Summary
Cole Medin walks through the next stage of his popular “AI second brain” project: turning a personal knowledge-base agent into a shared “team brain” that multiple collaborators can safely use. He explains why a simple Obsidian vault or flat markdown store breaks down at team scale, and lays out the architecture changes needed — centralized ingestion, expanded retrieval strategies, source citation, and, critically, permissioning so not everyone can see every document.
The video demonstrates a working example built with Oracle’s AI Database, showing how documents get labeled by source domain (e.g., “ops Slack” or “marketing wiki”) at ingestion time, and how row-level security policies and access tokens are used to enforce who can query what through an MCP server. Medin also details a “who knows” function that identifies the right person to follow up with based on document authorship.
Importantly, he argues teams don’t need to abandon their personal AI agents to build this — the team brain becomes a shared knowledge layer that individual agents query, preserving each person’s customized personal system. Viewers get a high-level but concrete blueprint for scaling a personal AI knowledge system into an enterprise-ready, access-controlled team resource, along with a linked working example they can hand to a coding agent.
📺 Source: Cole Medin · Published September 17, 2026
🏷️ Format: Hands On Build
