RagnarPitla Invoke the structured-interview skill.
Frame it as a discovery session: the user has an idea, a request from a customer, or a vague ask from a stakeholder, and does not yet have something that could be built, estimated or approved.
Two things to hold on to that the primitive does not assume:
The person in front of you may not be the person who decides. Ask early who has to agree to this. If a decision belongs to someone not in the room, do not let the user guess on their behalf. Record it as an open question with a named owner, and keep going. A discovery session that produces three named open questions is more useful than one that produces eight confident guesses.
Separate what is wanted from what is asked for. People arrive with a solution already chosen, usually the one they last saw demonstrated. Ask what happens today, what is wrong with it, and what would be different if this worked. Do that before discussing any Microsoft product by name. If the underlying problem does not need an agent, say so; that is a successful discovery session, not a failed one.
When the frontier is empty, produce a short written summary the user could send to a stakeholder without editing: the problem, the decisions taken and why, the open questions with owners, and what "done" means. Keep it in their vocabulary, not the product's.