I've taught 50 Claude trainings in the last 6 months, to engineers, execs, and operators at some of the best companies in the world. The part almost all of them still get wrong is how they set up their files. If Claude doesn't know you, your clients, or what you decided last week, every chat starts from zero. You end up explaining yourself over and over. My rough math: No file system, you're getting 20 to 30% of what Claude can do. A messy one, maybe 40 to 50%. A clean one your agents can move through fast, 80 to 90%. It doesn't matter if you code or not. If you write emails, build decks, manage clients, or run projects, it's the same problem and the same fix. It takes about 30 minutes, and it works in Claude Cowork, Claude Code, ChatGPT Work, and Codex. Your files are just folders on your computer, so any of them can read the same setup. It comes down to layers: Global instructions: for who you are and how you work. A top-level CLAUDE.md: that tells every new agent where things live. A CLAUDE.md in each client or project folder: so the agent gets up to speed without reading every file. A workspace map: so it goes straight to the right doc. Projects: so your chats build on each other. Set it up once and each new chat picks up where the last one left off. Learn it once and it works in every AI tool you use. I walked through my whole setup in this video, plus two free skills to build or audit yours (in comments below) Timestamps 00:00 You're only as good as your file system 02:00 What does Claude know about you? 04:11 The top-level CLAUDE.md 08:43 A CLAUDE.md in every subfolder 11:45 Free skills: create and audit your workspace 14:23 Global instructions 16:54 Projects stack your context 20:02 Before and after: an email 21:53 Before and after: slide decks 25:03 The 7 rules Enjoy 🍿