The moment you realize your perfectly organized Second Brain is useless. You spent months building it. PARA structure, atomic notes, progressive summaries. 2,000 ideas captured, categorized, cross-linked. You open a new chat with an AI assistant and start cold. "Let me explain how I organize my notes..." Again. Every session begins the same way. You burn tokens and patience re-establishing ground that never changed. Your vault, your carefully curated 2,000-note vault, lives on the other side of a conversation that has to start from zero. You are not querying a knowledge base. You are re-explaining your filing system to someone who forgot it. You have a category problem, not a tool problem. You built a database. A database is what you query through one interface, one session at a time. What you need is a substrate: a set of conventions that any number of agents can read, write, and operate on without renegotiating the rules each session. Claude Code solved part of this with CLAUDE.md. Persistent context loads at session start. But the moment you add a second agent, the cold-start problem returns in a new costume. I now run three agents against the same PARA vault. Terminal for deep work. WhatsApp for quick capture. Discord for structured collaboration. Each reads from and writes to the same files using the same conventions. The conventions live in `CLAUDE.md` at the vault root. Every agent reads it on start. All three append to today's daily note. None of them ask where to put it. The vault is the source of truth. Any agent must conform to it. Never the reverse. The shift is architectural, not technical. You stop asking "which tool do I use" and start asking "which agent should I send." Your Second Brain becomes a substrate: shared memory for agents to operate on. Once you build it that way, the tools stop mattering.