A useful assistant still changes the room

When an assistant becomes persistent, the social question is not whether it feels human. It is who can see its memory, who can correct it and how its presence changes the people around it.

OpenAI’s recent writing on AI use describes systems moving from short interactions toward longer, delegated work across departments and occupations. That transition changes the setting of assistance: AI is no longer only a private answer box, but a participant in shared workflows. OpenAI on agents and work

Presence creates expectations

A persistent assistant can remember preferences, carry context and act between meetings. Those conveniences also create expectations of availability and continuity. People may begin to treat an automated record as the room’s memory even when it is incomplete, selectively exposed or difficult to challenge.

Society needs better questions than whether an assistant is friendly. Who gets to inspect the memory? Who is excluded from it? Can a person refuse participation without losing access to a service? What happens when the system’s account of a conversation becomes the account that institutions trust?

The social design of assistants is therefore governance in miniature. A useful system should make its memory and authority visible enough that people can remain participants, not merely sources of training data for a room that has quietly acquired a machine secretary.