Texture already exists in human interaction. It’s part of what we actually mean when we say memory. The orientation you hold for a specific person in a specific moment, the weight of the room, the history between you, that’s all live information shaping how you receive what they say. The way we handle memory in AI systems is flat by comparison. It stores what was said and loses the orientation for each turn.
I think that’s why the prompt of “you are this person with this expertise” works as well as it does. It gives the model enough of a posture to be useful. But it only sets up one side. You’re orienting the AI, and you’re giving it almost no orientation of you.
The thing I keep circling is an orientation layer that holds the richness of a long conversation instead of compressing it back to flat text. Not more memory. Better-oriented memory.
When you talk to AI, you spend effort orienting it. Who’s orienting it to you?