How Can an AI Agent Ask a Human a Yes-or-No Question?
Send a two-option PingRoom Question, wait for a structured result, and keep No, expiry, cancellation, and no connection as separate states.
An AI agent can ask a person a yes-or-no question by creating a PingRoom Question with two explicit options, then waiting for the structured result. Use labels that describe the decision—Approve deployment and Deny deployment are safer than a context-free Yes and No.
Send enough context to decide
Include the proposed action, target, important constraints, and an expiry. Pass a correlation ID that ties the Question to the agent run. If detailed evidence lives elsewhere, include a safe authorized link rather than dumping logs or secrets into a lock-screen preview.
The agent can wait through the supported API flow, poll after a restart, or receive the result through a signed outgoing webhook. It should resume only when it receives a valid final state for the expected Question.
Model every outcome
No is an explicit human decision. Expired means no valid answer arrived in time. Cancelled means the request was withdrawn. A timeout or network error means the agent does not know the result yet. Do not collapse any of these into approval.
Use least-privilege agent credentials and keep them separate from the person’s PingRoom session. The downstream system must still authorize the real action; a notification response alone should never grant broad infrastructure access.
Questions are best for bounded choices with two to four predefined options. If the person must explain or edit the plan, hand off to an appropriate interactive surface.
Read the Question Protocol, Question state machine, and human approval guide.
PingRoom
The Ping that cuts through.
