If you've used Claude for writing emails or answering questions, you've used Chat. Cowork is something fundamentally different. It's an agentic desktop tool built into the Claude Desktop app that doesn't just tell you how to do work — it does the work itself, directly on your computer.
This guide explains what Claude Cowork actually is, how it works under the hood in plain English, and whether it's the right tool for what you do.
Cowork is not a chatbot
Regular Claude Chat works like a conversation. You ask a question, you get an answer. You paste in some text, Claude rewrites it. The interaction follows a simple loop: prompt in, response out.
Cowork breaks that model entirely. Instead of answering questions, it executes tasks. You point it at a folder on your computer, describe what "done" looks like, and Claude figures out the steps, builds a plan, and carries it out. It reads your files, creates new ones, organises folders, builds spreadsheets, and delivers finished work — all without you touching a single file yourself.
Think of the difference this way: Chat is like phoning a consultant for advice. Cowork is like having a capable colleague sitting at the desk next to you, working through a stack of folders while you get on with something else.
How Cowork differs from Chat and Claude Code
Anthropic offers three distinct ways to use Claude, and each serves a different audience:
Claude Chat is the conversational interface most people know. You type, Claude responds. It's excellent for brainstorming, drafting, research, and Q&A. But it can't touch your files, schedule recurring work, or execute multi-step workflows autonomously.
Claude Code is a command-line tool designed for software developers. It lives in the terminal, writes and edits code, runs tests, and manages git repositories. Powerful, but you need to be comfortable with a terminal to use it.
Claude Cowork sits between the two. It has the agentic power of Claude Code — file access, sub-agents, parallel execution, automation — but wrapped in a visual desktop interface that requires no technical background. It's built for knowledge workers: managers, analysts, consultants, operations professionals, and anyone whose job involves documents, data, and recurring processes.
The key distinction is agency. Chat gives you answers. Code gives developers a programming partner. Cowork gives professionals an autonomous colleague that does actual work on actual files.
The agentic architecture in plain English
When you give Cowork a task, it doesn't just generate text and hope for the best. It follows a structured execution pipeline:
1. Request analysis
Claude reads your instruction and examines the files in your working folder. It identifies what you're asking for, what resources are available, and what the end-state should look like.
2. Plan creation
Before touching anything, Claude builds a visible execution plan — a to-do list of every action it intends to take. You see exactly which files it will read, what it will create, and what it will move or modify. Nothing happens until you review and approve this plan.
3. Subtask decomposition
For complex work, Claude breaks the job into smaller pieces. If you ask it to "create a quarterly report from 50 CSV files," it might decompose that into: read all files, categorise by region, extract key metrics, build summary tables, create charts, assemble the final document.
4. Parallel execution with sub-agents
Here's where Cowork gets genuinely powerful. For tasks that can be split up, Claude spins up sub-agents — parallel workers that each handle a piece of the job simultaneously. Instead of processing 20 files one at a time, five sub-agents each process four files at once. A task that might take 30 minutes sequentially finishes in under 5.
5. Output delivery
Finished work is written directly to your file system. No downloading, no copy-pasting from a chat window. The Excel file, the organised folders, the summary document — they're just there, in your folder, ready to use.
This pipeline is what makes Cowork "agentic." It doesn't wait for you to tell it what to do at each step. You describe the destination; Claude navigates the route.
Key capabilities
Direct file access
Cowork operates inside a sandboxed virtual machine on your computer. You point it at a specific folder and it can read, edit, create, and organise files directly within that scope. Unlike Chat, where you manually upload documents one at a time, Cowork works with your actual file system.
The sandbox is the security mechanism. Claude can only access the folder you explicitly choose — it cannot wander into your Documents, Desktop, or anywhere else. Think of it as giving a colleague a key to one specific filing cabinet but nothing else.
Computer Use
When enabled, Computer Use lets Claude step outside the sandbox to interact directly with your desktop applications. It can click buttons, fill forms, navigate menus, and operate software just as you would — using your screen, mouse, and keyboard.
This is currently in research preview on macOS. It's powerful for tasks that span multiple applications (pulling data from a web app into a spreadsheet, for example), but it carries higher risk than sandbox-only file work and requires separate per-application permissions.
Skills, plugins, and connectors
Cowork connects to external tools through an ecosystem of integrations. Connectors link to services like Google Drive, Slack, and Notion. Skills are reusable task templates that encode specific workflows. You can even create custom skills for processes unique to your organisation — like a monthly reporting format or a client onboarding checklist.
Scheduled tasks
You can set Cowork to run tasks on a recurring schedule. A Monday morning data consolidation. A Friday afternoon folder cleanup. A daily email digest. This moves Cowork from "tool you use" to "colleague that handles the routine."
Projects and persistent memory
Projects group related tasks with their own files, context, instructions, and memory. Within a Project, Claude remembers previous conversations, your preferences, and the specific context of what you're working on. Standalone sessions, by contrast, start fresh each time.
Who Cowork is for
Cowork is specifically designed for non-technical professionals — people whose daily work involves documents, data, reports, and processes, but who don't write code and don't want to learn a command line.
If your job includes any of the following, Cowork is worth investigating:
- Preparing reports from multiple data sources (quarterly reviews, board packs, client summaries)
- Organising files that have accumulated without structure over months or years
- Analysing spreadsheets — consolidating, pivoting, summarising, and charting data
- Processing documents in batch (renaming, reformatting, extracting information from PDFs)
- Recurring administrative tasks that eat hours every week but follow the same pattern
Middle managers, senior leaders, operations teams, consultants, analysts, executive assistants — anyone who spends significant time on structured, repeatable knowledge work is the target audience.
What Cowork is not
It's worth being clear about the boundaries:
- It's not a cloud service. Cowork runs locally on your machine. If you close the Desktop app or your computer goes to sleep, the task stops immediately. There's no background processing.
- It's not covered by enterprise compliance tools (yet). Cowork activity is currently excluded from Enterprise Audit Logs, Compliance API, and Data Exports. Don't use it for regulated workloads until Anthropic addresses this gap.
- It's not infallible. Claude can misinterpret instructions, make errors in complex calculations, or produce output that needs correction. The plan review step exists for a reason — always read it before approving.
The principle that makes it work
The single most important thing to understand about Cowork is this: describe the outcome, not the steps.
Instead of telling Claude "open file A, then look for column B, then copy the values to file C," say "create a summary of all Q1 revenue by region from the CSV files in this folder and save it as an Excel file called Q1-Summary.xlsx."
You define what "done" looks like. Claude figures out how to get there. The more specific your description of the end-state — file formats, naming conventions, what to include, what to exclude — the better the results.
Getting started
Cowork is available in the Claude Desktop app on macOS for Pro, Team, and Enterprise subscribers. Open the app, switch to the Cowork tab in the sidebar, select a working folder, and give it a task.
If you want to build genuine proficiency — not just know which buttons to click, but understand the architecture, security model, and advanced capabilities well enough to be the go-to person on your team — the learning path on this site covers all six certification domains. And when you're ready to prove what you know, the certification exam is free and open to everyone.
Start with a low-stakes task. A folder that needs organising. A batch of files that need renaming. See the pipeline in action, get comfortable with the plan review step, and build from there.