The issues defined in the short paper make sense, and I like the idea of focusing on making sure that AI pilots move beyond the trial stage. But is a waterfall style 5 step process really the way to address this? Not in my experience. A much more agile, responsive approach is required!