Claude Cowork
Claude Cowork is Anthropic’s product for delegating complete pieces of work to Claude, as opposed to chatting through each step one at a time. The mental model shift is from question-and-answer to delegated work package.
Core difference from Chat
| Chat | Cowork |
|---|---|
| One question at a time | Complete work package handed over |
| User stitches steps manually | Claude plans and executes the work arc |
| Context is ephemeral | Projects accumulate memory and instructions |
| No connectors | Connects to email, calendar, Slack, Drive, CRM |
Building blocks
- Project — scoped workspace with standing instructions, connectors, scheduled tasks, and accumulated context memory
- Connector — integration to external apps (email, calendar, Slack, Drive, CRM) that Claude can read or act on
- Skill — reusable playbook for a recurring process; converts repeated prompting into a single invocable capability
- Plugin — packaged set of skills, connectors, and subagents around a role or process
- Claude in Chrome — browser-side integration for tools without dedicated connectors
- Scheduled tasks — time-triggered work, requires the desktop environment to be running
How to write a strong Cowork prompt
A good Cowork task specifies all five elements before execution begins:
- Deliverable — what exactly will be produced
- Inputs — which files, folders, or connected apps to use
- Scope — date range, folder boundary, or topic constraint
- Nuance — relevant context or constraints the human knows but Claude doesn’t
- Review boundary — what Claude should do vs. what requires human judgment
Context budget in Cowork
See ContextBudget. Managing context is a first-class discipline in Cowork:
- Use the smallest folder that contains the work
- Clear or compact context when switching work streams
- Convert repeating instructions into skills
Human review is non-negotiable
The file may be finished, but Cowork still requires human review of facts, business decisions, permanent actions, and sensitive outputs. Claude executes mechanical work and exposes evidence; the human makes consequential calls.
Related
- ContextBudget — managing context is a primary Cowork discipline
- ClaudeAgentSDK — Cowork is built on the same underlying agent loop
- AgenticGovernance — review discipline applies to Cowork deliverables
- ClaudeSDKEcosystem — Cowork’s position in the broader Claude ecosystem