1 paper
Jiawei Du, Jinlong Wu, Yuzheng Chen +3
Most LLM-based agent frameworks adopt a top-down philosophy: humans decompose tasks, define workflows, and assign agents to execute each step. While effective on benchmark-style ta…