Claude Agent SDK workshop — sample agenda
Abdellah Iraamane | Claude trainer and Udacity instructor

Sample duration: 1 day
Suggested group: 6–12 participants
Prerequisites: Basic TypeScript and AI fundamentals
Live delivery; format, preparation, fees and participant count agreed in the proposal.

1. Scope one useful workflow
Map inputs, decisions, tools and the point where a person should review the result. Define success before building.

2. Build the first agent
Use the Claude Agent SDK to run a narrow task. Inspect the interaction between instructions, tools and the agent loop.

3. Connect tools deliberately
Add a custom tool and discuss where MCP fits. Limit access to the capabilities the workflow needs.

4. Make outputs checkable
Define a structured response and validate it. Handle missing information and cases that need human review.

5. Evaluate and plan the next iteration
Test representative examples, inspect failure cases and identify what needs to change before broader use.

Takeaways
- A scoped agent prototype built during the session
- A custom tool or integration and its permission boundaries
- A structured output schema with validation
- Evaluation cases and a list of limitations to address

Preparation
Participants should know basic TypeScript and AI fundamentals. We confirm the environment, Claude API access, setup steps and any company restrictions in advance. Exercises use sample data; access and usage costs are clarified in the proposal.

Workshop details: https://abdellah-claude-training.airaamane.chatgpt.site/claude-agent-sdk-training/
Udacity course: https://www.udacity.com/course/agentic-engineering-with-claude-agent-sdk--cd14715

Independent workshop; not offered by or endorsed by Anthropic or Udacity.