記事一覧へ
【OpenClawでCEO補佐官を構築 — 人間のChief of Staffを超えたAIエージェント】
VCのRyan Sarverが構築したAI Chief of Staff「Stella」の全容:
1. メモリ層: 日次メモ(memory/YYYY-MM-DD.md) + MEMORY.md。フラットなMarkdownファイルで管理
2. Kaizen改善ループ: 毎週金曜にcronでOpenClawコミュニティをリサーチ。日曜に改善レビュー
3. ミーティング準備: 60分前にWhatsAppでブリーフ(過去の議事録、メール、アクション項目)
4. タスク管理: Markdown→Todoist同期。毎晩タスクスイープで遅延パターンを検出
5. ステークホルダーCRM: 全関係者の永続コンテキスト(関係履歴、コミットメント、最終接点)
6. メール/カレンダートリアージ: 経費レシート自動取得、旅程生成、フォローメール下書き
7. 運用リズム: 朝9時ブリーフ、夕方6時ラップ
設計原則: LLMは判断、スクリプトは決定的処理。この分離が信頼性の鍵。

OpenClawAIエージェントagent-opsbusiness-model
OpenClawでCEO補佐官を構築 — 人間のChief of Staffを超えたAIエージェント
♥ 66↻ 3
原文を表示 / Show original
How I built a chief of staff on OpenClaw that's better than any human I've hired.
I'm a VC in the middle of a fundraise, sitting on boards, helping portfolio companies, and angel investing on the side. What I have now is more capable than any human chief of staff I've ever worked with.
Key components:
1. Memory Layer: Daily notes (memory/YYYY-MM-DD.md) as raw log + MEMORY.md curated by the agent. All in flat markdown files, not a database.
2. Kaizen Loop: Weekly cron job scans OpenClaw community for new patterns. Sunday review of findings. System gets measurably better on a cadence.
3. Meeting Prep: 60 min before external meetings, brief arrives via WhatsApp with prior notes, email threads, action items. For LP meetings: pipeline stage, deck version, prior questions.
4. Task Management: Structured markdown as source of truth, important items synced to Todoist. Evening task sweep flags overdue items and patterns.
5. Stakeholder CRM: Persistent context on every person, company, project. Relationship history, commitments, last touchpoint.
6. Email/Calendar Triage: Auto-pulls expense receipts, generates travel itineraries, drafts follow-up emails.
7. Operational Rhythm: Morning brief at 9am, evening wrap at 6pm via WhatsApp.
Design rule: LLMs handle judgment, scripts handle everything deterministic.