Attention new products: Focus on your core feature
fliggo.posterous.com
fliggo.posterous.com
Again: We at JamLegend call it the minimum valid test. Others call it core features, but it's what you need to get a prototype to market.
You can either launch too early and it doesn't compare with other more mature services feature wise, or launch too late, and the software is a bit bloated or another service has got a massive lead by introducing one of your main features! (or something along those lines)
If you are building a new product, build the simplest thing possible to complete the use case. Don't worry about the fonts or ajax, don't worry about anything other than 'did we meet our core scenario -- did the user get what he/she wanted?' And then test this on the user -- does this help you?
If you are building a tweak or revision of some existing thing, go for the out-there features and see if those actually stick to users. If you are writing a new blogging platform, for example, it's expected that you'll be able to add, remove, and edit posts, and you'll have a great RSS feed. Users will assume you'll have that, and if you don't have it right now, you'll add it. In this case, build the parts that are truly novel. And then check this with your users -- is this something they want?
A good example is Zenter (YC startup acquired by google). They didn't build the "save" feature into their web presentation app until very late because it's assumed that you can save a presentation -- they instead focused on the cool stuff that people didn't expect -- collaborating on the web on your presentation, sharing it live, etc.
The bottom line is, build the minimum form of the unique thing first, and see if it sticks. Don't worry about being too simple -- your users will let you know if you don't meet their needs.