I agree. Go fast with a suboptimal architecture. If success arise, throw away version 1 and rebuild from scratch. Often is more effettive.