Getting Started with Claude Cowork for Everyday Work

Claude Cowork is Anthropic's mode for handing off multi-step work, not just one-off chat exchanges. It can work across local files and connected applications, draw on context from existing tools, and carry a project through several steps, files, and sources.

In this introductory session, participants explore how to choose suitable tasks, frame an assignment with the right context and deliverable, and review results before relying on them. Through demonstrations and guided exercises, participants practice turning simple prompts into well-defined work assignments and compare when Cowork makes sense versus a standard Claude Chat. Attendees leave with hands-on experience applying Cowork to their own everyday workplace tasks within their organization's policies.

Learners will have the opportunity to:

1. Compare Claude Chat and Claude Cowork and determine when each is a better fit.
2. Identify workplace tasks that are good candidates for delegation to Cowork.
3. Draft a Cowork assignment with a clear goal, context, source material, constraints, and expected deliverable.
4. Practice providing Cowork with appropriate files, folders, and connected sources.
5. Work through a multi-step workflow on a larger workplace assignment.
6. Explore the roles of connectors, skills, plugins, and organizational instructions.
7. Practice reviewing Cowork's output for accuracy, completeness, and appropriate use of sources.
8. Examine permissions and approval points before allowing Cowork to take actions.
9. Discuss which tasks call for human judgment or additional review.
10. Apply organizational policies for sensitive, confidential, or proprietary information.

Topics

Introduction to Claude Cowork: 
- Chat versus Cowork
- Moving from asking questions to delegating work
- What "agentic work" means in practice
- Example projects
- Capabilities and limitations

Choosing the right work for Cowork: 
- Good delegation candidates
- One-step versus multi-step tasks
- Multi-document and multi-app work
- Repeatable workflows, research and report preparation
- When Chat or a human is the better choice
- Tasks that should not be delegated

Giving Cowork a good assignment: 
- Goal and desired outcome
- Background and context
- Source materials, scope and boundaries
- Audience and format
- Defining "done" 
- Flagging what needs confirmation

Working with files, folders, and connected tools:
- Choosing a working folder, local files, or multiple documents
- Connectors at a conceptual level (Drive, email, Slack, or other approved services)
- Permissions and access boundaries
- Minimizing unnecessary access
- Organizing source material

Managing multi-step work: 
- How Cowork handles projects
- Breaking work into stages
- Reviewing intermediate results
- Clarifying and redirecting
- Adding context mid-task
- Checkpoints before consequential actions
- Moving from draft to final

Instructions, projects, skills, and plugins: 
- Global instructions and preferences
- Project context
- Skills as reusable workflows
- Plugins and connected capabilities
- Encoding repeatable processes

Reviewing, verifying, and approving work: 
- Checking facts against sources
- Spotting omissions and unsupported assumptions
- Verifying dates, names, numbers, and calculations
- Reviewing permission requests
- Knowing when not to approve
- Asking Cowork to revise or validate

Responsible and safe use: 
- Protecting confidential information
- Following Stanford AI-use policies
- Limiting access to what a task requires
- Recognizing unreliable output
- Avoiding over-reliance on autonomous work
- Keeping people in consequential decisions
- Knowing when subject-matter review is needed

Prerequisites: General familiarity with Claude, ChatGPT, or another generative AI tool is helpful but not required. No programming experience is needed. Suitable for business, administrative, operations, knowledge-work, and technical professionals who are new to Claude Cowork.