范文栋-BuildingtheFutureofMulti-agentWorkforce.pdf

1、Building the Future of Multi-agent WorkforceWendong Fan|Core Member of CAMEL-AITech Lead of Eigent AI范文栋CAMEL-AI Tech LeadAI工程师与技术负责人，专注于生成式人工智能（Generative AI）和智能体系统（Agentic Systems）。作为 CAMEL（16k Stars，全球首个基于LLM的多智能体框架）、OWL（19k Stars，NeurIPS 2025，GAIA#1）以及 Eigent（12k Stars，GitHub Trending#1）的核心维护者和主

2、要贡献者之一，建立了自主智能体协作的基础范式。结合在前沿多智能体研究与工业级AI应用方面的双重专长，覆盖完整价值链（供应链、生产、销售），致力于交付可落地的生产级运营解决方案。Agent from 1986Agents are mindless processesAgent by itself can only do some simple thingsJoining these agents in societies leads to true intelligenceWhat magical trick makes us intelligent?The trick is that ther

3、e is no trick.The power of intelligence stems from our vast diversity,not from any single,perfect principle.Marvin Minsky,The Society of Mind,p.308Agent in Reinforcement LearningThe agent-environment interaction in RLThe learner and decision-maker is called the AgentAgent interacts with an environme

4、nt through actionsEnvironment presents new states and rewardsFrom Specific to GeneralLanguage Models as AgentsLilian Weng:https:/lilianweng.github.io/posts/2023-06-23-agent/Key differences:Language as InputLanguage as OutputState and Action are expressed as natural languageGeneralizability.Language

5、Models as AgentsKey Features:-Memory:Manages chat history and context window-Tools:Supports both internal and external function calls-Step Loop:Handle task require multiple request with one stepMore Agents Between than one?Agent Evaluation ResultsCAMEL agents are better one agent 70%on 200 tasksGPT-

6、4 evaluation aligns with Human evaluationCAMEL(NeurIPS 2023):https:/arxiv.org/abs/2303.17760CAMELIdeaRole assignmentTask agentsChat agentsRole-Playing FrameworkCAMEL(NeurIPS 2023):https:/arxiv.org/abs/2303.17760CAMELWorkforceHierarchical ArchitectureTask Planning and DecompositionTask Channel Commun