For the last three months I’ve been building an AI second brain on top of Claude Code, and it’s gotten to the point where it saves me at least a dozen hours a week. In this video I walk you through the actual components: how I talk to it through Slack like a coworker, how it drafts replies in my voice, keeps up with my calendar/tasks, helps me ideate and write YouTube scripts, and how the whole thing compounds over time because it keeps learning from my real work. I start with the “elephant in the room”: why build your own instead of running something like OpenClaw. The core issue is what I call the lethal trifecta—private data access + untrusted content + an exfiltration vector—which second brains basically have by default. So the win is control: I keep the architecture simple (markdown + Python + Obsidian), define permissions up front, and evolve capabilities with a zero-trust mindset (read-only first, then carefully add write actions like Gmail drafts). Then I break down the architecture: a memory layer (soul.md/user.md/memory.md), hooks for session start and pre-compact logging, daily reflection + promotion into memory.md, RAG over daily logs in SQLite, skills as the “arms,” and a proactive heartbeat that checks my world every 30 minutes and pings me in Slack.