The "right information to the right agent at the right time" idea is exactly where I think agent memory is heading.
The tricky part is that "right" is not only semantic relevance.
In production, "right" also means:
This is where memory starts looking less like storage and more like runtime governance.
A memory system should not only retrieve what is similar.
It should decide what is allowed to become active.
That distinction matters a lot once agents are operating across long-running workflows or multiple roles.