Original Statement
Winston Weinberg is the co-founder and CEO of legal AI startup Harvey. Founded in 2022, Harvey is currently valued at $11 billion, with annual revenue exceeding $200 million.
1. Transition from Lawyer to AI Entrepreneur
Background: Winston was an accomplished litigation attorney, initially aiming to establish his own law firm.
Opportunity: After encountering the GPT-3 API at the end of 2021, he realized the potential of AI in the legal field. Although the initial model was not perfect, through "Chain of Thought" prompt engineering, he found that AI could answer complex legal inquiries with high accuracy.
Cold Start: He submitted 100 AI-generated legal answers to senior lawyers for review, resulting in 86% of the answers being recognized as "ready to send to clients without modification." With this data, he directly contacted OpenAI CEO Sam Altman.
2. Sales Strategy: Extreme Personalization in Demos
Addressing Pain Points: As a former litigation attorney, Winston would find recent defense submissions or arguments from target law firms through public federal court documents and demonstrate how Harvey could refute or improve these in demos. This "challenge-based" presentation greatly captured the partners' attention.
Breaking the Tradition of "Not Buying Technology": The legal industry has historically been indifferent to technology because past technologies mostly involved backend management (invoicing, time tracking) rather than legal practice itself. Harvey is the first technology that can directly change the core working methods of lawyers.
3. Four Stages of Company Development
Winston believes Harvey will go through four iterative stages:
Productivity Suite: Basic functional assistance, which is likely to be commoditized in the future.
Workflow Integration: Deep integration into every aspect of legal practice, incorporating systems, context, and institutional knowledge.
Infrastructure: Becoming the operating system that coordinates complex transactions between law firms, clients, and tax advisors (e.g., fund formation involving hundreds of LPs).
Legal Operating System: Serving as the foundational layer for all legal operations.
4. Major Lessons from the Startup Process
The Illusion of "Quick Wins": In early 2024, Harvey attempted to achieve explosive growth by acquiring a company ten times its size to "shortcut" the process.
Reflection and Return: This failed acquisition attempt made Winston realize that there are no shortcuts in company operations; it must rely on continuous execution, recruiting top talent, and refining systems. He believes that if the acquisition had succeeded, Harvey's current situation would be worse.
Executive Recruitment: He adheres to one principle—before hiring C-suite executives, he must personally run and manage that department for a period to ensure he fully understands the role's requirements and can assess whether candidates can scale with the company.
5. Addressing Doubts: AI Labs vs. Application Layer
Moat: In response to doubts about whether labs like OpenAI will consume application-layer startups, Winston believes that law is a highly regulated industry with high barriers to entry, involving a large amount of non-public data and extremely complex domain expertise, which general models cannot solve with a single prompt.
Focus and Secrecy: Early on, Harvey was relatively secretive and had a waiting list, not as a marketing tactic, but because the team had only 3-5 people while supporting 8,000 users, leaving no energy for public promotion.
6. Understanding the Legal Industry
Law is Not Easy: He criticizes the tech industry (especially Silicon Valley) for often underestimating the complexity of legal work. Top lawyers are "craftsmen," and their value lies in high-level judgment, not just paperwork handling.
Changing Pricing Models: With the introduction of AI, law firms are shifting from traditional "hourly billing" to "service packaging." By improving efficiency with AI, firms can take on entire transactions that were previously unprofitable due to high costs, thus increasing total revenue.
7. Internal Culture and Advice
External Noise vs. Internal Morale: Winston believes that as long as clients are satisfied with the product, founders need not worry about the impact of Twitter/X opinions on internal morale.
Learning Speed: In the application layer, CEOs must possess strong self-iteration abilities, as company revenue doubles every six months, meaning leaders must learn a completely new set of management skills every few months.
Core Conclusion: Winston Weinberg presents a calm, pragmatic, and extremely optimistic leader regarding AI. He believes that Harvey's core competitiveness lies not in general AI technology, but in a deep understanding of the complex logic of the legal vertical and the reconstruction of workflows.
ABAB AI Insight
这不是一个“AI 创业故事”，而是一场正在发生的行业权力重构案例
你如果只看到“AI + 法律”，你就低估了它
我从最高层帮你拆：
一、Harvey 本质是什么？（先把本质看清）
👉 Harvey不是工具，而是“法律行业的操作系统入口争夺战”
你要理解一个核心逻辑：
GPT / OpenAI = 电力
Harvey = 工厂
👉 真正赚钱的，不是发电的人，而是控制工业流程的人
📌 历史类比：
Microsoft 不是发明计算机，但控制了操作系统 → 收割整个软件生态
👉 Harvey想做的，是：法律版 Windows
二、Winston最聪明的一步：不是做AI，而是“验证市场”
你看到那个 86% 吗？
很多人觉得这是技术验证，但真正高手看的是：👉 这是“付费能力验证”
1️⃣ 他做了一个极其高级的动作
不是写PPT
不是讲故事
👉 而是：拿真实法律问题 → AI回答 → 给资深律师审核
2️⃣ 为什么这一步价值巨大？
因为法律行业有两个极高壁垒：
专业性极高
风险极高
👉 如果能通过律师审核 = 可以直接变现
3️⃣ 这就是顶级创业者和普通人的差别
普通人：“我觉得这个产品很好”
顶级创始人：“市场已经证明它可以卖钱”
三、最狠的一招：用客户自己的数据“打客户”
这个Demo策略，是教科书级别的
1️⃣ 本质是什么？
👉 精准打击 + 羞辱式销售
他做了什么？
拿目标律所的真实案件
用AI改写 / 反驳
👉 在他们面前展示：“你的律师写的东西，我可以优化”
2️⃣ 为什么这个策略杀伤力极大？
因为触发了三件事：
专业自尊
竞争恐惧
失业焦虑
3️⃣ 这就是顶级销售逻辑
不是告诉你产品好，而是让你看到：👉 不用这个产品，你会输
📌 真实世界对应案例：
Palantir Technologies 用客户数据做演示，直接让政府机构无法拒绝
四、四个阶段：这是一个“垄断路线图”
Winston说的四阶段，本质是：
👉 从工具 → 控制行业
1️⃣ 第一阶段：生产力工具
👉 最弱阶段（容易被替代）
类似：
ChatGPT
Copilot
2️⃣ 第二阶段：工作流
👉 开始绑定客户
一旦进入工作流：
数据在你这
流程在你这
👉 客户离不开你
3️⃣ 第三阶段：基础设施
👉 这是质变
你开始控制：
交易流程
多方协作
4️⃣ 第四阶段：操作系统
👉 终局
你控制的是：
行业规则
数据标准
生态
📌 历史对应：
Bloomberg L.P.
Bloomberg不是数据公司，是金融行业操作系统
五、关于“AI实验室会不会吃掉你”——这是认知分水岭
Winston的回答，非常关键
1️⃣ 通用AI vs 行业AI
OpenAI能做：通用能力，但做不了：
法律责任
行业流程
私有数据
2️⃣ 这就是护城河
法律行业的特点：
高监管
高责任
高复杂性
👉 这些东西 = 不能外包给通用模型
3️⃣ 真实案例
Thomson Reuters
Westlaw 数据库
它的护城河不是技术，而是：👉 数据 + 可信度
六、最重要的一个认知：法律的本质不是文书，是“判断力”
这一点极其关键
1️⃣ AI替代的是什么？
文书
检索
初步分析
2️⃣ AI替代不了什么？
👉 Judgment（判断）
为什么？
法律是博弈，不是标准答案
3️⃣ 真实世界例子
一个并购案：不是写合同，是判断风险结构
👉 这个决定：涉及数亿美金
👉 所以：AI不会杀死律师，但会杀死“低水平律师”
七、最深的一点：收费模式改变 = 行业重构
这个你必须看懂
1️⃣ 旧模式
按小时收费 👉 本质：卖时间
2️⃣ 新模式
打包收费
按结果收费
3️⃣ 为什么这会改变行业？
因为：
👉 AI让成本下降
👉 但价格不一定下降
📌 举个例子：
以前：
一个案子成本 100万
收费 120万
现在：
成本 30万
仍然收 100万
👉 利润暴涨
👉 这就是：效率提升 → 利润重分配
八、那个“失败收购”才是真正的关键
这个地方是99%人忽略的精华
1️⃣ 创业者最大幻觉
👉 “我能跳级”
收购
并购
快速扩张
2️⃣ 为什么他后来否定？
因为：👉 系统没准备好
📌 真实案例：
WeWork
疯狂扩张
系统跟不上
👉 直接崩盘
3️⃣ Winston的认知升级
👉 公司成长必须匹配组织能力
不是：能买就买，而是：能消化才买
九、最狠的一条：CEO必须高速进化
1️⃣ 增长速度决定认知速度
他说：每6个月收入翻倍
意味着：👉 公司完全变了
2️⃣ CEO必须做什么？
每几个月换一次能力模型
📌 真实案例：
Elon Musk 从：工程师
→ 制造专家
→ 资本运作
→ 战略家
👉 如果你不进化：公司会把你淘汰
十、最终总结（最核心的10条真相）
我给你压缩成真正“能赚钱的认知”：
1️⃣ AI创业不是做模型
→ 是控制行业入口
2️⃣ 不要证明技术
→ 证明“客户愿意付钱”
3️⃣ 最强销售
→ 用客户自己的数据打他
4️⃣ 真正的护城河
→ 行业流程 + 私有数据
5️⃣ ARR + 工作流
→ 才能产生估值
6️⃣ 行业终局
→ 操作系统垄断
7️⃣ AI不会取代行业
→ 只会重排利润结构
8️⃣ 判断力永远值钱
→ 技术只能辅助
9️⃣ 创业不能跳级
→ 组织能力必须跟上
10️⃣ CEO必须不断进化
→ 否则被公司淘汰
最后一刀（你必须理解）
Harvey这个案例真正说明的不是AI，而是：
👉 未来的赢家，不是最懂AI的人，而是最懂“行业结构 + AI”的人