Random Generation and the Endless Game

I accidentally learned about a game called Caves of Qud (pronounced "cud") by discovering the creator's twitter, and the game has thoroughly captured my interest.
It's a Nethack-esque roguelike (is that redundant?) that leans heavily on procedural generation. I'm a big fan of the theoretical "endless" game, so I picked it up.

 

I've made some realizations about how procedurally-generated games and worlds compare to handcrafted ones. Here's one conclusion: I think that procedural generation should ONLY be used for "background" information. Hand-crafted stories, characters and situations are a critical part of enjoying any RPG experience. For example, look at Zelda, Breath of the Wild. Every tree, bush and rock was placed there intentionally. Every part of the experience is optimized for maximum human interest and interaction, and it shows.
 

Qud is full of imaginative descriptions.

Now, let's look at another recent game, No Man's Sky. This game had a lot of hype that it didn't live up to, but it's still on the cutting edge of this kind of game idea. In theory, it's generating an entire universe! Unfortunately, when a computer is rolling a bunch of random variables, no matter how random they are, there eventually develops this feeling of "sameness" that is hard to escape from. 

 

Random generators don't create gameplay.
 

Diablo 3, one of my favorite games of all time, uses procedural generation to create environments, enemies and item drops, but the core gameplay loop is based around really satisfying character progression and optimizing loot. If the whole game had you exploring beautiful, randomly-created dungeons that lacked meaningful gameplay, the novelty would wear off in minutes instead of years.
 

In a game with such deep systems, almost anything can happen.

A game called SPORE attempted to game-ify several different generators, but from what I understand, that game evolved (ha) out of systems the developers were creating for fun. It was a sloppy attempt to assemble those generators into a coherent game, and it almost succeeded. I still enjoyed the game a lot, for a short while.
 

This brings me back to my core argument: Procedural generation is best when used to create "background" stuff. "Stuff" in this case is the terrain, cities, buildings, streets, rooms, decorations and other details that only exist to create a sense of "presence." Imagine I have a story that involves just three "hand made" NPCs, a dungeon and some items. A game developer only has to create those specific elements, lays out the flow of the plot, and then presses a button that fills in the extraneous, unimportant details of the game world. Now the player can move through the game and feel like the world is living and breathing, and the developer can focus on other things

 

Obviously that's way harder than it sounds, but hey. I'm a dreamer, and I've learned that a lot of things can be accomplished if your goals start out sky high. I believe that procedural generation, in some form or fashion, is the future of games.