Well, there's bludgeoning, and bludgeoning. Don't get me wrong, I don't think that there's a magical solution for everything, involving doing terrifying things to maps. That's not really practical, and no amount of clever automation can make every map work.I just don't see the appeal of trying to bludgeon a map into conforming to a mission via an automated process.
That said, I was thinking, as a simple first-try sort of thing, of doing a "north vs. south" automated scenario, where the player gets handed a lot of buildings and defenses somewhere near the middle (I can automatically find and place stuff on metal spots, using the stuff that avoids it, in reverse, so I could give 'em mines or what-have-you, on most maps) and the computer (not AI, as there doesn't seem to be a way to load an AI via LUA yet) would basically send waves from the other side of the map. Victory after three waves, of increasing difficulty (difficulty determined by how many things the player gets, which would then coorespond to how much stuff the computer gets) or death. That'd be pretty simple to do.
Doing stuff like, I dunno... convoy escort missions (I hate 'em, I hate 'em, I hate 'em, ever since Wing Commander... evil game-designer drek, every time)... would be fairly difficult, since the "convoy" would have to be told to go on a route that was possible (water and mountain detection? Oh, what fun, not.).
Fedex questing type stuff, though, so long as the game design had enough flex (go here, fight off the bad guys, collect more troops, listen to blather, go here) would probably be quite easy, as it'd mainly be a matter of having the script be "smart" enough to avoid bad places on the map (not easy, but not impossible, unless the map consists mainly of water and tiny islands, or is basically all mountains).
Stuff like, I dunno, historical recreations of a specific battle? Nope, can't do it that way. But, as XCOM demonstrated, it's quite fun to present players with random stuff. They can't get a guide for it, and the game designer then has to focus on making the actual mechanics of play interesting, instead of trying to be cute and giving people a puzzle to solve. I think that, in real-time, that'd be very, very interesting, if it was done well. I'll give it a go, when I'm done with the stuff I'm doing right now with World Builder- most of that coding work is directly applicable, and I think it could be fun.