企业从云数据仓库（cloud data warehouses）中学到了一课：他们把数据和算力都交给了别人，之后却眼睁睁看着自己业务中最具战略价值的资产——运营方式——变成了别人的杠杆。这才给 Iceberg（格式/生态）创造了机会。 “骗我一次就够了……” 如今的企业领导者已经意识到，公司需要一个全新的 AI 系统记录层（system of record），以上下文数据库（context database）**的形式来服务 AI agents。 上下文数据库分为两类（如第一张图）： 1. 运营型上下文数据库（Operational Context Databases） 用于存储标准操作流程（SOP）和组织内部知识，例如： * 客户打电话要求重置密码时应该怎么处理 * 法务在审查新客户的 NDA 时该如何走流程 * HR 如何回答新员工关于期权归属（vesting）的提问 这些流程都代表着企业的商业机密与知识产权，是关键资产。通过把这些内容从员工脑中提取出来并固化，可确保流程的连续性，并构建可持续、可防御的资产。 2. 分析型上下文数据库（Analytical Context Databases） 这是对“语义层（semantic layer）”的进一步演化： 它们存储对关键指标的定义与计算方式，例如： * 收入（revenue） * 客户获客成本（CAC） 语义层告诉 AI：“数据是什么”，而分析型上下文数据库则教 AI：“如何推理这些数据” AI 代理架构：协调器 + 反馈回路 Steven Talbot 最近在 Omni 上的 agentic analytics 架构文章指出： 系统中有一个协调器（coordinator）机制，会根据： * 当前的问题 * 返回的结果 * 已尝试的方法 * 来决定下一步使用哪些工具。 运营型与分析型上下文数据库的关键不在于“数据库”本身，而在于内部的反馈循环（feedback loops）。 Steven 的系统能够： * 在执行过程中动态调整 * 出错时自动重试 * 在得到有用结果时主动停止 这会形成一个不断提升准确率的循环：更高准确率 → 更高信任 → 更高采用率 → 更多反馈 → 更好的数据库。 拥有最佳反馈循环的企业，将构建出最有价值的上下文数据库。 上下文数据库 = AI 驱动的下一代流程自动化基础设施 上下文数据库将推动未来的流程自动化，体现 AI 在劳动体系中的真正价值。 这是 RPA（机器人流程自动化）的进化版： RPA + 流程发现（process discovery） + 非确定性（non-determinism） 非确定性对于 AI agents 的成功至关重要： * 它允许处理例外情况 * 避免第一代 RPA 系统遇到的失败模式 * AI agents 擅长摄取大量内容并对其进行推理 从人工上下文构建 → 自动化上下文平台：不可逆趋势 企业很快将从人工构建上下文的方式，全面转向自动化上下文平台。 上下文数据库未来将： * 以独立产品形式销售 * 或作为平台功能进行捆绑 最终，企业将从这一转型中受益：拥有可随时间不断进化、不断变得更好的系统。 Enterprises learned a lesson from cloud data warehouses. They handed over both data & compute, then watched as the most strategic asset in their business, how they operate, became someone else’s leverage, which created an opportunity for Iceberg. Fool me once… Leaders have recognized their companies need a new system of record for AI agents in the form of a context database. There are two different kinds of these context databases : (first image) Operational context databases store standard operating procedures & institutional knowledge : when a customer calls about resetting a password, when legal reviews an NDA with a new prospect, when HR answers questions about options vesting for a new hire. All of these processes represent trade secrets & intellectual property, which are key assets for a business. Capturing them from employees ensures continuity in processes & builds a sustainable, defensible asset. Analytical context databases are a semantic evolution of semantic layers : they contain definitions & calculations for metrics like revenue or customer acquisition cost. Semantic layers told AI what data meant. Analytical context databases teach AI how to reason about it. Steven Talbot’s recent piece on Omni’s agentic analytics architecture describes : a coordinator mechanism, which decides which tool to use next based on the question, the results, & what’s already been tried. The key to both operational & analytical context databases isn’t the databases themselves. It’s the feedback loops within them. Steven’s system adapts mid-flight, retries when things break, or stops when it has something useful to show. This creates an ever-improving cycle of accuracy. Accuracy creates trust. Trust creates adoption. Adoption creates more feedback. Companies that develop the best feedback loops will build the most valuable context databases. Context databases enable the future of process automation, representing the real promise of AI within the workforce. It’s the evolution of RPA (robotic process automation), but it’s RPA & process discovery injected with non-determinism. This non-determinism is essential for the success of AI agents. It allows for exception handling, forestalling one of the failure modes of the first generation of RPA. AI agents are excellent at ingesting large volumes of content & reasoning about them. The move from manual context engineering to automated context platforms is inevitable. Context databases will be sold as standalone products & bundled. Enterprises will come out of this transformation for the better : with evolving systems that improve over time.
目录
观点原文
企业从云数据仓库（cloud data warehouses）中学到了一课：他们把数据和算力都交给了别人，之后却眼睁睁看着自己业务中最具战略价值的资产——运营方式——变成了别人的杠杆。这才给 Iceberg（格式/生态）创造了机会。
“骗我一次就够了……”
如今的企业领导者已经意识到，公司需要一个全新的 AI 系统记录层（system of record），以上下文数据库（context database）**的形式来服务 AI agents。
上下文数据库分为两类（如第一张图）：
- 运营型上下文数据库（Operational Context Databases）
用于存储标准操作流程（SOP）和组织内部知识，例如：
- 客户打电话要求重置密码时应该怎么处理
- 法务在审查新客户的 NDA 时该如何走流程
- HR 如何回答新员工关于期权归属（vesting）的提问
这些流程都代表着企业的商业机密与知识产权，是关键资产。通过把这些内容从员工脑中提取出来并固化，可确保流程的连续性，并构建可持续、可防御的资产。
- 分析型上下文数据库（Analytical Context Databases）
这是对“语义层（semantic layer）”的进一步演化：
它们存储对关键指标的定义与计算方式，例如：
- 收入（revenue）
- 客户获客成本（CAC）
语义层告诉 AI：“数据是什么”，而分析型上下文数据库则教 AI：“如何推理这些数据”
AI 代理架构：协调器 + 反馈回路
Steven Talbot 最近在 Omni 上的 agentic analytics 架构文章指出：
系统中有一个协调器（coordinator）机制，会根据：
- 当前的问题
- 返回的结果
- 已尝试的方法
- 来决定下一步使用哪些工具。
运营型与分析型上下文数据库的关键不在于“数据库”本身，而在于内部的反馈循环（feedback loops）。
Steven 的系统能够：
- 在执行过程中动态调整
- 出错时自动重试
- 在得到有用结果时主动停止
这会形成一个不断提升准确率的循环：更高准确率 → 更高信任 → 更高采用率 → 更多反馈 → 更好的数据库。
拥有最佳反馈循环的企业，将构建出最有价值的上下文数据库。
上下文数据库 = AI 驱动的下一代流程自动化基础设施
上下文数据库将推动未来的流程自动化，体现 AI 在劳动体系中的真正价值。
这是 RPA（机器人流程自动化）的进化版：
RPA + 流程发现（process discovery） + 非确定性（non-determinism）
非确定性对于 AI agents 的成功至关重要：
- 它允许处理例外情况
- 避免第一代 RPA 系统遇到的失败模式
- AI agents 擅长摄取大量内容并对其进行推理
从人工上下文构建 → 自动化上下文平台：不可逆趋势
企业很快将从人工构建上下文的方式，全面转向自动化上下文平台。
上下文数据库未来将：
- 以独立产品形式销售
- 或作为平台功能进行捆绑
最终，企业将从这一转型中受益：拥有可随时间不断进化、不断变得更好的系统。
Enterprises learned a lesson from cloud data warehouses. They handed over both data & compute, then watched as the most strategic asset in their business, how they operate, became someone else’s leverage, which created an opportunity for Iceberg.
Fool me once…
Leaders have recognized their companies need a new system of record for AI agents in the form of a context database. There are two different kinds of these context databases : (first image)
Operational context databases store standard operating procedures & institutional knowledge : when a customer calls about resetting a password, when legal reviews an NDA with a new prospect, when HR answers questions about options vesting for a new hire.
All of these processes represent trade secrets & intellectual property, which are key assets for a business. Capturing them from employees ensures continuity in processes & builds a sustainable, defensible asset.
Analytical context databases are a semantic evolution of semantic layers : they contain definitions & calculations for metrics like revenue or customer acquisition cost.
Semantic layers told AI what data meant. Analytical context databases teach AI how to reason about it.
Steven Talbot’s recent piece on Omni’s agentic analytics architecture describes :
a coordinator mechanism, which decides which tool to use next based on the question, the results, & what’s already been tried.
The key to both operational & analytical context databases isn’t the databases themselves. It’s the feedback loops within them.
Steven’s system adapts mid-flight, retries when things break, or stops when it has something useful to show. This creates an ever-improving cycle of accuracy. Accuracy creates trust. Trust creates adoption. Adoption creates more feedback. Companies that develop the best feedback loops will build the most valuable context databases.
Context databases enable the future of process automation, representing the real promise of AI within the workforce. It’s the evolution of RPA (robotic process automation), but it’s RPA & process discovery injected with non-determinism.
This non-determinism is essential for the success of AI agents. It allows for exception handling, forestalling one of the failure modes of the first generation of RPA. AI agents are excellent at ingesting large volumes of content & reasoning about them.
The move from manual context engineering to automated context platforms is inevitable. Context databases will be sold as standalone products & bundled. Enterprises will come out of this transformation for the better : with evolving systems that improve over time.
相关观点
- “[硅谷] 是这个星球上最大、也最不稳定的原始资本主义培养皿。” 那么，一位长期生活在这里、并且取得成功的做空投资者能给出什么经验教训？ 作为 Crown Capital Management 的创始人，Scott Fearon 曾做空 200 多家公司的股票。在其著作《Dead Companies Walking（行尸走肉般的公司）》中，Fearon 将自己 30 年来与企业高管会面的经验，提炼为 六种失败模式： 尽管这些公司各不相同，但它们最终都失败了，因为其领导者犯下了我重点关注的六个共同错误中的一个或多个： 只从最近的过去学习。 过度依赖某一套成功公式。 误读或疏远了客户。 陷入群体狂热（泡沫/风潮）。 未能适应行业中的“板块级”结构性转变。 在身体或情感上脱离了一线运营。 从 SaaS 向 AI 的“急转弯”，放大了上述每一种错误。 1️⃣ 只从最近的过去学习 软件行业以 20 年为周期演进：大型机 → 客户端/服务器 → SaaS。 每一波浪潮都会带来牛市；当浪潮退去，增长放缓、估值倍数坍塌。 SaaS 的估值倍数已连续三年持平。 2️⃣ 过度依赖成功公式 销售方式已经发生变化： * 产品驱动增长（PLG）的爆发 * 数百万美元级大单的增多 * 许多围绕效率与配额的旧规则，已经不再适用。 3️⃣ 误读或疏远客户 客户希望成为 AI 原生（AI-native）。 他们愿意为培训与解决方案支付高额费用。 41% 的 AI 支出为新增预算，整体预算正在爆炸式增长。 如果继续销售 已无法解决客户痛点的软件，那就是通往高流失率的捷径。 4️⃣ 陷入狂热 AI 的炒作周期，制造了“赶快上线半成品功能”的压力。 宣布一个 AI 路线图 ≠ 交付真正的价值。 随着长期运行的 AI 代理在 2026 年进入劳动力体系，这种差异将愈发明显。 5️⃣ 未能适应板块级结构性转变 如果这句话在任何时候成立，那就是今天。 6️⃣ 在身体或情感上脱离一线 不每天使用 AI 的团队，会低估变化速度。技术进步太快，季度级战略复盘已经跟不上。 如果领导层没有每天亲自向 Claude 或 GPT 提问，那么他们已经落后。 认知偏差总是最难在自己身上察觉。在这个 AI 时代的关键节点，做空者提供的那面“镜子”，尤其值得一照。 “[Silicon Valley] is the biggest, most volatile petri dish of raw capitalism on the planet.” So what lessons does a successful short-seller living here have? As founder of Crown Capital Management, Scott Fearon has shorted the stocks of over 200 companies. In his book Dead Companies Walking, Fearon distills three decades of meetings with executives into six failure modes : Despite their differences, they all failed because their leaders made one or more of six common mistakes that I look for : 1. They learned from only the recent past. 2. They relied too heavily on a formula for success. 3. They misread or alienated their customers. 4. They fell victim to a mania. 5. They failed to adapt to tectonic shifts in their industries. 6. They were physically or emotionally removed from their companies’ operations. The hairpin turn from SaaS to AI amplifies each. Learning only from the recent past. Software moves in 20-year waves. Mainframe, then client/server, then SaaS. When these waves come, they create bull markets. As they wane, growth flattens & multiples collapse. SaaS multiples have been flat for three years. Relying too heavily on a formula for success. Sales motions have changed because of dramatic growth rates in product-led growth & multi-million dollar lands. Many of the rules around efficiencies & quotas no longer apply. Misreading or alienating customers. Customers want to be AI native. They are willing to pay huge sums for education & solutions. Budgets have exploded with 41% of AI spend net new. Selling the same software application that no longer solves the customer’s pain point is a path to churn. Falling victim to a mania. The AI hype cycle creates pressure to ship half-baked features. Announcing an AI roadmap isn’t the same as delivering value. This distinction will become increasingly stark as long-running agents enter the workforce in 2026. Failing to adapt to tectonic shifts. If ever this were true, it’s true today. Being physically or emotionally removed. Teams who don’t use AI daily miss the pace of change. The technology moves too fast for quarterly strategy reviews. If leadership isn’t prompting Claude or GPT every day, they’re already behind. Cognitive biases are always hard to see in ourselves. A short seller’s mirror is a useful one at this moment in AI.Tomasz Tunguz · 2026/01/16
- 自从我看到软件工程师使用 AI，我就开始羡慕了。 我看过他们把 15 到 20 个编码任务分配给一个 AI，然后去打桌上足球、喝咖啡，再回来检查 agent 的工作成果。 那个工程师相当于把自己的时间放大了 10 到 15 倍。 为什么商界人士不能也这样做？ 我看 Twitter 上的 AI 文章有很多问题想问，但没空深入研究。我常常从会议里走出来时手上就有五个任务要处理，还有内部实验想跑。 于是我把 Gemini 的命令行接入 Asana。 现在我可以像哈特曼军士那样下达命令，一群 agent 会在几分钟内回应，调用我构建的所有工具。 这和在浏览器里用 AI 完全不一样。 agent 可以访问我的整个系统：Asana、邮件、日历、CRM。它能读取我的文件，从工具中抓取数据，并替我执行操作。agent 实际上是在我的工作环境里行动。 我的产能从每天 10 个任务上升到 31 个任务。 当我在开会时，agent 正在替我工作。 今天早上我想到一个主意：对我的 Asana 使用情况进行统计分析。一个 agent 写了调用 Asana API 保存数据的代码。 另一个 R 脚本生成这些图表。我们在 Asana 的评论区来回讨论可视化呈现。 这些任务并不简单；很多任务有 30 到 40 条评论，代表对原始想法的迭代。 接着我用 SCQA 格式创建了大纲。 五分钟内，这篇帖子就被写出来了。 另一个 agent 检查文章确保符合我的写作风格。 这就是 AI 的力量：当我们处于“被动状态”时（我正在健身房），AI 在并行完成工作。 当我们给 agent 正确的工具，工作速度会瞬间改变。 现在商界人士也能做到软件工程师几个月前才发现的事情： 我们可以同时运行 15 条并行任务流，并把每天的产出提升 3 倍，而不用延长工作时间。 所谓的 “Agent 拐点（Agent Inflection Point）” 不再是未来，它已经来了。我们将在下周开源我们的工作。 Since I watched software engineers using AI, I've become jealous. I've seen them assign 15–20 coding tasks to an AI, play foosball, grab coffee, then return to evaluate the agent's work. That engineer has paralleled her time by 10x to 15x. Why can't a business person do the same? I read Twitter articles about AI and have questions but no time to delve deeper. I pop out of meetings with five tasks to process. I have ideas for experiments to run internally. So I hooked up Gemini command line to my Asana. Now I can bark out orders like Gunnery Sergeant Hartman and have a team of agents respond in minutes. Accessing all the tools I've built. This is different than accessing AI in the browser. The agent has access to my systems: Asana, email, calendar, CRM. It can read my files, pull data from my tools, and act on my behalf. The agent works within my environment. I've gone from producing 10 to 31 tasks per day. Agents are working on my behalf while I am in meetings. I had an idea this morning: run a statistical analysis on my Asana usage. An agent wrote code to hit the Asana API and save the data. An R script generated these charts. We went back and forth in the Asana comments on visualizations. These aren't simple tasks; many of them have 30 or 40 comments, representing iterations on the original idea. Then I created an outline in the SCQA format. Within five minutes, this post had been written. Another agent graded the post and ensured it matched my style. This is the power of AI: parallel work while we're passive (I was at the gym). When we give agents the right tools, the velocity of work changes instantly. Business people can now do what software engineers discovered months ago. We can spin up 15 parallel workstreams and 3x our daily output without working longer hours. The Agent Inflection Point isn't coming. It's here. We will be open-sourcing this next week.Tomasz Tunguz · 2025/12/06
- Hannah Stolberg 与 Team OS：把团队隐性知识变成 AI 可调用的组织操作系统Hannah Stulberg · 2026/04/17
- Evan Spiegel：拒绝被收购，不做流量生意，只押注下一代计算平台Evan Spiegel · 2026/04/16