In this video I answer the question I get more than any other: “Tom, aren’t you just depending on Claude now?” Fair question—and it forced me to get really precise about portability, data ownership, and what it means to actually own your knowledge system. My setup is simple on purpose: two folders on my computer (business and personal). AI handles the organization, filing, and connections—but the foundation lives locally in my folder structure, file names, and process docs. Claude is the brain, not the system. If Anthropic disappeared tomorrow, I’d lose the brain, but I wouldn’t lose a single file. I’d just point a new AI at the same folders and keep going. Then I make the controversial claim: this is more portable than an Obsidian vault. Yes, Obsidian is “local-first,” but the useful structure often depends on Obsidian-specific conventions—wiki links, Dataview frontmatter, plugins, graph metadata. Move the markdown elsewhere and the knowledge graph collapses. With AI-organized files, I’m optimizing for meaning: clean text, descriptive names, and comprehension over software quirks. My co-founder Paco independently arrived at the same pattern (I quit Heptabase; he’s using less Tana), and that’s the point: when the foundation is right, the AI brain can change without breaking your workflow.