lesson: never build your moat around optimizing the existing AI capability