used [38:48] to be very painful reintegration process [38:51] now can be done on exactly what the [38:54] client wants, which is hey, the client [38:55] wants a legal outcome. And Eve is made [38:58] just for that. [38:59] >> Yeah, so by doing that they get to [39:02] largely [39:03] work away from a lot of the change [39:05] management uh [39:07] work workforce changes, operation [39:09] changes cuz Eve handles a lot of that [39:11] for them. They can speed up their time [39:13] to making that acquisition worth it. The [39:16] funny thing is that there's this [39:17] movement underway to outcome-oriented [39:20] pricing. So like software SaaS used to [39:23] be per seat per month. Um and I used to [39:25] joke it was like the tall, grande, venti [39:27] model of software selling. It's like you [39:29] go to like any SaaS company's website. [39:31] This is why, you know, Claude can clone [39:32] them all so readily. They all look the [39:33] same. It's like $79 a month, $99 a [39:36] month, $129 a month, you know, tall, [39:37] grande, venti. Um and it was per seat. [39:40] >> Mhm. [39:40] >> And like look at a company like Zendesk, [39:42] which had $2 billion of ARR when they [39:44] were last a public company. They were [39:46] taken private by Permira and Hellman& [39:47] Friedman. And now they're like, "Uh-oh, [39:49] how many seats do you need? Like maybe [39:51] zero. We should We need to deliver [39:52] outcomes." So everybody in the software [39:54] world uh is kind of like in a lot of AI [39:57] covers like we deliver outcomes, we [39:58] charge for outcomes. The interesting [39:59] thing is that the form of law that you [40:02] focused on was always outcome-oriented [40:04] from the very beginning. Um but you have [40:06] other types of law that are not [40:08] outcome-oriented where I guarantee all [40:10] the clients would much, much, much [40:12] prefer that it was outcome-based. Right? [40:14] I mean, sometimes it's tricky like I [40:15] have a trust and estate attorney at [40:16] McDermott Will&Emery. She's very, very [40:18] good. She's like so expensive. Like I [40:21] guess my outcome The outcome is my [40:22] death, so I don't know if I want that. [40:24] But, you know, you would prefer to just [40:26] say, like, there was a fixed cost. But [40:28] there's often this data uh, you know, in [40:30] in their defense the reason to charge by [40:31] the hour is not so much to rip off the [40:33] client, but there's data asymmetry [40:36] between what I want in my head and [40:39] how I articulate it. Maybe the the the [40:41] scope of the project is changing over [40:43] time. But, you know, other forms of law, [40:46] I would argue they probably should look [40:48] more like plaintiff's law. They should [40:50] have outcome-oriented pricing as well, [40:51] >> Mhm. right? Or like immigration law, [40:53] that should be outcome-oriented. Like [40:55] everything should be outcome-oriented [40:56] pricing, but it's hard because there's a [40:58] data deficiency, if you will, in terms [41:00] of like how will I win this case on [41:03] behalf of the defendant? But I guess [41:05] like how do you think other forms of [41:06] like you could probably expand well [41:08] beyond plaintiff's law. [41:11] I don't know if you agree with me, but [41:12] like how do you think about just like [41:13] law in general changing? Because I kind [41:15] of feel like plaintiff's law was like [41:17] ahead of everything that we're seeing [41:19] right now. It's like everybody wants [41:20] outcomes. Nobody wants to pay for the [41:22] ingredients or the inputs. [41:24] This was the only kind of field probably [41:26] of business where it's like it was [41:28] always outcome-oriented, but kind of [41:30] shouldn't other parts of law work that [41:31] way? [41:32] >> That is our exact belief as well, right? [41:34] So like uh the the hard part has been it [41:37] is hard. You know, when when you're on [41:39] the defense side or even the plaintiff [41:40] side, the question around in your case [41:43] is it 10 hours of work? 100 hours of [41:45] work? [41:46] In many cases, you really don't know [41:48] until you take it on. [41:50] Um and for plaintiffs, that is why we [41:52] had to go so deep into each practice [41:54] area into achieving outcomes. And you [41:57] know, we were [41:58] doing what is now coined long horizon [42:00] agents like a last year. That's just [42:02] what was needed to drive results. And [42:04] now it's getting better and better to [42:06] actually drive legal outcomes. And [42:08] ultimately, I think [42:10] what is needed for democratizing of law [42:12] is from a set of facts, you need to know [42:15] how much compute resources you need to [42:18] arrive at a outcome. And that happens [42:20] through both aggregation of and better [42:22] usage of data and better usage of [42:25] actually running the compute to achieve [42:27] set outcome. And so we're doing both, [42:29] right? So like that is why we're [42:30] laser-focused on outcomes, and we want [42:33] to drive outcomes. And eventually, this [42:35] David versus Goliath uh thing is [42:37] happening everywhere. You know, it's not [42:39] just in in uh you know, big law or [42:43] in-house legal. Those are just where the [42:44] budgets are. So they actually can write [42:47] basically a blank check for an interim [42:49] period of time. And the [42:51] the forcing function around [42:54] how much is this actually worth in [42:55] outcomes to me [42:57] is not known, right? Like I mean, you [42:59] know this, like when you go fundraise [43:01] you might get hit with like a a 100k [43:03] check. What is the exact outcome of [43:05] that? Uh you know, like it's very hard [43:06] to know what the value of this, but that [43:08] is the only way they can price today. [43:10] >> Right. [43:11] >> Um and that's because they don't know [43:14] what the expected value of their work [43:15] actually is. [43:16] >> Well, that's the asymmetry as well. It's [43:17] like I don't know. It's like, "Okay, you [43:19] need to hire a lawyer for your series [43:21] C." [43:22] >> Man, what does that mean? Well, it could [43:24] be like very, very simple. It could be [43:26] very complicated. So, it's very hard to [43:28] have a flat rate. [43:29] >> That's right. [43:29] >> Um because they could get burned in they [43:31] can't get burned in the low direction, [43:33] right? It's like, "Oh, I charged you [43:33] $25,000. I spent an hour on this." But [43:36] they could get burned if it's like, "I [43:37] have to now spend 4,000 hours of [43:39] attorney time. I could lose money on [43:41] this." Which of course every single firm [43:42] they [43:43] Well, actually, like as you know, in [43:44] corporate law [43:46] these are lost leader products. And then [43:48] you try to make it up on [43:49] >> M&A or or IPO or something. [43:51] >> a little bit different than plaintiff [43:52] law. How do you guys charge? How did you [43:54] think Because I imagine you had a It's [43:55] not like on day one when you started the [43:57] business you knew exactly what you were [43:58] going to charge for like what type of [44:00] customer. I imagine your thinking has [44:01] evolved on this as well. [44:03] >> Absolutely. So, I mean, from actually [44:05] weirdly enough from day one [44:07] we started probably with the [44:10] with the you know, what was it? Tall, [44:12] grande [44:13] >> Tall, grande, venti. Yes. [44:14] >> venti model. That's that's what we [44:16] started with because we didn't know any [44:17] better, right? And then as we dug into [44:20] it we realized like what they care about [44:22] is outcomes. What they care about is how [44:25] do I serve more clients? So, that [44:27] actually forced us to be outcome based [44:29] from very early days. Uh you know, we're [44:31] talking like a year and a half ago, [44:32] which is pretty funny that's early in [44:35] today's world. [44:35] >> AI time. [44:36] Like dog [laughter] years times 100. [44:38] Yes. [44:38] >> Exactly. Um but that that is that is [44:41] kind of um [44:43] what happened to us, right? And there's [44:44] so much room to grow where now that you [44:47] know, we we talked about that that value [44:49] diagram of a third goes to attorneys, [44:52] half of that is labor that is half they [44:54] don't have to spend anymore. [44:56] Right? So, how do you actually turn that [44:58] into something more useful? And that [45:00] way, the best way to get more useful [45:01] work is actually to spend it weirdly on [45:04] more tokens and more more work that Eve [45:06] can do to make more money for your firm. [45:09] Right? And that is kind of the cycle [45:11] we're in today. And a lot of the market [45:13] dynamics that is helping us is exactly [45:15] this. Like, we actually believe [45:16] the uh billable model, I don't think [45:19] it's going to die completely, but I [45:20] think it's going to be on the way of the [45:23] seat-based model. Uh where for some [45:27] subset of law [45:28] that is probably the best way you can [45:30] charge for the work cuz it's very [45:32] difficult to know value until [45:34] an outcome many, many years later. [45:36] But for a lot of it, that's not true [45:38] anymore. [45:39] You know, for for a lot of it, you [45:40] actually know what the outcomes are. Um [45:42] and you should be able to take your work [45:45] down to an amortized amount and deliver [45:48] that back, right? Uh And that is what we [45:51] are betting on. Right? And this is why [45:53] we focused on delivering legal outcomes [45:55] from day one end to end. And doing that [45:59] in such a way that it is as simple as a [46:01] lawyer saying, "Here's this client I got [46:03] in. [46:04] Give it to Eve. Eve will pull you in [46:06] when you're needed and you're good to [46:07] go." [46:08] >> We're like, see, we we've talked about [46:10] how, [46:11] uh, plaintiff attorneys, like, they are [46:14] they are supply constrained, not demand [46:15] constrained or many of them are cuz they [46:17] have this infinite influx of of clients [46:19] that they can't serve because the math [46:21] doesn't really work in terms of, you [46:23] know, cost versus, you know, CAC versus [46:24] LTV. It's the the old fashioned thing [46:26] for any business. It applies to them as [46:28] well. You know, where where are your [46:29] constraints? Like, what what kind of [46:31] holds you back from growing faster? I [46:32] mean, you're obviously growing [46:33] incredibly fast. What are your rate [46:35] limiters? Um what what do you need as [46:38] inputs to grow faster as an output? [46:40] >> You know, it's it's really fascinating [46:42] because a lot of the legal benchmarks, [46:45] like even benchmarks that you share [46:47] publicly that Anthropic, OpenAI, many of [46:49] whom we're working very closely with are [46:51] working on, care a lot about like almost [46:53] single outcome. Right? Like, here's a [46:55] task. Did it kind of do mostly okay? And [46:58] the world is now moving into agentic [47:00] work. [47:01] For us, the only thing that ever [47:02] mattered was the agentic outcomes. [47:05] Right? So like now that is getting more [47:06] and more. [47:07] The end outcome here is today we're [47:10] still at the point with technology and [47:12] and processes where there's still a [47:15] decent amount of human touches needed [47:16] for a case. [47:18] But by doing more and more work we've [47:19] been drastically reducing that month [47:22] over month. And in our industry we need [47:24] to do that per case type. Right? So like [47:26] for every single case we're learning [47:28] that like a science. Like a how do you [47:30] actually resolve this case? And how do [47:32] you build a mini attorney brain that is [47:35] capable of crunching all these data [47:36] points to that resolution and driving it [47:40] across the board and working with all [47:41] the tools they have access to including [47:43] Claude and whatnot to drive that [47:45] outcome. [47:46] And that is a really hard problem. [47:48] Right? So we're largely limited by [47:50] engineers at this point. Like we have [47:53] a limitless number of things to do to [47:55] drive this forward. [47:57] And the other fun part is we have [47:58] demand. The moment we are able to [48:01] deliver on let's say [48:02] moving vehicle accident cases end to [48:04] end. [48:05] The amount of value that gets unlocked [48:07] for the world from that is in the [48:10] hundreds of billions of dollars. [48:11] And that's just one sub type. That's a [48:13] great question. Have you had a fully [48:15] autonomous end-to-end case? It's like [48:17] answer phone, onboard client, you know, [48:20] get out get get check from Geico. You [48:22] know, like has that happened end-to-end [48:24] Not yet, but we're close. Uh, that is [48:27] what that is why we need the engineers. [48:28] Right? So like to put things in [48:29] perspective, a single case has probably [48:33] 10,000 back and forths. [48:35] >> 10,000? [48:35] >> Something like that. Because [48:36] >> Wow. [48:37] >> Here's how you think about it. Right? A [48:38] client comes in, let's say it's a PI [48:41] case. Now you have medical records. So [48:43] now you have to actually outbound call [48:45] insurance companies and health care [48:47] companies to get hold of the records [48:50] that the the clients already had. [48:53] Right? So like that is actually that [48:55] takes 2 weeks [48:56] today. And with Eve outbound agents [48:58] where we actually have voice AI that can [49:00] do the outbound call. And funnily [49:02] enough, that's an area where we're [49:03] already seeing agents talking to other [49:05] agents and learning how to get past [49:06] Right. each other, which is hilarious. [49:08] Um And that is like one step that's [49:11] needed, right? And then from there, you [49:12] have to then create a document and [49:14] actually have a conversation with the [49:15] other side. So, all of this are task we [49:19] are com-completing end-to-end. So, today [49:21] there's a little bit of touch required [49:22] where someone makes an outcome better or [49:26] um does something [49:28] But exactly what we're building right [49:29] now is how do you actually drive this [49:31] whole thing end-to-end? [49:32] >> Yeah. [49:32] >> And we [49:34] we're seeing early signs of that where [49:36] it's actually live in some customers. [49:37] >> And if you've seen going back to the [49:39] defense side and their colossus of of [49:41] information, have you seen their [49:44] reaction [49:46] uh just in general change? So, as an [49:48] example of that, it's like, uh-oh, these [49:50] guys contacted me with this. I always [49:53] need to settle because they probably [49:54] have like pretty firm rules about this. [49:56] They don't just like randomly make it up [49:57] as you go, I would assume, because [49:59] they're dealing with this tens of [50:00] thousands of times, if not more, per [50:01] year. So, if I'm GEICO, it's like, oh, [50:04] you demanded this much, well, my first [50:06] attempt is going to settle here. But [50:08] like, if their costs go down, maybe that [50:11] changes, right? So, have have you seen [50:13] any change in reaction of your, you [50:15] know, not your counterparty per se, but [50:17] of your client's counterparties? [50:18] >> I would say largely still uh the defense [50:21] side very much still works on an hourly [50:24] business model. So, they haven't quite [50:27] caught up to the same pace as our [50:29] attorneys. So, like, there's a lot of I [50:31] guess pricing asymmetry here. What [50:33] defense used to do for this one one law [50:35] firm that we're working with is they [50:37] would wait they would push out discovery [50:39] request dates until a day before [50:42] Thanksgiving or a major holiday [50:44] >> Right. [50:45] >> and then send discovery there because [50:46] they would know the person would go on [50:48] vacation and be stuck with all this [50:50] workload over that vacation and they [50:52] wouldn't do a good job at it or would [50:54] rush it. [50:55] And now, what they used they because of [50:58] such a repeated tactic by them, they [50:59] created an agent on EVE to respond the [51:01] same day. [51:02] So that next time this came through like [51:05] they responded immediately with the [51:06] responses and the defense were like, [51:08] what is going on here? [51:09] >> Yeah. Yeah. [laughter] [51:10] The tactics don't work. [51:11] >> And and like we're seeing there's so [51:13] many variants of that that is happening [51:14] now which is causing [51:17] defense to realize, wait, is this this [51:19] is what happens when the other side has [51:21] resources? Right. Oh crap. And and [51:23] that's that's what's fun to see. [51:25] >> Well Jay, this is a fantastic [51:26] conversation. I learned a lot. Didn't [51:28] realize there were 10,000 touches every [51:30] time you went through a a case, but [51:32] really fascinating and delighted to be [51:34] your partner and bring Eve to the next [51:38] the next 10,000 firms. [51:39] >> Thank you. I mean we're we still have so [51:41] much more to go. We probably helped a [51:44] couple hundred thousand clients today [51:46] and we have hundreds of millions of [51:48] touch and I would love to be making a [51:50] difference in their life every day. So [51:51] if that's of interest, please come join [51:53] Eve. [51:54] >> Awesome. Thank you.