AgentFactory: Towards Automated Agentic System Design and Optimization
AgentFactory searches both model choices and workflow structures, judging them on performance, cost, and efficiency.
The paper presents a three-stage pipeline that uses LLMs as optimizers to find task-specific agent configurations. It evaluates the system on eight benchmarks across reasoning, coding, math, medicine, and finance. The authors report a 9.1% average gain over manual and automated baselines, with larger gains on MedQA and FinEval. ArXiv · AI/CL/LG's note
The paper presents a three-stage pipeline that uses LLMs as optimizers to find task-specific agent configurations. It evaluates the system on eight benchmarks across reasoning, coding, math, medicine, and finance. The authors report a 9.1% average gain over manual and automated baselines, with larger gains on MedQA and FinEval. ArXiv · AI/CL/LG's note
score 5