I No Longer Do Emails
The written word remains supreme but it's changing.
Email was fine when I first started using it. I’d get something from a friend or family member once or twice a week. I’d check it infrequently. The idea of “notifications” seemed silly. Since it was an analog of paper mail, you just checked it once a day or so, on your terms.
Just like any other piece of tech I’ve used, once it’s user friendly to the masses it ceases to be a useful piece of tech.
I am happy to finally be able to step away from “doing emails” I still correspond via email but I never open up my inbox or Mail.app anymore. I now fully interface with email correspondence through and LLM.
Filtering
It’s now impossible to filter my inbox. Once non-techies started to use email and my inbox became flooded with everything I tried to set up filters. My best one was anything with “unsubscribe” in the body got archived immediately. Once the emails started to get around that filter I’d declare email bankruptcy and move to a new address that only real humans knew of. That approach had its downsides.
Now I have a custom LLM agent that can read my emails. I ask it what’s important and it runs the inbox through a qualitative set of filters I’ve designed in its prompt to suss out:
Is the sender human?
If not does it require my attention?
If the sender is human - does the nature of their message line up with what I want to engage with?
I won’t get into the finer points of my prompt, less it be subverted, but I’ve honed it to a point that the filtering mechanism is a perfect proxy for how I used to sit down and look at emails. I used this approach in parallel with looking at my actual mail app for about a month and noticed zero difference in my LLM avatar and I when it came to filtering email.
Composing
I still write all of my own emails. I will probably continue to do this forever. When I send writing out into the world, it’s my writing, not AI-generated nonsense. My word represents me in the world. I want direct control over it and I want to be responsible for all of its faults too. This stubbornness still works with my LLM-mediated system. I type up replies to emails and then tell the LLM to dispatch them without modifying anything.
I will continue to use this approach for the foreseeable future and believe that this will largely be the pattern everybody uses in a few years once AI integration with existing web paradigms becomes standard and unavoidable. I’m just happy that email feels a bit more sane again, like it did in the 90s.