Proposes an overhaul of the agent memory feature to help the agent leverage learned experience to do a better job for the user over time.