I would also recommend reading the fowlers Is Design Dead file, as far as I understand its arguments, if you consider all flexible practices as a whole, then you get the freedom to make big changes and therefore you can develop the architecture.
Refactoring works most efficiently with continuous interaction, testing is improved with TDD and continuous integration ... I could continue. The evolving "architectures" are limited only if you cannot make the big changes required to correct the "errors."
In addition, I think that you have an architect as an interested person in the project, they tell user stories, which, in turn, are transferred to the architect.
It is also a good way to use pair programming with an architect working as part of a pair. In this context, an architect is not so much a devoted person as a hat that a member of the development team wears while pair programming.
I think that XP does not diminish the role of the architect (and architecture), it simply puts responsibility on all team members for the delivery and distribution of costs throughout the project life cycle.
[edit]
In other comments, do not be afraid of any preliminary planning, zero zero moment is a good time to try to outline a plan, just do not bring it strictly to a certain time scale.
Scott James
source share