Well if you're self disciplined it can be rather productive. If not you don't.
The question to me also is about time-frame and desired outcome. and other aspects.
Because if it's research and prototype development basically cowboy coding seams like a valid choice to me.
But for a structured business-process on a medium to large scale where software is the key/main product. It seams rather unpredictable.
as Matt Strom mentioned it seams wasteful, but than research always should be seen as wasteful/uncontrolled because you're going into new areas.
I think that's the point, cowboy is great for pioneers, researchers, free spirits.
But it's not a general attitude I would recommend for everyone.... but than I would not recommend anything these days. The team has to decide based on the target and usecase what's needed.
Cornerstones however are relevant... anyhow blah ... I think I made the point I wanted to make.