← Agentic AI
Lesson 4 of 4
Safety and evaluation
Watch, read the card, then practise. Guardrails · Evaluating agents
The video for this lesson is being recorded.
Meanwhile, the notes below and the 60-second cards cover the same ideas.
Practise this topicGuardrails
Guardrails limit what an agent can do, so mistakes and attacks can't cause real damage.
Remember: The more an action can't be undone, the more checks it needs.
Evaluating agents
Evaluate agents on whole tasks: did they succeed, how many steps, what did it cost, and was anything unsafe?
Remember: Same success rate can hide very different costs and risks.
Try the example in Code Lab
allowed = {"search", "read_file"}
requested = ["search", "delete_file", "read_file"]
print([t for t in requested if t in allowed])