Nine short sections. You'll end with a Monday brief that reads your real calendar and inbox and lands before you sit down.
You already know the chat window. You type, it answers, you copy the answer somewhere else. Cowork is the other mode. You describe an outcome, walk away, and come back to finished work.
The difference is that Claude can use things. Your calendar. Your inbox. Your files. It runs multi-step work on its own and hands you the result instead of instructions for doing it yourself.
Cowork is on paid plans only. If you are on the free plan, this whole guide is theory until you upgrade.
It also uses more of your usage allowance than chat does, because the work is heavier. Use chat for questions. Use Cowork for jobs.
Do this now: open Claude, look at the message box, and switch it from Chat to Cowork. Same box, different mode. There is no separate app to find.
Before you give Claude access to anything, decide how much it checks with you. There are three modes and you can change them any time from the message box.
Pauses and asks before every action. Slow, and correct while you are learning what this thing does.
Keeps working without stopping at each step, and reviews each action for safety before it runs.
Does not pause and does not check. Only for things you completely trust.
Start on Manual. You will watch it work for a week, get bored of approving things, and move to Auto. Do not start on Skip because someone on YouTube told you to.
Global instructions are standing instructions that apply to every Cowork session. Not one chat. All of them, until you change them.
Set these before you do anything else and every output afterward gets better.
Go to Settings > Cowork, click Edit next to Global instructions, and paste in who you are and how your business works. This is where the context lives now. You are not going to re-explain your offers in a chat window ever again.
Keep it under a page. What you sell, who you serve, what you are focused on right now, and how you want to be talked to.
I'm [name], and I run [business] solo. What I sell: - [Offer]: [what it is, price, who it's for] - [Offer]: [what it is, price, who it's for] Who I serve: [describe them like a real person, not a demographic] What I'm focused on right now: [one thing, this quarter] Numbers that matter: [where your sales live, and what "on track" means as a real number] Shape of my week: [which days are for what, which days you don't work] How to work with me: - Direct answers, not options lists, unless I ask - Draft first, ask questions after - Never invent facts, stats, or testimonials about my business - Tell me when something looks bad. Don't soften it.
Fill it in fast. You will edit it a dozen times in the first month, and that is the system working.
Global instructions tell Claude about your business. Connectors give it access to your actual week.
Without connectors, Claude can only tell you things you already told it. With them, it can tell you what is actually happening.
You control connectors from the "+" menu in the message box, or from the Connectors page in your settings.
Connect two things and stop:
Ignore the rest for now. Every connector you add is another thing that can go wrong while you are still learning what good output looks like. Add the third one when you have a specific job for it.
If you track revenue somewhere with a connector available, add that as your third. If your sales live in a platform without one, skip it. The brief below handles a missing revenue source without breaking.
On desktop, Claude can read and write files directly on your computer. No uploading, no downloading. It writes the finished document into your folder and you open it like anything else.
You do not need this to get started. It is the point where Cowork stops feeling like software and starts feeling like a coworker.
Two notes. The desktop app has to be open when a task touches local files, even though the session itself runs in the cloud. And when you select a local folder, you can add folder instructions, which are context that applies to work in that folder specifically. Global instructions are about you. Folder instructions are about that project.
Deletion is protected. Claude has to ask before it permanently deletes anything, and you have to approve it.
A project is a workspace for related work, with its own files, instructions, and memory.
The memory part is the reason to care. Work inside a project builds on itself instead of starting cold every time.
You do not need this on day one. You will want it around week three, when you notice you keep re-explaining the same launch to the same assistant. Make a project then.
A skill is a set of instructions Claude loads when the job matches. Think of it as an SOP that gets read automatically instead of one you have to remember to open.
The test for whether something should be a skill: you have explained it more than twice, and you will explain it again.
Not everything needs one. A skill for a task you do once a year is a file you will forget you wrote. Start with one, which is the next section.
This one uses everything above: your global instructions for context, your connectors for real data, and a schedule so it runs without you.
Paste the block below into a Cowork task and say: save this as a skill. Claude handles the rest.
--- name: monday-brief description: Produce my Monday morning brief. Use when asked for a Monday brief, weekly brief, or my week ahead. --- # Monday Brief Write my brief for the week. Five minutes to read, maximum. If a section has nothing worth saying, cut the section. A short brief is a good brief. Use my global instructions for business context. Check what you remember from previous briefs so you don't repeat yourself. ## 1. Last week's numbers Pull last week's revenue from the source in my context. Show the number, how it compares to the week before, and whether it puts me on track against my goal. Include month to date and days left in the month. If I don't have a revenue source connected, skip this section entirely. Do not estimate and do not substitute a different number. ## 2. Still open Check my email for anything from the last week that needs a decision or a reply from me and hasn't had one. One line each: who, what they want, what happens if it keeps waiting. Skip newsletters, receipts, and notifications. If nothing is hanging, say so and move on. Do not pad this to look useful. ## 3. The week ahead My calendar for the next seven days, grouped by day, times included. Flag anything that needs prep and say what the prep is. Then the number that matters: how many hours of unbroken working time are left after the meetings. Compare that against the priority in my context. ## 4. The read Four or five sentences. What kind of week this is, based on the calendar, the open items, and the numbers. Name the one thing that matters most this week, and the one thing most likely to eat the week if I let it. Be honest. If the numbers are down, say the numbers are down. If my calendar is full and my stated priority has no time on it, say that plainly. This section is worthless if it's nice. ## Rules - Write like a sharp colleague who read everything before I got in - No greeting, no motivational close, no offer to help further - Never invent a number, a sender, or a meeting. Missing data gets skipped, not filled in. - If a section is thin because something isn't connected, say which one, once, at the end ## After writing If I have a local folder connected, save the brief as briefs/YYYY-MM-DD.md using today's date.
Do not wait until Monday to find out whether it works. Start a Cowork task and say: run my monday-brief skill.
Read what comes back. If it is wrong, it is almost always your global instructions, not the skill. Fix those and run it again.
Ask for it in the same Cowork chat, in plain words: run my monday-brief skill every Monday at 7am. Approve it when Claude asks, and it is set.
Scheduled tasks run in the cloud, so your computer does not need to be awake and the app does not need to be open. The brief is waiting when you sit down.
Whatever you have now explained three times.
You catch yourself typing the same context again, and you turn it into a skill or a scheduled task instead. That is the method.
Mine started here, with one brief. It now runs four brands with no employees.
A few that tend to come second: a weekly content brief, a customer follow-up sweep, an inbox triage that drafts replies without sending them.
Build the next one when the first one is boring. Not before.
Once Claude is writing for you, the next problem is that it writes like everyone else's Claude. This skill fixes most of it. Install it the same way as the brief.
Be picky filling it out. The banned list does the heavy lifting.
---
name: my-voice
description: My writing voice rules. Load this skill whenever writing anything in my name - emails, social posts, blog posts, product descriptions, captions, or replies.
---
# My Voice
## How I sound
[Three or four honest descriptors. Example: "Casual but competent. I talk
like a friend who happens to know a lot. I use short sentences. I'm funny
in a dry way, never goofy."]
## Rules
- Write in first person
- Short paragraphs, 1-3 sentences
- No em dashes
- Contractions always: "you're," not "you are"
- [Your rule: emoji policy, capitalization quirks, sign-offs]
- [Your rule]
## Words and phrases I actually use
[Steal these from your own past posts and emails: "no shame," "here's the
thing," "low-key," whatever is actually yours]
## Banned words and phrases
Never use: game-changer, unlock, unleash, elevate, dive in, delve, seamless,
effortless, revolutionize, "in today's fast-paced world," "look no further,"
excited to announce, empower, supercharge, "friendly reminder"
[Add the ones that make you cringe]
## Banned structures
- No "It's not just X, it's Y" sentences
- No three-item parallel lists in a row ("Fast. Simple. Powerful.")
- No rhetorical questions as openers
- No motivational wrap-up endings
## Sample of my real writing
[Paste 2-3 short pieces you actually wrote and like. This is the single
most powerful part of the skill.]
The version I actually use is bigger. It has years of my own tells in it, the phrases I catch myself deleting, and the sentence shapes that give AI away even when the words are fine. I turned it into a skill anyone can install.
Yes. Cowork is on Pro, Max, Team, and Enterprise. There is no free version of this, and I would rather tell you now than after you have read the whole page.
The brief still runs, it just gets thinner. With nothing connected it can only work from your global instructions, which means it can only tell you things you already told it. Connect your calendar first. That single connector is the difference between a summary of your own notes and an actual read on your week.
Only for tasks that touch files on your computer or use your browser. Cowork itself runs on desktop, web, and mobile, and scheduled tasks run in the cloud whether your machine is on or not.
Your global instructions are too vague, or your calendar is not connected. Those two account for nearly every bad first brief. Go back to section 3 and make the priority line specific enough that a stranger could tell whether you spent your week on it.
The setup is about 30 minutes. The brief itself saves you the Monday morning hour you normally spend deciding what to work on. The bigger payoff comes over weeks, as you add a system every time you catch yourself explaining the same thing twice.
No, and do not try it this weekend. It does production work so you can spend your time on decisions and quality control. You stay the owner. It is the best employee you have ever had, and it is still an employee.
My exact system for creating branded Instagram carousels and static posts with Claude instead of dragging boxes around Canva at 11pm.
Check it out →I document exactly how I run four businesses solo with AI systems. New builds, real numbers, no hype.
Subscribe →Do not wait until Monday. Open a Cowork task, say run my monday-brief skill, and see what comes back today.
Back to the brief