This is from the "doesn't scale" quadrant, but if you are not confident that your bot will behave well, shouldn't you supervise everything it does closely until you become confident?