Here's something from a very recent August 2026 study that I think more people should know.
Researchers tested different ways of giving AI agents access to memory. And they found something counterintuitive:
More retrieved memory can actually hurt. Why?
Because memory takes up attention.
If the agent retrieves 30 old pieces of information when it only needed 3, some of the useful context gets buried. This is called memory routing.
Basically:
"Which memory system should I use, and how much should I retrieve for this task?"
That's a very different idea from:
"Give the AI access to everything it remembers."
The best memory system isn't the one that remembers the most. It's the one that knows what to bring back