Last week I audited my own setup and found Grok Bot had a pile of its own files and Claude had a stack of memory files. I had both move everything into my workspace. Now Grok Bot keeps only pointers to it. If you use more than one AI, each one keeps its own memory. Tell Claude about a new client on Monday, and your Grok Bot still won't know on Tuesday. You end up repeating yourself, and two AIs give you two different answers to the same question. This lesson is for you if you use two or more AIs, like Claude or ChatGPT plus a personal agent such as Grok Bot or Muse. Using only one? Start with last week's lesson, Give your next hire a map. A lesson in four steps: get every AI you use reading from one workspace, so you stop telling them the same thing twice. What you'll do: 1. Move what each AI knows into your workspace. 2. Turn off built-in memory. 3. Give your personal agents one rule. 4. Ask them all the same question. YOUR FILES HAVE ONE HOME. YOUR AIS' NOTES DON'T. Last week's lesson gave your work one home: a workspace, with a map that tells any AI where to look. That settles where your files live. Each AI still keeps its own notes about you, though. That memory is just files too, stored inside each app, where your other AIs can't read them. TWO KINDS OF MEMORY - Each app's own memory. Claude's memory, ChatGPT's saved memories and the notes a personal agent like Grok Bot or Muse keeps are all stored inside that app, outside your workspace. One AI can't see what another remembers. - Your workspace is one main folder you own, with your files and folders inside and a map that says where things are. Every AI you point at it reads the same thing. So keep what matters in your workspace, and let each AI's memory hold only a pointer to it: a short note that says where to look. STEP 1: MOVE WHAT EACH AI KNOWS INTO YOUR WORKSPACE Run this in every AI you use that can open your workspace, personal agents included. For one that can't, like ChatGPT in a browser, ask it: "Write out everything you remember about me, plus my custom instructions." Then paste the answer into one that can.

Hi all! Sharing a lesson I sent out this week so you get it too! Please let me know what you think. A new hire's first week is a scavenger hunt: laptops, Google Drive, Dropbox, email, Slack, the CRM and someone's head. A person asks around until they find what they need. An AI agent can't ask around. It works with what you hand it, then guesses or stops. The fix is one page that says where everything is. The map is one short file. It lists each kind of work and where it lives. Your agent reads the map first, then goes straight to the right place. Connections give an AI access. A map gives it a route through your work. I did this for an online education business. Their team now asks AI a member's question and gets an answer from their own material. This is a lesson in two parts: map where your work lives in about 10 minutes, then move it into one home. PART 1: MAKE THE MAP IN ABOUT 10 MINUTES You need the Claude desktop app on a paid plan. It can look inside folders on your computer and drives you've connected. Step 1. Add your folders. Start a new chat, or a new task in Cowork. Click + and add the folder where most of your work lives. Files in Google Drive? Connect it under Customize, then Connectors. Step 2. Paste this prompt. It only reads names. Nothing gets moved or opened. --- I want to make a map of my work so an AI assistant can find things on its own. 1. Look through the folders and drives I've shared with you. Read the file and folder names only. Don't open, move, change or delete anything. 2. Then ask me these questions, one at a time: - What does my business do, in one sentence? - Where else does my work live that you can't see? For example email, texts, Slack, a CRM, my phone or paper. - Optional: what do people here know that isn't written down anywhere? 3. Write a one-page file called CLAUDE.md in the top folder, in plain English, with: - What the business does - Where each kind of thing lives - What isn't written down yet Show me the file before you save it.

This comparison comes from 20-plus hours with each agent this past week, running a solopreneur business on one and then the other. The business runs on Muse now. The part this group will care about: switching only worked because everything lives in an ICM-style workspace. Price: Muse wins easily. Free is 100M tokens a week, and the $16 Power plan gives 500M a week. On Grok Bot the $20 plan was gone in about a day. About half went to one job checking email every 5 to 10 minutes. The account worked up to the $200 plan before X's new $30 bundle came out. Every run re-reads its context, and runs of more than 15 steps made up half the cost. Smarts: Grok Bot, for now. With Muse's Spark models, all writing and real planning went to Claude. Grok Bot handled more of that itself. Since the upgrade, Muse says it runs on Wonder, Meta's new model. A test is under way. The workspace did the heavy lifting. One AGENTS.md at the top holds shared rules and a routing table. Each folder has a CONTEXT.md map. Each bot has a job description file. Each recurring job is a file too. In either agent the scheduled task is one sentence that names the job file. Run records go back into the repo. After the switch, the new agent saw every job and every past run. Some bots became skills. The research bot followed the same procedure every time, so it became a skill. The relationships bot ended up calling the relationships skill for everything. What didn't move on its own. Chat histories needed an export from each bot. One job file was out of date because Grok Bot hadn't saved its changes back. Social media connections have to be set up again on each platform. Has anyone else here moved an ICM workspace between agents? What broke? Full write-up: https://opendoorlearningai.com/articles/grok-bot-vs-muse/

Your setup should outlive your agent. This deck is how I run my business across several agents from one folder of plain files. This is the starter I'd hand anyone. 1. Gather everything into one repo. Chats, notes, drives, tool memories. 2. Sort it into domains. One folder for each area of your work. 3. Write the front door, AGENTS.md. A routing table that says for this kind of work, start here. 4. Give each domain a short CONTEXT.md. What lives here and how to work here. 5. Give each fact one home. Then run the walk test. Open a fresh agent and ask a real question, like "When is my next talk, and what do I still owe the organizer?" Three hops or fewer means it works. More hops, or a guess, means a route needs fixing. Rerun it every time you switch. The part I haven't solved is running several agents at once. Each one builds its own island. If you've built bridges between yours, I want to hear how. The full deck is attached as a PDF.

I just recorded a walkthrough of this but I want to lay it out here too. I've been building an instructor kit. A full folder architecture (yes, the same method I teach) designed to help people in this community create real courses. Not loose tutorials. Not "here's a tip" posts. Structured, formatted courses that fit into what we're building here and actually teach someone something useful. The kit has my writing rules, my lesson templates, a course outline system, sample lessons, and a prompt you can hand to Claude or any AI to help you write content that matches the community format. You can use it with Claude Code and a full folder setup, or you can just copy and paste a single file into any AI chat and start building. Both paths work. 🎯 Why I'm doing this. This community has grown fast. 12,500+ members. The courses I've built so far cover the foundation: folder architecture, the abstraction series, Claude Code, prompting frameworks. But I'm one person. There are people in this community who know things I don't. Who have built things I haven't. Who could teach topics I can't cover with the depth they deserve. The way I see it, Clief Notes should operate more like a university than a course platform. Sections and departments, each run by someone who actually has depth in that subject. Your course, your name on it, your expertise. I provide the structure, the audience, the format, and the quality control. You bring the knowledge. 💰 What's in it for you right now. I want to be straight about this. Early on, the main thing I can offer instructors is free VIP access for as long as your course is live and maintained. That gets you into every tier of the community, the live sessions, the recordings, the Discord, everything. As this grows and I can see what kind of attention and value the instructor courses bring in, I want to move toward profit sharing or payouts for instructors whose courses are driving real engagement. I'm still working out exactly how that looks. I'd rather figure it out with real data than promise something I can't deliver yet. What I can promise is that this will not stay volunteer-only forever. The goal is for this to be worth your time in every sense.