This from a lean manufacturing consultant:

Find the simple path to what works and follow it, always looking for a simpler path.

Patrick D. Smith

 "Without test-first and refactoring, clients think
 they must assemble as many program requirements as
 they can afford to have written. This effort snarls
 all relative business priorities together, making
 Scope Control impossible. It obscures opportunities
 for simplification. Designing and implementing many
 features all at once is very hard, leading to our
 industry's reputation for very large failures. Putting
 tests in front of development's inner cycle permits an
 outer cycle of incremental feature growth. That
 relieves the Customer of the responsibility to predict
 the future and guess which complete set of features
 will maximize productivity."

.. Juergens answer:

 Nice description.
 Sometimes I try to explain that to non technical people
 with the following picture:

 If you have only one chance to get your wishes on a
 list, it is like Christmas for a child. You make sure
 you get every little wish on that list and hope for the best.

 If you are sure that you can get your wishes on the list
 at any time. You will just put the most important ones
 there which come to your mind easily.

 > Surely, you're not calling design documents you built in the middle
 > of a project "up-front" design?

Let’s talk about "project" and "up-front" for a second.

In the old world I came from, a project had a feedback loop. This feedback loop could be considered to have covered design to unit testing, roughly a period of 6 months to a year on many projects. In other words, I would get feedback on my design 6 months after I made it.

Now let’s enter the fractal nature of XP.

My design to unit test feedback loop is the duration of a card in most cases. In some cases it’s as small as design/refactor/new test/new code/refactor/ (which may be a scope of minutes). In some cases it may be as large as an iteration (after all, we didn’t pick the cards in this iteration at random, we had a plan). It may even have been as large as a release plan.

The difference is that I get my feedback quickly and the design I do at any given stage is as small as it needs to be instead of as large as it can be. But I still do design "up-front". I have a plan before I leave the release meeting. I have a plan before I leave the iteration meeting. I have a plan before I even start refactoring before that first unit test. I have to make the same decisions I would have made in the BDUF, the only difference is that I make them as late as possible. In short, I make them just before I do the task that requires that decision to have been made.

XP is fractal. It’s possible to think about an XP project as a large collection of projects, each small enough to be written on 3*5 cards. And I do design for every one of those projects up front.

Amongst other things this article tells you when best to have children :-). [groups.yahoo.com/group/extremeprogramming/files/when%20should%20we%20test.pdf]

 Inside my head, I sometimes refer to Del.icio.us as the Google In Reverse.
 Google has amassed a solid mound of ranked and twined web sites. The
 standings shift about with caution, the behemoths are tough to dethrone.
 And if I ask for Ruby, the answers in place may hold through the end of
 the year.
 ...
 Del.icio.us is perfect! The activity bred by competitive linking would
 be enhanced by the sharing of richer media.
 ...
 Better client software is needed to make this happen.