CLAUDE COWORK

What Is Claude Cowork?

Cowork is Anthropic’s agentic task mode in Claude Desktop. You give Claude an outcome, then it plans and executes multi-step work with local file access, code execution in an isolated VM environment, and optional tool/plugin support.

What Cowork Is (and Is Not)

Cowork is not just a longer chat window. It is designed for delegated execution: Claude can split work into subtasks, run parallel sub-agents, and deliver outputs directly to your files. You can still steer mid-task, but the operating model is “task handoff + supervision,” not “message-by-message prompting.”

How To Actually Use Cowork

1. Open Claude Desktop and switch from Chat to Cowork.
2. Describe the outcome, inputs, and constraints.
3. Review Claude’s proposed plan, then allow execution.
4. Monitor progress indicators and steer if needed.
5. Review outputs before distribution or action.

Desktop and Plan Requirements

• Claude Desktop on macOS or Windows x64 (Windows ARM64 not supported).
• Paid Claude plan: Pro, Max, Team, or Enterprise.
• Active internet connection for the full session.
• The Claude Desktop app must stay open during task execution.

How Projects Work With Cowork

Projects are persistent workspaces with instructions, knowledge files, and context memory. Cowork can operate as the execution layer for these contexts: project assets provide grounded inputs, while Cowork handles the multi-step processing and output generation.

How Plugins, Skills, and Connectors Fit Together

In Cowork, plugins are workflow bundles that can package skills, connectors, slash commands, and sub-agents. Skills specify how to perform a task; connectors provide data/system access; sub-agents distribute complex work in parallel.

Best-Fit Task Types

• Multi-file synthesis (briefing packs, meeting packets, issue digests)
• Cross-tool outputs (spreadsheet analysis into presentation artifacts)
• Long-running research and structured report generation
• Recurring task runs via scheduled Cowork jobs
• Complex tasks where parallel sub-agents reduce turnaround time

Safety and Compliance Boundaries

Anthropic explicitly warns that Cowork has higher risk because it is agentic and internet-capable. Be cautious with browser access, plugin installs, and MCP permissions due to prompt-injection and expanded action scope. Use trusted sources, review permissions, and avoid sensitive/regulated workflows.

Why PromptedWork Cowork Workflows Are Useful

PromptedWork Cowork workflows are designed for higher-friction administrative tasks where execution depth matters: collecting files, comparing sources, producing polished outputs, and reducing manual window-switching.