Exercises: AI Agents

Concept Drills

• Explain tool calling.
• Explain agent vs workflow.
• Explain memory vs state.
• Explain why write actions require confirmation.

Build Drills

• Create a read-only data lookup tool.
• Create a write tool with preview mode.
• Add input schema validation.
• Add role-based tool authorization.
• Add audit logs.
• Add max-step protection for an agent loop.

Failure Drills

• Attempt unauthorized tool access.
• Attempt prompt injection against tools.
• Attempt duplicate write execution.
• Attempt ambiguous user requests.

Output

• Agent tool registry.
• Authorization matrix.
• Audit log examples.
• Agent evaluation cases.