>How can this not be world-changing? For one, the principal-agent problem. Who's building, running, and maintaining these agents? I doubt that these are runnin

>How can this not be world-changing? For one, the principal-agent problem. Who's building, running, and maintaining these agents? I doubt that these are running locally and tweaked by the user themselves. The business model is to sell these things as a service running in the cloud. That means the operator has final say on how these systems are implemented. Sure, your helpful AI agent today may remind you to pick up some soda at the store. But if the last decade+ of "enshittification" has shown, it's only a matter of time before your helpful AI agent will remind you to pick up some delicious Mountain Dew (TM) at the Piggly Wiggly (TM) and use your American Express (TM) card to earn Roblox (TM) points. And even if we assume these agents don't eventually come to serve masters other than the end-user, the internet is already a user-hostile place. It's entirely reasonable to believe it'll be agent-hostile too, where APIs and UI designs adapt to trick AI agents with "dark patterns" to do things that aren't necessarily in their users' best interests. Techno-utopianism might feel good, but economics and money-making will drive how these things actually play out.