1 paper
Michael Wu, Arquimedes Canedo
LLM agents that cache recovery suggestions from API errors can skip re-derivation in later episodes, spending fewer tokens and fewer model calls on constraints they have already le…