Trust and recovery

Users trust an agent when they can see what it did and undo it.

Checklist

• Can the user see what the agent did, in order?
• Can the user undo the last action, or all actions of one task?
• Does the interface say when the agent is unsure?
• Does an error say what went wrong and what the user can do next?

Recovery patterns

• Undo. Reverse the last action with one click.
• History. Keep a log of agent actions that the user can read.
• Safe default. When in doubt, the agent asks instead of acting.