More thoughts on agent memory
I think “memory” is the wrong term altogether. It mostly reflects people anthropomorphizing agents. What we should be talking about is lineage and heritage, because agent sessions come and go and do not persist across long spans of time. The closest thing an agent has to memory is like books or oral history for humans. I have no recollection of the Civil War because I did not exist when it happened; I can only read or hear stories about it. An agent can save things it encounters to memory files, skill files, or harness instructions, but those only persist while the underlying system exists, and they function as “memories” only while that agent session is running. When a new session starts, it does not “wake up” and wonder what it did yesterday. It receives its primary instructions, then can consult the records left by its lineage.
