AI agent memory reportedly isn't automatically portable between models

A user summarizing a LinkedIn paper describes tests in which one model inherited another's memory. The summary says fixed-format memory barely changed across the two models, while free-form notes were less reliable. Rewriting compressed notes could not recover facts already lost.

For retrieval, the post says mixing old and new embeddings—data representations used for search—in one index recovered only 4.96 of the 11.90-point gain from fully rebuilding it. The user recommends treating model upgrades as memory migrations: test compatibility, rebuild indexes fully, prefer structured memory where it fits, and retain protected raw history when policy allows.