> Large language models can’t do much with raw notes or scattered documents. Actually, they thrive with raw notes and scattered documents. I hoard project info

> Large language models can’t do much with raw notes or scattered documents. Actually, they thrive with raw notes and scattered documents. I hoard project information in Obsidian with deliberately very little structure: everything lives in a single directory, with optionally one category as a frontmatter value, and very light linking. I do not have to think when adding to this vault at all, it's a low-friction drop box. But an agent in this infodump can do absolute wonders using just search.