That's what we do, using the Reply-to header, in case people hit reply. But the From header (which is what users see) still has to be one of ours. This is a requirement of SendGrid, AWS SES, etc., and it'd look pretty spammy otherwise.
Also, I just checked the latest notification I received from Facebook (someone posted on my wall) and its headers were
I can't recall which service, it's right on the tip of my tongue, but I've seen it done before that such a reply would end up being routed to the correct user. In the simplest case you get "X Commented on your Post!" sent from your service's account but with reply-to set to X's address. If you wanted to obscure X's address, you could process replies to your service's address and route them accordingly.
So then maybe you could give second thought to sending these mails in the first place. For me, they are the first thing I filter to /dev/null when I sign up for social-like service. I have in the ballpark of 300 people in my circles - so I just have to filter the updates like these in order to be able to actually use the mail account :-)
Maybe I am a super-extravert with a ton of similarly hyperactive friends, so take this with a grain of salt :)
However, I've had to introduce noreply for a lot of transactional emails for a reason I didn't expect.
If you send an email like:
...even if you mention that replying to the email will come to you, not X, many users will instinctively reply. E.g., Then you feel bad that you're intercepting personal emails.I switched to noreply and put a help email link in the signature (which is ridiculed as the height of too muchery in the article).
If anyone's got a better solution, I'd love to hear it.