“Researchers propose Memory Worth, a new metric for managing which memories AI agents should keep or discard as their task environments change. Unlike static importance scores, this approach uses outcome feedback to dynamically govern memory quality, helping agents adapt more effectively when task distributions shift.”