Obviously you have multiple agents justify why they picked a certain response and then create another agent that picks the solution with the best justification.
touché
touché