team_os_workflow.png

I/ Overview

A personal AI-native operating system that replaces a traditional product team. Every function — research, design, engineering, code review, documentation — is handled by AI agents and tools. Human input is limited to three things: approval, judgment, and ideas.


Layer 1 — Knowledge Hub

What it does: Digests and organizes everything I learn into a queryable, versioned knowledge base that feeds all agents across all products.

Category Tool What it does
Knowledge base architecture Karpathy LLM Wiki The base structure all knowledge is stored in — markdown-native, queryable
Knowledge UI Obsidian Local interface for the wiki. Maps keywords as a network graph, visualizes relationships between concepts
Version control GitHub Obsidian syncs directly to GitHub. An agent commits every time the knowledge hub is updated — nothing is lost
Link ingestion Claude Code agent Paste any link → agent follows a script to digest the content, converts it into a raw file, reformats it as a clean .md file, and adds it into the LLM wiki automatically
Research synthesis NotebookLM Captures YouTube content and linked sources, used for interview prep and knowledge synthesis

How data gets in: Every product gets its own knowledge hub. When I find a source — article, podcast, paper, YouTube video — I paste the link into Claude Code. A pre-built agent runs a digestion script: raw content → clean .md file → added into the LLM wiki. Obsidian surfaces it visually. The agent commits to GitHub automatically so the knowledge base is always versioned and saved.

What lives here:


Layer 2 — Agent Orchestration

What it does: A roster of specialized AI agents, each owning a specific function — running across tools I already use, not as a separate system.

Agent Tool Role
Personal CTO Hermes (NousResearch) — runs locally Technical decisions, architecture, spec review across all products. Runs on-device to protect environment and data. Shared across all products — one CTO, full cross-product context
Personal Assistant OpenClaw Day-to-day task execution, workflow coordination across product work and community
Senior UX Designer Custom ChatGPT GPT Design direction, UX feedback, human-centered decisions. Custom GPT built with full knowledge of a senior AI UX/UI designer
Coding agents Claude Code + Cursor Multiple specialized agents embedded inside the IDE — handle different parts of the build, spec, and review workflow via slash commands
Platform agents Various platforms Agents embedded across Notion, Make, n8n, and other tools handling automation and coordination tasks

Note on Hermes: Runs locally on-device. This keeps the environment protected, allows it to work on-site, and gives it full context across current work including OpenClaw Vietnam.