I'm usually using RACE framework when initiating the request. Role = act like... [persona] Action = what will AI do Context = explain the context Expectation = details about what should be the end product (doc, table, summary, etc) Then I also use Silvia's process for large requests.