He mentioned in passing that writing a typical README requires just the right amount of planning upfront. It was tangential to his main point, about documentation, but I think it's a good observation.

It avoids both planning extremes: writing an incredibly detailed spec upfront without any feedback from prototyping (waterfall), and diving in without any planning and expecting tests to magically do your design for you (naive TDD).

It's good to having an articulate summary of your project, but for planning, the README is just a prop. It could just as well be "whiteboard-driven design" or whatever.