AIMM Hub

Mastermind recap

Desktop Folder Is Your Workspace

· AIMM · 90 min

Facilitators: Lou D'Alo

AI Mastermind | Knowledge Entrepreneurs Edition

“You were filling it up with content, and you didn’t leave enough room for intelligence.” — Lou, on why stuffing files into a conversation makes Claude dumber, not smarter

This Week in 30 Seconds

  • The desktop folder is your new workspace — Lou walked Joanna through replacing the “paste everything into a chat” habit with a project folder, a CLAUDE.md index, and source URLs treated as ground truth. It kills hallucinations and gives you portable memory across conversations.
  • One-off or forever — Dirk and Don both wrote board-level documents with AI. The real lesson wasn’t the writing — it was knowing when to do it in a throwaway chat versus when to build a skill that gets smarter every time you run it.
  • Don’t let the tool run your process — Don named the most expensive AI mistake companies are making right now, and located the one job AI can’t take: supplying the wisdom that tells the AI what it’s for.
  • The build barrier just collapsed — Donald turned his book into a custom reading app instead of renting BookFunnel. Lou bypassed his email platform with a Gmail MCP. The “I can’t build that” excuse is officially dead.
  • Build a board that disagrees with you — Kasimir is building digital twins of real board members; Lou is building AI councils that red-team your thinking. Both are chasing the same prize: an advisor that won’t just tell you you’re brilliant.

The Desktop Folder Is Your New Workspace

The single most useful shift in this session: stop treating an AI conversation like a filing cabinet and start treating a folder on your disk as the workspace. Joanna came in carrying a genuinely hard problem — she’s self-representing in a formal complaint to Ontario Small Claims Court and the financial regulator, drafting legal documents that run to 150 pages, pulling citations from the Insurance Act, FSRA rules, and the UDAP regulation. Volume plus precision plus zero tolerance for a hallucinated citation. She’d been living inside ChatGPT and Claude threads, fighting the model as it quietly changed dates and words across documents.

Lou’s reframe was the heart of the hour. The problem wasn’t Claude’s intelligence — it was that she’d been pouring source documents directly into the conversation. “You were filling it up with content, and you didn’t leave enough room for intelligence.” A conversation has a fixed memory budget. Every PDF you paste eats into the space the model needs to actually think. The fix is to separate the conversation from the library.

Here’s the pattern Lou built live. Create a folder on your desktop for the project. Drop in everything relevant — the purchase agreement, your correspondence, a plain-language description of your case. Then create a CLAUDE.md file, which is just a markdown text file of instructions, and use it as an index: this file holds the purchase agreement, the legal claim is in legal-claim.md, the case context is here. Point Claude at the folder, and from then on it pulls only the files it needs to answer a given question — managing its own context instead of you stuffing it manually. Ten files or a thousand, it doesn’t matter, because only the relevant ones load into memory at any moment.

The anti-hallucination move is the part legal and research-heavy members should steal immediately. Instead of letting Claude roam the open web, Lou wrote a line into the CLAUDE.md telling it to only pull regulatory information from a specific URL and its subfolders, to treat that website as ground truth, and to never use pretraining or internet search unless explicitly asked. Now every citation is anchored to the real source. He flagged Notebook LM as the even-stronger option for this: feed it the regulatory URLs, and it answers only from those sources, with references — you can verify nothing was invented.

Two more durable tips surfaced. First, the folder persists across conversations — when a thread gets long and noisy, start a fresh one, point it at the same folder, tell it to read the CLAUDE.md, and it picks up with full context and zero re-uploading. Think of it as memory that lives on your disk, not in the chat. Second, for large or numerous PDFs, have Claude convert them to markdown once: a markdown file is roughly a hundredth the size, it’s greppable without loading into memory, and it retrieves faster and more reliably. Lou was clear this last one is an optimization — don’t bother until volume forces it.

Joanna kept circling one worry: “I’m still the brains.” Lou’s answer was exactly right for legal work — yes, and you should be. Stay in control of anything consequential, demand inline citations so you can spot-check each one, and have it reviewed. The workspace doesn’t replace your judgment. It just stops you from wasting that judgment on file management.

💡 What This Means for You

You don’t need a 150-page legal case to use this. Pick any recurring project — client deliverables, a course module, your newsletter. Make one folder. Drop your reference material in. Write a five-line CLAUDE.md that says what each file contains and where to find the truth. That’s the whole trick. The moment you stop pasting documents into the chat window, every conversation you have gets noticeably smarter, because you’ve left room for the intelligence to actually work.

Next session: Thursday, June 4, 2026