Lowenstein Sandler Chair Says AI Slashes Legal Costs, Spurring Work Volume

Lowenstein Sandler主席称AI大幅削减法律成本，刺激业务量增长 成本更低的AI辅助法律研究拉低了诉讼与专利申请成本，进而引发杰文斯悖论，推高对律师的整体需求。 Lowenstein Sandler 主席 Gary Wingens 探讨了人工智能如何推动法律行业从基础文档自动化走向高阶“思考伙伴”。尽管 AI 大幅缩减了尽职调查和专利起草等劳动密集型任务所需的时间，使成本降低最高达 70%，但客户仍需为专家的把关与审核认证付费。Wingens 认为，成本下降将触发杰文斯悖论，释放以往因费用过高而望而却步的诉讼及复杂交易需求。尽管外界对初级律师培训和小时费率上涨存有顾虑，研讨嘉宾最终认为，AI 最终将扩大法律业务需求并增加行政管理开销。 生成式 AI 正超越过去的文档自动化工具，成为律师的核心思考伙伴与协作工具；大型企业客户不仅使用内部 AI 初拟法律文书，还实现了客户与律所双方 AI 智能体之间的直接对接与审查交互。 (10:00) AI 能够将复杂尽职调查等项目的执行成本降低多达 70%，使律所能将此前因费用过高而被搁置的项目变现，通过降低诉讼和专利申请的门槛催生出更多法律工作，印证了“杰文斯悖论”。 (22:30) 律所并未因自动化缩减初级律师招聘，而是通过 AI 模拟谈判和全所交易数据库进行实训，但法律界也警惕过度依赖 AI 带来的认知退化风险，尤其是模型在基础加法计算和格式排版上的天然硬伤。 (25:30) 执业过失保险承保机构与企业客户的态度在两年内发生 180 度逆转，从视 AI 为潜在风险转变为要求律所必须使用 AI 确保履职尽责；得益于单工时生产力的跃升，2025 年大型律所的平均小时费率逆势大涨 10.1%。 (45:00)

Lowenstein Sandler chair Gary Wingens argues that artificial intelligence will not shrink the legal sector, but will instead trigger Jevons paradox—where steep drops in legal costs expand aggregate demand while redirecting human labor toward high-level quality control. Jevons Paradox and Expanding Demand for Legal Services (33:30) Rather than reducing overall legal work, efficiency gains are unlocking previously unviable projects. Wingens highlights a complex trust-agreement review that a client initially rejected at a traditional $10 million estimate; once AI reduced the cost to $3 million, the client approved the engagement, creating revenue that previously did not exist. Similarly, automated search and review tools that lower electronic discovery budgets from $2 million to $200,000 make corporate disputes and civil rights lawsuits far more economical to file. This expansion extends to transactional and intellectual property practices. When corporate research and development teams began deploying internal AI tools, their invention disclosures increased by 400%, subsequently multiplying the volume of patent applications sent to outside counsel. While institutional clients are bringing more routine documentation like ISDA derivatives in-house, podcast hosts Joe Weisenthal and Tracy Alloway observe that lower generation costs threaten to spur automated claims, counter-claims, and patent challenges that will keep human lawyers in high demand. Preserving Margins Through High-Level Quality Control (22:30) Law firms are managing to protect profitability even while passing substantial cost reductions to clients. In massive due diligence matters, AI replaces the laborious first layer of human document review by extracting contract terms directly into multi-column spreadsheets. However, clients still lack the internal capacity and legal expertise to interpret and validate the results, leaving them willing to pay for law firms to provide specialized quality control and legal certification. This dynamic is transforming billing economics rather than destroying them. Although total billable hours on certain standardized tasks have fallen to roughly 30% of their historical levels, the remaining hours represent higher-level analytical counsel. Citing data from The American Lawyer, average hourly rates at the largest U.S. law firms climbed 10.1% in 2025 despite CPI inflation hovering around 3%, reflecting the fact that each individual attorney billable hour has become substantially more productive when augmented by AI. Enterprise Tech Architecture and Shifting Risk Norms (40:00) Rather than building frontier foundation models from scratch—which Wingens estimates would cost upwards of $1.5 billion—Lowenstein Sandler relies on commercial legal platforms like Harvey and Microsoft Copilot. These tools function as fine-tuned Retrieval-Augmented Generation (RAG) layers over frontier models such as Claude, ChatGPT, and Gemini, combining broad model intelligence with proprietary firm playbooks, pleadings, and deposition transcripts. Crucially, enterprise platforms maintain private offline environments to ensure sensitive corporate information does not enter public training sets and destroy attorney-client privilege. The regulatory and insurance environment has rapidly inverted to support this infrastructure. Wingens reports that two years prior, underwriters at Lloyd's of London and corporate clients questioned whether AI adoption created malpractice liability or prohibited its use on their matters. Today, both insurers and clients actively mandate AI utilization to enforce baseline standards of legal care and reduce unnecessary billing overhead. Junior Lawyer Training and the Risk of Passive Trust (25:30) Law firms are adjusting their recruitment and professional development strategies rather than shrinking early-career ranks. Lowenstein Sandler has maintained its historical hiring pipeline, securing summer associate classes through 2027. Because AI eliminates entry-level grunt work like manual document blacklining, the firm trains associates through deal simulators and centralized dashboards that index historical firm transactions, allowing trainees to practice drafting seller-favorable provisions in real time. However, Wingens warns of significant cognitive and developmental trade-offs. Tedious manual tasks historically provided mental downtime and built the meticulous attention to detail required to catch costly clerical or contractual errors. Because current AI systems remain prone to basic math and formatting mistakes, senior attorneys caution that junior lawyers risk falling into a trap of passive trust—accepting machine outputs uncritically instead of applying rigorous independent scrutiny. A major unresolved question is whether law firms face future margin pressure as AI vendors shift from flat enterprise subscriptions to usage-based token pricing, especially if venture capital subsidies fade. Additionally, hosts Weisenthal and Alloway caution that the technology may ultimately fuel corporate administrative creep, functioning as a Jevons paradox that expands internal bureaucracy and workplace busywork rather than delivering genuine leisure.

彭博《Odd Lots》节目指出，人工智能并未如预期般削减法律行业的工作机会，反而通过降低服务成本与法律门槛引发了“杰文斯悖论”，显著推高了总体业务需求。Lowenstein Sandler律师事务所主席Gary Wingens表示，AI正作为高阶思维伙伴重塑法律工作流程，促使律所将低层繁杂任务自动化，并将服务重心转向具有更高定价能力的人工质量控制与战略咨询。 技术演进与计费模式的韧性 (0:00) 法律行业历来经历过多次技术颠覆，从早期的人工抄写员过渡到打字员，每一次变革都曾伴随对岗位存续与计时计费模式崩溃的担忧。尽管长期以来行业尝试推广按项目计费、封顶收费等替代性计费安排（AFA），但按小时甚至十分钟为单位的传统计时模式依然占据主导。Gary Wingens解释称，这种现象源于律所与客户之间根深蒂固的不信任：当律所提供固定报价时，客户往往担心自己支付过多，宁愿查看详尽的小时工时明细；而对于IPO等高度标准化的业务，法规强制公开披露法律费用，使得市场价格具有透明度，形成了明确的价格基准。 重构尽职调查与工作流程 (10:00) 与以往仅能处理文本比对的机械性软件不同，现代大语言模型扮演了人类律师“思维伙伴”与联合驾驶员的角色，能够协助合伙人推演跨境税务架构，或在专利申请中跨学科引入生物学概念以拓宽保护范围。在大型尽调项目中，AI工具实现了初级工作流的全面重构。例如一项涉及数千份信托协议审查的复杂尽调项目，此前完全依靠初级律师人工逐级审核，原估算报价高达约1000万美元而遭客户放弃；随后律所利用AI完全替代第一层人工审阅，将关键字段自动提取至百列表格中，最终使项目总成本大幅削减70%至300万美元。律所律师仅负责第二层质量控制（QC）与专业认证，从而确保了可观的利润与业务落地。 初级律师培养与认知风险 (25:30) 面对技术淘汰初级职位的担忧，Wingens指出Lowenstein Sandler依然维持正常的招聘规模，甚至已锁定2027年的暑期实习生名额，但对新人的技能要求发生了转移。律所通过引入实时提示词培训等模拟环境，训练年轻律师直接利用AI审查并购协议。然而，这种跳过底层枯燥工作的培训模式也带来了认知风险。正如学习寿司制作需要长时间洗米或音乐家反复练习音阶以建立肌肉记忆，初级律师过早依赖机器可能丧失对标点、基本数学错误及细微格式纰漏的敏锐度。此外，AI缺乏真实商业谈判中识别对手策略的情感敏锐度，初级人员容易陷入对机器输出过度信任的心理陷阱。 杰文斯悖论与法律需求爆发 (33:30) AI对边际成本的压缩正引发典型的杰文斯悖论。电子证据开示（E-Discovery）成本从过去的200万美元降至约20万美元，大幅降低了诉讼发起门槛，使得民权诉讼与商业纠纷的立案意愿同步激增；在知识产权领域，企业研发团队由于在内部使用AI，技术披露与专利申请量增长了400%，直接推高了整体外部法务需求。与此同时，大型机构将常规衍生品协议审查收归内部，而外部顶尖律所依靠处理复杂市场交易的能力维持了定价权。根据《The American Lawyer》2025年数据，美国头部律所平均小时费率飙升了10.1%，远超同期约3%的CPI通胀率，反映出经AI赋能后单个计费工时所承载的产出与商业价值大幅提升。 技术栈架构与监管环境转向 (40:00) 律所当前的技术栈主要依赖Harvey等深度集成的企业级平台，结合微软Copilot以及检索增强生成（RAG）架构，严禁律师使用消费级工具以防破坏律师-客户特权。行业监管与保险风向也在短期内彻底逆转：劳合社等过失责任保险承保人从两年前担忧使用AI带来执业风险，转变为如今将使用AI视作律所履行尽职审查义务的必要标配。然而，律所面临的核心瓶颈在于合伙人内部的“各凭本事赚钱”文化，企业级AI的推行要求合伙人将各自私有的交易模板与经验知识共享给中央系统，以便将明星律师的专业经验编码固化为全所通用的办案指引。 未来法律服务的真实成本仍面临变数，尤其当下AI使用成本高度依赖科技巨头与风投资本补贴，一旦供应商全面转向按Token消耗计费并转嫁能源网与算力支出，律所运营成本或将重估。此外，生成门槛的降低可能催生大量琐碎诉讼与专利争议，导致人类法律从业者陷入应对更多低质行政对抗与自动化官僚内耗的新困境。

## Background 本期节目由彭博（Bloomberg）旗下的播客频道《Odd Lots》于 2026 年 7 月 13 日在 RSS 及网络平台发布。节目简介指出，在检索数据库与审阅合同等案头工作被 AI 自动化的背景下，法律行业看似将面临严峻冲击，但 Lowenstein Sandler 律所主席兼合伙人 Gary Wingens 在访谈中指出，技术降本反而可能引发法律层面的“杰文斯悖论”，带来更多的诉讼、交易与业务需求，同时节目还深入探讨了律所的按时计费模式及初级人才培养等议题。 ## TL;DR 节目颠覆了“AI 将使律师失业”的传统认知，揭示出 AI 作为资深律师的“思维伙伴”，正通过降低尽调与诉讼成本（如尽职调查成本暴跌 70%）触发“杰文斯悖论”，激活原本因费用高昂而被放弃的海量边际案件与专利业务。律所并未因工作耗时缩短而走向衰亡，反而因单兵单位时间输出的商业价值跃升，促成了小时费率的大幅上涨与最终解决方案总价的相对下调。在落地层面，中型律所坚决放弃自研底层大模型的资本陷阱，转向采购具备严密保密特权隔离和法律 RAG 的专业应用（如 Harvey），同时也不得不直面初级律师失去底层苦活历练的“练琴困境”与明星合伙人私有知识共享的机制阻力。此外，节目还警示了底层 Token 资本补贴退潮后的账单休克风险，以及 AI 普及可能反向诱发行政官僚主义与职场内卷的潜在威胁。 ## Atmosphere & style 对谈保持了彭博财经节目一贯的犀利、敏锐与幽默感，在机锋频出的问答中不断拆解法律商业逻辑的底层悖论。两位主持人 Tracy 与 Joe 屡屡直击痛点，追问 AI 启发是否属于“轻度消遣的智力幻觉”以及资本退潮后的算力代价；嘉宾 Gary Wingens 则以老牌大所掌舵人的坦诚与清醒接招，用极具画面感的行业掌故（从手拿红黑笔比对文本，到寿司学徒淘米和琴童练音阶的比喻）解构技术变局。整场对话既充满对前沿技术浪潮的宏观推演，又弥漫着对职业传承退化与官僚内卷膨胀的现实焦虑，兼具知性博弈的张力与极度务实的行业纵深感。 ## Bulletin - `23:35` Lowenstein Sandler 律所披露，一项原本因报价过高而搁置的信托协议尽职调查项目，在借助 AI 自动化初审后综合成本骤降 70%，反向激活了客户签单并为律所贡献 300 万美元收入。 - `33:15` 法律服务领域显现“杰文斯悖论”，单次服务单价大幅缩减并未砸掉律所饭碗，反而激发了研发端 4 倍的专利申请增量以及海量低标的诉讼与审查需求。 - `51:40` 2025 年全美大型律所平均小时费率同比暴涨 10.1%（同期 CPI 仅涨约 3%），AI 赋能大幅推高了单兵单位工时的产出含金量，呈现小时单价走高而客户实际采购总价走低的双向演变。 - `55:20` 律所正面临由 VC 算力补贴退潮带来的“账单休克”风险，企业级法律 AI 正加速从包月制转向按 Token 计费，未来真实使用成本或飙升至预期的 90% 甚至 130%。 ## Key takeaways - `10:00` 法律界对办公技术的演进并不陌生，但生成式 AI 与传统打字机或比对软件的本质区别在于其扮演着全员律师的“思维伙伴”（Co-pilot），直接介入复杂交易与策略推演。 - `12:15` 初创企业单靠 ChatGPT 起草的投资协议往往致命缺失市场博弈认知与后续资本运作铺垫，商业交互正迅速走向“客户 AI 智能体与律所 AI 智能体直接对话”的深水区。 - `15:00` AI 无法无中生有地发明全新金融或法律架构，其核心价值在于极高维度的策略碰撞与方案淘汰，与人类律师协同形成高效的架构试错飞轮。 - `23:35` 在尽调等流程中成本缩减 70% 不等于律所价值被掏空，剩余 30% 工时升维为专业度更密集的高阶质控，客户买单的核心是律所的权威背书与法律出证。 - `25:20` AI 消灭的是逐字核对等枯燥杂务，律所不仅未缩减初级律师招聘，甚至已锁定数年后的实习编制，但年轻律师跳过底层“苦活磨炼”后如何培养商业直觉成为全行业新难题。 - `27:45` 年轻律师的实操培训正转向“模拟器实战”，通过在指尖瞬间调取全所沉淀数十年的历史交易与诉讼超级知识库，打破了过去穿行走廊找合伙人请教的壁垒。 - `33:15` 法律服务验证了杰文斯悖论：电子证据开示与文书审查成本大幅下降，使曾经在经济账上不成立的微型维权、存量索赔及海量专利申请迅速具备可行性并出现井喷。 - `35:50` 大企业法务内部正借助 AI 加速吸收常规衍生品协议等程式化业务，但外部顶尖律所凭借跨行业案例的宏观市场认知与独立客观判断构筑起高阶护城河。 - `40:00` 律所依赖 Harvey 等垂类应用层而非直接调用通用底层模型，关键在于必须设立极度严苛的数据隔离层以誓死捍卫“律师-委托人保密特权”，并嵌入针对判例卷宗微调的专用 RAG 检索。 - `47:00` 执业过失保险承保人对 AI 的态度在两年内彻底反转：从严厉防范律所使用 AI 违规，转变为将使用 AI 进行双重核验视作防范执业过失的必须标配。 - `50:00` 顶尖合伙人的“独家秘籍”与集中化 AI 知识库共享之间存在深刻的组织文化冲突，天价挖角明星律师的逻辑正转变为将其专属交易智慧与品牌背书算法化。 - `53:30` 中型律所技术路线极其务实，绝不在内部斥巨资从零训练基础大模型，而是将专有交易数据以 Playbook 或 RAG 形式接入成熟前沿模型。 - `01:00:00` AI 带来的更深层悖论是职场与行政开销的膨胀：当所有人都用 AI 将 8 小时工作压缩至 2 小时后，为争夺晋升大家被迫创造出更多繁文缛节与琐碎内耗。 ## Detailed walkthrough ### `00:36` 法律行业的技术迭代与焦虑 Joe Weisenthal 与 Tracy Alloway 讨论了人工智能对法律行业的潜在颠覆。对于普通人而言，法律似乎是最容易被 AI 颠覆的典型领域——检索数据库、合同审查、日常案卷整理等机械重复的研究工作看起来极易被自动化。 Tracy 指出，尽管 AI 势头强劲，但法律界经历技术变革早已不是新鲜事。从手工誊抄的文牍员（scriveners）到打字员与键盘录入，每一次办公技术的跃升最初都被视作对法律行业的巨大冲击，但行业最终都平稳度过了调整期。此外，法律行业内部长期存在标准表单和模版合同，规范化操作早有渊源。 围绕计费模式，两人展开了探讨：既然律师普遍按小时（甚至按十分钟为最小单位）向客户收费，他们又有什么动力去引入提高效率、缩减工时的工具？Joe 认为，即便新技术不必然导致总用工量骤减，其对工作技能范式的重构也是不可逆的；过去精通判例汇编检索或熟练操作 LexisNexis 是核心技能，如今这一门槛正被技术推平。 ### `03:52` 官僚系统中的“杰文斯悖论”与行业现状 针对 AI 带来的效率提升，Tracy 提出了“行政官僚体系中的杰文斯悖论（Jevons paradox）”猜想：当处理成本下降时，整体需求与纠纷规模反而可能暴增。例如，保险公司利用 AI 自动化驳回理赔申请，而申请人随即调用 AI 生成申诉材料反击，最终导致大量“AI 智能体互相博弈并生成海量诉求”，迫使人类在终审环节承担更多调解和整理工作。技术并未让坐拥电脑的专业人士感到轻松，反而普遍加剧了工时压力与职业倦怠。 随着顶级律所（如 Kirkland & Ellis）开始大规模构建内部专属信息技术与 AI 基础设施，在后 ChatGPT 时代步入第四个年头的当下，AI 究竟如何实际改变法律实务，成为行业亟待厘清的核心议题。 ### `05:07` 对话 Lowenstein Sandler：业务版图与按时计费之谜 Lowenstein Sandler 律师事务所主席兼合伙人 Gary Wingens 参与讨论。Wingens 介绍了该所的基本体量与核心定位：拥有约 400 名律师，总部主要位于纽约地区。在法律界，400 名律师的规模属于“大所”（Big Law），但有别于 Kirkland & Ellis 这类超大型律所。该所业务主要深耕于私募资本市场，核心客户群涵盖私募股权基金（PE）、风险投资机构（VC）、对冲基金，以及科技和生命科学领域的成长型企业。 关于长期饱受诟病的“按时计费制（billable hours）”，Wingens 坦言其荒谬性：无论是律师还是客户都讨厌这套机制，客户购买的是“商业解决方案”，没有任何客户进门是专门来采购“计费工时”的。过去 15 年来，行业内一直在倡导替代性计费模式（Alternative Fee Arrangements，简称 AFA），如按项目固定打包计费、设定费用上限（cap fees）或区间费用机制（collars），但这类模式始终未成为绝对主流。 ```mermaid flowchart TD accTitle: Legal Billing Disconnect accDescr: Diagram illustrating the friction between clients wanting business solutions and law firms billing by the hour. subgraph ClientNeed["客户诉求"] A["商业解决方案"] --> B["追求确定性与价值匹配"] end subgraph MarketFriction["信任摩擦与惯性"] B --> C{"报价博弈与信任缺失"} C -->|"打包价偏高则转回按时计费"| D["传统工时计量模式"] C -->|"信息公开/标准化市场"| E["隐性价格共识"] end subgraph Outcomes["计费实践"] D --> F["按小时/单位时间收费 (主流)"] E --> G["替代性计费方案 (AFA/封顶价)"] end ``` Wingens 认为，法律之所以成为少数迟迟未能全面转向价值定价的专业服务行业，核心原因在于“律师与客户之间的信任壁垒”。当律所提出一个固定总价时，客户往往会怀疑价格过高，宁愿退回按小时计费，通过对明细账单（timesheet）的审查来确认律师到底投入了哪些实质性劳动。 - **市场公允区间的隐性共识**：尽管替代性协议未必以书面形式锁定，但市场上存在明确的价格参照。例如 IPO 项目，因监管要求公开披露支付给律所的法律服务总费用，收费具有极高的市场透明度和横向可比性。 - **复杂交易的浮动范围**：在风险投资融资轮次中，不同轮次、不同条款及奇异税务结构的复杂性会导致服务费有所波动，但在业内对常规项目的价格区间依然存在明确的默契与心理底线。 ### `08:44` 法律实务的技术跨越：从文字处理到智能化 对于 AI 究竟是一次渐进式的工具升级还是底层范式的断代飞跃，Wingens 明确表示这是一场“相当巨大的变革（pretty huge change）”。他将这种变化分为两大维度：其一是业务流程与系统层面的效率提升，其意义绝不亚于早年微软 Word 的引入。回顾早年作为律所一年级初级律师的经历，文书修订、格式校对和对比文本等机械劳动消耗了惊人的时间，而软件和自动化工具彻底重构了该维度的生产效率。 ### `10:00` 从手工比对到“思维伙伴”：技术如何重塑律师生产力 谈及法律行业的自动化演进，加里·温根斯（Gary Wingens）回忆起职业生涯早期的经历：当年他曾趴在桌上一页一页地手工比对合同文本的修改痕迹（blacklining），而客户则按小时为这些机械劳动支付高昂费用。然而仅仅两年后，文本比对软件横空出世，原本耗费数小时的人工核对被缩短至大约 90 秒，律所也再无法就此项工作按小时计费。这种纯粹由软件驱动的“效率提升”，开启了法律服务成本逐步下降的先河。主持人半开玩笑地插话询问，他是否曾像业界著名的诉讼案例那样“因漏掉一个逗号而导致客户损失 500 万美元”，温根斯笑称自己虽难免漏掉过标点符号，但万幸未曾引发过任何灾难性后果。 温根斯强调，生成式 AI 与以往所有办公软件有着本质区别。早期的技术跃升仅仅聚焦于“降本增效”，让客户能以更低的总成本完成交易或推进诉讼；但今天的 AI 正在律所内部充当全员律师的“思维伙伴”（thought partner）与“副驾驶”（co-pilot）。它不再只是机械地节省打字或排版时间，而是直接介入复杂问题的推演过程，在案头工作展开的最初阶段就显著提升了法律工作成果的质量。 ```mermaid flowchart LR accTitle: 法律科技演进与人机协作模式 accDescr: 展示从传统手工时代、软件提效时代到生成式AI时代律师工作模式的演变。 subgraph Phase1["传统阶段"] A["人工手工比对文本<br/>按小时计费 / 效率极低"] end subgraph Phase2["软件时代"] B["自动化比对工具<br/>90秒完成 / 纯效率提升"] end subgraph Phase3["AI 时代"] C["AI 思维伙伴 (Co-pilot)<br/>协助策略构思与推演"] D["律师把关与深度验证<br/>产出高质量法律成果"] C <--> D end Phase1 --> Phase2 --> Phase3 ``` --- ### `12:15` 风投融资中的律师价值与“AI 对话 AI”的现实 在剖析具体的商业应用场景时，主持人问及传统风险投资（VC）融资中律师的核心角色，以及 AI 如何作为思维伙伴介入。温根斯作出了背景说明：尽管他本人是结构化融资领域的专家，但他所在的洛文斯坦·桑德勒律所（Lowenstein Sandler）拥有全美规模最大的风投法律团队之一。在典型的初创企业股权融资中，创始人往往缺乏交易经验，律师的首要职责是扮演导师，为其拆解交易架构，并提供关于种子轮、A 轮、B 轮或 C 轮市场常规商业条款的全局认知。更重要的是，顶级创投律师深谙投资机构及其法务顾问的谈判习惯，能够协助创始人在当前轮次中前瞻性地布局后续轮次，乃至终局的 IPO 或并购重组。 温根斯指出，很多初创企业客户拿着由 ChatGPT 自动生成的投资协议草案找上门来，这些文件往往致命地缺失了市场纵深经验、对交易对手博弈心理的把握以及对后续资本运作的铺垫。与此同时，客户层面的 AI 应用正在迅速两极化： - **企业内部预先起草**：部分《财富》50 强超大型跨国企业已熟练使用内部 AI 工具生成合同初稿、起诉状或答辩状，再交由外部律所进行纵深修改。 - **客户与律所的双向 AI 校验**：在律所西海岸办公室的专利申请代理业务中，顶尖科技客户正在使用他们自己的 AI 工具审查律所提交的成果并生成修改意见；而律所本身也在客户知情的前提下，借助 AI 辅助撰写专利申请文件。整个业务交互正在迅速迈向“客户的 AI 智能体与律所的 AI 智能体直接对话”的新常态。 --- ### `15:00` 国际税法构架与人机协同的边界 针对主持人的追问——既然 AI 能够辅助处理复杂策略，它是否能在结构化融资中凭空构思出一套全新的资产证券化（ABS 或 CMBS）架构？温根斯明确给出了否定回答：“我不认为它能无中生有地发明一种全新的融资结构。”AI 的长处不在于横空出世的系统创新，而在于极高维度的策略碰撞与方案试错。 他以律所内一位顶尖的国际税法合伙人为例进行了生动阐述。该合伙人长期为家族办公室和跨国集团设计跨境资产配置路径，其核心诉求是以最少的税收中转环节将资金从国家 A 合规调配至国家 C。在日常实务中，这位合伙人会使用诸如 Claude 这类大语言模型来推演和压力测试各种极端架构设想： - **思路验证与淘汰**：利用 AI 迅速提供反馈，不仅确认某些假设的可行性，更重要的是迅速推翻、淘汰（invalidate）不切实际的想法。 - **人机交替迭代**：合伙人将与 Claude 深度探讨后整理出的分析方案移交给初级律师（Associate），指示其进行严谨的判例与成文法检索，最终由专业律师完成合规审查与优化。 这种“资深合伙人-AI-初级律师”的三方迭代飞轮，往往能催生出单凭人类律师想不到、单靠 AI 也绝不可能独立构想出的精巧税收架构。 --- ### `16:45` 质疑与验证：客户未经提示的真实反馈 主持人提出了一针见血的质疑：与聊天机器人来回交流很容易让人产生一种“我正在构思新策略”的智力幻觉，这种所谓的启发究竟是真正创造了增量价值，还是仅仅是一种“看起来很像在工作的轻度消遣”（work-ish or work-light）？ 温根斯对此坦然承认，单纯在税法领域确实很难量化这种价值，但他拿出了在帕洛阿尔托（Palo Alto）办公室调研专利代理业务时获得的客观实证。律所引入 AI 工具辅助律师撰写专利申请书已逾半年。该团队的专利律师绝大多数是计算机或电气工程背景出身，虽然专业功底扎实，但知识结构相对垂直；而通用 AI 工具吸收了全门类工程学与自然科学的知识网络，能够在撰写技术交底书时将生物学、材料科学等多学科视角无缝融合进来，从而起草出保护范围更广、权利要求更严密的专利申请文件。 最让温根斯确信技术价值的，是客户未经引导的主动反馈。在拜访西海岸数家顶尖科技公司法务主管时，对方在未被提问的情况下自发表示，他们明显观察到洛文斯坦·桑德勒律所在过去六个月中提交的专利申请书质量“越来越好”，不仅显著拉开了与市场上其他传统律所的差距，更展现出了跨维度的技术前瞻性。温根斯总结道，能够经得起挑剔客户的主动点赞，正是 AI 绝非“自我感动式工具”的最硬核证据。 ### `20:00` 中场商业赞助：定制周边、AI 投资代理与医疗系统数字化 4imprint 赞助了本期节目。无论是团队装备还是定制促销礼品，4imprint 提供了涵盖潮流服饰、高端饮水具、科技周边、托特包与活动赠品等数千种选择，可根据受众、用途及预算定制企业 Logo 与外观，众多产品免收初始设置费，并辅以 360 度交付保障与快速出货。随后播出了 Public.com 的赞助信息，该投资平台现已支持用户使用自然语言构建专属 AI Agents（例如设定“当恐慌指数 VIX 触及 25 时买入标普 500 看跌期权”或“当闲置现金超出阈值时自动转入直接指数化投资”），无需频繁手动盯盘与点击操作，平台亦向用户开放了完整的账户 API 读写权限。最后播出了联合健康旗下 Optum 的赞助，该机构致力于将患者护理、药房服务与全人健康（涵盖全科医生、心理支持与日常就医）进行数字化打通，旨在通过技术减少医护人员的繁琐杂务，让医生将时间还给患者，并进一步压缩处方药的获取成本与流程。 ### `23:35` 尽调成本暴跌 70%：AI 时代律所如何捍卫高单价与定价权 节目话题切回正题，主持人抛出了一个核心疑问：AI 究竟会如何影响律所实际的定价权？在彭博法律（Bloomberg Law）此前刊发的一篇题为《AI 将为初级律师带来更有价值的工作》的专栏文章中，低enstein Sandler 律所主席兼合伙人加里·温根斯（Gary Wingens）曾披露过一个极具冲击力的真实案例——一项曾因报价过于高昂而被迫搁置的尽职调查项目，在短短一年后借助 AI 辅助重新启动时，整体成本竟然骤降了整整 70%。主持人直击要害：如果技术能砍掉七成的成本，客户难道不会要求吞下所有的成本红利吗？在工时大幅缩水的现实下，律所究竟靠什么保住自己的利润率？ 温根斯解释称，成本缩减 70% 绝不等于工作量归零，律所依然在投入过去 30% 左右的工时，但这些留存下来的工时已经彻底完成了“升维”，变成了单价更高、专业度更密集的高阶劳动。他具体拆解了那起涉及“审阅数千份信托协议以梳理各方义务”的尽职调查案件：三年前如果纯靠人工操作，律所必须投入大批初级律师通读全部合同，并在第一轮海选之上额外配备资深质控团队（QC），过程枯燥繁冗且总价惊人。如今，律所直接剥离了第一层审阅苦力，将其全权交给“机器人”。AI 会把提取的所有合同要素梳理成一张整齐漂亮的“百列表格（100-column spreadsheet）”，而人类律师则退居第二道关卡，专注于高阶质控。 ```mermaid flowchart TD accTitle: 尽职调查流程变革 accDescr: 对比传统模式与AI辅助模式下尽调分工与工作重心的转移 subgraph Traditional["传统尽职调查流程"] T1["初级律师大部队<br/>（逐页人工通读数千份信托协议）"] --> T2["资深律师质控团队（QC）<br/>（高层重复抽检与核对）"] T2 --> T3["交付客户<br/>（100% 密集工时与高昂报价）"] end subgraph AIAssisted["AI 辅助尽职调查流程"] A1["AI 自动化首轮审阅<br/>（解析海量合同并生成百列表格）"] --> A2["专业律师高阶审查与质控<br/>（敏锐识别逻辑漏洞与异常值）"] A2 --> A3["律所权威法律背书交付<br/>（成本下降 70%，聚焦高溢价工时）"] end ``` 这种质控并非简单的校对，而是需要深刻理解金融交易底层逻辑与行业运作方式的资深律师去掌舵。质控人员必须具备敏锐的专业嗅觉，一眼看破 AI 输出结果中哪些地方“不合常理”并追根究底。温根斯强调，企业客户自身既缺乏足够的人手，也缺乏跨案例的深厚经验来亲自处理这套流程；客户付钱买单的本质，是买律所对最终结果的专业背书与认证（"certify or say this is good output"），他们始终需要外部专家的眼睛作为安全防火墙。 - 审阅标的物：涉及数千份复杂的信托协议（trust agreements），需逐条解析各签约方的法律责任与履约义务。 - 自动化产物：AI 工具能将非结构化的法律文本高效映射并整理为包含 100 列字段的规整电子表格。 - 价值重心转移：初审层完全自动化，律所向客户交付的不再是“阅读时间”，而是“异常甄别能力”与“出证背书”。 ### `25:20` 初级律师的危机：“抽掉梯子”还是告别苦工？ 针对外界普遍弥漫的悲观情绪——即合伙人会借助 AI “抽掉梯子”（pull up the ladder），大幅削减法学院毕业生的招聘名额——温根斯给出了不同的历史视角。他回忆起自己初入职场的年代，年轻律师的大量时间都耗费在“拿笔在纸质文件上手动画红黑线比对文本修订（blacklining documents by hand）”这种机械琐事上。当比对软件横空出世彻底终结了这项苦役时，行业不仅没有消亡，初级律师依然有做不完的工作，而温根斯本人的职业生活反而变得丰富有趣得多。他坚信，技术首先消灭的永远是平庸乏味的苦力劳动，初级律师将得以在职业生涯更早的阶段接触到更具智力挑战性的核心业务。 不过温根斯也坦承，律所目前面临的一个真正挑战在于：整个法律界尚未彻底摸索出一套“在完全绕过传统底层苦差事的前提下，如何高效培养合格成熟律师”的新方法论。但这并未阻碍律所的招募步伐。低enstein Sandler 不仅没有收缩招聘，反而维持着常规的人才吸收节奏： - 暑期实习与全职聘用：律所当前正常接收在读法学生参与暑期实习项目（summer associates），秋季将迎来完整编制的第一年全职初级律师团队。 - 长期招聘时间线：该所甚至已经超前锁定了 2027 年夏季的实习生班底，在招募数量上未现丝毫退缩。 - 选拔核心转向：招聘的硬性名额虽未改变，但律所评估候选人所需具备的技能模型（skill set）已经开始发生深远变化。 ### `27:45` 新一代律师的培养革命：从“跑走廊请教”到全所知识库与“模拟器实战” 谈到未来年轻律师入行后的工作与培训模式，温根斯勾勒出一幅从“机械苦工”转向“实战模拟器”的图景。以律所最经典的初级通识培训——“并购交易解剖（anatomy of an M&A transaction）”为例，过去老派的教学方式是带年轻律师机械地通读合并协议、股权收购协议（SPA）或资产收购协议（APA）的具体法条条款。而现在，律所要求所有学员进入培训室时必须随身携带笔记本电脑，并同步开启律所部署的 AI 法律工具。 在当下的培训课堂上，合伙人会直接抛出高难度的博弈场景：“现在把这个合同条款放进工具里，观察它给出的反馈；接着向 AI 输入指令，假设你正代表卖方利益，让它为你生成一个极度有利于卖方的谈判对抗条款。”这种人手一台设备、与 AI 实时交锋的交互式训练，正全面取代被动的死记硬背。不仅如此，律所正在构建并将很快向全员开放一个覆盖全所数年积累的超级知识库，里面完整沉淀了事务所近年来经手谈判过的每一笔并购案、每一场诉讼和每一桩破产清算案的底层条款与博弈细节。 温根斯一针见血地指出，客户之所以愿意向顶级律所支付每小时令人咋舌的高昂费率，核心在于律所坐拥客户永远无法独立企及的“市场行情认知（market knowledge）”。对于绝大多数企业管理层来说，并购往往是一生仅此一次的重大决策；但对于律所而言，这不过是日复一日的日常作业，客户雇佣律师正是为了获知最真实的“当下市场公认条款”。过去，年轻律师若想弄清某个特殊条款在业内如何落地，必须硬着头皮穿过长长的办公走廊去敲合伙人的门，挨个询问“有谁以前做过类似的案子”；而现在，AI 技术能让任何一名第一年入职的年轻律师，在指尖瞬间调取整个律所数十年沉淀的集体智慧。 ### `30:00` 基础杂活的消失与青年律师的“练琴”困境 当所有历史交易文档与案卷都可以汇聚在一个数字化控制面板中供随时调取时，律师确实能更快、更聪明地掌握全局，但这也引发了全新的认知负荷问题。过去，低阶律师花费大量时间逐字进行文档核对（blacklining），这种纯体力脑力活看似枯燥，却也是一种难得的“心理放空与喘息期”。更关键的是，法律界长期信奉一种近乎“学徒制”的职业训练逻辑：新人必须亲身去跑腿送交 ISDA 主协议、手动审查成百上千份信托契约。正如经典纪录片《寿司之神》（Jiro Dreams of Sushi）中展示的那样，学徒要在做寿司前先花上数年甚至十几年时间学习淘米；又如音乐家在登台前必须日复一日枯燥地练音阶。正是这种严苛且机械的重复劳动，磨炼了律师对细节的极端敏锐度——在动辄涉及巨额商业利益的文本中，漏掉一个标点符号可能就会造成数百万美元的损失。 面对这种基础技能可能退化的担忧，Gary Wingens 坦言自己同样深感焦虑，并且从行为心理学的角度来看，目前业界并没有完美的解决路径。尽管如今有人提议使用模拟器来模拟商业交易与谈判，但模拟系统根本无法替代真实交易中的人际博弈。在现实业务中，你必须去亲自面对谈判桌对面的各色人物——无论是态度恶劣凶悍的对手，还是表面和颜悦色、背地里却能悄无声息将你套牢的笑面虎。人际交往中的直觉、现场触感以及对客户与对手的把握，是任何工具都无法虚拟出来的。 随着 AI 技术的突飞猛进，日常文档起草工作极可能在不久的将来全面转交机器处理，但这反而对审阅者提出了更隐蔽的要求。AI 经常犯低级错误，甚至基本的算术和标点运用都会出现致命漏洞；如果年轻一代律师过早产生技术依赖，盲目信任算法生成的现成答案，而放弃了更高层面的独立审视与批判性思考，那将是非常危险的职业陷阱。 - **学徒训练的比喻**：正如学徒先学淘米、琴童反复弹奏枯燥的音阶，传统律师从校对文档和核对条款中建立对文字和逻辑的敬畏感。 - **细节决定巨资**：在复杂法律协议中，一个遗漏的逗号都可能引发价值数百万元的合同纠纷，而 AI 在处理这些基础运算与标点细节时经常犯低级错误。 - **人际博弈的不可替代性**：交易谈判的精髓在于人性的试探与交锋，无论是应对蛮横的对手还是提防看似亲切的“笑面虎”，都必须依靠真实的实战积淀。 ### `33:15` 诉讼与交易中的“杰文斯悖论”：降价如何撬动巨量需求 在金融与市场分析的语境下，法律成本的大幅下降引发了一个核心命题：AI 是否会触发法律领域的“杰文斯悖论”（Jevons paradox），即单次服务单价的大幅缩减，反而刺激出几何级增长的总体需求？在此前讨论民权法律的播客中，嘉宾曾提到许多受害者遭遇警察不当执法等侵害时，在法理上完全站得住脚，但由于预估的赔偿金额过低，扣除昂贵诉讼费后在经济账上根本无法成立，正义不得不向成本妥协。而在商业并购与公司法律服务中，同样的逻辑也在时刻上演。 ```mermaid flowchart TD accTitle: 法律服务中的杰文斯悖论机制 accDescr: 描述法律技术降本如何推动案件和业务总量反弹的机制循环 subgraph 传统高成本模式 A["高昂诉讼/尽调单价"] --> B["客户放弃边际案件/审查"] B --> C["律所该项收入为零"] end subgraph AI驱动的杰文斯效应 D["AI大幅压缩底层工时"] --> E["法律单价大幅下调 (如降低70%)"] E --> F["原本不经济的诉讼/交易被激活"] F --> G["案件总量激增，律所总营收增长"] end ``` Gary 明确证实，杰文斯悖论在法律市场完全适用。他举例说明，当律所此前为客户的数千份信托协议审查报出传统价格时，客户直接拒绝了，因为潜在下行风险与巨额账单不成正比，律所从该项目中获得的收入为零；但当 AI 工具将综合成本压缩了约 70% 之后，原本搁置的诉讼或审查项目便瞬间具备了经济可行性。用一个假设数据来比喻：原本一千万美元的报价换来的是零成交，而降至三百万美元后，客户立刻爽快签单，律所也实打实入账了三百万美元。单价崩盘不仅没有砸掉律所的饭碗，反而释放了庞大的沉睡市场。 过去 25 年间，电子证据开示（e-discovery）技术的泛滥让复杂诉讼成本水涨船高，任何像样的大型商业纠纷光是开示调查费用就轻易消耗数百万美元，使诉讼门槛高不可攀。倘若 AI 能将两百万美元的开示成本骤降至二十万美元，整个诉讼策略的计算方程式将被彻底改写。不仅民权律师能够承接更多标的微小却意义重大的案件，大型跨国企业也会重新激活那些曾经被权衡放弃的正当索赔与诉讼主张。 ### `35:50` 专利井喷与企业法务：外部律所的生存新格局 法律需求扩张的另一个生动案例来自于专利申请业务。随着 AI 的辅助，专利撰写与申报流程的成本正在急剧下降。低阶成本的下挫同时在企业端产生了连锁反应：某大型企业客户两周前告知律所，因为企业内部研发团队也全面接入了 AI，科学家与工程师产出发明的效率暴增，内部提出的专利申请立项需求竟然翻了整整四倍。虽然外部律所承接每一件专利申请收取的单价在走低，但面对四倍业务量的井喷，律所的整体业务体量和营收空间反而被大幅扩容。 随着大量程式化工作被 AI 接管，大企业法务部与外部顶级律所之间的权力天平也必然发生倾斜。如果一家大型公司内部拥有几名经验丰富的顶级律所离职背景律师，配合高效率的 AI 生产力工具，他们完全可以将过去大量外包的繁琐杂活就地消化。比如银行业早就将大部分 ISDA 衍生品协议起草和审阅收归内部法务，未来对冲基金等买方机构也会进一步将这些流程性文书内卷化，不再斥巨资外聘专业律师团队处理初级合规与合同事务。 然而，外部律所的核心护城河并不会因此坍塌。Gary 强调，真正的价值正在向不可替代的高阶认知集中。大企业法务往往深陷于自身企业与行业的单一视角，而顶级外部商业律所横跨数百家客户与不同交易架构，掌握着全局性的“市场深度与行业纵深见解”（market breadth and market knowledge）。当面对前所未有的重大危机、关键交易结构设计以及需要高度独立性的战略决策时，企业依然必须仰仗外部律所提供超然的第三方客观洞察。在后半段，节目还插入了一段关于定制化促销周边供应商 4imprint 的商业赞助播报，强调其全品类、零安装费与全流程交付保障服务。 - **研发与专利申请翻倍**：AI 辅助不仅降低了法务处理成本，更促使工程师发明产出提速，带动专利申请需求增长至过去的四倍。 - **基础工作加速“内卷化”**：大型银行与基金客户正借助内部法务与自动化工具，将更多类似 ISDA 协议的日常文书牢牢留在公司内部。 - **外部律所的核心护城河**：由跨行业服务所带来的宏观市场认知、深厚案例积淀以及独立客观的判断力，是单一公司内部法务团队难以自建的壁垒。 ### `40:00` 律所技术栈剖析：为什么法律行业需要 Harvey 这一层封装？ 在探讨律师日常工作中到底如何接入 AI 时，访谈转向了具体的底层技术栈。虽然律师们经常提到在与 Claude 或 ChatGPT 交互，但市场上一批像 Harvey 或 Legora 这样建立在前沿基础模型之上的垂类应用公司正迅速崛起。对于“律所是否真的需要这一层专门的应用封装”的疑问，Gary Wingens 直言 Lowenstein Sandler 目前尝试的工具可能“多得有点过头”，随着行业成熟最终会收敛，但律所当前是一个非常明确的“Harvey 重度用户”（Harvey Shop），几乎全体律师都在日常业务中高频使用 Harvey，同时在日常办公套件中配合使用 Microsoft Copilot。 针对非法律专业人士最常产生的疑问——既然底层都是 GPT-4/5 或 Claude 这类前沿大模型，为什么不直接调用底层模型，而要多买一层 Harvey 这样的软件？Gary 指出了几大核心差异： - **模型可自由切换**：在 Harvey 等平台内部，用户本身就可以自由选择背后的基础驱动模型（比如在 Claude、ChatGPT 或 Gemini 之间切换）。 - **极度严苛的安全与特权防火墙（Security Layer）**：这是律所的立身之本。Harvey 提供了完全隔离的数据环境，保证内部数据绝不外流至公开互联网，甚至可以开启完全不连外网的会话。Gary 特别警示，很多客户并未意识到，如果直接在消费级的 Claude 或 ChatGPT 中输入案件细节，极可能会直接丧失极为关键的“律师-委托人保密特权”（Attorney-Client Privilege）。 - **团队协作与项目手册（Playbooks）**：相比单人面对对话框，专业平台支持团队间共享提示词手册、工作流项目和办案模板。 ```mermaid flowchart TD accTitle: 法律垂直 AI 应用层架构与价值 accDescr: 展示前沿通用大模型如何通过法律应用层转化为律所安全可用的业务工具 subgraph FrontierModels["底层前沿大模型"] M1["OpenAI GPT"] M2["Anthropic Claude"] M3["Google Gemini"] end subgraph AppLayer["Harvey / 法律垂直应用层"] SEC["安全与伦理隔离层<br/>(保护律师-委托人保密特权)"] RAG["法律专用 RAG 检索层<br/>(针对判例、合同与证词微调)"] SHARE["团队协作与 Playbook 沉淀"] end subgraph LawFirmUse["律所高价值落地场景"] L1["并购协议复杂条款对比"] L2["诉讼证据与证词笔录智能检索"] L3["起诉状/答辩状依据引用匹配"] end FrontierModels --> SEC SEC --> RAG RAG --> SHARE SHARE --> LawFirmUse ``` ### `45:00` 法律微调 RAG 的威力：从海量证言中秒级捕捉胜诉证据 在安全底座之上，Harvey 这类平台的核心技术优势在于构建了一套针对法律业务深度微调的“检索增强生成”（RAG, Retrieval-Augmented Generation）层。Gary 自嘲说自己也不懂底层算法原理，但这个概念“听起来很酷”，而且它在业务现场的作用立竿见影。通用大模型在面对复杂的并购（M&A）协议条款演进、或是对证人宣誓证词笔录（Deposition Transcripts）进行细微法律语境理解时常常力不从心，而垂直 RAG 工具则开箱即用。 以诉讼业务为例，Gary 描绘了诉讼律师以往的典型痛点：一个案件中堆积如山的诉状、答辩状、开庭动议以及律师之间数千页的来往信件和漫长的问供笔录。以往为了在起草动议时找到一句支撑己方论点的关键证词，初级律师需要耗费几十甚至几百个小时手动逐行翻阅比对。而在法律 RAG 系统中，律师只需将所有案件卷宗作为一个独立项目载入，撰写诉状时提问系统，AI 就会以极高精度直接定位并提取出最能印证论点的宣誓引文。虽然通用前沿模型原则上也能做，但经过法律语料精细微调的系统不仅更懂行规与潜台词，还能更高效地沉淀并复用整个团队的最佳实践。 - **法律专用 RAG 优势**：深度针对法律文书格式与行业术语进行微调，天然契合交易合同与诉讼材料审查。 - **证词智能溯源**：在数十万字的问供记录中准确定位支撑律师诉求的判例与口供，节约大量繁琐人工查找时间。 - **知识资产固化**：项目内的检索逻辑与提示流程可以固化为 Playbook，供整个团队平滑复用。 ### `47:00` 态度大反转：两年前被视作违规隐患，如今不让用 AI 才是失职 在数据合规与执业风险控制方面，法律行业对 AI 的态度在过去两年间发生了戏剧性的逆转。Gary 分享了自己作为管理合伙人与律所的执业过失保险（Malpractice Insurance）承保人——英国劳合社（Lloyd's of London）核保团队打交道的亲身经历： - **两年前的核保审查**：承保团队满脸警惕，反复质询：“你们律所没有在让律师用 AI 吧？这会不会给我们带来巨大的赔付风险？” - **现在的核保审查**：同一个承保团队的口风彻底颠倒，严厉追问：“你们有要求律师必须使用 AI 来复核工作吧？你们得确保办案万无一失。” Gary 指出，主流认知已经将 AI 辅助确立为行业“标配”，不用 AI 反而成了专业尽责层面可能存在漏洞的隐患。商业客户的心态也经历了完全一致的急转弯：18 个月前，企业法务还在严厉下令“禁止在我们的案子里用 AI”；如今则是拿着放大镜质问律所“你们怎么还没全面接入 AI 把账单费用打下来”。 但底线依然不可触碰：律所对律师三令五申，AI 生成的内容必须进行严密的人工双重核查。 Gary 措辞严厉地表示，在向法院递交的诉状或代理意见中引用“虚构、幻觉生成的虚假案例判决”，没有任何托词与借口。如今市场上有如此多的交叉验证工具，“出现虚假判例在法庭上被当场揭穿，实在是太令人无地自容了（too embarrassing）”。即便个别律师在赶截稿时间的一小时内慌乱提交导致系统校验崩溃，这种低级错误也绝对不能被容忍。 ### `49:00` 明星合伙人与 AI 知识共享的隐秘博弈 访谈最后抛出了一个直击法律行业薪酬与组织机制痛点的尖锐矛盾。主持人提到他曾向高盛 CEO David Solomon 提过类似问题：当前不论是华尔街对冲基金还是顶尖律所，都在涌现大量薪酬和签字费极其夸张的“超级明星合伙人（Superstar Hires）”。律所往往砸下重金从竞品那里高薪挖人。 然而，AI 赋能律所的前提，是全所必须建立一个无私贡献个人经验、交易模板与实战 Know-How 的知识库，从而让初级律师能够通过系统迅速获得大师级的 Playbooks。主持人质疑：一个靠个人独门秘籍和专属经验拿到数百万甚至上千万年薪的顶尖律师，凭什么愿意把自己的底牌和核心模板全部上传到公共 AI 系统里？如果每个明星合伙人都选择打个人算盘、留一手防范被替代，那么律所所谓“集中化 AI 赋能”的飞轮与特许经营价值（Franchise Value）是否根本无法真正运转？这一质疑直切组织文化与激励分配的核心矛盾。 ### `50:00` 合伙人自主性与知识共享的文化冲突 在律所引入 AI 的过程中，最棘手的往往不是技术难题，而是深刻的“人性与文化”挑战。律师通常具有极高的独立性和自主性，顶尖合伙人更是习惯了“别教我怎么办案”的工作模式。节目中提到一个生动的例子：有初级律师曾在周末接到合伙人从钓鱼船上打来的电话布置任务，这种“自己打猎自己吃”（eat what you kill）的文化在行业内根深蒂固。Lowenstein Sandler 内部常将这种机制称为“放马奔腾”（let the horses run），即鼓励合伙人自由发挥企业家的进取活力，客户也非常买账。然而，AI 系统的运行逻辑恰恰相反——它要求每位顶尖合伙人把个人的独门秘籍、实务经验和私有知识毫无保留地共享给系统，这在文化上构成了巨大的冲突。 这种矛盾也催生了合伙人薪酬与价值认同的新逻辑。如今法律界频频出现类似 NBA 或 NFL 明星球员级别的天价年薪，引进这些超级巨星合伙人的一部分动机，或许正在于将其专属的实务智慧与交易战绩吸纳进律所的 AI 工具中。当律所能够向客户展示其 AI 平台内置了某位行业大咖近期主导的交易条款与框架时，客户会深感信服。这种“将大咖名字印在算法上”的模式，不仅强化了律所的技术溢价，也在进一步推高超级明星律师的身价。 ### `51:40` 计费工时悖论：费率暴涨与最终交付成本下降 针对法律行业是否会因为 AI 普及、定价承压和案件不确定性降低而最终放弃按小时计费（billable hours）的疑问，嘉宾认为按项目计费或按结果计费（outcome pricing）确实会迎来真正的发展契机。然而，顶尖律所的按小时计费模式非但不会消失，其名义费率反而在加速飙升。 根据《美国律师》（The American Lawyer）公布的 2025 年数据，全美大型律所的平均小时费率同比暴涨了 10.1%，而在利润率排名前 20 的顶级律所中，涨幅甚至更为惊人。在当年消费者价格指数（CPI）仅上涨约 3% 的背景下，这种费率与通胀之间史无前例的巨大鸿沟看似不合常理，但其背后的真正推手正是 AI：由于 AI 工具的强力赋能，律师每个计费工时的产出效率与含金量发生了质的飞跃。 ```mermaid flowchart LR accTitle: AI推动下的法律服务计费双向变动 accDescr: 展示AI如何同时推高单小时费率并降低客户最终采购成本 subgraph Efficiency ["单小时价值重塑"] A["引入 AI 工具"] --> B["单位工时产出与含金量跃升"] B --> C["小时费率大幅上涨\n(如 2025 年涨 10.1%)"] end subgraph Delivery ["实际交付总账"] B --> D["解决业务问题所需工时大幅缩减"] D --> E["客户购买商业解决方案的最终总价下降"] end ``` 这种演变带来了一种双向分化：律所名义上的小时单价在水涨船高，但客户真正购买的并不是工时，而是商业问题的解决方案；由于完成同一项复杂事务所耗费的时间被极度压缩，客户最终支付的实际总成本反而在持续走低。 - 2025 年全美大型律所平均小时费率上涨 10.1%，远超约 3% 的 CPI 涨幅，盈利排名前 20 的头部律所增幅更甚。 - 计费小时单价攀升与客户实际总支出下降并行不悖，核心变量是单兵生产力因技术加持而指数级提高。 ### `53:30` 技术栈路线抉择：绝不从零自研基础模型 面对行业内关于开源模型深度微调与商业模型外挂 RAG（检索增强生成）的讨论，以及桥水基金（Bridgewater）等机构和超级巨头 Kirkland & Ellis 在技术自研上的动作，中型律所展现出了极其务实的技术路线。低特律所这样拥有 400 名律师的机构，绝不会试图在内部从零训练一个所谓的“Lowenstein LLM”。 从经济账来看，从头预训练一个顶级基础大模型所需资金高达 15 亿美元左右。即便是业内巨头 Kirkland & Ellis 宣布在五年内投入 5 亿美元，按照这个资金体量也需要持续投资 15 年才勉强够用。因此，中型律所合理的切入方式是将自有的专有交易数据接入前沿模型，构建基于自身实务积累的定制化解决方案与工作手册（例如打造属于自己的并购实操指南 Lowenstein M&A Playbook），或者在现有成熟模型基础上做白标（white label）包装，而非盲目投入资本黑洞。 ### `55:20` 补贴迷雾散去：Token 定价冲击与法律版杰文斯悖论 律所乃至整个商业界在 AI 应用上正面临一场潜在的“账单休克”（sticker shock）。目前企业级法律 AI 大多采用类似“自助餐”（all-you-can-eat）的打包包月收费，但整个行业正加速转向按 Token 用量计费（例如 Claude 正率先推行这一模式）。当前许多律所之所以觉得 AI 帮其节约了 30% 的成本，很大程度上是因为底层算力成本被外部风险投资人巨额补贴了；一旦补贴退潮、云厂商按实际用量追求每 Token 利润，加上电网与能源瓶颈推高算力边际成本，AI 的实际使用成本可能会直接飙升到原本预期的 90% 甚至 130%。 截至 7 月 8 日，律所管理层对真正的技术落地成本依然缺乏清晰的可见度。嘉宾透露，据其律所 IT 主管统计，当月前三天工作日（扣除 7 月 4 日美国独立日假期）他个人使用 Claude 产生的费用仅为 36 美元，这还抵不上他两分钟的计费工时。但在补贴终结后，两年后的真实定价走势仍是未知数。 主持人随后展开了一段深刻的商业类比与隐忧探讨： - **资本补贴的繁荣假象**：当前的 AI 极像当年的 Uber、外卖平台甚至页岩油产业——早年均靠风险资本或公开资本市场的巨额亏损补贴换取用户习惯，一旦要求真实盈利，服务价格就会大幅反弹，需求随之受到抑制。 - **好消息与坏消息（杰文斯悖论）**：如果未来 Token 成本没有反弹而是持续暴跌，技术成本的雪崩式下降将诱发法律界的“杰文斯悖论”（Jevons Paradox）。好消息是人类律师绝不会失业，未来依然有大把工作可做；坏消息是，这些新增工作很可能是原本根本不需要处理的上百起恶意微型诉讼、骚扰性维权以及无休止的专利无效反击战，整个社会可能因此陷入更庞大而内耗的法律泥潭。 ### `01:00:00` AI、企业官僚主义扩张与职场焦文斯悖论 当人们把 AI 视作生产力提效工具时，很可能会导向另一种现实：AI 并非解放人类，而是被用来无休止地扩张官僚主义。这种扩张不仅限于滥诉等无谓诉讼，还会蔓延至人力资源等行政管理领域。节目中打趣道，甚至一直想写一篇题为《人力资源部门是如何吞噬美国经济的》的文章。 如果一家公司的每位员工都用上了类似 Claude 这样的大语言模型工具，初期的体验确实非常惊艳：原本需要 8 小时才能完成的繁重工作被压缩到了 2 小时。然而，一旦所有争夺晋升机会的同事也都把 8 小时的工作压缩到了 2 小时，职场竞争的平衡便被打破。为了在升职竞争中脱颖而出，员工只能去创造更多工作、寻找更多任务，最终导致工作总量持续膨胀。AI 普及后大家都能轻松躺平的乌托邦景象，在职场现实面前显得极为遥远。 ```mermaid flowchart TD accTitle: 职场 AI 效率与内卷循环图 accDescr: 展示 AI 压缩工时后如何引发竞争升级并导致官僚主义膨胀的焦文斯悖论循环 subgraph Efficiency["效率跃升"] A["引入 AI 工具 (Claude 等)"] --> B["8 小时工作量压缩至 2 小时"] end subgraph Competition["竞争与内卷"] B --> C["竞争对手同样压缩工时"] C --> D["为争夺晋升寻找更多新任务"] end subgraph Expansion["官僚膨胀"] D --> E["行政开销与琐碎工作暴增"] E --> F["职场焦文斯悖论"] F -.-> A end ``` - **真实生活中的提效微光**：尽管整体工作量并未减少，但大模型在处理特定复杂流程时确实能减轻认知负担。例如在达美航空（Delta）网站上需要在最后一刻将往返机票临时更改为多段票时，网页操作逻辑极其繁琐，询问 ChatGPT 具体分步指南后，AI 精确指出了点击右下角特定入口等全套流程，成功缓解了心理上的繁复压力（"psychic tax"）。 - **终极结论**：AI 在未来带来的更可能是一场针对行政开销（administrative overhead）与琐碎事务（busy work）的“焦文斯悖论”（Jevons Paradox）——单位效率越高，整体制造出来的无用繁杂工作反而越多。 --- ### `01:01:25` 节目尾声与制作团队致谢 主持人 Tracy Alloway 与 Joe Weisenthal 结束了本期深入探讨律所与 AI 关系的讨论，并向听众总结了本期节目及频道订阅信息。节目团队感谢了制作人员 Carmen Rodriguez、Dashiell Bennett、Kale Brooks 以及 Kevin Lozano 的幕后付出。听众被邀请前往 Bloomberg 网站订阅每日简报，加入官方 Discord 讨论社区，或在各大播客平台留下好评，彭博订阅用户亦可在 Apple Podcasts 上享受无广告收听体验。 ## Notable quotes > `15:00` 我不认为它能无中生有地发明一种全新的融资结构。 — 加里·温根斯 (Gary Wingens) > `23:35` certify or say this is good output — 加里·温根斯 (Gary Wingens) > `27:45` 现在把这个合同条款放进工具里，观察它给出的反馈；接着向 AI 输入指令，假设你正代表卖方利益，让它为你生成一个极度有利于卖方的谈判对抗条款。 — 加里·温根斯 (Gary Wingens) > `47:00` 你们律所没有在让律师用 AI 吧？这会不会给我们带来巨大的赔付风险？ — 英国劳合社 (Lloyd's of London) 核保团队 > `47:00` 你们有要求律师必须使用 AI 来复核工作吧？你们得确保办案万无一失。 — 英国劳合社 (Lloyd's of London) 核保团队 > `47:00` 出现虚假判例在法庭上被当场揭穿，实在是太令人无地自容了（too embarrassing）。 — 加里·温根斯 (Gary Wingens) > `50:00` 放马奔腾（let the horses run） — 加里·温根斯 (Gary Wingens) > `50:00` 别教我怎么办案 — 加里·温根斯 (Gary Wingens) > `01:00:00` 人力资源部门是如何吞噬美国经济的 — Tracy Alloway / Joe Weisenthal ## Mentioned Joe Weisenthal, Tracy Alloway, Gary Wingens, Lowenstein Sandler, Kirkland & Ellis, LexisNexis, Microsoft Word, ChatGPT, Claude, Harvey, Legora, Microsoft Copilot, Gemini, GPT-4, GPT-5, 4imprint, Public.com, Optum, UnitedHealth Group, Lloyd's of London, David Solomon, Goldman Sachs, Bridgewater Associates, The American Lawyer, Delta Air Lines, Carmen Rodriguez, Dashiell Bennett, Kale Brooks, Kevin Lozano, Jiro Dreams of Sushi ## Video description <p>It seems obvious that among the many industries that AI might disrupt, the legal profession might face some of the most adverse outcomes. When clerical, research-based tasks like searching through databases and reading contracts are automated, what is left for lawyers to do and how might they justify all those billable hours? In this episode we speak with Gary Wingens, chair and partner at the law firm Lowenstein Sandler. He talks about how his firm is using AI and why he thinks the technology could end up increasing legal work for lawyers as costs come down, creating a sort of &ldquo;Jevon's paradox&rdquo; for lawsuits, deals and litigation. We also talk about the billable hours model and training junior talent.<br><br>Read more: <a href="https://www.bloomberg.com/news/articles/2026-07-07/ai-legal-startup-norm-valued-at-1-2-billion-in-funding-round?utm_medium=referral&utm_source=podcast&utm_campaign=odd_lots&utm_content=article">AI Legal Startup Norm Valued at $1.2 Billion Funding Round</a><br><br>Only Bloomberg.com subscribers can get the Odd Lots newsletter in their inbox &mdash; now delivered every weekday &mdash; plus unlimited access to the site and app. <a href="https://www.bloomberg.com/subscriptions/oddlots?in_source=oddlotspodcast" data-link-cookie="1" data-destination="https://www.bloomberg.com/subscriptions/oddlots?in_source=oddlotspodcast">bloomberg.com/subscriptions/oddlots</a><br><br></p><p>See <a href="https://omnystudio.com/listener">omnystudio.com/listener</a> for privacy information.</p> [Source](https://podtrac.com/pts/redirect.mp3/tracking.swap.fm/track/UVBrz8bN8aM2Xe47PEPu/traffic.omny.fm/d/clips/e73c998e-6e60-432f-8610-ae210140c5b1/8a94442e-5a74-4fa2-8b8d-ae27003a8d6b/bf18344e-5dd0-4852-be82-b483013dfb22/audio.mp3?in_playlist=982f5071-765c-403d-969d-ae27003a8d83)

[15:00] 加里·温根斯 (Gary Wingens): 我不认为它能无中生有地发明一种全新的融资结构。 [23:35] 加里·温根斯 (Gary Wingens): certify or say this is good output [27:45] 加里·温根斯 (Gary Wingens): 现在把这个合同条款放进工具里，观察它给出的反馈；接着向 AI 输入指令，假设你正代表卖方利益，让它为你生成一个极度有利于卖方的谈判对抗条款。 [47:00] 英国劳合社 (Lloyd's of London) 核保团队: 你们律所没有在让律师用 AI 吧？这会不会给我们带来巨大的赔付风险？ [47:00] 英国劳合社 (Lloyd's of London) 核保团队: 你们有要求律师必须使用 AI 来复核工作吧？你们得确保办案万无一失。 [47:00] 加里·温根斯 (Gary Wingens): 出现虚假判例在法庭上被当场揭穿，实在是太令人无地自容了（too embarrassing）。 [50:00] 加里·温根斯 (Gary Wingens): 放马奔腾（let the horses run） [50:00] 加里·温根斯 (Gary Wingens): 别教我怎么办案 [60:00] Tracy Alloway / Joe Weisenthal: 人力资源部门是如何吞噬美国经济的

[45:00] Gary Wingens: Two years ago, they were worried about, 'Are you letting your lawyers use AI and is that creating risk for us?' Now they're asking, 'You let your lawyers use AI, right? To make sure they're doing things right?' [55:00] Joe Weisenthal or Tracy Alloway: It turns out there's going to be plenty of human labor in the future. Bad news: there's going to be 100 more frivolous lawsuits and patent counter-challenges that you never had to do.

[0:00] Whatever your goal, trade show giveaways, client gifts, or team gear, 4imprint has the promo products to match. With thousands of options, from apparel and drinkware to tech and totes, it's easy to find the right fit for your brand and budget, with standout choices at every price point. And with their 360-degree guarantee, you can be 4imprint certain your order will show up just right, right on time. Explore more at 4imprint.com. 4imprint. 4certain. The thing about AI for business? It may not automatically fit the way your business works. At IBM, we've seen this firsthand. But by embedding AI across HR, IT, and procurement processes, we've reduced costs by millions, slash repetitive tasks, and freed thousands of hours for strategic work. Now we're helping companies get smarter by putting AI where it actually pays off, deep in the work that moves the business. Let's create smarter business, IBM. When you're running a business, the best days are the ones where priorities stay on track. For midsize and large companies, risk can affect multiple parts of the organization at once, from property and liability to cyber and regulatory challenges. At that level managing risk becomes an ongoing discipline At the Hartford the focus is on helping businesses manage risk before it turns into something more disruptive And when losses do happen, that work is paired with insurance coverage shaped by years of underwriting, risk engineering, and claims experience. Learn more at thehartford.com slash risk mitigation. Policies provided by Hartford Fire Insurance Company and its property and casualty affiliates, Hartford, Connecticut. Bloomberg Audio Studios, Podcasts Radio News. Hello and welcome to another episode of the Odd Lots podcast. I'm Joe Weisenthal. And I'm Tracy Alloway. So Tracy, I think like maybe a year and a half ago, we did that episode with Joel Wertheimer talking about AI and law. It does feel like law specifically is one of those areas where it's very easy for the normal person to imagine how AI could be very disruptive. Would you say that's fair? Yes. I would say however and full disclaimer so my husband used to be a lawyer a corporate lawyer I would say that like standardized forms and templates have existed in the legal profession for many many years And law has also gone through many, many technological revolutions. So we've gone from like scriveners. Do you know that term? Scriveners? Oh, yeah, yeah, yeah. To like keyboardists. Totally. And the advent of like typists was supposed to be this massive hit for – not a hit, but like this massive deal for the legal industry. and everything ended up like navigating through it pretty much. I guess they probably at each one of those waves, there probably was some extreme period of anxiety about disruption and how would billing still happen. And of course, one of the questions that comes to mind in the legal profession is like, well, why would the lawyers want to reduce the number of hours that they can do on a case given that famously they're paid by the hour? I think actually, aren't they like paid by the 10 minutes technically? technically. Sometimes, yeah. But it does feel like... So maybe, like, I'm not saying and I don't even have an intuition per se that AI would, like, upend or reduce the amount of legal work that needs to get done. But those technological revolutions that you mentioned in the past probably did change the industry quite a bit So regardless of like total employment or job prospects etc it feels inevitable that change is going to come At one point it was probably like a skill to be able to know case law or know how to search LexisNexis Like that almost certainly is going to change. So one thing I've been thinking about, and this gets into the productivity debate, is I'm starting to come around to the idea of a Jevons paradox in just general bureaucracy, basically. So you already have AI that's like being used by insurance companies to push back against claims and then you have claimants using AI to then push back against the insurance companies and I feel like the future could just be a bunch of bots like talking to each other and filing different claims and I guess legal disputes. And then humans still have to sort it all out at the end. Yeah and maybe we don't get more productive. I don't know. Other people have said this. everyone's a lot of people like in tech coding all of these computer jobs right now a lot of people seem to be feel very stressed and overworked like i don't think you like look at a situation someone who has a job that primarily involves on their computer is like oh i'm feeling like i'm really i have it easy these days everyone feels uh very stressed but anyway i'm very interested in this topic there was a big story several months ago one of the major law firms kirkland and ellis like talking about like building out more information [5:00] how it's infrastructure. So I think it's a good time to sort of get a state of the market of like, okay, like we're what almost four years into the post-chat GPT world. Where do we actually stand in terms of what AI could do for legal work? And then also just like how it's changing the profession, if at all. Yeah, let's do it. All right. I'm very excited to say we do, in fact, have the perfect guest. Someone who's been talking a lot about this, someone who's written about this topic, someone who is at a very substantial law firm where you can actually see some of this taking place in real time. We're going to be speaking with Gary Wingens. He is the chair of Lowenstein Sandler. So, Gary, thank you so much for coming on Odd Lots. Delighted to be here. Thanks for having me. Just to set the scene, why don't you give us a little bit of an idea of what the firm is, how big it is, your role there, and what kind of work gets done there? Again, thanks for having me. Lowenstein Sandler, about 400 lawyers, primarily located in the New York area. Is 400 big law at that point? 400 is big law. It's not huge law. You mentioned Kirkland earlier. They're much larger. We very firmly focus on clients in the private capital sector private equity venture hedge funds technology companies and life sciences businesses Why has the billable hours thing lasted as long as it has? Because everyone seems to complain about this. Like the lawyers complain that it's terrible. The clients complain that it's terrible. And it does seem to lead to sometimes not the most efficient outcomes. Everyone agrees it's terrible, as you said. yet it remains the dominant model. I have never met a client who has come to me asking to buy billable hours, right? Nobody ever wants to buy billable hours. They want to buy business solutions. And it happens that billable hours is how we measure what the bill will turn out to be. There has been, at least for the past 15 years, there's been a pretty strong movement toward what are called alternative fee arrangements, where project-based pricing, cap fees, collars, things like that, has not really taken hold, which is surprising. I think law is one of the last segments in professional services firms that has not moved to some kind of value or project pricing And I not exactly sure why perhaps it because lawyers and clients don really trust each other that much And when we propose a you know here how much it costs to do this kind of public offering, sometimes clients say, well, if you're going to propose that amount, I'd rather pay you by the hour because I want to see what you're actually doing. On the other hand, when clients ask for alternative fee arrangements, we almost always offer them. Is there a tacit understanding, though, of like, OK, maybe it's billable hours and the alternative fee arrangement doesn't get formally written down. But is there a general tacit understanding of like, OK, here is a IPO. We're going to raise a billion dollars or here is a venture capital deal where a company is going to raise $500 million. How much that is going to cost? Is it well understood? This is how much this kind of job costs. And if it's wildly different, some alarm bells would go off. Yeah. So the IPO market is easy because as part of your disclosure in doing an IPO, you disclose how much you pay the law firms. And so there is a known market and there a known comparative price A venture round is a little more challenging because there are lots of different details but there a common understanding about the range of rates. And then things go from there if there are funky tax issues or things like that. But for most areas, there's kind of a common range. So getting back to the topic at hand, AI, how much of a step change is this in terms of the actual work process for a corporate lawyer? Because as I mentioned before, standardized templates exist. Law firms have had software for due diligence for ages. A lot of them have knowledge libraries where they sort of share information or you have experts that act like little encyclopedias for law. How big a change is this? I think it is a pretty huge change. So there are two ways it's changing. One is what you're talking about, kind of the process and systems and using AI to become more efficient. Just like when Microsoft Word came out and it became much more efficient for you to kind of do your own documents or compare right. When I was a first year associate, I'm old, when I was a first year associate, [10:00] I was hand blacklining documents and clients were paying me by the hour to hand blackline documents. Within two years, you know, we had blacklining software and they no longer paid me by the hour. And it took roughly 90 seconds to blackline a document. So there's there's the efficiency piece, which definitely is starting to make it to make the things that clients want to buy less expensive. Did you ever miss a comma that resulted in like $5 million in extra charges? Not that I'm aware of. Okay. I have to ask. That's a famous case, right? Right. But I've definitely missed commas. Everybody has. But I'm not aware that it's resulted in anything bad happening. Hopefully, nobody listening to this has the other side of that. But the other way AI is working in law firms right now is it's acting as a thought partner and it is making us better at our jobs and, to use a term, acting as a co-pilot for all of our lawyers in helping think through problems. So as that thought partner we never had that before That a dramatic difference that AI is providing that these other technological advances did not They only went to the efficiency piece and allowed us to produce what clients want to buy, a deal or litigation for a lower total cost. But we're now, I believe, producing better work product from the get-go. Okay, let's go back to a VC financing round. So why don't you explain, even say pre-AI, what was the role of the lawyer or the law firm in a typical financing round? Why are their lawyers involved? What are they brought in to do? And then maybe sort of give us a concrete example of today what it means or how a lawyer who is in one of these deals can use AI as a, quote, thought partner. Yeah. So first, I have to give you a bit of a disclaimer. I'm a structured finance lawyer, not a venture capital lawyer, but we have one of the largest venture practices in the United States in our firm. So I'm familiar with it. Right. What's the role of the lawyer? So often when our client is the company side or the founders they never done anything like this before And we are educating our client on how deals work They rely on us for market knowledge of kind of what typical deal terms are in a venture financing, whether it's a series C, A, B, C. And because we are so active in the space, we usually know the funds that are investing. We know their counsel. And we can add, I believe, a lot of value to the client in the way we can negotiate and structure the deal, knowing also what the next round is going to look like and the round after that and their ultimate exit, whether it's through an IPO or an M&A deal. And we can help them get their structure right from the get-go so that they'll get all those future rounds right. And when clients come to us with kind of their chat GPT produced documents. I think Echo, you know, they're missing all of that nuance, the knowledge of the market and the other humans in the deal and what they need for future transactions. So happen a lot. Clients coming to you with chat GPT produced stuff in the venture space Yeah I not surprised Right And in a number of areas I mean we have some very sophisticated clients who have the same AI tools we have in the legal profession and some Fortune, for example, a Fortune 50 client that basically produces their first drafts of whether it's a contract or a complaint or an answer in a litigation, they'll produce that using their AI tools first, send it to us, and want us to take that and run with it. On the other end, in another area of our practice, we do a fair amount of patent prosecution work, particularly out of our West Coast offices. We represent some extremely sophisticated technology companies who will have their AI tools reviewing our work and providing us with AI produced comments. Now, we're also using AI tools to create these patents with full knowledge of our clients. So we're almost at the point where their AI agent is talking to our AI agent. That's an interesting development as well. For sure. On the AI co-pilot idea, since your expertise is in structured research, [15:00] finance. How far does this actually go? As in, could you envision asking an LLM to produce a brand new structure for, I don't know, ABS or CMBS or pick your structured finance poison? When you talk about AI, it can help with strategy and complexity. I don't believe that it can come up with a new structure. Okay. But I have a partner who is a real expert in international tax, for example. And he will use AI tools to – and he's always coming up and he represents a lot of family offices and global businesses. And he will come up with – and he spends all of his time structuring how these family offices work. How do you get money from country A to country C with a minimum number of tax hops along the way? He will use AI to test out some of his theories and it will give him some feedback. And he'll work with the AI to come up with new structures. He will then hand the report that he gets out of say Claude and hand it to an associate and say okay now run this down and actually do the research to figure out if this is right But let's work on this structure together. So it increases the horizons. It validates some stuff. What's the opposite of validating? Not validate. Knocks out some ideas. Invalidates, right? Right. And then we still have, you know, a real human who has a law degree, run it down and double check and see if we can make it even better. And usually the iterative process between the A.I. and the human and the associate and the A.I. and the partner come up with some really cool structures that they wouldn't have thought of on their own and that the A.I. wouldn't have thought of on its own. This is going to sound like a rude question. How do you know? Because I think it's like I can see going back and forth with a chatbot as a way to like stress test ideas or sort of like, you know, just sort of like generate these new ideas. And so like I guess on some level I not surprised that by interacting with say Claude a really good tax lawyer could like explore some new potentials or like feel out this possibility But on the other hand it also easy to go back and forth with the chatbot and create the illusion that you're finding something new that you also could have like intuitively arrived at yourself because if you were one of the most brilliant tax lawyers in the world. So how do you sort of establish that this is in fact adding value as opposed to just work-ish or work-light. So in the tax example I just gave you, I have no idea how. But I can tell you, because I was out in our Palo Alto office just the week before last and was talking to our... I'll go back to the patent example, because that's discreet. Right? And it's... And there's a discreet set of steps. We are using an AI tool to help our lawyers draft patent applications. They have been telling me, we've been doing this for a little over six months, that the primary benefit is that it creates a better patent application because while almost all of our patent lawyers are engineers of some type mostly software or electrical engineers the knowledge that the AI tool has is of all engineering right and of all sciences and arts and it able to bring in a biologist's perspective, perhaps just to use something, and create a broader application. I was like, okay, that's kind of cool. I was then visiting with a couple of our clients out there who we do patent work for. And on their own, they have told us that they really appreciate how we're using the technology, that we are ahead of most of the other firms that they use right now, and that they have noticed over the past six months how our applications have gotten better and better. So I didn't ask them that question. Unprompted, they have said that. So I was really excited about that because that was validation because I asked the same question that you do. He was like, how do you know it makes it any better? And do clients notice? But clients noticing, you know, makes a difference. [20:00] up on team gear, finding the right promotional products makes all the difference. 4imprint offers thousands of options, from on-trend apparel and premium drinkware to tech, totes, and giveaways, so you can find the right fit for any audience, purpose, or budget. You can customize it all, your logo, your message, your look, and many items come with no setup charge to help you save. And if you're really watching the bottom line, you'll find standout choices at every price point so you can make a real impact while staying on budget. Plus, you'll get expert help, fast turnaround times, and their 360-degree guarantee. So you can be 4imprint certain your order will arrive on time and look exactly right. Whatever your goal, 4imprint makes it easy to find your perfect promo match. Explore the possibilities today at 4imprint.com. 4imprint. 4certain. Support for this show comes from Public.com. If you're actively involved in your portfolio, you probably catch yourself repeating the same actions. buying the dip manually sweeping idle cash putting on a hedge on public you can now create ai agents that handle all these tasks on your behalf just describe what you want to do in plain english like if the vix hits 25 buy a put option on the s 500 or if my cash balance goes above move the excess into my direct index You approve the workflow and your agent handles the rest Monitoring the market, watching for your conditions, and executing your strategies exactly as defined. An investing platform driven by your intent, not just your clicks. You can also get full read and write access to your account via the public API. Go to public.com slash market and fund your account in five minutes or less. That's public.com slash market. Paid for by Public Investing. Brokered services by Open to the Public Investing, Inc., member FINRA and SIPC. Advisory services by Public Advisors, LLC, SEC Registered Advisor. Complete disclosures available at public.com slash disclosures. Let's talk about health care for a second. It doesn't always work the way people expect it to. If you've ever waited on a prescription refill or had a hard time getting the care you needed, you know the feeling. The system should just work better for everyone. That's exactly what the people at Optum are trying to do every day. They a health care company linking patient care and pharmacy services and using data and technology to drive the whole system so care is connected not complicated for patients and providers Things like making it easier to get care that looks at the whole person from primary care doctors to mental health support and even in care and then using technology to make sure they all work together. Technology designed to help doctors spend less time on busy work and more time with their patients. And those prescriptions? Optum is working to bring costs down, save patients money, and make it easier to get refills. Little by little, Optum is helping make healthcare work as one for everyone. Head to business.optum.com to see how. One thing I'm really interested in is the impact of AI on actual pricing power for law firms. And you wrote a really good piece for our colleagues over at Bloomberg Law called AI will give junior lawyers better work, one of our legal insights. And you mentioned a specific number in there. You cite a project where you were going to do it, but you decided not to because it was too expensive. but then a year later or so with the assistance of ai you said that the cost had come down 70 which is absolutely huge but i i guess my question is if costs are coming down by that much then why don clients accrue all of those cost savings how do lawyers actually protect their margins in that scenario So a couple of ways First off the clients don necessarily have the same knowledge deep industry knowledge that their outside lawyers do in reviewing the work because while the cost has come down 70%, it hasn't come down 100%, right? So we're still doing 30% of the hours that we might have done a few years ago or something like that. But those hours that we're putting in it are much higher level hours. The project that I was talking about in that article is basically a due diligence project where the assignment was to review thousands of trust agreements for the obligations of various parties in the agreements. Usually laborious, kind of boring, tedious. And the price we quoted or the cost that we quoted to do it, I think it was like three years ago, was based on humans doing all of the work and a QC team on top of the first round. We can basically take out... [25:00] out that first layer of review and assign that to the robot, which then puts it all in this beautiful 100 column spreadsheet to show us all of the fields. And we're doing the QC layer. And the QC layer requires legal knowledge of how these transactions work and some experience that you can spot where the output doesn't make sense and you go back and double check it. So I think the clients still value having a law firm basically certify or say this is good output in a way that they wouldn't want to do themselves. They don't have the staffing to do it themselves and they want outside eyes on it. So what does this mean specifically, though, for early career lawyers? Right. Like the fear is or one version of this is like, that's great for the senior lawyers, pull up the ladder, hire a few law grads, et cetera, because you can now do this. Right. Like, is that what does that mean for the people who three years ago would have been doing this? You know as you as you said not the most exciting tedious work Right Again when I started I was blacklining documents by hand And somehow when the machines took that job away from me I still had plenty of work to do And first-year lawyers behind me had plenty of work to do. But my life got more interesting. So I think that the tedious jobs are going away first. Junior lawyers, I think, are going to be doing more interesting work sooner. We have not fully figured out yet, and we will solve, but we haven't fully yet, how to train people without going through that tedious work. Are you hiring 1Ls and 2Ls? We are hiring summer. We have summer associates this summer. We will have a full slate of first year associates who join us in the fall. We have already hired our summer associate class for the summer of 2027. And we have not changed our hiring patterns. We are starting to change the skill set we're looking for. And what you need to be successful in this profession I think will change a little bit Well can you talk then about what is that skill set and what are they going to be doing when they get when they arrive there on their first day after law school What is the skill set they need to first get in the door? And then what are they going to do once they're in the door? I think no matter what, you're going to need some training on the area of practice you're in. And there's going to be less of the grunt work kind of training. and we're ultimately going to have more simulator type training or, you know, what we've already started doing, you know, a very basic, a very standard, like first year training curriculum at a law firm is like the, uh, the anatomy of an M&A transaction where you actually go through the provisions of a merger agreement or an act or a stock purchase agreement or an asset purchase agreement. And you go through it, you know, those provisions. We're now going to, what we've started doing is make sure everybody brings their laptop with them to the training and that they have their ai tools open at the time at the same time and as part of the training you're saying okay now take this provision and put it in this tool and see what the responses are see uh tell it you representing a seller uh and give me a seller favorable provision to negotiate So we going to do a lot more of that training hands with the tools where people will get to see that We will have, they will have in front of them that they don't have now, but they will very soon, our knowledge set of every similar deal that we have done, say, in the past couple of years that have been fully negotiated, they will be able to see the terms of every deal we've recently done. Clients care a lot about our market knowledge. They care what are the reasons that they're willing to pay law firms at the exorbitant rates we charge is because you have at the market knowledge that they can't possibly have on their own because for many of them, this is a one-time deal. For us, we do it every day, and they expect us to know what the market terms are. Using this technology, I can make sure every junior lawyer has access to all of the knowledge of our entire firm in doing similar types of deals or litigation or bankruptcy proceedings as opposed to just having to walk through the halls and try to find somebody who's done it before. [30:00] and just talk to them about what they've done. You will now have access to basically a dashboard with all prior documents. And I think that makes you much smarter, faster. We could have a separate conversation about the cognitive load from that, right? There was a lot of value in my having downtime to black line, right? It was like relaxing. This was going to be my next question, which is lawyers have to be very detail-oriented, right? And one of the arguments for having to do all this grunt work like hand deliver ISDA agreements or review trusts and things like that was that it teaches you to focus. Yeah. And it teaches you to look out for that missing comma that's going to cost you like $5 million. You know, I was thinking about that as you were saying that documentary on Netflix from like 15 years ago about the sushi guy. Or it's like all these chefs who are like, okay, I'm going to teach you to make sushi. We first spend 15 years learning how to clean rice or whatever it is, and then we get to sushi. Or thinking about musicians, and it's like it's no fun to do scales, right? But you do it and you build some sort of deep understanding And that feels like the equivalent of what the black lining Yeah So so what happens if people aren doing as many repetitive tasks I worry about that as well And I not I not sure what the answer is from a behavioral psychology perspective because I think you right Doing those scales is important. I don't know. Yeah. We're talking about simulators to simulate doing a deal. there's no you know there's nothing like actually doing a deal right and actually negotiating with somebody on the other side who's a jerk or somebody on the other side who's really nice but you know bamboozles you by being so nice and the human element of this all is still really important and the human interaction with your client with the adversary you know you need to get that experience and you need to have the touch and feel. With the advances in AI just over the past six months or a year, you can see a time when the drafting is largely delegated to the machine. You still have to be careful. As you said the commas you know I can see AI screwing up commas AIs don know how to add right And they make silly mistakes all the time So you got to learn to be careful I do worry about it becoming too easy to just fall into the trap of trusting what it tells you rather than thinking above it. So I don't know the answer to your question, but I have the same concern you do. I want to get into the various tools and infrastructure you have, because we're a markets and finance podcast, so people want to know what the trade is, obviously. But before we get into that specifically, when we did our last episode about AI law, our guest, Joel Wertheimer, he's a civil rights lawyer in New York City. And one of the points that he made is that sometimes clients come with him with cases, often maybe like suing the city over police misconduct or something like that. and the client has a good case on, might have a good case on, or you're aware of, a client might have a good case on paper, but the potential damages are so small that it's uneconomical to pursue it, even though clearly there's a legitimate case. If we're sort of, as Tracy mentioned, the Jevons paradox of all this do you see this as like okay you collapse the price of certain types of legal work Does that expand the number of theoretical cases or I guess you not deals that oh let's take a look at this deal. There's some like minimum upfront cost that's going to be the same. Let's take a look at this deal that then compensates for the reduced number of hours that you're charged? I think so. I think that the Jevons paradox applies to legal. In the example I gave you with the client who had thousands of trust agreements that needed to be reviewed, when we gave him the first quote, we didn't say we didn't want to do it. The client said, at that price, we're not going to do it because the downside risk isn't high enough to justify the price. But when the price came down by 70%, they said, oh, yeah, at that price, we'll do it. So at and let's just use fake numbers at 10 million dollars, they said, no, we're not going to do it. And we had our revenue was zero when it came down to three million. They said that's worth it. And we had three million of revenue. So that's the Jevons paradox in action. Over the past 25 years, e-discovery has ballooned [35:00] The cost of litigation, to your point, right? Electronic discovery run amok has made it virtually impossible to bring a sophisticated litigation for under a few million dollars in legal fees. And in just the discovery expenses alone are millions of dollars. If we can bring, say, two million of discovery costs down to 200,000, it does change the calculus for both. on this. Lawsuits are a lot more economical to file. Everyone is going to love that aspect of AI. Isn't that fantastic? But you raised it in a virtuous example with a civil rights lawyer, but I think it's also true for large corporates who were abandoning things before. So yeah, You can debate whether it a social good to have more litigation but if I speaking on behalf of a law firm that has a large litigation department I do think that the potential for increased litigation does outweigh the loss of really boring mundane hours reviewing documents I go back to the patent example I gave you earlier where we do see over time the cost of being able to prosecute a patent application coming down dramatically through the use of AI. And that same client who told me two weeks ago that they've really noticed an increase in the quality of a work product also told us that because they're using AI in their own work, they're seeing a quadrupling in the number of inventions that their engineers and scientists are coming up with. And the number of patent requests that they're getting internally from their engineering team has quadrupled. And we're going to ultimately be able to lower the cost of each of those. So the clients are able to do more and create more economic activity, which is virtuous. And a lot of that will require some legal work, even though each unit of legal work will cost less. I expect we'll be doing more of it. So, you know, that should preserve the ability of lawyers to to do OK economically Well how do you see AI shifting I guess the balance of power or balance of work between external law firms versus in lawyers Because you could imagine a scenario where I'm a big company, I have a couple or maybe five ex-big law lawyers who work for me, and because so much of the work is now automated, they can use AI, they don't need to go to a big law firm to do a lot of the grunt work, as you put it earlier. Yeah. And I think you're going to see the other side of the Jevons paradox is you will see larger clients that have dedicated legal teams keeping more of the routine legal work in-house. You mentioned ISDAs before. Most banks, for example, have brought all their ISDA and derivatives work in-house. You'll probably see more of that, and you'll see more end users on the fund side doing that internally rather than sending it out, although we have a terrific derivatives group. You may see that in some other areas as well. But again where you really need market breadth and you need market knowledge I think outside law firms will still be able to provide a lot of value and a lot of outside kind of independent perspective and independent advice that is hard to do inside Whether you're planning a big tech event, launching a new campaign, or just stocking up on team gear, finding the right promotional products makes all the difference. 4imprint offers thousands of options. from on-trend apparel and premium drinkware to tech, totes, and giveaways, so you can find the right fit for any audience, purpose, or budget. You can customize it all, your logo, your message, your look, and many items come with no setup charge to help you save. And if you're really watching the bottom line, you'll find standout choices at every price point so you can make a real impact while staying on budget. Plus, you'll get expert help, fast turnaround times, and their 360-degree guarantee, so you can be 4imprint certain your order will arrive on time and look exactly right. Whatever your goal, 4imprint makes it easy to find your perfect promo match. Explore the possibilities today at 4imprint.com. [40:00] for imprint for certain support for the show comes from public.com if you're actively involved in your portfolio you probably catch yourself repeating the same actions buying the dip manually sweeping idle cash putting on a hedge on public you can now create ai agents that handle all these tasks on your behalf just describe what you want to do in plain english like if the vix hits 25 buy a put option on the s&p 500 or if my cash balance goes above twenty thousand dollars move the excess into my direct index. You approve the workflow and your agent handles the rest. Monitoring the market, watching for your conditions, and executing your strategies exactly as defined. An investing platform driven by your intent, not just your clicks. You can also get full read and write access to your account via the public API. Go to public.com slash market and fund your account in five minutes or less. That's public.com slash market. Paid for by Public Investing. Brokered services by Open to the Public Investing, Inc., member FINRA and SIPC. Advisory services by Public Advisors, LLC, SEC Registered Advisor. Complete disclosures available at public slash disclosures Let talk about healthcare for a second It doesn always work the way people expect it to If you ever waited on a prescription refill or had a hard time getting the care you needed you know the feeling The system should just work better for everyone. That's exactly what the people at Optum are trying to do every day. They're a health care company linking patient care and pharmacy services and using data and technology to drive the whole system so care is connected, not complicated, for patients and providers. Things like making it easier to get care that looks at the whole person, from primary care doctors to mental health support and even in-home care, and then using technology to make sure they all work together. Technology designed to help doctors spend less time on busy work and more time with their patients. And those prescriptions? Optum is working to bring costs down, save patients money, and make it easier to get refills. Little by little, Optum is helping make healthcare work as one for everyone. Head to business.optum.com to see how. Let's talk about your tech stack, so to speak. You mentioned lawyers talking to Claude. I know that there are applications or companies like Harvey, which are sort of like built on top of some of the frontier models There are questions about do you need these application layers I also interested in whether like when it comes to ingesting your own internal data whether, like, open source AI models will play a role in that. Why don't you give us, like, the overview of your firm? Like, what is the sort of essence of your tech stack that you're using? So you actually want me to name products? Yeah, sure. Yeah, why not? Okay. Yeah, you mentioned Claude, but I'm really good. I did mention Claude, but we also – so we use a lot of different products at the moment, probably too many. And over time, that will narrow. We are Harvey Shop. Okay. We – almost all of our users – our lawyers use Harvey regularly. We also use Microsoft Copilot in the Outlook suite or the Office suite. Can you, for our non-lawyer listeners and me, and I assume most of us, what does a Harvey offer that is not offered when someone interfaces with the model directly? With the model being a frontier model? Yeah so instead of using Fable or OPIS 5 or GPT 5 what happens when I use a model that powers Harvey Within Harvey and I think this is true of Legora as well Legora is the other big one Yeah You can pick which frontier model you want to use You can choose Chat or Claude for example And I think Gemini and some of them as well. But what a Harvey adds is, number one, a security layer that is much more robust than the other models. and that is super important for a law firm. We're able to control our own data set and make sure that it's not going up to the internet and that our client information is not going out to the internet at all. We have sessions that do not even connect to the internet. It is much more robust and one of the things that our clients often don't realize is that if they're using, especially a consumer grade CLAW or ChatGPT, They often are losing attorney-client privilege by asking a consumer model their questions. We are, number one, focused on the security layer and the ethics layer, and that's super important. And a Harvey or a Legora or Westlaw has co-counsel, all are very security conscious. So it gives us that. [45:00] allows us to have playbooks that we can share among our team so that people can see the projects that we're building, which I think is harder to share in some of these other models directly. And what at least Harvey provides is basically a rag layer on top of the AI frontier model. Retrieval augmented generation. Wow. Okay. I don't know. Yeah, I don't know. I barely even know what it means, but I think it's cool to say. It's a rag layer. Right. That the that has been trained on legal stuff. OK. Right. So it's much more fine tuned to law. And out of the box, it comes with some of the things lawyers like to do, like look at clauses, merger and acquisition agreement. It has a better understanding of nuance in reviewing deposition transcripts. But we're seeing, you know, some of the main uses are creating a project for if you're a litigator. I give some litigators some airtime to hear not just talk about deals loading all of the pleadings the briefs all of the transcripts and the back and forth between lawyers into a project in a Harvey and then when you are drafting a pleading or a brief it able to retrieve for you the quotes in the deposition transcripts that support what you trying to argue So rather than spending, you know, tens or hundreds of hours trying to find those, it's using the intelligence to find it for you. Claude could do that as well, but Harvey does it, I think, a little better. And it's collecting knowledge and allowing you to share the playbooks more easily. Yeah, I wanted to ask you about the security and privacy aspect of all of this. So how much of AI adoption in the legal industry is actually, I mean, law is a very heavily regulated industry itself. So how much AI adoption is, I guess, constrained by things like malpractice risk, you know, data concerns, accountability? Is that a real limitation for you? So two years ago, in my role, I talked to our malpractice carriers and our underwriters out of Lloyd of London and go visit with them And they always ask questions about the things that they worried about Two years ago they were worried about are you you know are you letting your lawyers use AI? And, you know, is that creating risk for us? Now they're asking, you let your lawyers use AI, right? To make sure they're doing things right, right? Because it's becoming part of mainstream and it is becoming an assumption that you're not going to be doing some of these tasks without having AI assist you in doing them. So it's going from a insurer's saying, oh, don't use that stuff until it's proven, to you have to be using it. And the clients are doing the same thing, right? Clients that 18 months ago were saying, don't use AI for our work, are now saying, well, you've got to be using AI to try to cut down the cost of the work. So we've seen that shift over the past two years pretty dramatically. That has changed. But clearly, you know, we emphasize to our lawyers all the time that you got to be double checking this stuff. And there is no excuse for filing a brief or pleading in a court that cites you know made up hallucinated cases They just too embarrassing It is it is embarrassing We have too many tools at our immediate disposal to be able to catch that And yeah systems break down and somebody trying to draft something and file it in an hour and they get hallucinate cases slip in, but it's not. So I want to ask you a question. I asked Goldman CEO David Solomon this on an episode. One of the things we're seeing actually in a lot of industries, including the legal industry, is like the rise of like these superstar hires. Right. And you see them like reported in various publications. Some big name lawyer gets some crazy salary to go to or signing bonus or whatever, kind of like, you know, like we see in some of the hedge fund space. When I see these things and when I think about AI and you mentioned it's like, oh, here's this knowledge pool or here's every we're going to, the junior lawyers are going to get access to all these templates. And I'm thinking like, if I'm this guy, like, do I want to share all my templates and knowledge with the firm? But if I don't, if everyone thinks that, then these tools don't really work, right? If you don't have a culture of like contributing insight and knowledge back to the firm, there's no franchise value and you're definitely not going to get anything out of AI or in terms [50:00] like meaningful, I would think. Maybe we'll get some. Does this create any tensions? And when you're thinking about the firm, maybe not this year, but down the road, how to make sure that the interests of the firm are aligned with the interests of the partners? That is a huge issue. It's not a technological issue. It's a human issue, right? Lawyers tend to be highly autonomous creatures, right? And successful law firm partners tend to be even more highly autonomous and generally take the view, you're not going to tell me how to practice law. My husband used to get calls from when he was a junior lawyer, he would get calls from one of the partners out on their fishing boat on weekends, assigning work. Very autonomous. Your husband worked at our firm? No, not at all. You think like an equal you kill environment in many cases. Yeah, right. And many people would say, especially at our firm, that one of the secrets to our success is that we, a term we use often is we let the horses run, right? You let people go do their thing and clients love that and love the entrepreneurial energy And now with AI we saying we need you to share all your knowledge So the culture clash there is significant Interesting And I think that's a real challenge. Now, when you talk about the NBA-type hires or NFL-type salaries that you're seeing, perhaps some of the motivation for that is getting their knowledge into the systems, right? If you can get some of these superstars and kind of put their name on your AI tools and say that, you know, so-and-so recently did a deal that had these terms, you know, how's the client's going to be super impressed by that? And maybe that's part of the value add. And maybe that creates even higher comp numbers for superstars. I want to go back to, I guess, my first question about billable hours. In the age of AI, if we're seeing pricing actually compressed, if we're seeing a sort of maybe democratization of legal tools, maybe uncertainty over case results starts to collapse as well. Would you maybe finally see a shift away from the billable hours thing Maybe So and you should Right We should be able to move to project pricing or outcome pricing I think that that will finally take off There will still be a lot of instances where hourly rates apply. And I think you're going to see a continuing huge increase in hourly rates at top tier firms. And I think we're already seeing it. If you look at data from 2025 as published by the American Lawyer, average hourly rates in 2025 went up by 10.1% at the largest firms in the United States. And at the top 20 firms by profitability, they went up even more. And that doesn't make sense in a year when CPI went up by around 3%. I've never seen that kind of gap between CPI and average rates, one of the things that might explain it is that the billable hour has actually become more productive and more valuable because of AI tools. And I think you'll see that continuing at an ever faster pace. So I would expect those hourly rates will continue to go up yet the price for the things that clients actually want to buy which are solutions to their business problems are going to come down because they will take far fewer hours I want to ask you one more question about your tech stack There's a couple other things that are sort of happening. One is obviously this idea that, well, if you use an open source model, unlike, say, Claude, then you can really train, bake all your knowledge into the model itself. And so you're not just sort of doing the retrieval augmented generation where the model essentially is like combined with a search engine, but it's actually like baked in and you can actually do that. And you see, there was a really interesting article, paper from Bridgewater, et cetera, recently, where they talked about doing this. And then of course, like just in general, setting aside open source versus closed source and model, It's like we do see this thing and we mentioned Kirkland Ellis in the beginning. Eventually, you hit a scale where it's like, no, you want to just, you know, you don't just want to be, you know, build your own Harvey or whatever it is and actually internalize some of the infrastructure spend and like actually truly customize this, etc. And arguably, it's never been easier to build software, etc. I'm curious, like, okay, as a $400. [55:00] person firm, obviously, in your view, getting a lot of value already out of AI. What comes the point where you would think, you know what, we want to start building some technology in-house, including perhaps a homegrown version of one of the models trained on your proprietary data? So we definitely want to connect these proprietary models to our data And use that to help train the model. But training from scratch a model, Kirkland announced they're spending $500 million over five years. It costs something like a billion and a half dollars to train a model. One and a half billion. So even that would take Kirkland 15 years of investment. I don't see us really training our own models. I see us doing some customized solutions on top of existing models. I see us certainly doing our playbooks so that there'll be the Lowenstein Sandler M&A playbook for that's customized for our firm and our market knowledge. But I don think we going to really build our own as much as modify and customize Lowenstein LLM That right That not happening We always I mean we may you know we might white label an existing one and come up with a fancy name, but I don't think that's realistic for a firm of our size. I don't think it's even realistic for a Kirkland. So we'll, you know, we'll see how that shakes out. But I don't see us building our own. I do want to mention one thing, though, that's related to both of your points. All of these models that law firms have been using have been at the enterprise level have been basically all you can eat pricing. Yeah. Right. They're all going to be switching to token based pricing. You know, Claude is doing it first. But do you think you could see sicker shock? I was like, oh, I know what I thought we were getting. I thought we were able to do all these trusts at 30 percent of the cost. Right. But it turned out that was subsidized. It turns out there's actually going to be 130 percent of the cost or 90 percent of the cost. So I don't think we yet know what the actual cost is if it's not subsidized by outside investors. I don know that we really know what the optimal combination of labor and capital and robot capital is because we don know the true cost especially if we you know kind of hit this energy grid wall where it just gets more and more expensive And today, July 8th, you do not yet have visibility from any of your partners into like the true cost of this stuff. I mean, if it's like if it if they have to make a profit per token. Right. I we don't know. I know from my IT person who I was talking to yesterday that so far this month, my use of Claude has used up $36. But that's what July 4th weekend in there, right? And it was only on the 7th or only three business days that I was actually using it. So that's not so much. Two minutes of your billable time. Yeah, exactly. But I don't know where that pricing is ultimately going to go. and so we don't know how that story is going to play out two years down the road. Gary Winkins, thank you so much. That was a really helpful conversation. Really appreciate it. Great being here. So thanks for having me. Tracy it really interesting that we really don even know anything about the economics of AI yet because of like how much like the you know these all you can eat models and like what how that all going to shake out I don't know. Well, it reminds me a lot of, you know, what happened with Uber and the food delivery apps where basically venture capital was subsidizing everyone's ability to order food at home. And then when it actually came time to prove the business model and generate some profit, we saw the cost of food delivery go up and use of it go down. Yeah. Also, same with oil. Yeah. And the fact that like capital markets subsidized losses for years and years with oil. And now we're getting, you know, similar examples there. And then on the flip side. So let's say that like, you know, the historical price is like token prices continue to drop at a very high rate. On the flip side, we get a lot more lawsuits. And so on the flip side, like this is like, OK, this is, you know, the Jevons paradox is like good news. It turns out there's going to be plenty of human labor in the future. Bad news. There's going to be 100 more frivolous lawsuits and patent counter challenges that you never had to do. And that's what you're doing now. This is what I worry about on a wider scale. [60:00] Because if we talk about AI as this productivity enhancement, you could have a situation where AI is just used to expand bureaucracy forever and ever and ever. So not just frivolous lawsuits, but like maybe human resources, things like that. I always wanted to write an article about how like human resources ate the U.S. economy. Yeah. Well, think about like, you know, you could imagine like, OK, every worker at a firm gets access to Claude or something like that. And on day one, it's like, oh, this is great. I just condensed eight hours of labor into two hours of labor, et cetera. But then suddenly, all of the other people that they're working with who are fighting for the same promotions also just condensed eight hours of labor into two hours. And so it's like, well, I'm trying to get that promotion. You just find new work. And you just find more and more work. that scenario in which like all of us suddenly feel like life is easier because of ai like i gotta say that feels like you know what i did actually have a um i needed to change a flight recently and change a um a round trip ticket to a three ticket at the last minute and that was like a very daunting thing to me to do on the Delta website And so like I asked like Chad GPD the steps like what, and it like was really helpful. It was like, go to this page in the bottom, right. There will be a thing. This is what you click to like change one leg of the thing. And like, it worked exactly as it. So it's like, there was an example where it's like, okay, this actually eased some psychic attacks. But generally speaking, this future where it's like life just gets much easier because of AI feels like at a minimum, it's a long way off. No, the future is a Jevons paradox for like administrative overhead and busy work. That's that's how I feel at the moment. Shall we leave it there? Let's leave it there. This has been another episode of the All Thoughts podcast. I'm Tracy Alloway. You can follow me at Tracy Alloway. And I'm Joe Weisenthal. You can follow me at The Stalwart. Follow our producers, Carmen Rodriguez at Carmen Armand, Dashiell Bennett at Dashbot, Kale Brooks at Kale Brooks, and Kevin Lozano at Kevin Lloyd Lozano. And for more OddLots content, go to Bloomberg.com slash OddLots where we have a daily newsletter and all of our episodes. And you can chat about all of these topics 24 seven in our discord discord slash OddLots And if you enjoy OddLots if you like it when we talk about the future of the legal industry then please leave us a positive review on your favorite podcast platform And remember, if you are a Bloomberg subscriber, you can listen to all of our episodes absolutely ad-free. All you need to do is find the Bloomberg channel on Apple Podcasts and follow the instructions there. Thanks for listening. Thank you. We be right back at every price point And with their 360 guarantee you can be 4imprint certain your order will show up just right right on time Explore more at 4imprint 4imprint 4certain Healthcare doesn't always work great. If you've ever waited on a refill or couldn't schedule an appointment, you get it. That's the kind of stuff Optum is changing. They're using data and technology to integrate patient care, pharmacy, and everything else. So healthcare is connected, not complicated. What's that look like? Cheaper prescriptions that are easier to get and care that looks at the whole person how you need it. Optum is helping make health care work as one for everyone. Learn more at business.optum.com. Get the news you need in just 15 minutes. Start your day with Bloomberg Daybreak, the podcast with a global view on the stories that matter. I'm Nathan Hager. And I'm Karen Moscow. Join us each morning for curated stories on current events, politics, business, and foreign relations. Plus one conversation on the day's biggest developments all in just 15 minutes. Subscribe to Bloomberg Daybreak for a precise, thoughtful take on the stories that matter. Listen to Bloomberg Daybreak each morning on Apple, Spotify, or anywhere you listen.