We experiment with the strategy of developing composable AI agents with slightly tempered autonomy. The resulting agent exhibits vastly improved reliability, and performance.| /research