So I’d have the bookmarkable friendly URIs be capable of returning both human-friendly and machine-readable representations via Accept, and maybe also have them respond to programmatic access (by PUTing/POSTing with a request body with XML or JSON or something like that), but I’d have a separate URI space to handle the browser-based editing UI.

My plan is to provide RESTful access via the Accept header types of text/xml and application/xml, and web access (menus, CSS) for application/xhtml+xml and text/html. The problem that I'm trying to solve now is that Firefox uses the following accept header: Accept: text/xml,application/xml,application/xhtml+xml,text/html;q=0.9,text/plain;q=0.8,image/png,*/*;q=0.5. I guess this means my application will always have to provide HTML of some sort, unless the script asks only for XML, in which case I provide XML, in a "Just the facts, Ma'am" [1] format.

1. Obligatory link to Snopes for the pedants in the crowd. The quote is actually "All we want are the facts, ma'am."