What specifically about them do you think is new? I don't know Clojure so I can't judge it, but the idea of "agents" goes back decades, and languages such as Erlang that have been around for a long time have implemented various versions fairly well.
Perhaps there is something new, but the space is fairly well covered by lesser-known-but-usable languages, if unheard of in the mainstream. (I use that phrase to draw a distinction against purely academic languages; implementing things there hardly counts.)
Well, message passing actors have their own theory and calculus (based on particle physics, actually), and there have been "agents" in the form of interprocess mobile code, but the idea of pure functions as messages and the recipient (the so called "agent") being just a dynamically typed blob of data with an associated thread and possible associated watchers - that's new. I don't think it has a theory, yet.
Perhaps there is something new, but the space is fairly well covered by lesser-known-but-usable languages, if unheard of in the mainstream. (I use that phrase to draw a distinction against purely academic languages; implementing things there hardly counts.)