I've just finished the very early alpha of my semantic bookmarking tool. It's currently set up just for me (more on that in a minute). I'm using Notation3 for the semantic markup, and it's extremely simple. 
Basically, the process is quite easy at the moment. When I want to bookmark something, I press a button in my browser. Up comes a page with a box for bashing in N3, with some already pre-filled (the Dublin Core title element). I fill it in, then choose either the "submit" or "submit to delicious" button at the bottom of the page.
Either option then adds the semantic markup to a specified file. This is my 'scratchpad' file. It's stored as raw N3. Occasionally, I go in and do a diff of the current scratch file with a 'public' N3 file. If there's anything that needs changing, I change it and update the public copy and then transform that in to RDF XML. I can then run SPARQL queries over that and display the results however I like.
What's nice about it is that it's a kind of geeky notepad for surfing. I can quite easily dream up a new field to add and type it in. I can work out the relationships between URLs as I bookmark them, or return to them later.
I'm honing this idea of "individual iterative development". I've got an itch that I want to scratch - ie. I want a semantic bookmarking tool because del.icio.us tags don't scale for me. So I've built a little hack that does the job for me. As I discover annoyances with it, I'll fix those annoyances as quickly as I can. If I keep going in that manner, within a few months I'll have something ready to ship.
It's a bit like how you can choose to write a 60,000 word book from start to finish, or you can write lots of bits and pieces and then stick it all together in to a product afterwards. I haven't got time to do a big software construction project. So I'm just piecing together bits until something comes out to be highly cool. Software is my "20% time" for the next few months while I finish my degree, so this development model seems to be a good idea.
Laziness and ennui sets in if you have code that doesn't work. Having code that does something is a huge motivator. In fact, I'd say that the 'boil the ocean' projects that have been coming out of academia in the RDF/SemWeb space are probably the primary reason that RDF/SemWeb haven't happened yet (running code, however primitive, is the best defence against vapourware and the accusation of pipe dreamery).
Similarly, if I build something for me and get the actual guts of the thing working first, then I can do the boring stuff later. For a semantic bookmarking tool, it is the actual bookmarking interface and storage which counts most and the login system least. I've got some good ideas on how to do that - OpenID for one. But it's more important to build the piece that has value first then worry about the rej... [sorry, the rest of this post was cut off due to some XML parsing magic]
