James at Grognardia started it off a megadungeon discussion about how impossible it would have been for Gygax to publish Castle Greyhawk.* Both Joseph at Greyhawk Grognard and Rob at Bat in the Attic raise an interesting counterpoint: although a real, living megadungeon cannot be published in traditional tournament-style module format, what was needed all along was tools and training on how to run a megadungeon off a basic map and a few sparse notes. Megadungeons, or at least the famous ones, were off-the-cuff, modified in response to player actions. They weren't static.
Now, ever since my crude first attempt at random dungeon stocking, I've been working on random tools potentially usable on the fly, although not specifically for megadungeons. I'm not certain anyone can actually teach people how to improvise room contents: we can give some advice, but really it's just something you have to do over and over until you get the hang of it. However, I think I have one or two comments that could kickstart a discussion on how to create a living megadungeon.
For one, start with broad strokes and only fill in details where needed. Published modules often start at the bottom, describing the shape of the room when entry and filling in all the details about the floor, the walls, items in the room, and so on. To improvise, you need to start at the top, describing the entire dungeon or a specific area with a short comparison to something you know from your own experience or research: "The dungeon is an extensive network of burrows dug in hard-packed dirt, with occasional roots".
As the party moves through the dungeon, you use that simple image as a guide to to your description: "Some dirt falls loose in this section of tunnel, almost getting in your eyes"; "Earthworms are poking through the tunnel walls here." You use what you'd imagine could happen in that environment to fill in incidentals, and if the players choose to interact with those details, you use that as a springboard for more ideas. Do they try to dig out some earthworms for potential future use? You could go with either "what if digging will weaken the walls here?" or "what if these aren't normal worms?" and give a small chance that something dangerous or annoying happens as a consequence of their actions: there's a chance of a minor cave-in from digging, or that the worms are blood-sucking parasites. Go with what feels right at the time.
Point Two: describe individual areas in terms of how they are different from their surroundings, again using a top-down approach. Area 3 might just be listed as "giant badger's lair" in your notes, along with stats for the badger and any treasure. You fill in the details by improvising off that word "lair", exactly as you did for the general dungeon description. What do you think of when you imagine an underground animal's lair? There might be some bedding, dry grass it's drug in. Maybe loose bits of fur or clumps of dung. None of these need to be written down beforehand, but any of them could be given a small chance of being useful or creating a complication.
Point Three: in keeping with all this talk about improvised risks, any time one player suggests doing something and another objects "what if X happens?" set a small chance that X will happen. Hell, don't even wait for interplayer arguments: any time a player says they are preparing for some specific, possible bad event, roll for a slight possibility it will happen. Let the players fill in the dangers of the dungeon. For minor bad things, give a 1 in 6 chance. Major random badness should be a 2 on 2d6 or a 3 on 3d6; the more horrible the unplanned event, the rarer it should be.
I've got some more ideas, but I'll save them for a future post.
*(Side note: I think "Schrödinger's Dungeon" is an interesting metaphor for a megadungeon -- each room is both alive and dead at the same time until a room is observed -- but a better metaphor would be "The Tao of the Megadungeon" -- the megadungeon that is published is not the true megadungeon.)