What is Claude Cowork?

Claude is not one product anymore. It has become an ecosystem of tools, agents and building blocks that fit together. When you understand the whole, a very different way of working opens up. This series starts with the piece most people meet first after chat: Claude Cowork.

The Claude toolbox

Anthropic has built three ways to work with Claude. Same brain, different interfaces.

Claude.ai is the chat you already know. You write, the AI answers. For many tasks it works just like other chat tools.

Claude Code arrived in February 2025. An AI agent in the terminal that reads your codebase, writes code, runs tests and commits. Despite the name, it is used as much for analysis, research, documents, slides and content as for code. It is also the tool behind our Claude Code bootcamps.

Claude Cowork is the newest addition, launched in January 2026. Think of it as Claude Code in a friendlier wrapper. Same agentic architecture, same ability to work independently over many steps, but instead of a terminal you point Claude at a folder on your machine and describe what you want done. You also get Claude inside PowerPoint and Excel.

In short: chat means you talk with AI. Cowork means you delegate to AI.

From chat to delegation

In Cowork you describe an outcome. The agentic loop handles the rest: Claude gathers context, acts, and verifies the result. If something is off, it adjusts course and tries again, until the task is done. You can step in whenever you want, but you do not have to.

Point Claude at a folder and say "organise these files by project". Go make a coffee, and come back to a tidy folder structure. Or: "Write a report based on these notes." Claude reads the notes, finds the patterns and delivers a finished file on disk, not a draft in a chat window.

Three things make Cowork different from chat:

1. File system access. Claude sees your files directly. Whole folders of documents, spreadsheets and images, not just the text you paste in.
2. Multi-step execution. Claude plans a sequence of steps and runs them as one continuous piece of work that can last for hours.
3. Persistent context. Because Claude works with files in a folder, the context builds up naturally over time. Project knowledge, previous work and reference material are all there next time.

Chat means you copy from the chat. Cowork means you open a finished file.

What people use it for

Research and analysis. "Research the 100 companies in this spreadsheet, find key figures, competitors and recent news, and deliver a structured report." Claude reads the file, works through the list and builds the analysis.

Document production. "Here are the meeting notes from the last two weeks and the numbers from the spreadsheet. Produce a finished quarterly report with an action plan." No copy-paste, no manual assembly.

Content production. "Write a blog post based on the research in this folder, then break it down into posts for LinkedIn and the newsletter." Cowork already has the sources, so you skip the briefing.

Dashboards and micro-apps. "Build a dashboard from the sales numbers in this spreadsheet." Claude reads the sources directly and delivers a working HTML file. Same principle for small tools: a price calculator, a timesheet form, an onboarding quiz.

Getting started

Step 1: Get Claude Desktop. Download the app for Mac or Windows at claude.com/download. You need a paid plan (Pro, Max, Team or Enterprise).

Step 2: Activate Cowork. In the chat you will see a small folder icon. Click it and pick a folder on your machine. Start with something simple. This is your sandbox: Claude can read and write inside it, nothing outside.

Step 3: Your first delegation. Pick a task where your files are the whole point. For example:

The folder Suppliers contains 50 PDF invoices. Read them all, extract vendor, invoice number, amount and due date. Collect everything in a spreadsheet and produce a short summary presentation.

Press enter and watch Claude work. The finished work lands in your folder.

The mental shift

Most people use AI to iterate: draft, feedback, adjust, run again. That works, but you are still driving the process. The shift with Cowork is to ask a different question. Not "how can AI help me with this task", but "do I need to do this task at all?"

Anthropic calls it vibe working. You describe an outcome, go do something else, and come back to finished work. You can still stay in the loop and give feedback along the way, like a manager who sometimes reviews the work and sometimes hands it over completely.

Tips for the first days

• Be specific about the outcome, not the process. Say what you want to end up with and let Claude find the way.
• Start with something you actually need done. The real aha moment comes from delegating real work.
• Check the output. The results are often impressive, but verify important work.
• Build on what works. Found a structure you like? Use it as the starting point next time.

Is it safe?

A fair question when you give an AI access to your files. On Team and Enterprise plans Anthropic does not train on your data. On Pro and Max, training is off by default, but confirm it in Privacy Settings. Tasks run inside a sandboxed virtual machine with network traffic through a controlled proxy, and write access is limited to the folder you chose. Keep to one project folder at a time, and do not point Claude at your home directory.

Next in the series: skills, or how you turn Claude from a generalist into a specialist that knows your methods, your tone and your workflows.

First published in Norwegian on LinkedIn.