Hmm, I think you might want to give a bit more info on this up-front. It's actually really quite interesting and insightful. Basically the idea is that if you come up with an idea for something that you think might have unforeseeable problems (e.g. it depends on other people's code/systems that you may not be familiar with yet), then you just start coding immediately so you can (hopefully) run into the problems quickly and solve them. Then when you've done a few iterations of this exploratory, trial-and-error process, you start over with the lessons you've learned and do a proper structured (planned) version of the feature.
This technique is particularly interesting to me as I've just read the "tips for the one man programmer" article linked from here a while back and it stressed planning quite a bit. At the time I agreed with it wholeheartedly. However I think this is an interesting twist on that. It is a way to potentially combine the best aspects of the true "cowboy coder" and the more methodical, "professional" approach.
That being said I'm not actually a developer so I'm not sure how valuable my opinion is.
I'd love to hear what some real coders think of this though.