We currently have an answering machine with ringer turned off. Every day (or so), we try to see if it's fielded any calls. Then, when we have time, listen to the messages and decide how we'll handle them.
[This is why folks don't *rely* on our phone for anything urgent]I've been working on an algorithm to passively let us "tell" the answering machine which calls we want to take and which to ignore. Once taught, then let *it* decide to delete messages before even bothering us with them! *And*, know when to deliberately find us for messages that are particularly important: "Please stay on the line while I attempt to locate Don..."
I've written some software to implement an NNTP "agent" -- sits between my Thunderbird "client" and (remote) NNTP "server"... acts like a client to the server and a server to my client! It examines posts (parsing the headers *and* content -- so, *it* fetches all messages from the server) and, based on how I've interacted with that "agent" (acting as a client), decides which messages it wants to "bother me with".
So, for example, if I never look at messages from Joe Whackjob, it soon learns not to *show* me any of Joe Whackjob's messages! This is analagous to a telephone system learning "he never takes phone calls from Joe Whackjob so, when Joe calls, don't even bother ringing the phone *or* taking a message!"
This is what a (good) secretary would do -- learn how you want calls handled and then implement that policy *for* you to save you the effort of doing it yourself.
Of course, with NNTP, I can look at the content and make smarter decisions than just looking at the headers. With telephone calls, I can only look at CID (which can be forged), time of day, and other authentication mechanisms -- i.e., I can only get an idea as to *who* is calling, not WHAT they are calling ABOUT!
[So, a close friend can call and try to SELL ME CAR INSURANCE and I wouldn't know ahead of time that it was going to be a sales pitch :< ]