Our company has five distinct process competencies: UX, web, desktop, data-mining and QA with a fixed number of engineers in each role. The tools that each of these uses are quite different: photoshop, PHP, QT/C++, Hadoop, Watir/Sakuli.
I've been trying to set WIP limits on each of these roles based on development capacity, but we realise that our engineers are also involved in the specification phase, but not in a clearly defined way. On a given day a dev will be doing work on a specified ticket, and also working with a product manager on creating the specification backlog.
I'm guessing that in the end the important thing is to visualise the work in flight, the flow and to see where there are constraints.
My understanding is that Kanban is a map of the software production process, and as with any map it has to leave out some fine grained details. The important thing is that it must give a good representation of the aspects that are critical to delivery.
Should I be overly concerned about the split nature of the work that our devs are doing?