Daniel Crenna is building a DSM language for designing software robots to play poker, as part of the Made In Express competition:

I would build a poker bot simulator, where people could design their own bots using a simple control-based interface. You could share your bots with other people easily via XML and use the simulation environment to battle it out, or go head-to-head with the bots to improve your game. I would use C# for odds calculation, plug-in interfaces for the bots, and the GUI, and SQL Express to store hand histories and some standard cached tables for the core bot AI. The project will rely on clean, attractive custom controls to hide the "dirty work" from the user who may not have a lot of exposure to AI techniques.

According to Larry O'Brien's article, Daniel's system is a domain-specific visual language. That reminds me of a cool DSM language in I've seen in MetaEdit+ a few years back. It was for designing software bots that fight it out in a virtual arena, something like a bit'n'pixel version of Robot Wars. Some interesting statistics about the development of this DSM solution were described in Kalle Korhonen's article in the 2002 OOPSLA DSM workshop. The main article about it was in that year's 6th World Multi-Conference on Systemics, Cybernetics and Informatics, but sadly that seems not to be available online.

Note that was the 6th, not the notorious 9th WMSCI, which had the scandal over the randomly generated papers that were accepted. Of course, accepting randomly generated papers is appalling, but what about generated papers in general? If we can built programs to play chess, and DSMs that generate poker bots, why not a DSM for scientific articles? If a good indication of the possibilities for automation is that large portions of the output are currently build by copy-pasting by hand, this could be the perfect domain! :-)