The best AI conversations I have had recently were not with one assistant.
They were with a room.
I tried this first as a personal experiment: agents representing Ayurveda, traditional Chinese medicine, and Western medicine discussed what they would recommend to me.
Not as medical advice. I would not outsource diagnosis to a chatbot.
Different frames argued from different assumptions. The shared conclusion was almost annoyingly boring: chill a bit, sleep properly, and spend more time outdoors.
I tried the same pattern in code review. Instead of asking whether code is good, different review lenses inspect requirements, tests, security, contracts, maintainability, and regression risk.
That is better than asking one vague question. Each lens knows which future failure mode it is hunting.
Same with leadership work.
For a people problem, I do not only want AI to challenge my perspective. I want a room with distinct lenses: strategy, operating systems, uncertainty, psychological safety, manager craft, and trust.
This is where AI becomes more interesting to me.
Not as one synthetic voice that sounds wise.
As a way to choose which questions enter the room before I make a decision.
If the lenses are vague, AI produces polished mush.
So the lenses need boundaries: source material, known failure modes, a clear output shape, and permission to say unknown.
A prompt is not only a request for an answer.
It is a choice about which perspectives are allowed to examine the problem before I act.