返回列表
🧠 阿头学 · 🪞 Uota学 · 💬 讨论题

把 Claude Code 当执行助理：你要的不是工具，是“工作操作系统”

AI 助理的拐点不在“更会聊天”，而在它能把你的碎片输入（语音/会议转录/临时指令）自动沉淀成可检索的组织知识，并且直接代你执行（Slack/Calendar/Linear）——你只管说人话。

2026-01-28 原文链接 ↗
阅读简报
双语对照
完整翻译
原文
讨论归档

核心观点

  • 从“我维护系统”变成“系统维护我” 作者不设计目录、不做模板，直接让 Claude 在一个新目录里自组织：实现细节不可见，使用体验像雇了个永不掉线的执行助理。
  • 会议后处理是最硬的 ROI：转录→纪要→行动项→更新上下文 关键不是写总结，而是自动把 owners、状态、团队信息写进长期知识库，让“下次准备”变成检索问题。
  • 工具接入（Rube/MCP）决定上限：能执行才叫助理 只会建议的 LLM 还是搜索引擎；能发 Slack、改日历、拉 Linear 状态，才是“代你做事”。
  • 并行会话=并行工作流：把注意力从切换成本里赎回来 多个 Claude Code tab 同时跑不同流，像给自己配了多个执行分身。
  • 机构知识会复利，但也会放大风险 全量上下文+自动执行意味着：权限、审计、误发、隐私泄露的代价也会同步变大；需要明确的“哪些可自动、哪些必须确认”。

跟我们的关联

### 👤ATou

  • 海外增长/品牌本质是高频决策+复盘：如果能做到“每次 campaign/渠道/素材的决策记录自动化”，你会在 3 个月后拥有别人抄不走的增长记忆。
  • 你可以把这套思路抽象成「Growth OS」：每天 morning brief（focus/待办/风险）+ 会后自动沉淀 + 自动发周报/PRD/对外叙事草稿。

### 🧠Neta

  • 这是一种很强的 context engineering demo：把输入（转录/指令）结构化落盘，把输出（行动/沟通）工具化执行，中间用检索串起“跨会话连续性”。

讨论引子

1. 你愿意把哪些“执行权限”交给 AI（发消息/改日历/建工单）？哪些必须人确认？ 2. 这类系统的关键护城河是模型更强，还是你的“组织记忆库”更强？ 3. 当知识库增长到 1 万/10 万行时，你更担心检索不准，还是维护成本反噬？

我让 Claude Code 给我打造了一名执行助理：现在我作为 CTO 的工作是什么样

用近乎超人的专注与力量，同时管理 10 名工程师、交付代码，并在 C 级层面运转。

大约三周前，我在 ZAR（@zardotapp）正式全职担任 CTO。我新建了一个目录，给 Claude Code 下了一个简单的提示：

为我创建一个基于 Markdown 的系统，让我能定期运行你——Claude Code——从而成为尽可能最世界级的 CTO。我打算把你当作我的个人执行助理和 CTO 专家。请按你认为合适的方式，把所有内容记录并组织在一系列文件夹中。

我没有设计文件夹结构。我没有创建模板。我只是让 Claude 自己搞定。

我来告诉你，生活在未来是什么样子。

结果

我每天每一场会议都准备得无比充分。

我始终清楚自己的优先级。任何时刻我都能重新找回重心，或者弄明白接下来该聚焦什么。

当有人问起上个月某个决定时，我有完整的上下文。不是模糊的记忆，而是我们当时真正考虑过的替代方案、为何选择了现在这一项，以及涉及了哪些人。

我以前雇过执行助理。很优秀的那种，年薪六位数。有些简直他妈的厉害。（嘿 Mare Contrare，想你！！！）

Claude 更强。比她们所有人。加起来都强。抱歉，各位姑娘。

这个系统给了我超人的能力。我同时管理 10 名工程师、亲自写代码交付，还要以 C 级高管的身份运转。系统把一切都记录下来，在我需要的时候把我需要的东西提上来，并负责执行（发 Slack、更新日历、跟踪行动项），而我无需频繁切换上下文。

我不想再维护一个需要我伺候的工具。我只想跟一个助理对话，让它把一切都办妥。不必思考文件该放哪儿，不必记得去更新文档。只要与一个了解我整个组织、能代表我行动的 AI 自然对话即可。

它实际上如何运作

我始终在这个目录里至少开着一个 Claude Code 会话。白天只要有并行的工作流——比如为多场会议做准备、研究候选人或写博客——我就会再开一些额外的标签页，同时跑多个会话。

而且我不只是打字。只要超过一两句话，我就尽量记得用 Wispr Flow 语音输入。我真就是像跟助理说话一样，对着 Claude 进行意识流输出。“早晨同步。”“准备和 Daniel 的 1:1。”“把这个 PR 的摘要发到 Slack。”感觉就像有了 Jarvis，只不过我仍然被限制在一个终端窗口里。

我已经迫不及待想要摆脱终端的束缚，能在一天之中持续不断地和它聊天了。这显然就是发展方向。

这就是我现在一个普通工作日的样子。

我说一句“早上好”，Claude 就会：

读取我本周的聚焦重点

检查待处理的行动项

获取我今天的真实日程（通过 Rube/MCP 集成）

告诉我哪些事情需要关注

30 秒搞定。我对今天的样子一清二楚。

每次会议后：我只要粘贴会议转录

我用 Gemini 把会议转录成文字。通话结束后，我把转录内容粘贴到 Claude 里。就这么简单。（一两天内我就会把这一步自动化。）

Claude 会自动：

生成会议纪要

提取带负责人的行动项

更新团队名册，把关于成员的新信息补充进去

更新任何其他相关的上下文文件

我不用告诉它该做什么。我不需要考虑内容该放哪儿。它自己就处理好了。

“准备和 Daniel 的 1:1”

在 1:1 之前，我会说“准备和 [name] 的 1:1”。Claude 会读他们的历史记录，回顾近期笔记，检查未完成的行动项，并建议要覆盖的讨论主题。

我带着完整上下文走进每一场 1:1。我敢说我的团队一定注意到了。

“把这条更新发到 Slack”

我需要在工程频道同步一个重要的 PR 或决策。我会说“把这条发到 Slack 的 #engineering”，然后把内容给 Claude。它就发了。完事。

我还让它通过 Slack 代我和所有直接汇报对象沟通，请他们用我的会议预约链接设置周期性 1:1。（他们并不知道那是 AI。）

Twitter 也一样。日历邀请也一样。任何我通过 Rube 连接的服务都一样。

Rube 集成绝对不可或缺。它让我能完全访问日历、Slack、Twitter、Linear 以及所有包含组织信息的其他系统。任何时候我只要说一句“给我下一场会议的笔记”，Claude 就知道该怎么做。

“记录关于 ______ 的决策”

当我做出一个决策时，我会说“记录关于 X 的决策”。Claude 会和我讨论上下文与选项，创建一份结构化的决策记录，并链接到相关背景。

三个月后当有人问“我们为什么从 X 切换到 Y？”时，我有完整的理由记录。不是只有结论，而是当时考虑过的替代方案，以及我们为什么否掉它们。

回头看上面这些，我可能还在低估它。我可以跟这玩意儿聊我的一天，聊我需要做的决策。我可以向它征求建议，并确信它是在汲取人类汇聚的智慧，同时又能访问我的所有相关上下文。这和随手打开 ChatGPT 问个问题完全不是一回事。

我从来不用去想的那部分

有意思的是：我根本不知道 Claude 是怎么组织这些文件的。我的意思是，我当然可以去看。但我没必要。那才是重点所在。

有文件夹。有 Markdown 文件。有一套 Claude 自己想出来的结构。它能用。我也不用去想。

对比 Notion：你总是在纠结“该做成页面还是数据库？”或“该放在哪个 workspace？”；又或者因为半年前选的分类法不再合适而不断重组。

而在这个系统里，实现细节是不可见的。我只需要跟 Claude 说话，它就把一切都处理好。这就是它全部的价值主张。

真实例子

它是对话式的。我不用填表。我不用更新电子表格。我只是像跟一个人类助理说话一样交流。

它能完全访问我的工作工具：日历、Slack、Twitter、Linear。我可以说“我明天日程是什么？”或者“把这条发到 Slack”，它就能直接完成。这就是 Rube/MCP 集成带来的能力，价值无可替代。

它能跨会话保持上下文。所有内容都在版本控制下的 Markdown 文件里。Hooks 负责让这个文件夹的内容与其后端仓库保持同步——那只是我 Github 账号下的一个个人仓库。Claude 能引用几周前的决策；能对比多场 1:1 的笔记；能发现那些我本不会注意到的模式。

系统会随着时间变得更聪明。每次对话都会增加上下文。每个决定都会形成一个参照点。每次团队更新都会让它对“这些人是谁、在乎什么”形成更丰满的画像。

当我需要做战略决策时，我可以说“我们做过哪些类似的决策？”然后拿到真正相关的历史，而不是一页可能有关的文档搜索结果。

它也能应对并行工作。多个 Claude Code 会话意味着我可以同时为三场不同的会议做准备——每场都在自己的标签页里，并且拥有完整上下文。

真实案例

招聘：我粘贴候选人的简历或 LinkedIn。Claude 会更新招聘管道信息，基于团队短板建议面试问题，并通过阅读近期 1:1 来理解团队真正需要什么，从而帮我为筛选电话做好准备。

招聘进展：“把 Stephen 接受我们 offer 的消息发到 Slack 的 #leadership。”完成。我不用切到 Slack。我不会打断思路。Claude 会处理。

绩效问题：比如某位工程师表现不佳。我会说“带我梳理一下和 [Engineer Name] 相关的历史”。Claude 会拉出 1:1 笔记，展示担忧点的演化模式，并引用那份关于期望未达成的决策记录。等到我最终不得不让他离开时，我不是凭模糊的感觉行事——我有反复出现的模式的书面证据。我让 Claude 把一份报告发到公司私密 HR Slack 频道里一个新的 thread，用于讨论离职交接。

规划：“现在阻碍生产力的前三个问题是什么？”Claude 会阅读近期会议纪要，检查待办行动项、Linear 上项目与 issue 的状态，并呈现真实瓶颈，而不是我凭直觉猜测的瓶颈。

比其他系统更好

大多数知识管理系统之所以失败，是因为维护它们本身成了第二份工作。你得记得更新；得把东西归类；得思考系统本身，而不是思考你的真实工作。

这个系统之所以有效，是因为我从来不用去思考系统。我只需要思考我的工作，而系统会把它作为自然对话的副产品捕捉下来。

它不是一份需要我维护的待办清单，也不是一套需要我持续更新的 wiki。它只是 Claude：永远在那里、永远在听、永远在整理，随时准备把我需要的东西准确地拿出来。

这对你意味着什么，亲爱的读者

如果你是一个 CTO 或技术管理者，需要同时扮演多个角色，一边带团队一边交付代码，并且被期望在数十条并行工作流之间始终保持完美上下文——这已经是入场门槛了。

如果你是任何类型的知识工作者——这现在同样是入场门槛。

不一定是我这套系统。Claude 构建的是适合我的大脑和我的角色的东西。基于你的需求，它会为你构建另一套不同的东西。

但你需要某个系统，它能够：

在不需要你刻意思考的前提下捕捉上下文

维护会随时间复利增长的组织知识

让你能完全访问你真正的工作工具

让你用自然语言工作，而不是表单与字段

随着并行工作流扩展

我的建议是：

先开始。在一个全新的目录里打开 Claude Code，告诉它你需要什么，让它自己想出结构。用一周看看会发生什么。如果你是工程师，就把它放进一个私有 Git 仓库；否则就放到 Google Drive、iCloud 或 Dropbox 目录里，让它自动备份。

把工具接上。Rube/MCP 集成是必需的。单是日历访问就值回设置成本。能在不切换上下文的情况下说一句“发到 Slack”或“安排一次会议”，会彻底改变工作方式。

多开会话。一个会话很好；并行开三个会话来处理多条工作流时，你才会真正感受到那种超能力。

真实数字（前三周）发生得非常快。到第三周，你的运转水平就会达到过去不可能的程度。

真实数字（三周后）

让我给你看看这在实践中是什么样。下面是我系统里的真实数据：

已处理并归档 82 份会议纪要

仅 1 月就开了 47 场会议（每天 2+ 场）

18 次有完整上下文和后续跟进记录的 1:1

35 个带负责人和状态的行动项跟踪文件

全公司 23 名成员——每个人都被跟踪记录，共 264 行细致上下文（背景、优势、职业目标、最近对话）

维护了 9 份上下文文档（公司战略、架构、工具、关键相关方）

累计捕捉 11,579 行组织知识

这些都是三周内完成的。与此同时我还在交付代码。与此同时我还要在 CEO 和 CPO 一起的 C 级层面运转。

它能扩展吗？当然能扩展。哪天我觉得它开始卡顿了，我就会说点类似的话：“嘿，你好像因为数据太多有点变慢了，不如你自己想办法处理一下。”

问题解决。

而这还只是版本 1。现在我仍然受限于终端窗口。但通过 Wispr Flow 的语音输入，已经让我感觉像在和 Jarvis 对话。未来会是全天持续不断的对话——不是“使用工具”，而是拥有一个了解你组织一切、能代表你行动的助理。

而且模型只会越来越强。想象一下用 Opus 5 或 6 来做这件事；再配上更好的工具与集成；再来个可穿戴设备？哇。这只是时间问题，大概今年晚些时候就会发生。

它发生的速度也比大多数人意识到的更快。现在还有人争论 AI 到底有没有用。如果他们不赶紧跟上，这帮人就完蛋了。

最终，每个人都会拥有类似的东西。要么你现在就弄明白，获得复利般的收益；要么你会在竞争对手构建出真正能持续、能扩展的竞争优势时落后。ZAR 的每个人都被鼓励采用这种系统，我也在投资请顾问手把手带他们，帮他们跨过最初摸索的学习曲线。我的所有直接汇报对象都被要求采用它；而且既然他们是工程师，就应当自己摸清门道，把它变成他们试验最新前沿技术的个人游乐场。

你也亲自试试

说真的，打开 Claude Code，建一个全新的目录，然后说：“构建一个基于 Markdown 的操作系统，让我像世界级的 _______ 那样工作。”把空白处换成你的角色。看看它会给你做出什么。

未来已经到来，只是分布还不均匀。

关于我：我是 Obie Fernandez，ZAR 的 CTO。我写过多本书，包括 "The Rails Way" 和 "Patterns of Application Development Using AI."。我做软件已经 30 年了，我对我们将要去的方向的兴奋程度，比以往任何时候都更强。

关于 ZAR：我们在为新兴市场打造金融科技基础设施。我们获得了 a16z crypto、Dragonfly、Solana 的创始人以及一大批顶级投资人的支持。如果你是一位资深 Ruby on Rails 产品工程师，想和一个已经摸索出如何用 AI 全面自动化业务的团队一起攻克难题，我们正在招聘。你知道怎么联系我。

链接: http://x.com/i/article/2013947760773390336

相关笔记

Managing 10 engineers, shipping code, and operating at the C-level simultaneously, with superhuman focus and power.

用近乎超人的专注与力量，同时管理 10 名工程师、交付代码，并在 C 级层面运转。

About three weeks ago I started as full-time CTO at ZAR (@zardotapp) . I opened a fresh directory and gave Claude Code a simple prompt:

大约三周前，我在 ZAR（@zardotapp）正式全职担任 CTO。我新建了一个目录，给 Claude Code 下了一个简单的提示：

Create me a markdown-based system where I can regularly run you, Claude Code, that lets me be the best world-class CTO possible. I'm planning to use you as my personal executive assistant and CTO expert. Document everything in a series of folders as you see fit.

为我创建一个基于 Markdown 的系统，让我能定期运行你——Claude Code——从而成为尽可能最世界级的 CTO。我打算把你当作我的个人执行助理和 CTO 专家。请按你认为合适的方式，把所有内容记录并组织在一系列文件夹中。

I didn't design a folder structure. I didn't create templates. I just asked Claude to figure it out.

我没有设计文件夹结构。我没有创建模板。我只是让 Claude 自己搞定。

I'm here to tell you what living in the future looks like.

我来告诉你，生活在未来是什么样子。

The results

结果

I'm prepared for every single meeting. Every day.

我每天每一场会议都准备得无比充分。

I always understand my priorities. At any moment I can get centered or figure out what to focus on next.

我始终清楚自己的优先级。任何时刻我都能重新找回重心，或者弄明白接下来该聚焦什么。

When someone asks about a decision from last month, I have the full context. Not a vague memory. The actual alternatives we considered, why we chose what we did, and who was involved.

当有人问起上个月某个决定时，我有完整的上下文。不是模糊的记忆，而是我们当时真正考虑过的替代方案、为何选择了现在这一项，以及涉及了哪些人。

I've had executive assistants in the past. Good ones, making six figures. Some of them were fucking awesome. (Hey Mare Contrare, miss you!!!)

我以前雇过执行助理。很优秀的那种，年薪六位数。有些简直他妈的厉害。（嘿 Mare Contrare，想你！！！）

Claude is better. Than all of them. Put together. Sorry gals.

Claude 更强。比她们所有人。加起来都强。抱歉，各位姑娘。

This system gives me superhuman capabilities. I'm managing 10 engineers, shipping code myself, and operating as a C-level executive simultaneously. The system keeps track of everything, surfaces what I need when I need it, and handles execution (posting to Slack, updating calendars, tracking action items) without me having to context switch.

这个系统给了我超人的能力。我同时管理 10 名工程师、亲自写代码交付，还要以 C 级高管的身份运转。系统把一切都记录下来，在我需要的时候把我需要的东西提上来，并负责执行（发 Slack、更新日历、跟踪行动项），而我无需频繁切换上下文。

I didn't want another tool to maintain. I wanted to just talk to an assistant and have it handle everything. No thinking about where files go. No remembering to update documents. Just natural conversation with an AI that knows my entire organization and acts on my behalf.

我不想再维护一个需要我伺候的工具。我只想跟一个助理对话，让它把一切都办妥。不必思考文件该放哪儿，不必记得去更新文档。只要与一个了解我整个组织、能代表我行动的 AI 自然对话即可。

How it actually works

它实际上如何运作

I keep at least one Claude Code session open in this directory at all times. During the day, I'll fire up extra tabs with additional concurrent sessions as I have parallel workstreams, like preparing for multiple meetings, researching candidates, or writing blog posts.

我始终在这个目录里至少开着一个 Claude Code 会话。白天只要有并行的工作流——比如为多场会议做准备、研究候选人或写博客——我就会再开一些额外的标签页，同时跑多个会话。

And I don't just type to it. If it's more than a sentence or two I try to remember to use Wispr Flow for voice input. I literally just stream of consciousness to Claude like it's my assistant. "Morning sync." "Prep for 1:1 with Daniel." "Post a summary of this PR to Slack." It feels like having Jarvis, except I'm still limited to a terminal window.

而且我不只是打字。只要超过一两句话，我就尽量记得用 Wispr Flow 语音输入。我真就是像跟助理说话一样，对着 Claude 进行意识流输出。“早晨同步。”“准备和 Daniel 的 1:1。”“把这个 PR 的摘要发到 Slack。”感觉就像有了 Jarvis，只不过我仍然被限制在一个终端窗口里。

Can't wait until I can just chat with it continuously throughout the day without the terminal constraint. That's obviously where this is going.

我已经迫不及待想要摆脱终端的束缚，能在一天之中持续不断地和它聊天了。这显然就是发展方向。

Here's what a normal day looks like right now.

这就是我现在一个普通工作日的样子。

I say "good morning" and Claude:

我说一句“早上好”，Claude 就会：

Reads my weekly focus

读取我本周的聚焦重点

Checks pending action items

检查待处理的行动项

Fetches my actual calendar for today (via Rube/MCP integration)

获取我今天的真实日程（通过 Rube/MCP 集成）

Tells me what needs attention

告诉我哪些事情需要关注

Takes 30 seconds. I know exactly what my day looks like.

30 秒搞定。我对今天的样子一清二楚。

After any meeting: I just paste the transcript

每次会议后：我只要粘贴会议转录

I use Gemini to transcribe meetings. After a call, I paste the transcript into Claude. That's it. (I'll have an automated solution for this within a day or two.)

我用 Gemini 把会议转录成文字。通话结束后，我把转录内容粘贴到 Claude 里。就这么简单。（一两天内我就会把这一步自动化。）

Claude automatically:

Claude 会自动：

Creates meeting notes

生成会议纪要

Extracts action items with owners

提取带负责人的行动项

Updates team roster with anything new learned about people

更新团队名册，把关于成员的新信息补充进去

Updates any other relevant context files

更新任何其他相关的上下文文件

I don't tell it what to do. I don't think about where things go. It just handles it.

我不用告诉它该做什么。我不需要考虑内容该放哪儿。它自己就处理好了。

"Prep for 1:1 with Daniel"

“准备和 Daniel 的 1:1”

Before a 1:1, I say "prep for 1:1 with [name]". Claude reads their history, reviews recent notes, checks pending action items, and suggests topics to cover.

在 1:1 之前，我会说“准备和 [name] 的 1:1”。Claude 会读他们的历史记录，回顾近期笔记，检查未完成的行动项，并建议要覆盖的讨论主题。

I walk into every 1:1 with full context. I'm pretty sure my team notices.

我带着完整上下文走进每一场 1:1。我敢说我的团队一定注意到了。

"Post this update to Slack"

“把这条更新发到 Slack”

I need to update the engineering channel about a significant PR or decision. I say "post this to #engineering on Slack" and give Claude the message. It posts it. Done.

我需要在工程频道同步一个重要的 PR 或决策。我会说“把这条发到 Slack 的 #engineering”，然后把内容给 Claude。它就发了。完事。

I've had it communicate with all my direct reports for me via Slack, asking them to setup recurring 1:1s with my meeting maker link. (They didn't realize it was an AI.)

我还让它通过 Slack 代我和所有直接汇报对象沟通，请他们用我的会议预约链接设置周期性 1:1。（他们并不知道那是 AI。）

Same with Twitter. Same with calendar invites. Same with any other service I've connected via Rube.

Twitter 也一样。日历邀请也一样。任何我通过 Rube 连接的服务都一样。

The Rube integration is absolutely essential. It gives me full access to Calendar, Slack, Twitter, Linear, and everything else that contains organizational information. At any moment I can say "give me notes for the next meeting" and Claude just knows what to do.

Rube 集成绝对不可或缺。它让我能完全访问日历、Slack、Twitter、Linear 以及所有包含组织信息的其他系统。任何时候我只要说一句“给我下一场会议的笔记”，Claude 就知道该怎么做。

"Log decision about ______"

“记录关于 ______ 的决策”

When I make a decision, I say "log decision about X". Claude discusses context and options with me, creates a structured decision record, links to relevant context.

当我做出一个决策时，我会说“记录关于 X 的决策”。Claude 会和我讨论上下文与选项，创建一份结构化的决策记录，并链接到相关背景。

Three months later when someone asks "why did we switch from X to Y?", I have the full rationale documented. Not just the decision, but the alternatives considered and why we rejected them.

三个月后当有人问“我们为什么从 X 切换到 Y？”时，我有完整的理由记录。不是只有结论，而是当时考虑过的替代方案，以及我们为什么否掉它们。

Looking at the above, I'm probably still underselling it. I can talk to this thing about my day, about decisions I need to make. I can ask it advice, and know that it's drawing upon the collected wisdom of humanity, but with access to all of my relevant context. This is not the same as popping into ChatGPT and asking it a question.

回头看上面这些，我可能还在低估它。我可以跟这玩意儿聊我的一天，聊我需要做的决策。我可以向它征求建议，并确信它是在汲取人类汇聚的智慧，同时又能访问我的所有相关上下文。这和随手打开 ChatGPT 问个问题完全不是一回事。

The part I don't ever think about

我从来不用去想的那部分

Here's what's interesting: I have no idea how Claude organized the files. I mean, I could look. But I don't need to. That's the entire point.

有意思的是：我根本不知道 Claude 是怎么组织这些文件的。我的意思是，我当然可以去看。但我没必要。那才是重点所在。

There are folders. There are markdown files. There's some structure Claude came up with. It works. I don't think about it.

有文件夹。有 Markdown 文件。有一套 Claude 自己想出来的结构。它能用。我也不用去想。

Compare that to Notion where you're constantly deciding "should this be a page or a database?" or "which workspace does this belong in?" or reorganizing things because the taxonomy you picked six months ago doesn't fit anymore.

对比 Notion：你总是在纠结“该做成页面还是数据库？”或“该放在哪个 workspace？”；又或者因为半年前选的分类法不再合适而不断重组。

With this system, the implementation is invisible. I just talk to Claude and it handles everything. That's the entire value proposition.

而在这个系统里，实现细节是不可见的。我只需要跟 Claude 说话，它就把一切都处理好。这就是它全部的价值主张。

Real examples

真实例子

It's conversational. I don't fill out forms. I don't update spreadsheets. I just talk like I'm talking to a human assistant.

它是对话式的。我不用填表。我不用更新电子表格。我只是像跟一个人类助理说话一样交流。

It has full access to my work tools. Calendar, Slack, Twitter, Linear. I can say "what's on my calendar tomorrow?" or "post this to Slack" and it just works. This is what the Rube/MCP integration provides and it's priceless.

它能完全访问我的工作工具：日历、Slack、Twitter、Linear。我可以说“我明天日程是什么？”或者“把这条发到 Slack”，它就能直接完成。这就是 Rube/MCP 集成带来的能力，价值无可替代。

It maintains context across sessions. Everything is in markdown files under version control. Hooks handle keeping the contents of this folder in sync with its backing repo, which is just a personal repo on my Github account. Claude can reference decisions from weeks ago. Compare notes from multiple 1:1s. Surface patterns I wouldn't have noticed.

它能跨会话保持上下文。所有内容都在版本控制下的 Markdown 文件里。Hooks 负责让这个文件夹的内容与其后端仓库保持同步——那只是我 Github 账号下的一个个人仓库。Claude 能引用几周前的决策；能对比多场 1:1 的笔记；能发现那些我本不会注意到的模式。

The system gets smarter over time. Every conversation adds context. Every decision creates a reference point. Every team update builds a richer picture of who people are and what they care about.

系统会随着时间变得更聪明。每次对话都会增加上下文。每个决定都会形成一个参照点。每次团队更新都会让它对“这些人是谁、在乎什么”形成更丰满的画像。

When I need to make a strategic decision, I can say "what similar decisions have we made?" and get actual relevant history, not a search results page of possibly-related documents.

当我需要做战略决策时，我可以说“我们做过哪些类似的决策？”然后拿到真正相关的历史，而不是一页可能有关的文档搜索结果。

It scales with parallel work. Multiple Claude Code sessions mean I can prep for three different meetings simultaneously, each in its own tab with full context.

它也能应对并行工作。多个 Claude Code 会话意味着我可以同时为三场不同的会议做准备——每场都在自己的标签页里，并且拥有完整上下文。

Real Examples

真实案例

Hiring: I paste a candidate's resume or LinkedIn. Claude updates its recruiting pipeline information, suggests questions based on gaps in our team, and preps me for the screening call by reading recent 1:1s to understand what the team actually needs.

招聘：我粘贴候选人的简历或 LinkedIn。Claude 会更新招聘管道信息，基于团队短板建议面试问题，并通过阅读近期 1:1 来理解团队真正需要什么，从而帮我为筛选电话做好准备。

Recruiting updates: "Post an update about Stephen accepting our offer to #leadership on Slack." Done. I don't context switch to Slack. I don't lose my train of thought. Claude handles it.

招聘进展：“把 Stephen 接受我们 offer 的消息发到 Slack 的 #leadership。”完成。我不用切到 Slack。我不会打断思路。Claude 会处理。

Performance issues: Say an engineer isn't doing a great job. I say "walk me through the history with [Engineer Name]". Claude pulls up 1:1 notes, shows me the pattern of concerns, references the decision record about expectations not being met. When I eventually have to let them go, I'm not doing it based on vague feelings. I have documented evidence of repeated patterns. I tell Claude to post a report to a new thread on the the company's private HR slack channel to discuss offboarding.

绩效问题：比如某位工程师表现不佳。我会说“带我梳理一下和 [Engineer Name] 相关的历史”。Claude 会拉出 1:1 笔记，展示担忧点的演化模式，并引用那份关于期望未达成的决策记录。等到我最终不得不让他离开时，我不是凭模糊的感觉行事——我有反复出现的模式的书面证据。我让 Claude 把一份报告发到公司私密 HR Slack 频道里一个新的 thread，用于讨论离职交接。

Planning: "What are the top three things blocking productivity right now?" Claude reads recent meeting notes, checks pending action items, status of projects and issues on Linear, and surfaces actual bottlenecks rather than what I think the bottlenecks might be based on hunches.

规划：“现在阻碍生产力的前三个问题是什么？”Claude 会阅读近期会议纪要，检查待办行动项、Linear 上项目与 issue 的状态，并呈现真实瓶颈，而不是我凭直觉猜测的瓶颈。

Better than other systems

比其他系统更好

Most knowledge management systems fail because maintaining them is a second job. You have to remember to update things. You have to organize things. You have to think about the system instead of thinking about your actual work.

大多数知识管理系统之所以失败，是因为维护它们本身成了第二份工作。你得记得更新；得把东西归类；得思考系统本身，而不是思考你的真实工作。

This system works because I never think about the system. I think about my work, and the system captures it as a side effect of natural conversation.

这个系统之所以有效，是因为我从来不用去思考系统。我只需要思考我的工作，而系统会把它作为自然对话的副产品捕捉下来。

It's not a todo list I have to maintain. It's not a wiki I have to keep up to date. It's just Claude, always there, always listening, always organizing, always ready to surface exactly what I need.

它不是一份需要我维护的待办清单，也不是一套需要我持续更新的 wiki。它只是 Claude：永远在那里、永远在听、永远在整理，随时准备把我需要的东西准确地拿出来。

What this means for you, dear reader

这对你意味着什么，亲爱的读者

If you're a CTO or technical manager operating across multiple roles, managing a team while shipping code, and expected to maintain perfect context across dozens of concurrent workstreams, this is table stakes.

如果你是一个 CTO 或技术管理者，需要同时扮演多个角色，一边带团队一边交付代码，并且被期望在数十条并行工作流之间始终保持完美上下文——这已经是入场门槛了。

If you're any kind of knowledge worker, this is now table stakes.

如果你是任何类型的知识工作者——这现在同样是入场门槛。

Not my exact system necessarily. Claude built what works for my brain and my role. It would build something different for you based on your needs.

不一定是我这套系统。Claude 构建的是适合我的大脑和我的角色的东西。基于你的需求，它会为你构建另一套不同的东西。

But you need some system that:

但你需要某个系统，它能够：

Captures context without requiring conscious effort

在不需要你刻意思考的前提下捕捉上下文

Maintains institutional knowledge that compounds over time

维护会随时间复利增长的组织知识

Gives you full access to your actual work tools

让你能完全访问你真正的工作工具

Lets you work in natural language instead of forms and fields

让你用自然语言工作，而不是表单与字段

Scales with parallel workstreams

随着并行工作流扩展

Here's what I'd suggest:

我的建议是：

Just start. Open Claude Code in a fresh directory. Tell it what you need. Let it figure out the structure. Use it for a week and see what happens. If you're an engineer put it in a private Git repo. Otherwise do it in a Google Drive, iCloud, or Dropbox folder so that it's automatically backed up.

先开始。在一个全新的目录里打开 Claude Code，告诉它你需要什么，让它自己想出结构。用一周看看会发生什么。如果你是工程师，就把它放进一个私有 Git 仓库；否则就放到 Google Drive、iCloud 或 Dropbox 目录里，让它自动备份。

Connect your tools. Rube/MCP integration is essential. Calendar access alone is worth the setup. Being able to say "post to Slack" or "schedule a meeting" without context switching is game-changing.

把工具接上。Rube/MCP 集成是必需的。单是日历访问就值回设置成本。能在不切换上下文的情况下说一句“发到 Slack”或“安排一次会议”，会彻底改变工作方式。

Run multiple sessions. One session is good. Three concurrent sessions for parallel work is when you feel the superpower.

多开会话。一个会话很好；并行开三个会话来处理多条工作流时，你才会真正感受到那种超能力。

The actual Numbers (Three Weeks In)happens fast. By week three, you're operating at a level that wasn't possible before.a

真实数字（前三周）发生得非常快。到第三周，你的运转水平就会达到过去不可能的程度。

The actual numbers (three weeks in)

真实数字（三周后）

Let me show you what this looks like in practice. Here's the real data from my system:

让我给你看看这在实践中是什么样。下面是我系统里的真实数据：

82 meeting notes processed and filed

已处理并归档 82 份会议纪要

47 meetings in January alone (2+ per day)

仅 1 月就开了 47 场会议（每天 2+ 场）

18 documented 1:1s with full context and follow-up

18 次有完整上下文和后续跟进记录的 1:1

35 action item tracking files with owners and status

35 个带负责人和状态的行动项跟踪文件

23 team members everyone in the company, tracked with 264 lines of detailed context (background, strengths, career goals, recent conversations)

全公司 23 名成员——每个人都被跟踪记录，共 264 行细致上下文（背景、优势、职业目标、最近对话）

9 context documents maintained (company strategy, architecture, tools, stakeholders)

维护了 9 份上下文文档（公司战略、架构、工具、关键相关方）

11,579 total lines of institutional knowledge captured

累计捕捉 11,579 行组织知识

That's in three weeks. While also shipping code. While also operating at the C-level with the CEO and CPO.

这些都是三周内完成的。与此同时我还在交付代码。与此同时我还要在 CEO 和 CPO 一起的 C 级层面运转。

Will it scale? Yes, of course it will scale. If at some point it seems to be bogging down, I'll say something like "Hey you seem to be bogging down due to too much data, why don't you do something about it".

它能扩展吗？当然能扩展。哪天我觉得它开始卡顿了，我就会说点类似的话：“嘿，你好像因为数据太多有点变慢了，不如你自己想办法处理一下。”

Problem solved.

问题解决。

And this is just version 1. Right now I'm constrained to a terminal window. But voice input via Wispr Flow already makes it feel like talking to Jarvis. The future is continuous conversation throughout the day. Not "using a tool." Just having an assistant who knows everything about your organization and can act on your behalf.

而这还只是版本 1。现在我仍然受限于终端窗口。但通过 Wispr Flow 的语音输入，已经让我感觉像在和 Jarvis 对话。未来会是全天持续不断的对话——不是“使用工具”，而是拥有一个了解你组织一切、能代表你行动的助理。

Also the models are only going to get better. Imagine this with Opus 5 or 6. With better tools and integrations. In some sort of wearable? Wow. It's just a matter of time, probably later this year.

而且模型只会越来越强。想象一下用 Opus 5 或 6 来做这件事；再配上更好的工具与集成；再来个可穿戴设备？哇。这只是时间问题，大概今年晚些时候就会发生。

And it's happening faster than most people realize. There are still people out there debating whether AI is useful or not. They're so fucked if they don't get with the program.

它发生的速度也比大多数人意识到的更快。现在还有人争论 AI 到底有没有用。如果他们不赶紧跟上，这帮人就完蛋了。

Everyone will have something like this eventually. Either you figure it out now and get the compounding benefits, or you fall behind while your competition builds competitive advantages that actually persist and scale. Everyone at ZAR is being encouraged to adopt this kind of system, and I'm investing in consultants to hold their hand and get them past the initial learning curve of figuring out. All of my direct reports are required to adopt it, and since they're engineers they are expected to figure it out and make it their personal playground for experimenting with the latest cutting-edge techniques.

最终，每个人都会拥有类似的东西。要么你现在就弄明白，获得复利般的收益；要么你会在竞争对手构建出真正能持续、能扩展的竞争优势时落后。ZAR 的每个人都被鼓励采用这种系统，我也在投资请顾问手把手带他们，帮他们跨过最初摸索的学习曲线。我的所有直接汇报对象都被要求采用它；而且既然他们是工程师，就应当自己摸清门道，把它变成他们试验最新前沿技术的个人游乐场。

Give it a try yourself

你也亲自试试

Seriously, just open Claude Code in a brand new directory, and say "build a markdown-based operating system that makes me operate like a world-class _______". Fill in the blank with your role. See what it comes up with.

说真的，打开 Claude Code，建一个全新的目录，然后说：“构建一个基于 Markdown 的操作系统，让我像世界级的 _______ 那样工作。”把空白处换成你的角色。看看它会给你做出什么。

The future is here. It's just not evenly distributed yet.

未来已经到来，只是分布还不均匀。

About me: I'm Obie Fernandez, CTO at ZAR. I've written multiple books including "The Rails Way" and "Patterns of Application Development Using AI." I've been building software for 30 years and I'm more excited about where we're going than I've ever been.

关于我：我是 Obie Fernandez，ZAR 的 CTO。我写过多本书，包括 "The Rails Way" 和 "Patterns of Application Development Using AI."。我做软件已经 30 年了，我对我们将要去的方向的兴奋程度，比以往任何时候都更强。

About ZAR: We're building fintech infrastructure for emerging markets. Backed by a16z crypto, Dragonfly, the founders of Solana, and whole slew of other top-tier investors. If you're a senior Ruby on Rails product engineer who wants to work on hard problems with a team that's figured out how to fully automate its business using AI, we're hiring. You know how to reach me.

关于 ZAR：我们在为新兴市场打造金融科技基础设施。我们获得了 a16z crypto、Dragonfly、Solana 的创始人以及一大批顶级投资人的支持。如果你是一位资深 Ruby on Rails 产品工程师，想和一个已经摸索出如何用 AI 全面自动化业务的团队一起攻克难题，我们正在招聘。你知道怎么联系我。

Link: http://x.com/i/article/2013947760773390336

链接: http://x.com/i/article/2013947760773390336

相关笔记

I told Claude Code to build me an executive assistant. This is what my work as CTO looks like now

  • Source: https://x.com/obie/status/2013955736292704342?s=46
  • Mirror: https://x.com/obie/status/2013955736292704342?s=46
  • Published: 2026-01-21T12:43:59+00:00
  • Saved: 2026-01-28

Content

Managing 10 engineers, shipping code, and operating at the C-level simultaneously, with superhuman focus and power.

About three weeks ago I started as full-time CTO at ZAR (@zardotapp) . I opened a fresh directory and gave Claude Code a simple prompt:

Create me a markdown-based system where I can regularly run you, Claude Code, that lets me be the best world-class CTO possible. I'm planning to use you as my personal executive assistant and CTO expert. Document everything in a series of folders as you see fit.

I didn't design a folder structure. I didn't create templates. I just asked Claude to figure it out.

I'm here to tell you what living in the future looks like.

The results

I'm prepared for every single meeting. Every day.

I always understand my priorities. At any moment I can get centered or figure out what to focus on next.

When someone asks about a decision from last month, I have the full context. Not a vague memory. The actual alternatives we considered, why we chose what we did, and who was involved.

I've had executive assistants in the past. Good ones, making six figures. Some of them were fucking awesome. (Hey Mare Contrare, miss you!!!)

Claude is better. Than all of them. Put together. Sorry gals.

This system gives me superhuman capabilities. I'm managing 10 engineers, shipping code myself, and operating as a C-level executive simultaneously. The system keeps track of everything, surfaces what I need when I need it, and handles execution (posting to Slack, updating calendars, tracking action items) without me having to context switch.

I didn't want another tool to maintain. I wanted to just talk to an assistant and have it handle everything. No thinking about where files go. No remembering to update documents. Just natural conversation with an AI that knows my entire organization and acts on my behalf.

How it actually works

I keep at least one Claude Code session open in this directory at all times. During the day, I'll fire up extra tabs with additional concurrent sessions as I have parallel workstreams, like preparing for multiple meetings, researching candidates, or writing blog posts.

And I don't just type to it. If it's more than a sentence or two I try to remember to use Wispr Flow for voice input. I literally just stream of consciousness to Claude like it's my assistant. "Morning sync." "Prep for 1:1 with Daniel." "Post a summary of this PR to Slack." It feels like having Jarvis, except I'm still limited to a terminal window.

Can't wait until I can just chat with it continuously throughout the day without the terminal constraint. That's obviously where this is going.

Here's what a normal day looks like right now.

I say "good morning" and Claude:

Reads my weekly focus

Checks pending action items

Fetches my actual calendar for today (via Rube/MCP integration)

Tells me what needs attention

Takes 30 seconds. I know exactly what my day looks like.

After any meeting: I just paste the transcript

I use Gemini to transcribe meetings. After a call, I paste the transcript into Claude. That's it. (I'll have an automated solution for this within a day or two.)

Claude automatically:

Creates meeting notes

Extracts action items with owners

Updates team roster with anything new learned about people

Updates any other relevant context files

I don't tell it what to do. I don't think about where things go. It just handles it.

"Prep for 1:1 with Daniel"

Before a 1:1, I say "prep for 1:1 with [name]". Claude reads their history, reviews recent notes, checks pending action items, and suggests topics to cover.

I walk into every 1:1 with full context. I'm pretty sure my team notices.

"Post this update to Slack"

I need to update the engineering channel about a significant PR or decision. I say "post this to #engineering on Slack" and give Claude the message. It posts it. Done.

I've had it communicate with all my direct reports for me via Slack, asking them to setup recurring 1:1s with my meeting maker link. (They didn't realize it was an AI.)

Same with Twitter. Same with calendar invites. Same with any other service I've connected via Rube.

The Rube integration is absolutely essential. It gives me full access to Calendar, Slack, Twitter, Linear, and everything else that contains organizational information. At any moment I can say "give me notes for the next meeting" and Claude just knows what to do.

"Log decision about ______"

When I make a decision, I say "log decision about X". Claude discusses context and options with me, creates a structured decision record, links to relevant context.

Three months later when someone asks "why did we switch from X to Y?", I have the full rationale documented. Not just the decision, but the alternatives considered and why we rejected them.

Looking at the above, I'm probably still underselling it. I can talk to this thing about my day, about decisions I need to make. I can ask it advice, and know that it's drawing upon the collected wisdom of humanity, but with access to all of my relevant context. This is not the same as popping into ChatGPT and asking it a question.

The part I don't ever think about

Here's what's interesting: I have no idea how Claude organized the files. I mean, I could look. But I don't need to. That's the entire point.

There are folders. There are markdown files. There's some structure Claude came up with. It works. I don't think about it.

Compare that to Notion where you're constantly deciding "should this be a page or a database?" or "which workspace does this belong in?" or reorganizing things because the taxonomy you picked six months ago doesn't fit anymore.

With this system, the implementation is invisible. I just talk to Claude and it handles everything. That's the entire value proposition.

Real examples

It's conversational. I don't fill out forms. I don't update spreadsheets. I just talk like I'm talking to a human assistant.

It has full access to my work tools. Calendar, Slack, Twitter, Linear. I can say "what's on my calendar tomorrow?" or "post this to Slack" and it just works. This is what the Rube/MCP integration provides and it's priceless.

It maintains context across sessions. Everything is in markdown files under version control. Hooks handle keeping the contents of this folder in sync with its backing repo, which is just a personal repo on my Github account. Claude can reference decisions from weeks ago. Compare notes from multiple 1:1s. Surface patterns I wouldn't have noticed.

The system gets smarter over time. Every conversation adds context. Every decision creates a reference point. Every team update builds a richer picture of who people are and what they care about.

When I need to make a strategic decision, I can say "what similar decisions have we made?" and get actual relevant history, not a search results page of possibly-related documents.

It scales with parallel work. Multiple Claude Code sessions mean I can prep for three different meetings simultaneously, each in its own tab with full context.

Real Examples

Hiring: I paste a candidate's resume or LinkedIn. Claude updates its recruiting pipeline information, suggests questions based on gaps in our team, and preps me for the screening call by reading recent 1:1s to understand what the team actually needs.

Recruiting updates: "Post an update about Stephen accepting our offer to #leadership on Slack." Done. I don't context switch to Slack. I don't lose my train of thought. Claude handles it.

Performance issues: Say an engineer isn't doing a great job. I say "walk me through the history with [Engineer Name]". Claude pulls up 1:1 notes, shows me the pattern of concerns, references the decision record about expectations not being met. When I eventually have to let them go, I'm not doing it based on vague feelings. I have documented evidence of repeated patterns. I tell Claude to post a report to a new thread on the the company's private HR slack channel to discuss offboarding.

Planning: "What are the top three things blocking productivity right now?" Claude reads recent meeting notes, checks pending action items, status of projects and issues on Linear, and surfaces actual bottlenecks rather than what I think the bottlenecks might be based on hunches.

Better than other systems

Most knowledge management systems fail because maintaining them is a second job. You have to remember to update things. You have to organize things. You have to think about the system instead of thinking about your actual work.

This system works because I never think about the system. I think about my work, and the system captures it as a side effect of natural conversation.

It's not a todo list I have to maintain. It's not a wiki I have to keep up to date. It's just Claude, always there, always listening, always organizing, always ready to surface exactly what I need.

What this means for you, dear reader

If you're a CTO or technical manager operating across multiple roles, managing a team while shipping code, and expected to maintain perfect context across dozens of concurrent workstreams, this is table stakes.

If you're any kind of knowledge worker, this is now table stakes.

Not my exact system necessarily. Claude built what works for my brain and my role. It would build something different for you based on your needs.

But you need some system that:

Captures context without requiring conscious effort

Maintains institutional knowledge that compounds over time

Gives you full access to your actual work tools

Lets you work in natural language instead of forms and fields

Scales with parallel workstreams

Here's what I'd suggest:

Just start. Open Claude Code in a fresh directory. Tell it what you need. Let it figure out the structure. Use it for a week and see what happens. If you're an engineer put it in a private Git repo. Otherwise do it in a Google Drive, iCloud, or Dropbox folder so that it's automatically backed up.

Connect your tools. Rube/MCP integration is essential. Calendar access alone is worth the setup. Being able to say "post to Slack" or "schedule a meeting" without context switching is game-changing.

Run multiple sessions. One session is good. Three concurrent sessions for parallel work is when you feel the superpower.

The actual Numbers (Three Weeks In)happens fast. By week three, you're operating at a level that wasn't possible before.a

The actual numbers (three weeks in)

Let me show you what this looks like in practice. Here's the real data from my system:

82 meeting notes processed and filed

47 meetings in January alone (2+ per day)

18 documented 1:1s with full context and follow-up

35 action item tracking files with owners and status

23 team members everyone in the company, tracked with 264 lines of detailed context (background, strengths, career goals, recent conversations)

9 context documents maintained (company strategy, architecture, tools, stakeholders)

11,579 total lines of institutional knowledge captured

That's in three weeks. While also shipping code. While also operating at the C-level with the CEO and CPO.

Will it scale? Yes, of course it will scale. If at some point it seems to be bogging down, I'll say something like "Hey you seem to be bogging down due to too much data, why don't you do something about it".

Problem solved.

And this is just version 1. Right now I'm constrained to a terminal window. But voice input via Wispr Flow already makes it feel like talking to Jarvis. The future is continuous conversation throughout the day. Not "using a tool." Just having an assistant who knows everything about your organization and can act on your behalf.

Also the models are only going to get better. Imagine this with Opus 5 or 6. With better tools and integrations. In some sort of wearable? Wow. It's just a matter of time, probably later this year.

And it's happening faster than most people realize. There are still people out there debating whether AI is useful or not. They're so fucked if they don't get with the program.

Everyone will have something like this eventually. Either you figure it out now and get the compounding benefits, or you fall behind while your competition builds competitive advantages that actually persist and scale. Everyone at ZAR is being encouraged to adopt this kind of system, and I'm investing in consultants to hold their hand and get them past the initial learning curve of figuring out. All of my direct reports are required to adopt it, and since they're engineers they are expected to figure it out and make it their personal playground for experimenting with the latest cutting-edge techniques.

Give it a try yourself

Seriously, just open Claude Code in a brand new directory, and say "build a markdown-based operating system that makes me operate like a world-class _______". Fill in the blank with your role. See what it comes up with.

The future is here. It's just not evenly distributed yet.

About me: I'm Obie Fernandez, CTO at ZAR. I've written multiple books including "The Rails Way" and "Patterns of Application Development Using AI." I've been building software for 30 years and I'm more excited about where we're going than I've ever been.

About ZAR: We're building fintech infrastructure for emerging markets. Backed by a16z crypto, Dragonfly, the founders of Solana, and whole slew of other top-tier investors. If you're a senior Ruby on Rails product engineer who wants to work on hard problems with a team that's figured out how to fully automate its business using AI, we're hiring. You know how to reach me.

Link: http://x.com/i/article/2013947760773390336

📋 讨论归档

讨论进行中…