A lot can be done, and I've come across some great examples. Originally I'd been thinking I'd do writer's planner, but for some reason yesterday or the day before poemranker popped into my head. A scaled down version of the site to just whip through random poems (optionally leaving comments on them).
This somehow, while I was starting to prepare a 320x480 mockup of what the site might look like, snowballed into my redoing how the comments looked and worked. The threaded comment model I devised at the time was horribly rendered, with all sorts of mark-up errors, and it was ugly as hell to boot. I simplified it and beautified it in the span of 30 minutes (and I remember spending hours, hours UPON hours, just trying to get the logic right... but this logic was right, and quick, and took no serious thought... what was wrong with me!?).
Then I applied the same thing to the suggestions nesting.
Then I went through pages and made them validate (though if I give it a proper doctype the rendering gets all #(*$&(#'d up, so leaving it without a proper doctype for now).
CSS practically didn't exist when I started poemranker. And then it did, and I started learning it second-hand (I was a programmer, not a front-end guy), and I solidly did NOT get how css was supposed to work. I made workarounds to work around the limitations I was seeing inherent in the system. Fixed some of those, too (I defined a base set of fonts in java, then used that variable throughout a stylesheet... yeah. ok. fail.)
So that's where my last four or five hours went, somehow, and I've still barely just got a mock-up of what the iPhone web-app's main screen would be.
And to think, I had a livejournal back when poemranker started, practically. Journal entries didn't have tags back then, though. My usage of the poemranker tag apparently only goes back to 2005.