Claude Cowork: Your AI Working Partner

Apr 6, 2026 · 16 min read

What you'll need: A paid Claude plan (from $17/month), the Claude Desktop app, and about 30 minutes.

Last Tuesday I was on a coaching call. Fully present, not distracted. While I was in the conversation, Claude was on my desktop drafting a programme summary from session notes I'd dropped into a folder that morning. When the call ended, I had a notification: task complete. The summary was in my outputs folder, written in my voice, structured the way I structure things, referencing the frameworks I actually use.

I didn't prompt it during the call. I didn't copy and paste anything. I assigned the work from my phone before the session started, and Claude ran it on my machine while I did the work that required me to be human.

That's Claude Cowork. Not a chatbot. A working partner.

What Makes Cowork Different

Most AI tools are conversations. You type, they respond, you copy the output somewhere useful. Cowork breaks that pattern entirely.

Cowork is a mode inside the Claude Desktop app that gives Claude access to your actual computer. Your files, your apps, the whole machine. You describe what "done" looks like, and Claude plans the work, executes it, and saves the result. You can walk away. You can be on your phone in another city. Claude works in the background on your desktop and sends you a push notification when it's finished.

This is not a faster way to prompt. It's delegation.

Here's what that looks like in practice:

It works with your files. Point Cowork at a folder and it can read everything in it, create new files, organise existing ones, and save finished deliverables where you tell it to.

It runs while you do other things. Assign a task and close the lid on the conversation. Claude keeps working. You get a notification when the output is ready.

It dispatches from your phone. Sitting in a coffee shop, waiting for a flight, between meetings. Open the Claude app on your phone, assign a task, and your desktop does the work.

It connects to your tools. Gmail, Google Calendar, Notion, Slack, Google Drive, GitHub. Cowork reads from and writes to the services you already use through built-in connectors.

It operates your apps. Through computer use, Claude can interact with Excel, Chrome, and other applications on your machine. Not just file access. Actual application control.

It runs on a schedule. Set up recurring tasks (a weekly email summary, a daily priority check, a content calendar review) and Cowork executes them automatically. Your computer needs to be awake. That's the only requirement.

It comes with pre-built skills. Plugins for marketing, sales, design, finance, data analysis, and more. Each plugin includes a set of skills: specialist capabilities that Claude applies to your specific work. Think of them as departments in a company, each with domain expertise.

It works in parallel. Sub-agents handle multiple parts of a complex task simultaneously. Research one thing while drafting another.

The gap between this and a regular AI chat is the same gap between texting a colleague a question and having someone on your team do the work.

Why Context Is the Whole Game

Cowork has the capability. But capability without context produces generic output. An AI that can do anything but knows nothing about you is a talented stranger.

This is where most people stop. They install the app, point it at a folder, type a task, get back something technically correct but completely impersonal, and conclude that AI is useful but limited.

The problem isn't the tool. The problem is the tool has no idea who you are.

The Human Stack is a five-layer context architecture that solves this. Identity at the base. Systems and methodology above that. Then projects, then your current priorities, then your raw thinking. Each layer feeds AI output quality. Skip a layer and the quality drops.

Layer	What It Holds	Without It
Identity	Voice, values, positioning, constraints	Output sounds like a different person every session
Systems	Frameworks, methodology, IP	AI reinvents your approach from scratch each time
Projects	Goals, phase, decisions, stakeholders	Generic advice for a generic project
Pulse	This week's priorities, blockers, energy	Wrong priorities, wrong timing
Thinking	Journal, insights, raw material	No institutional memory

The good news: Cowork's architecture maps directly to these layers. You don't need to learn a new system. You just need to know where each layer lives.

How to Set Up Cowork (With Your Human Stack)

Step 1: Get Cowork Running (5 minutes)

Download the Claude Desktop app from claude.ai/download. You need a paid Claude plan. Pro ($17/month annual, $20/month monthly) includes Cowork. Max (from $100/month) gives you more usage. Install the app and sign in with your Claude account.

When you open the app, you'll see a toggle at the top: Chat and Cowork. Chat is the familiar back-and-forth conversation. Cowork is what we're setting up. Click Cowork.

You'll see an empty workspace with a text box at the bottom for typing tasks. On the left sidebar: your task history. On the bottom left: your name and settings. That's your home base.

You'll know it's working when you see "New task" and "Scheduled" options in the left sidebar, and a folder icon near the text box at the bottom.

Step 2: Build Your Identity Layer (10 minutes)

Go to Settings > Cowork > Global Instructions. This is your Identity layer. It loads automatically before every session you run. Set it once.

This is where you define who you are. Your expertise, your voice, the standards you hold your work to. It maps to the Amplify OS Self dimension: the foundation everything else builds on.

Include 2-3 samples of writing you're proud of. Add examples of writing you hate, the kind of output you never want to see. ("Never use corporate jargon." "Never open with 'In today's fast-paced world.'" "Never use bullet points where a sentence would do.") The more specific you are about what good and bad look like, the less editing you do on the other side.

If you're not sure where to start, I built a free tool that walks you through the key questions and produces your Identity layer for you. It runs on ChatGPT (a different AI platform), but the output is designed to be pasted straight into your Cowork global instructions.

Use the Amplify OS Builder to create your Identity layer.

Think of this like inserting a SIM card into a phone. Before the SIM, the phone works but doesn't know who you are. After, it's yours. Your global instructions are that SIM. Every session starts knowing who you are.

You'll know it's working when you run your first task and the output reflects your voice and preferences without you having to explain them in the prompt.

Step 3: Give Cowork a Folder (2 minutes)

Before you give Cowork a task, you need to point it at a folder on your computer. In the Cowork tab, you'll see a folder icon or a prompt to select a working directory. Click it, choose a folder, and Cowork can now read and work with everything inside it.

There's no required folder structure. Use whatever organisation already makes sense for your work. You can point Cowork at different folders for different tasks: a client project folder for client work, a content folder for writing, your main workspace for general tasks. You choose each time you start a session.

Here's the key: Cowork reads every file in the folder you give it. So if you drop reference material in there (your audience profile, your programme outline, your brand guidelines), Claude has that context before you say a word. The more relevant material in the folder, the better the output.

If you want a starting point, here's a simple structure that works well. It mirrors the Human Stack layers: who you are, how you work, what you're building.

My Workspace/
  about-me.md
  how-i-work.md
  Client Project A/
    project-brief.md
    audience-profile.md
    session-notes.md
  Content/
    brand-voice.md
    topics-and-ideas.md

The .md extension is Markdown, a simple text format that AI reads well. You can create these in any text editor. If Markdown feels unfamiliar, plain .txt files work too.

Nothing here is required. There are no magic folder names. This is just a clean starting point. Organise in whatever way makes sense for how you think. The only thing that matters is that the files are in the folder you point Cowork at.

You'll know it's working when you start a new task and see the folder name displayed in the Cowork interface, confirming Claude has access.

Important: Cowork has real read/write access to whatever folder you share. It can create, modify, and delete files inside it. Treat it the way you'd treat handing a capable new team member the keys to a filing cabinet: give them what they need for the job, not everything you own. Don't share your entire Documents directory. Keep sensitive files (passwords, financial records, credentials) elsewhere. Start contained, expand as trust builds.

Step 4: Add a Project Brief to Your Folder (10 minutes)

Global instructions tell Cowork who you are. The project brief tells it what you're working on.

Create a file in your project folder. Call it anything that makes sense: project-brief.md, about-this-project.md, whatever fits. Use the .md (Markdown) extension if you can, or .txt if that's simpler. The name doesn't matter. What matters is the content: a short description of the project, who's involved, what the goals are, and how you want Claude to work.

Twenty lines is plenty. Here's an example:

Client Onboarding Programme

What this is:
12-week group programme for coaches integrating AI into their practice.
Delivery via Zoom. 30 participants. Starts March 2026.

Other files in this folder:
- programme-structure (module breakdown)
- participant-profile (audience details)

Working rules:
- Read all files in this folder before starting any task
- Save deliverables here with descriptive filenames
- Match my voice from global instructions
- Ask clarifying questions before making assumptions
- Never delete files without explicit approval

The brief is a pointer, not a dump. It tells Claude what this project is and where to find the detail. Keep your reference material in the same folder so Claude reads it all automatically.

You'll know it's working when Claude references details from your brief and other files in the folder without you having to mention them in your prompt.

How This Maps to the Human Stack

What you've just set up is the Human Stack in practice:

Identity = Your global instructions. Who you are. Set once. Loads every session.
Systems + Strategy = Your project brief and reference files in the folder. What this project is, how it works, where it's headed.
Task = The prompt you type each time you give Cowork something to do.

Three layers. Each one has a clear home. Identity lives in settings. The project context lives in the folder. The task lives in your prompt.

Step 5: Connect Your Tools and Add Skills (5 minutes)

Cowork can do a lot with just files and folders. But it becomes significantly more useful when it can reach your other tools.

Connectors let Cowork read from and write to services you already use. Click Customize (bottom left of the Cowork interface), then Connectors, then the + button to browse what's available. You'll see Gmail, Google Calendar, Google Drive, Notion, Slack, and others. Click one to connect it. Each connector asks you to sign in to that service and grant permission. Once connected, Cowork can reference those tools in any task.

Start with one or two. Gmail and Google Calendar are good first picks because they unlock the scheduled routines you'll set up later.

Plugins are pre-built collections of skills. Click Customize, then Browse Plugins. You'll see options for marketing, sales, design, finance, and more. Each plugin gives Cowork specialist capabilities for that type of work. Install one that matches what you do most. You can always add more later.

Once a plugin is installed, its skills show up when you type / in the task box. You don't need to memorise them. Cowork is smart enough to use the right skill based on what you ask for.

You'll know it's working when you type / and see a list of available skills, or when Claude uses a connected service (like pulling from Gmail) without you having to set anything up in that moment.

Five Cowork Workflows to Try Today

Don't just read about this. Pick one and run it.

1. The Voice Test

Once your global instructions are set, open a new Cowork session and type:

Read my context files. Then write a short LinkedIn post about a challenge I'd typically face in my work. Match my voice exactly.

Read the output. Does it sound like you? If it reads like a generic AI post, your Identity layer needs more specificity. Add writing samples. Add anti-patterns. Run it again. This single test tells you how strong your foundation is.

2. Dispatch from Your Phone

Download the Claude app on your phone (iOS or Android) and sign in with the same account. Open it and assign a task that references files on your desktop:

Summarise everything in this folder into a one-page brief. Save the result as client-summary.md.

Walk away. Do something else. When you get the notification, check the output. This is the workflow that separates Cowork from every other AI tool: you don't need to be at your desk for the work to happen.

3. The Messy Folder

Find a folder with 30+ disorganised files. Point Cowork at it:

Organise these files into sensible subfolders. Rename them with consistent naming conventions. Create a log of every change you make so I can review it.

What takes you an hour takes Cowork minutes. And because it has your global instructions, the naming conventions and folder logic match how you actually work.

4. From Raw Notes to Finished Output

Drop rough notes, a meeting transcript, or client feedback into a project folder:

Synthesise everything in this folder into a structured summary. Include key themes, decisions made, and action items. Save the result as a new file in this folder.

When it's done, archive the raw files. The project folder stays clean. The deliverable is ready to use. This is where the partnership becomes practical. You provided the raw material. Claude shaped it. You review and refine.

5. The Scheduled Routine

This one requires the Gmail connector from Step 5. Once it's connected, set up a recurring task:

Every Monday at 8am, check my Gmail for anything flagged urgent over the weekend. Summarise it and save the summary to my workspace folder.

Your computer needs to be awake (there's a toggle in Cowork's settings to prevent sleep). Beyond that, it runs on its own. This is the bridge from "useful tool" to "daily system."

Common Problems (and What They Mean)

What Happens	What It Means	What to Do
Output is generic	Identity layer is thin on voice	Add more writing samples and anti-patterns to global instructions
Output misses the point	Task brief was vague	Be specific about what "done" looks like before you assign the work
Technically right but wrong tone	Voice section needs work	Add examples of "never write like this"
Claude asks good questions before starting	Everything is working	Let it run. This is the goal.
Excellent on first pass	Identity layer is dialled in	This is what partnership looks like.

From One Task to a Daily System

One solved problem is useful. A repeatable workflow that runs every week without you thinking about it is something else entirely.

Three moves:

Stop doing it manually. You just proved the machine can handle it. Let it. Every hour you spend on production mechanics is an hour stolen from the work that actually requires your presence: the client call, the strategic decision, the creative work that only you can do.

Systemize the workflow. Save the prompt that worked. Structure your recurring prompts using the ACT framework to get consistent results. Set up the folder structure so files flow in and outputs flow out. If it recurs, use Cowork's scheduling to automate it.

Succeed by doubling down on your genius work. What does this free you to do? Not more tasks. Better presence. Deeper thinking. The compound return comes from redirecting your attention to the work that only a human can do, now that the production mechanics are handled.

This is the S3 Framework in practice. Stop what drains you. Systemize what scales you. Succeed by focusing where you are irreplaceable.

The people who get the compound return refine their global instructions after every session and build one workflow at a time. Not ten. One. Get that one dialled in, then add the next.

Start Here

Three things:

Build your Identity layer. Use the Amplify OS Builder (a free tool I built that walks you through the key questions). Paste the result into your Cowork global instructions.
Write one project brief. Pick your most active project. Twenty lines in a text file, dropped into the folder. Point Cowork at it and run a task.
Refine after. Whatever the output got wrong, fix it in your global instructions or your project brief. That's the compound return.

The setup takes an afternoon. The return shows up every day after that.

Your wisdom and your judgment stay yours. The machine handles the production mechanics. That's the partnership.

Launch the Amplify OS Builder and start building your AI working partner.