Claude Cowork: the complete guide to automating your daily work
Cowork turns Claude into a personal assistant that reads your files, manages your calendar and executes workflows. Complete guide: folders, instructions, connectors, skills, Dispatch and scheduled tasks.
Claude is no longer just a chatbot
If you’re using Claude in your browser, you’re only using a third of what it can do. Chat is fine. But it’s like using Excel only for making lists.
With the desktop app (macOS and Windows since April 2026), Claude offers three modes: Chat (what you know), Code (the terminal for developers), and Cowork. The latter is what we’re covering today.
Cowork is Claude Code without the terminal. Same power, same ability to read and create files, connect to your apps, execute multi-step workflows. You just don’t need to touch a command line.
You describe what you want. Cowork does it.
Requirements: a paid Claude subscription (Pro, Max, Team or Enterprise) and the Claude desktop app installed.
Folders: everything starts here
The first thing Cowork asks is to choose a folder. A real folder on your machine, not something in the cloud.
This is the foundation. Cowork reads files in that folder, creates new ones, works with what’s there. Think of each folder as a department:
- A
marketing/folder for content generation - A
finances/folder for accounting - An
executive-assistant/folder for daily management
You can also start with an empty folder. Cowork will fill it with markdown files, notes, and scripts as you go.
Concrete example
Imagine a youtube/ folder with two subfolders: transcripts/ (15 video transcriptions) and analytics/ (last 90 days of data).
Open Cowork, select this folder, and ask:
What are my 15 video titles? What’s the performance trend?
Cowork scans the subfolders, reads the files, and responds in seconds. It can go as deep as needed.
Instructions: giving Claude context
Each folder can have its own instructions. When you open a folder in Cowork, a claude.md file is created automatically.
Same principle as the CLAUDE.md in Claude Code. A markdown file in natural language that tells Claude what the folder contains and how to work with it.
This YouTube folder contains my video transcriptions and analytics.
Be direct and analytical. I want to study these videos to understand
what works and what doesn't.
The benefit: these instructions are read automatically every session in that folder. No need to repeat context.
You can get more granular: a claude.md in the parent folder, another in transcripts/, another in analytics/. Each with level-specific instructions.
Global instructions vs. folder instructions
Folder instructions define what this department does. Global instructions define who you are.
Global instructions are configured in Claude settings (Settings > Personal preferences). They apply everywhere: Cowork, Chat, Code. They contain:
- Your identity: role, company, domain
- Your communication style: direct, constructive criticism
- Your goals: what you’re working on right now
- Your schedule: non-negotiables, time constraints
- Your avoidances: what you never want to see in responses
Claude reads both layers together. Global instructions + folder instructions. Result: an assistant that knows you AND understands each project’s specific context.
Connectors: plug in your apps
Cowork can read your files. Connectors give it access to your apps.
Gmail, Google Calendar, Google Drive, Slack, Notion, Linear, GitHub, Figma, the Microsoft ecosystem… Over 38 connectors available as of April 2026. Setup is simple: standard OAuth, connect, authorize.
For each connector, you choose permissions:
- Read: view events, read emails
- Write: create events, draft emails, modify documents
Example: with Google Calendar connected, ask “What do I have tomorrow?” and Cowork reads your actual calendar. Not a simulation, your real appointments.
Combined with folders, this becomes powerful. Your executive-assistant/ folder has access to your calendar, email, and local files. Claude can cross-reference all these sources for a complete briefing.
Skills: automate what you do often
A skill is a markdown file in your folder that describes a step-by-step workflow. Like a job description for an employee.
Write it once, and Claude executes it every time you type the associated command.
Example: /daily-brief
A skill triggered when you type /daily-brief:
- Check timezone and current time
- Get all calendar events for today
- Scan emails received since yesterday
- Identify those needing a response
- Present everything in a structured format
Claude builds the skill for you. Describe what you want in natural language, it generates the file, tests it, and installs it. You can even see test results before validating.
The best part: Claude has a built-in skill that creates skills. Say “I want a skill that does X” and it asks the right questions (frequency, output format, data sources) then generates everything.
Possibilities
- Email management: sorting, draft replies, follow-ups
- Meeting prep: agenda + attendees + related docs
- Business tracking: pipeline, invoices, prospects
- Content generation: LinkedIn posts, summaries, reports
- Research: competitive intelligence, data aggregation
If you do something more than once, it’s a skill.
Plugins: pre-packaged skills
Plugins are all-in-one bundles: skills + connectors + commands, grouped by theme.
Since March 2026, Anthropic offers a plugin marketplace by category: Legal, Marketing, Finance, HR, Productivity. Install, customize, use.
Each plugin contains exactly the same structure we’ve been building: markdown files with instructions and workflows. The difference is someone else did the structuring work for you.
You can also:
- Create your own plugins and share them
- Import plugins from GitHub or by URL
- Customize any installed plugin
Great starting point if you don’t know where to begin. Install a plugin, study how it’s built, then adapt it to your needs.
Scheduled tasks: autopilot mode
This is the feature that changes everything. Until now, you type the command. Scheduled tasks remove that step.
Configure a skill to run automatically: every hour, every day, every week. The /schedule command is all you need.
Example: hourly email scan
/schedule Every hour, scan my inbox.
If someone needs a reply, draft a response.
Never send without my approval.
Claude creates the task, schedules it, and runs it in the background. You work in peace, and when a reply is needed, Claude notifies you with a ready draft.
Constraint: your computer must be on and the desktop app open. If your machine is asleep when a task is due, Cowork will catch up automatically when it wakes. You can see all scheduled tasks in the sidebar with their status and next execution time.
Dispatch: control Claude from your phone
Launched March 2026, Dispatch is the feature that transforms Cowork into a true personal assistant.
The concept: one continuous conversation between your phone and your desktop. Send an instruction from your phone, Claude executes it on your computer with access to all your files, connectors and plugins, then sends you the result.
How it works
- Open Cowork on your desktop or phone
- Click “Dispatch” in the side panel
- Send your instruction from any device
Dispatch isn’t a mobile agent. It’s a remote control for your desktop. When you send a task from your phone, Claude works on your computer using files, connectors and plugins already configured.
Usage examples
- On the subway: “Summarize my 10 unread emails and flag anything urgent”
- In a meeting: “Block my calendar from 2-4pm and tell Slack I’m unavailable”
- On the road: “Compile the weekly report with GA4 data and save it as draft”
You can even run multiple simultaneous Cowork sessions from your phone, each with its own context and connectors.
Availability: Pro and Max plans. Not yet available on Team and Enterprise.
Computer Use: Claude controls your screen
Since April 2026, Cowork can interact directly with apps on your computer (macOS and Windows). If no connector exists for the app you need, Claude can drive it visually.
The priority order is clear:
- Connectors first (Gmail, Slack, Drive…)
- Browser next
- Computer Use as a last resort
Examples: updating an Excel spreadsheet, navigating an internal dashboard, filling out a web form. Claude always asks permission before accessing a new app.
Warning: computer use is in “research preview”. Anthropic recommends not using it with sensitive data (financial info, passwords, confidential documents). Start with simple tasks before increasing complexity.
Cowork vs Claude Code: which one?
| Cowork | Claude Code | |
|---|---|---|
| Interface | Graphical desktop app | Terminal CLI |
| Target audience | All knowledge workers | Developers |
| File access | Selected folder | Entire project |
| Connectors | 38+ (Gmail, Slack, Notion…) | MCP servers |
| Skills | Markdown files, guided creation | Markdown files, manual creation |
| Dispatch | Yes (phone → desktop) | Yes (phone → desktop) |
| Git | No | Native |
| Hooks | No | Yes (pre/post actions) |
| Price | Included in Pro/Max | Included in Pro/Max |
The simple rule: if your task involves code and Git, use Claude Code. For everything else (project management, content, admin, business ops), Cowork is better suited.
Going further
- Install Claude Code: the complete guide
- CLAUDE.md: best practices
- The 4 key concepts of Claude Code
- Claude and Claude Code FAQ
- How I built an AI copilot
Claude Cowork is available in the Claude desktop app on macOS and Windows. Download at claude.ai/download. Last updated: April 2026.
Pierre Rondeau
Developer and indie builder. I build products and automations with AI. Creator of Claude Hub.
LinkedIn