Post

It often strikes me how many proposed uses for agents rely on the assumption that they're the only ones with them, they'll be ahead, & nobody will adapt policies accordingly. "Spam insurance to get denials appealed!" like insurance wouldn't be using agents to mass-reject those same appeals

but like... why. why would you assume this is something that can actually, economically, function. the technology can always produce more spam and slop than it can produce useful output, the bad output always outpaces and drives out the good.

The question is whether the hour of dealing with a support bot and a phone tree and just "being on hold" that is necessary to deal with the cable company exists because your cable company can't do better than that, or because putting a time tax in front of users is how they scare off requests.

immanuel kant warned us about this

isn't it basically everyone paying to add friction to the system... up until that friction melts the system? As you say, it doesn't really seem viable to actually produce outcomes, because it's just going to turn things off over time

Possible futures for agents -- This all falls apart (likely) Distributed identity follows these things around so you only interact with well behaved agents (unlikely but interesting) Walled gardens again (they hope)

bsky.app/profile/mcmi...

Cloudflare wet dream