The impetus for my current game was the WM approach. With my player base and limited flexibility I wasn't able to make it work but I did take much of it to heart. I shipwrecked my players, described their initial surroundings, and made them initiate the action.
As we have progressed and they explore the world they learn things. They just discovered and map and a lighthouse. They still have an island. At the lighthouse they discovered there were some missing people. My players can pick up the missing people thread, continue to explore the island or figure out where they want to go next from the world map they discovered. I don't care which thing they do. While on the island they encountered a tribe of Maori Cannibal Halflings. The halflings and the lizardfolk on the island had been in a standoff. Due to my player interference they aren't any more, they are actively at war.
Should my players go back into the jungle they will find evidence of that war. They might become embroiled in it, they might not. In either case it's being fought due to their interference in the world.
If my players decide to bash together a small boat they can do that and leave the island behind.
You can have both an open world with decision making by the players, and NPCs with plots. The players just have to be the ones making the decisions. There is no 'railroading' in my game. I don't have a vested interest in the outcome of any decision they make. If they uncover something and ignore it, so be it. If they uncover something and take up the plot then you get a more traditional game for a while but they always have the option to get distracted by something else.
The WM game works because of the player base, and the competition between players. It is almost purely site based though. In the Smoke Woods is a tribe of goblins that worships fire, a druid ring, several animal lairs, and the ruins of an elven city. The groups may compete to explore those sites, or keep them secret from each other, but they aren't doing anything except kicking down the door and exploring the elven ruins as a site.
Maybe one group trades with the goblins while another goes to way. Now the goblins have some kind of motivation that you as the DM have to deal with in a reasonable and realistic way so that the world makes sense to the players. That is akin to having NPCs in the world with motivations.
One of the most valuable things I did though was send them the link to the WM articles and ask them to read them. It put them in the proper mindset to be initiators of the action rather than waiting for joe in the bar to ask them to do something.
Mathius: go read the Alexandrian hexcrawl articles linked earlier in this thread. Step 1: make a hex map; step 2... step 3: make a hex map key; step 4: profit!
All kidding aside making a map key has helped me immensely for prep. I have a hex map (insanely large so I doubt all of it will see play) but I started keying hexes likely to be explored. This helped me key 8 of the 12 "known areas of interest" on the surface at the megadungeon. Between the hexes I've keyed so far and these 8 areas, I have enough gaming to last the entire campaign.
The tip I got from the Alexandrian? Small write-up, and wing it.
For example in one of the hexes I grabbed the GMG and rolled on the "Scenic Spots" table. I ended up with "Village Pond." How the heck do I have a village pond in the middle of a wilderness hex of grassland/hills?
So I gave it some thought. I've said the whole area in my campaign was overgrown by supernaturally charged wilderness. So there's the ruins of a village and a very large village pond. At one time it'd been scenic, but now it's marshy, overgrown and surrounded not only by the fieldstone bones of some of the dugout huts of the village but also gnarled old shrubs and trees, such that you can't see very well directly into the pond.
My writeup states: "a great, marshy pond is surrounded by ruined cottages and gnarled scrub. Some of the huts are crudely thatched and repaired." In parentheses I added a note about evil Halfling cultists worshipping an aberrant god (Rovagug?) and the pond being home to some Grindylow.
So a few sentences, and I'm done keying the hex. Simple right? Then if the PCs venture there I give them the first two sentences. If they investigate, I wing the halflings. If they stay with it, maybe they discover the truth about the halflings' beliefs. Conflict would ensue then, perhaps followed by a run-in with the grindylow.
If they DO encounter the grindylow, I'd drop a bunch of clues. These will hopefully lead the players to the conclusion I'd like them to find: the grindylow were once villagers here. They turned to a dark power when cataclysm struck and were cursed into this hideous form. The secret of their power and curse is somewhere under the fetid water of the pond.
What're the clues? What's the source of the secret? What needs to happen to stop it all? I have no idea, but I'd probably wing it.
I don't know where i read it (probably the Alexandrian), but looting the internet should definately be added to winging it. For example the one page dungeon contest produces loads of handy one page instant dungeons.
Some links:
one page dungeon contest homepage
Hexcrawl part 10: Stocking the Hexes (at the Alexandrian)
Search the Alexandrian for hex crawl
... even more dungeon resources links
While linking to this excellent thread multiple times I feel the need to add some points to the plotting villain problem.
The Alexandrian (surprise!) has excellent articles on node based design which is a technique I think one could leverage for the plotting villains in a sandbox. Design and interconnect the nodes that villains drive and see where the players intersect and influence them.
I believe Razor Coast does this offering 3 major driving villainous plots for the players to interact with if they choose.
You could argue that this only creates more and different rails but it is definitely better than the monorail. And having a sufficient number if rails would make a campaign still a sandbox IMHO.