I suppose we can only accurately define a problem when there are no big unknowns, so I wouldn't prescribe readme based development in every scenario - a quick and dirty exploration of the space which can then be scrapped is more important sometimes.