From specifying to training

Teaching people People are already used to reusing information, though they typically either retype or cut-and-paste it. Teaching people to describe these relationships will likely have to build on their knowledge of existing structures, not the wholesale replacement of all they've known and used for the last hundred years or even hundred days.
Teaching computers We already have a few basic tools for describing transformations, but they're presently designed with the computer in mind, not the person. Connecting people to these systems is pretty difficult.
A more limited problem set than 4GL To take advantage of the kind of distributed development these approaches promise, we're going to have to let people who don't know a lot about programming describe data pathways. This didn't work well in some previous attempts, but I hope that letting people model documents with which they are familiar works better than modeling abstract business processes.
Debugging and diagnostics are critical Users and developers will need to be able to see information about what happened in both (their) individual transactions and (potentially shared) sequences of transactions within their field of responsibility.