October 6, 2026
Curious About Muse but Don't Know What to Type?
Type exactly that. A plain guide to starting with AI when you're already overwhelmed.
The empty box
Meta launched its personal agent, Muse, in September, and suddenly people who have never touched AI are curious about what an agent could do for them. (Where these agents go next is the subject of The Year AI Comes Into the Room.) Then they open it, or ChatGPT, or Claude, and meet an empty box and a blinking cursor, and they have no idea what to type. So they type something small, like a recipe question, get a decent answer, and decide it’s a slightly better search engine, or they close the window and go back to the pile.
The vocabulary doesn’t help either. People hear “Codex” or “Claude Code” or “OpenClaw,” and something in them closes, because it sounds like a world with a dress code nobody gave them.
The pile is the insurance form, the calls you keep meaning to make, and the work you’re already behind on, and every piece of it is a little knot of complexity. Complexity kills more good intentions than laziness ever has, because it’s hard to start anything when you can’t see where it ends. This technology is best at exactly that, taking a tangle and handing it back simpler. But to someone already drowning, the tool looks like one more thing to learn.
People also tell me they don’t have time for this. I understand, and I’d push back gently. You don’t need a weekend or a course, just fifteen minutes and one real task, the thing actually sitting on you this week. If those fifteen minutes take one thing off the pile, you’ll find the next fifteen on your own.
Tell it you’re overwhelmed
Open whatever you have and say, “I don’t get this. I don’t know where to start. Here’s what my week looks like, and I’m drowning.” Then tell it about your week, at whatever length it comes out, because long, rambling, half-formed input is better input. If it misreads you, say “that’s not what I meant,” and it adjusts. The complaint is the prompt, and it’s the most honest one you’ll ever give it.
Talk instead of typing
I talk to mine through a dictation app called Wispr Flow. (That’s a referral link: a friend who signs up through it gets a free month of the Pro version, and Wispr credits me for the introduction.) Talking lets you say far more than you’d ever type, and the machine takes in all of it: the aside, the frustration, the problem you mentioned twice without noticing. Inside a Project, or in a tool that keeps a memory, that’s how it learns your patterns.
Let it interview you
Then give it one job, and let it do the asking. I’ve written before about why the question is the real skill, and this is the heart of how I onboard people. You can paste this in word for word:
I’m overwhelmed and I need help figuring out where to begin. Interview me about my 80/20. Ask me one question at a time and wait for my answer. I want to find the small part of my week that’s causing most of my stress, and the small part of my work that produces most of what matters. Ask me where my hours actually go, what I dread opening, and what I avoided this week and what avoiding it cost me. When you’ve heard enough, tell me in one sentence the one bottleneck you think matters most.
Where it lands usually surprises people, because the bottleneck turns out to be the inbox they check forty times a day, or the one recurring form that eats every Sunday night, or the conversation they keep putting off. The tangle narrows to one named thing, and a person can move one named thing.
Give it a memory
It forgets more than you’d expect. Some tools now keep a memory of you, and Muse is built around one, but most conversations still start close to zero. That’s why the people who get the most from it keep a few documents about themselves and paste them back in, or keep their work inside a Project, which is just a chat that keeps your files and instructions attached. The documents are the memory.
If you’d like a head start, I put together a small free folder called the Inner Exodus Starter Kit, and you can download it here. Its interview goes wider than the one above: it learns how you talk, builds a master document about your life, and fills in a map of your 80/20 with you. Drop the files into a Claude or ChatGPT Project, paste the harness page into the instructions box, and say, “Read the kit and interview me.”
Keep your own judgment
It will state wrong things with total confidence, because it doesn’t know when it doesn’t know. Treat any specific fact, number, date, or quotation as a guess until you’ve checked the source yourself. It’s excellent at thinking with you and unreliable at remembering for you.
It also tends to agree with you. These systems are trained to please, and a machine that always agrees is only a mirror. I’ve written about my own trouble with that. Tell it plainly, “You’re being too agreeable. Push back on me,” and it will.
And when it starts doing things for you, keep one rule: it drafts, you send. Anything that spends your money or speaks in your name should wait for your yes. Beyond that, none of this is graded. You can’t break it, and if you dislike where a conversation went, you can start a new one.
Walk through the door that’s open
The tools look like a maze, but there are really four doors, and you can stop at whichever one fits. The first is plain chat, the free version of Claude or ChatGPT, and it covers most of what most people need. The second is a Project.
The third is for when you want it to do the thing itself: chat is question, then answer, and this mode is goal, then result. In ChatGPT it’s the switch that says Work, and in Claude it’s Cowork. If Muse is what made you curious, this is its door, and it’s a fine place to begin: it’s free to start, it lives in a text thread, and it’s rolling out in the US now.
The fourth door is what I use, a harness built in Claude Code, which wraps routines and self-improvement loops around the model so that a correction I make once stays made. Most people shouldn’t begin there. It earns its complexity only when at least two of these are true: there are real files on your computer it needs to work with, the same task comes around every week, you want it to act and not just answer, or you’ll actually maintain it. If that isn’t you yet, stay in chat. A person who succeeds in the chat window comes back for the harder tools on their own, and a person who drowns in setup decides the whole thing isn’t for them.
Where the rest of the complaint goes
With everyone I teach, I draw the same line, and it comes down to whether the answer requires an interior life. A machine can tell you what the calendar says, what the document says, and what your options are. It can’t tell you whether you’re avoiding a hard conversation because it’s imprudent or because you’re afraid. It has no fear of its own to recognize yours by.
Some of what comes out once you start talking won’t be about the pile at all. It’ll be grief, or fear, or the sense that you’re failing the people you love, and that part has another address. The Vatican’s 2025 note on AI warns that if we replace our relationships with God and with others with interactions with technology, “we risk replacing authentic relationality with a lifeless image” (Antiqua et Nova 63). Bring it to a person who loves you, or a priest, or a good therapist.
And bring it to God, who has always been willing to hear a complaint. The Church counts lament as real prayer, one of the forms the Psalms hand us (Catechism 2588). David, hiding in a cave and sure that “no one cares for my soul” (Ps 142:4), prayed, “I pour out my complaint before him; I tell my trouble before him” (Ps 142:2).
So tell the machine you’re overwhelmed, and let it find the question and sort the pile. Then pour out the rest where David poured his, without tidying it first.
If this gave words to something you have been carrying, share it with someone who needs it.
The Inner Exodus is free, and I intend to keep it that way. If these essays feed you, you can become a patron. A paid subscription buys nothing extra. It simply keeps the door open for the next reader. Thank you for walking this with me.
More from The Inner Exodus