Evidently there is a technique in which a small piece of Javascript is embedded in a hyperlink on a webpage.  The hyperlink can be dragged onto the toolbar of the browser and executed later.

The cool thing is that the bookmarklet has access to the HTML DOM of the document currently in the browser and can modify or read the contents of the document.  The hyperlink Javascript does not have to be huge.  It can append the current document's DOM with a reference to a script stored on a server that can be much longer and more sophisticated.

I've wanted a way to blog pages I find without always needing to type up all of the stuff that goes around it.  This way I could grab the page title, address, and maybe some metadata as well.  Or how about harvesting the hyperlinks on someone's blog to analyze the intersection with others'?  You could almost begin to draw some conclusions about interests and community based on who links to who.  (Not that I'm smart enough to pull code like that off.  I've never studied AI algorithms, so I'd be reinventing the square wheel, I'm sure.)

Jon Udell gives a great example of a bookmarklet that allows him to request books that he finds on Amazon from his local library.  Now that's integration!  The concept of an agent connecting two diverse entities is called an intermediary.