Episode 5

Your first week with Muse

This is the companion guide to the video where I let Meta’s Muse log in and work as me. It has the day one connection list, the two ways to log in, the four rules I set before an AI agent signs in as me, and the handoff brief I use to onboard one. If you have not watched yet, start there.

Episode 5Watch on YouTube

The one idea

Treat Muse like a new chief of staff who has their own computer. It signs in as you, so you keep the sign in steps, you approve what goes out, and you judge it by how it handles its own mistakes. Everything below follows from that.

Step 1: connect the basics

Muse has its own computer with its own web browser, separate from the Chrome or Brave windows on your screen. It reaches services two ways: built in connections, which take a tap each, and that browser, for everything else. I connected Gmail, Calendar, Drive, and WhatsApp in about fifteen minutes, including the time I spent asking it questions.

DAY ONE: WHAT TO CONNECT

Connect these first. Muse sends you a link the first time it needs each one, so you can also let it ask.
- Gmail (inbox triage, drafts)
- Google Calendar (events, reminders)
- Google Drive, Sheets, and Docs (files, trackers, documents it creates for you)
- WhatsApp (a separate thread from the desktop app)

For everything else, it uses its own browser.
- Sites without a built in connection, such as Craigslist or LinkedIn
- Forms on a company's website

Muse tells you what it can and cannot reach. Ask it: "Is this the full list of what you can connect to, or can you work with other services some other way?"

Step 2: choose how it logs in

For any site without a built in connection, Muse needs to sign in as you. You have two choices. Pick the secure card for sites that accept a saved login, and take over the browser when a site refuses it or asks for a CAPTCHA or a second factor.

THE TWO WAYS TO LOG IN

1. SECURE CARD. Muse shows a card. You type your username and password into it. It goes straight to storage the agent cannot read.

2. TAKE OVER ITS BROWSER. Muse opens the login page in its own browser. You open the live view in the Muse app, sign in yourself, and end the takeover. The session usually stays signed in for later tasks.

Either way, you never paste a password into the chat.

One thing to test on day one: the live browser view only appears inside the Muse app, on the screens that support it. When I asked Muse for a link I could open from anywhere, it told me plainly that there isn’t one. Try a takeover once while nothing depends on it, so you know it works before the day you need it.

Step 3: set the four rules before it signs in as you

These are the rules I set. Paste them into your first message so the agent follows them from the start. The fourth one matters more than it looks: when I connected the Mac app, Muse told me it can read and write files on my computer, and that staying in one folder is a rule it follows, not a limit the system enforces.

MY FOUR RULES FOR AN AI AGENT THAT SIGNS IN AS ME

1. Approve everything that goes out. The agent drafts. I approve. Only then does it publish, send, or message anyone.

2. Passwords go in the secure card, never in the chat.

3. Start with reversible work. Research, drafts, listings I can delete. Nothing that cannot be undone.

4. Decide what it can reach. If it has access to files on my computer, I give it one folder and tell it to stay there. I only keep things there that I am okay with.

BEFORE ANY OF THIS: if I work for a company, I check my company's AI policy before I connect work email or shared drives to a consumer AI agent.

Muse is a consumer product. If you work at a company, check its AI policy before you connect work email or a shared drive. For what to do after a mistake, see the agent checklist from Episode 4.

Step 4: onboard it with a handoff brief

A new agent knows nothing about you, so give it a one page brief. This is the structure I use: the role, the job, how we work, and the tone. Fill in the brackets, paste the whole thing in as your first message, and it will confirm each connection before it does anything.

YOU ARE MY [WORK / PERSONAL] ASSISTANT. My name is [YOUR NAME]. [One or two sentences of context: what I do, and what is going on right now.]

YOUR JOB: [The one job, in one sentence. Example: help me sell items on Facebook Marketplace and Craigslist.]

HOW WE'LL WORK:
1. I'll send you [what you will send: photos, receipts, documents, links] and the details that matter [price, deadline, audience].
2. Draft everything for my approval BEFORE it goes out. State clearly whether it will publish live or stay a draft.
3. Never publish, send, or delete anything without my explicit confirmation.
4. Keep everything factual. Use only what I told you. Never invent details, specs, or history. If something is unclear, hold that item and ask me.
5. Keep a running tracker in a Google Sheet: what is done, what is pending, and what is waiting on me. Share it with [YOUR EMAIL].
6. FIRST THING: check that each connection you need works. If one is broken, tell me right away instead of guessing.

TONE: short, practical, no fluff. Confirm each step. Do not narrate your process.

Step 5: tell it a rule once

Muse keeps a memory of what you tell it. When you notice yourself repeating an instruction, turn it into a standing rule, and ask it to confirm what it saved.

STANDING RULE PROMPT

From now on, [the rule]. Save this as a standing rule in your memory so you follow it every time, and confirm what you saved.

EXAMPLES I HAVE USED
- "Always share every Google Sheet or Doc you create with [my email], unless I say otherwise."
- "Any calendar invite from [my other business address], accept it automatically. I created it."
- "Never use em dashes in anything you write for me."

Should you run two agents?

Start with one. I run two, a work agent and a personal agent, for these reasons:

Separate context. Each agent builds its own memory, rules, and saved logins for its own area.
Separate credentials. My work logins stay away from the personal documents I hand over.
Trust. My personal agent runs on a Facebook account with years of history, which makes buyers more comfortable meeting a stranger.

There is a cost. Muse is one agent per Meta account, and a Facebook profile can only live in one Meta account at a time, so a second agent means a second account. Each account has its own usage allowance, so check the plan first. And the second agent starts with a blank slate, which is what the handoff brief above is for.

Want an AI agent set up for your own work?

We design and deploy AI agents with the approvals, the access limits, and the handoffs built in from the first day.

Start a Conversation