When there’s too much information, how you work with it decides everything.
I spent a long time trying to build the perfect Obsidian — splitting everything into atomic notes, linking, tagging, cross-referencing. In practice, finding and recalling what I needed was hard: too many fragments, not enough meaning.
LLMs changed that. I stopped looking at notes as an encyclopedia and started looking at them as context an LLM can work with. Now I have fewer notes, but each one is heavier — whole topics with ideas, links, summaries and conclusions; directories actively used to keep related material together.
Claude Code has become something like a notes manager for me: it helps me plan, search, connect, write. I just describe the task — it finds the right fragments, the relevant prompts, the older thoughts.
Claude Code is a personal preference, but the same approach works with any agent. The point is that Obsidian becomes a piece of context that’s written not only for me but also for the agents.
The main thing I learned: there’s no universal system. Everyone has a different thinking style, attention span, and approach to structure.
My notes aren’t a “clean knowledge base” — they’re a slightly structured chaos. I don’t have the discipline to maintain strict systems, but AI tools finally make it possible to live with that.