Assessment: AI Agents

Explain

• When is an agent appropriate?
• When is a deterministic workflow better?
• How do you secure tools?
• What is excessive agency?
• How do you evaluate an agent?

Build

• Design a safe read tool.
• Design a safe write tool.
• Design an audit log schema.

Debug

• The agent loops forever. What do you add?
• The agent chooses the wrong tool. What do you inspect?
• The agent attempts an unauthorized action. What should happen?

Completion Gate

• Can build tool calling.
• Can enforce tool authorization.
• Can evaluate agent behavior.