---
title: "Multi-Agent Specialized Team (Solo Founder Setup)"
type: source
tags: [multi-agent, openclaw, solo-founder, telegram]
date: 2026-04-15
---

## Source File
- [[raw/Agent/usecases/multi-agent-team.md]]

## Summary
- 核心主题：一人公司通过多个专业化 AI Agent 组建虚拟团队，实现 24/7 自动运转
- 问题域：创始人身兼数职导致上下文切换成本高、精力耗散、无法深度工作
- 方法/机制：多 Agent 分工 + 共享内存 + Telegram 统一入口 + 定时主动汇报
- 结论/价值：Agent 团队比单一 Agent 更高效，分工专业化是关键

## Key Claims
- 单一 Agent 无法同时处理战略、代码、营销等多领域任务而不快速填满上下文窗口
- 共享记忆（GOALS.md/DECISIONS.md）+ 私有上下文组合是多 Agent 协作的核心机制
- 定时主动任务（scheduled daily tasks）是 Agent 团队产生真实价值的飞轮
- 从 2 个 Agent 开始（lead + specialist），再按瓶颈扩展至 4 个

## Key Quotes
> "Personality matters more than you'd think: Giving agents distinct names and communication styles makes it natural to 'talk to your team' rather than wrestle with a generic AI" — Trebuh on X
> "Start with 2, not 4: Begin with a lead + one specialist, then add agents as you identify bottlenecks"

## Key Concepts
- [[Multi-Agent Hierarchy]]：Supervisor（战略 Lead）+ Worker（领域专家）+ Validator 的层级分工
- [[共享内存模式]]：共享 GOALS.md/DECISIONS.md + 私有 notes，实现共同上下文与专业积累
- [[Telegram 路由]]：单一 Telegram 群聊，通过 @mention 分发到不同 Agent
- [[定时主动任务]]：Agent 不等待指令，而是按日程主动 surface insights

## Key Entities
- [[Trebuh]]：Solo founder，4 Agent 团队（ Milo/Josh/Marketing/Dev）实践者，OpenClaw Showcase 案例
- [[OpenClaw]]：多 Agent 控制平面，支持 sessions_spawn/sessions_send 通过 Telegram 协调
- [[Claude Opus]]：Milo（Strategy Lead）使用，擅长战略规划和综合协调
- [[Claude Sonnet]]：Josh（Business）使用，快速分析能力匹配数字驱动任务
- [[Gemini]]：Marketing Agent 使用，强于网页研究和长上下文分析

## Connections
- [[Multi-Agent System Reliability]] ← 扩展 ← [[Multi-Agent Hierarchy]]
- [[OpenClaw]] ← 支撑 ← [[Multi-Agent Hierarchy]] 的执行层
- [[Agentic AI]] ← 上位概念 ← [[Multi-Agent Hierarchy]]

## Contradictions
- 与 [[Multi-Agent System Reliability]] 冲突：
  - 冲突点：Multi-Agent System Reliability 强调可靠性（多数投票/对抗辩论/淘汰制），Solo Founder Setup 强调快速交付和主动性
  - 当前观点：专业化分工 + 主动汇报优先于冗余可靠性机制
  - 对方观点：高风险任务需要 Validator/Hierarchy 等可靠性保障机制

## Related Links
- [Trebuh on X](https://x.com/iamtrebuh/status/2011260468975771862)
- [OpenClaw Showcase](https://openclaw.ai/showcase)
- [Anthropic: Building Effective Agents](https://www.anthropic.com/research/building-effective-agents)