Tag
This paper proposes the Agent-Editing World Model (AEWM) to enhance LLM agents by modeling how reasoning and actions shape future task progress, achieving significant performance improvements on benchmarks across domains like search and software engineering.