Everyone wants an AI second brain right now. And almost everyone who builds one watches the same thing happen: their AI gets dumber. Here is why, and what actually works.
The second brain that works is boring. Notes in a knowledge base. AI pulls them through MCP. Freshness flags. Provenance on every edit. Short curated notes, not transcripts. No graph pipelines, no personality vectors.
@paulg Shipping fast and knowing what to ship used to be the same skill. AI broke that link.
When anyone can ship in an afternoon, speed stops being the moat. The constraint moves up the stack to judgment: what to build, for whom, and why.
@Chicky_Think@physicsandpicts Obsidian is local first. Querying it with an LLM from mobile needs a host. I build https://t.co/8WgS3rUacj for this: hosted knowledge base + MCP server.
@lucasteske@vmesel@avelinorun@logseq Perfect use case for Hjarni. Obsidian import is easy, and the AI reads and updates your notes directly over MCP. Ping me if you want a hand getting your game design db across.
@aiglesias_@TheThetaGuy@rezoundous Basically my old setup, minus syncing the vault. I build Hjarni: one hosted KB, MCP built in, single retrieval point for every harness.
Every time you use Claude:
• Your best prompts stay buried
• Your insights live in chats
• Your workflows disappear
You’re rebuilding your brain every day.