I build AI-native tools, agent workflows, and personal operating systems.
我在构建 AI 原生工具、Agent 工作流，以及个人操作系统。

I care about software that actually enters daily life: tools that help people think clearer, execute with less friction, and turn messy context into repeatable systems.

我关心的不是“看起来很 AI”的 demo，而是真正会进入日常生活和工作流的软件：帮助人更清楚地思考、更低摩擦地行动，并把混乱上下文变成可复用系统。

Right now, I’m especially interested in one question:

What happens when Codex, Feishu, GitHub, CLI tools, and agent skills become one continuous operating layer for a person?

我现在最感兴趣的问题是：

当 Codex、飞书、GitHub、CLI 工具和 Agent skills 连接成一个人的连续操作层，会发生什么？

• Agent-first workflows: turning real human workflows into structured, callable infrastructure. Agent-first 工作流：把真实的人类工作流变成结构化、可调用的基础设施。
• Personal operating systems: using tasks, docs, journals, reviews, and automations to support self-direction. 个人操作系统：用任务、文档、日志、复盘和自动化支持自我驱动。
• Productized judgment: packaging taste, diagnosis, growth thinking, and execution rules into reusable tools. 判断力产品化：把审美、诊断、增长思维和执行规则封装成可复用工具。
• AI-native product management: learning how to go from idea, build, launch, feedback, and monetization as one loop. AI 时代产品经理：学习如何把 idea、开发、上线、反馈和赚钱闭环连起来。

The common thread is simple:

I like products that reduce coordination cost, not just manual effort.

这些项目背后的共同点很简单：

我喜欢减少协作成本的软件，而不只是减少手工操作的软件。

A lightweight self-evolution system centered on action, reality feedback, thought snapshots, and weekly pattern review.

一个轻量的自我进化系统，用行动、现实反馈、想法快照和每周模式复盘来追踪长期变化。

Recently, it has grown from a journaling tool into the GitHub memory layer of a wider personal operating system:

最近它已经不只是一个记录工具，而是在变成一套更大的个人操作系统里的 GitHub 长期记忆层：

• Feishu Task/Todo handles execution. 飞书 Task/Todo 负责具体执行。
• Feishu Docs/Wiki preserves context and thinking. 飞书文档/知识库保留上下文和思考。
• Codex routes, distills, and automates with permission. Codex 负责分流、提炼和授权后的自动化。
• GitHub stores long-term patterns and system rules. GitHub 保存长期模式和系统规则。

An agent-first CLI for BOSS 直聘 workflows, from job search and filtering to recruiter operations and reporting.

一个围绕 BOSS 直聘工作流设计的 agent-first CLI，覆盖职位搜索、筛选、招聘流程支持和报告生成。

The product idea here is that AI should not sit on top of a workflow as a chat box. The workflow itself should become structured enough for agents to operate.

这个项目的核心判断是：AI 不应该只是挂在工作流上的聊天框，工作流本身就应该变得足够结构化，让 Agent 可以直接操作。

• A good AI product is not “AI + interface”; it is a redesigned workflow. 好的 AI 产品不是“AI + 界面”，而是重新设计过的工作流。
• Agents need structure, memory, tools, and permissions, not just better prompts. Agent 需要结构、记忆、工具和权限，而不只是更好的提示词。
• CLI, docs, tasks, skills, and automations are all product surfaces. CLI、文档、任务、skills 和自动化，本质上都可以是产品界面。
• Personal systems are worth building because real workflows are always messy first. 个人系统值得被认真构建，因为真实工作流一开始永远是混乱的。
• Product taste matters more in the AI era, not less. AI 时代，产品审美不是变得不重要，而是更重要。

• building a Feishu + Codex + GitHub personal COO system 构建飞书 + Codex + GitHub 的 Personal COO 系统
• turning repeated work into reusable agent skills 把重复劳动沉淀成可复用的 Agent skills
• learning global growth, distribution, and monetization loops 学习出海增长、分发和商业化闭环
• designing small AI-native products that can actually be used every day 设计真正能被每天使用的小型 AI 原生产品

Build tools that earn a place in someone’s daily workflow.
做那种值得进入用户日常工作流的工具。

Make good judgment operational.
让好的判断力变成可执行系统。

Keep the interface simple, but make the underlying system serious.
界面可以简单，但底层系统要认真。

Ship, observe reality, and let feedback update the system.
发布，观察现实反馈，然后让系统随之更新。

• GitHub: @Fmaverick

If you’re building AI tools, agent workflows, or opinionated software for real human systems, I’d love to see what you’re working on.

如果你也在做 AI 工具、Agent 工作流，或者服务真实人类系统的强观点软件，我很愿意看看你在做什么。