Chat vs Cowork: what's the difference?
The question that comes up in nearly every lesson. Both are powerful, which is exactly why it confuses people. It is not about power. It is about how you work: you steer Chat, you delegate to Cowork.
Chat is not just a chatbot. It reaches your Gmail, Calendar, files and tools, and so does Cowork. What changes is the mode. In Chat you drive every turn. In Cowork you hand over the whole job.
Steer it, or hand it over
Chat
You steer, turn by turn.
A conversation. You ask, you read, you ask again. Nothing happens until you send the next message.
Reach for it when- You have a quick question.
- You are thinking something through.
- You want a draft, fast.
- You want to go back and forth.
Like asking a sharp colleague at their desk.
Cowork
You delegate, then step away.
A self-contained workspace. Point it at a whole folder, give it a goal, and let it run multi-step work while you do something else.
Reach for it when- The job is worth real effort.
- It spans many files or sources.
- You want a finished deliverable back.
- You would rather manage the outcome than the keystrokes.
Like handing a job to a capable employee.
Cowork is your first step into agentic work
Chat answers. Cowork acts.
"Agentic" just means the AI takes a goal and does the steps itself: reads the files, works through them, uses its tools, instead of you prompting every move.
Cowork is where most people first feel that shift, the jump from AI that talks to AI that does. You give it a goal and manage the outcome, not the keystrokes.
Your Cowork project is a library, not a folder
This is the part most people under-use. A project is not just somewhere to drop a file. It is a growing library Claude reads from, and that is what turns it into a real thinking partner rather than a stranger you brief from scratch each time. The scale is bigger than it feels.
Several long documents open in front of Claude at once. Around 500 pages on older models, up to roughly 2,500 on the newest.
When a project outgrows the desk, Claude searches it instead of holding all of it at once. It scans the shelves and pulls the pages your question needs, which is what lets a project hold far more.
Figures from Anthropic's own docs: the context window a conversation holds, and how a project reaches past it with retrieval.
So load it up
Past examples, spreadsheets, PDFs, notes, whatever the job touches. The more real context the project holds, the more it works like a colleague who has read everything you have. Most people give it two pages and wonder why it feels generic. Give it the library.
One question to decide
Ask yourself: am I doing this with it, or handing it over? Doing it with it, turn by turn, is Chat. Handing the whole job over is Cowork.
Or let Claude pick. Paste this into Claude.ai Chat with your task in the brackets.
Copy thisI have a task: [describe your task]. Should I do it in Claude Chat, steering it turn by turn, or hand it to Cowork to run as a delegated job? Give me one clear pick and a one-line reason.
API keys, usage limits, data safety, Pro vs Max, and the other questions that come up in nearly every lesson live on one page: the FAQ →
- Not about power: both reach your tools, files and connectors.
- Chat: you steer, turn by turn. A conversation. Quick and light.
- Cowork: you delegate, then step away. A workspace pointed at a whole folder, running multi-step work.
- The shift: Chat answers, Cowork acts. Cowork is your first step into agentic work.
- The library: a Cowork project holds far more than you think. Load it up, and it works like a colleague who has read everything.
The one question when you are stuck: am I doing this with it, or handing it over?
Building software or making visuals? Those are the other two surfaces, Claude Code and Claude Design, on the same plan. Ready to pick how much power a job needs? That is the Leverage Ladder → New to the framework? Start with AI Fluency →