In spite of popular belief (which is often erroneously attributed to Agile methodologies) it is still required to make an architecture prior to hacking away at the codez. So, on which axis will the system have to be able to grow? Which potential problems can we deal with off-line? Which type of issue occurs so often that absolutely must be automated? What are the uptime-requirements? How easy should it be to update the system? What does that mean for the architecture? Lots of questions. I can find the answers and process them into a system architecture.
I will always remain a techie at heart and therefore must stay grounded in reality by regularly doing implementation work myself. It keeps any designs I make grounded in reality and it’s the best way to mentor less experienced team-members: lead by example. Plus, it’s very enjoyable.