I just saw Dot in action here at work the other day - generating flow charts in SVG on the fly from a database of screen specs and interactions/workflows. Interesting and positively plump with possibilities. Generating and maintaining these kind of documents has been an irritation for me for years - particularly as the information they capture is essentially locked up and only accessible visually.
Implementing a workflow has always meant essentially starting over - with the document serving as a reference only. By flipping this on its head and making your flows *output*, rather than input, your application logic can stay fluid and be developed independantly. Ideally you would be able to round-trip - so you could model an interaction or workflow in a tool like Visio or Omnigraffle - export or merge into your XML that captures the application logic in its entirety, test, tune, and re-output to a Visio/Omnigraffle importable format for futher refinement. As I understand it this capability exists...
Any opportunity to defer the permanent entanglement of application, business, presentation logic allows for more testing, refinement and interation, at less cost. We don't necessarily need DotML to do this, but it will help. Once it becomes XML, half the work is already done for me.