Triggers
Activate on: "automate my company", "agents for my business", "replace team with AI", "company structure with agents", "which skills do I need".
Different from: startup (methodology) and business (strategy). This is about building the organization itself.
Core Flow
- 1. Discovery — What does the company do? What functions exist?
- Mapping — Each function → existing skill, custom skill, or hybrid
- Sequence — Quick wins first, customer-facing later
- Iteration — Run → review → adjust → expand
Discovery Questions
Ask before recommending anything:
- - What's the core value you deliver to customers?
- What functions exist today? (sales, support, ops, finance, marketing, legal, HR)
- Where does work pile up? What's the bottleneck?
- What do you actually want to do yourself?
Function Mapping
For each function, determine approach:
| Approach | When |
|---|
| Install existing skill | Common function (email, CRM, support) |
| Create custom skill |
Unique to this business |
| Hybrid | Existing skill + company-specific rules |
| Human + agent assist | Needs judgment, agent handles prep |
See functions.md for common mappings.
Building Sequence
Build in order of impact, not org chart:
- 1. Internal ops — Low risk, clear inputs/outputs
- Support — After internal proves reliable
- Sales — After support is stable
- Strategy — Agents assist, humans decide
See patterns.md for organizational structures.
Iteration Protocol
After each function is delegated:
- 1. Run 1-2 weeks with human oversight
- Review: what did the agent miss?
- Adjust scope based on errors
- Reduce oversight only when stable
Never hand off completely on day one.
See iteration.md for tracking template.
Learning System
As the company evolves, capture:
- - Decisions made and why
- Adjustments to agent scope
- What worked vs what failed
- New skills needed
This becomes the company's operational memory.
Red Flags
Stop and reassess:
- - Automating trust-building → agents assist, humans close
- Delegating legal/compliance decisions → agents draft, lawyers approve
- No clear function boundaries → define before automating
- Expecting 100% automation immediately → set realistic timeline
触发条件
激活条件：自动化我的公司、为我的业务配置智能体、用AI替换团队、智能体公司架构、我需要哪些技能。
区别于： 初创公司（方法论）和业务（战略）。本技能聚焦于构建组织本身。
核心流程
- 1. 发现 — 公司做什么？存在哪些职能？
- 映射 — 每项职能 → 现有技能、自定义技能或混合模式
- 排序 — 先速赢，后客户导向
- 迭代 — 运行 → 审查 → 调整 → 扩展
发现性问题
在推荐任何方案前先提问：
- - 你为客户提供的核心价值是什么？
- 目前存在哪些职能？（销售、支持、运营、财务、市场、法务、人力资源）
- 哪些环节工作积压？瓶颈在哪里？
- 你真正想亲自处理什么？
职能映射
针对每项职能，确定处理方式：
| 方式 | 适用场景 |
|---|
| 安装现有技能 | 通用职能（邮件、CRM、支持） |
| 创建自定义技能 |
该业务独有需求 |
| 混合模式 | 现有技能 + 公司特定规则 |
| 人工+智能体辅助 | 需判断力，智能体负责准备工作 |
常见映射方案参见 functions.md。
构建顺序
按影响程度而非组织架构图构建：
- 1. 内部运营 — 低风险，输入/输出清晰
- 支持 — 内部运营验证可靠后实施
- 销售 — 支持稳定后实施
- 战略 — 智能体辅助，人类决策
组织架构模式参见 patterns.md。
迭代协议
每项职能委派后：
- 1. 人工监督运行1-2周
- 审查：智能体遗漏了什么？
- 根据错误调整范围
- 仅当稳定后减少监督
切勿在第一天完全移交。
跟踪模板参见 iteration.md。
学习系统
随着公司发展，记录：
- - 决策内容及原因
- 智能体范围的调整
- 有效与无效的做法
- 所需的新技能
这将成为公司的运营记忆库。
警示信号
遇到以下情况需暂停并重新评估：
- - 自动化建立信任 → 智能体辅助，人类完成交易
- 委派法律/合规决策 → 智能体起草，律师审批
- 职能边界不清晰 → 先定义再自动化
- 期望立即实现100%自动化 → 设定现实时间表
