You explain who you are and what you're working on to Claude from scratch every single time 🔥 Claude Cowork gives it a permanent workspace with memory, access to your folders and services, and scheduled tasks — it grinds away while you sleep 🤖
What it is and who it's for
Claude Cowork is Anthropic's agentic work mode: the same agentic architecture as Claude Code, but without the terminal. This isn't "one question, one chat": Claude takes a multi-step task, works inside the folders you connect on your computer and in the connectors you've already set up, and brings the result back for review. Projects have their own files, instructions and memory — no re-explaining the context every time. You state the outcome you want, not a step-by-step prompt. Who it's for: people who want to delegate office routine to a real assistant rather than "ask a neural net" — founders, marketers, product managers, ops people.
Key features
- Projects as permanent workspaces — their own files, links, instructions and memory.
- Access to the local folders you connect yourself: Claude reads and writes only inside them instead of roaming your whole drive.
- The Claude connectors you've already set up (per the official docs — Notion, Slack, Asana and others; email, drive, calendar — if you have them connected).
- Scheduled tasks on any cadence: they run remotely, in the background, even when your device is offline.
- You state the outcome you want, not a step-by-step prompt, and come back to a finished draft.
- Cross-platform: web, desktop and mobile (mobile is in beta).
How to start in 5 minutes
- Open Cowork mode on a paid plan (web/desktop) and create a project for your task; write the project instructions: who you are, who it is, what it must never do without your explicit permission.
- Connect the folder you need on your computer and the connectors you've already set up (for example Notion / Slack / email / calendar) — Claude gets access only to what you open up.
- Set up your first scheduled task (a morning email digest, say) in "read-only/draft" mode and check how it performs.
When to take it, when not to
✅ Take it if you want to delegate office routine — email, reports, digests, keeping Notion up to date — to an assistant with memory and access to your folders and services. ✅ Take it if you already pay for Claude Pro/Max/Team — Cowork is part of the plan, no extra charge. ❌ Don't take it → "Claude Code" is better if the job is writing and editing code in a repo rather than everyday office routine. ❌ Don't take it → "Manus" or "Goose" is better if you need a fully autonomous web agent for your own scenarios, or an open-source solution you run yourself. ❌ Don't take it → plain "MCP" is better if all you need is connectors to services inside your own app, not a Claude work mode.
The honest price
paid. Cowork has no separate price tag — it's included in the paid Claude plans: Pro, Max, Team, Enterprise. It's not available on the free tier: per the official Cowork page, it's rolling out to "all paid users" (web and desktop, mobile in beta). So you pay for a Claude subscription and get Cowork as a mode inside it. Exact plan prices are on Anthropic's side and change over time.
Gotchas
- Paid plans only — there's no such mode on the free tier.
- You're handing the assistant access to folders, email and services — that's sensitive data. Write down the "what it must never do without permission" rule from day one.
- ADVISORY: that rule is an instruction to the agent, not a hard technical gate. The model can make a mistake or read it differently, so verify sensitive actions (sending, deleting, publishing) yourself.
- The agent acts autonomously on a schedule: a bad instruction means an automated mistake that repeats itself. Start with read-only tasks and drafts, not "send it yourself".
- Access is granted per folder and per service — double-check what exactly you've connected before you put a task in the background.
- Anthropic changes the feature and connector lineup often (the mode is still rolling out) — check the official page before you set things up.
🤖 Prompt accelerator
"You are my personal assistant in Claude Cowork. The context about me and the rules are in the project instructions. Hard rule: don't send, delete or publish anything without my explicit confirmation — show me a draft first. Every morning at {TIME}, put together a digest: unread email from the past 24 hours on the topic "{TOPIC}", the 3 main tasks for the day from my calendar, and a list of changes in {PROJECT/NOTION}. Deliver it as a single message, tone — {STYLE}."