随着 AI 技术从"工具化"向"自主化"严谨，智能体（Agent）正在成为企业应用大模型的重要形态。那 么，如何优化 Agent，让它变得更可信、更好用，最终能够成为企业优秀的"数字员工"？ 近日 InfoQ《极客有约》X AICon 直播栏目特别邀请、RBC senior application support analyst 马可薇担 任主持人，和值得买科技 CTO 王云峰、商汤科技大装置事业群高级技术总监鲁琲、明略科技集团高级 技术总监吴昊宇一起，在AICon 全球人工智能开发与应用大会 2025 北京站即将召开之际，共同探讨如 何提升企业 Agent 的"可信度"。 部分精彩观点如下： 以下内容基于直播速记整理，经 InfoQ 删减。 定义 Agent 的技术边界 马可薇：很多人觉得 Agent 就是 Chatbot 加了几个插件。但从技术架构视角看，当系统目标从"对话"变 成"行动"，你们认为技术栈上产生的最大一个质变是什么？ 完整的过程包括：模型接收任务，判断应采取的行动，感知外界、接收反馈，并基于反馈不断调整规 划。这与过去单纯的 chatbot 模式有巨大差异，其技术复杂度和对生态的要求都远高 ...

Agent Protocol Landscape - The industry is moving towards interoperability through three open protocols for agentic frameworks [1] - These protocols create a universal language for agents, enabling different frameworks to work together [3] Key Protocols - AG-UI (Agent-User Interaction) facilitates bidirectional communication between agent backends and frontends, enabling interactive agent experiences within applications [1][2] - A2A (Agent-to-Agent) is a protocol for multi-agent coordination, task delegation, and intent sharing across systems [3][5] - MCP (Model Context Protocol) is the standard for agents connecting to tools, data, and workflows [5] Interoperability and Integration - Protocols eliminate the need for point-to-point integrations, allowing developers to build to protocols instead [3] - Frameworks like LangGraph, CrewAI, and Agno can be integrated into the same frontend without rewriting UI logic [3] - CopilotKit unifies the entire stack into one framework, simplifying the implementation of these protocols [4] Example Workflow - A LangGraph agent retrieves data via MCP, delegates analysis to a CrewAI agent via A2A, and streams results to a React app via AG-UI [6]

Agent Protocol Landscape - The industry is converging on three open protocols for agent interoperability: AG-UI (Agent-User Interaction), MCP (Model Context Protocol), and A2A (Agent-to-Agent) [1][2] - These protocols are complementary layers of a stack, not competing standards, facilitating a universal language for agents [2] - Protocols enable integration of frameworks like LangGraph, CrewAI, and Agno into the same frontend without rewriting UI logic [3] Protocol Functionality - AG-UI enables bidirectional connection between agentic backends and frontends, creating interactive agents within applications [1][2] - MCP standardizes how agents connect to tools, data, and workflows [2] - A2A facilitates multi-agent coordination, enabling task delegation and intent sharing across systems [2][5] Framework Integration - CopilotKit unifies the entire protocol stack into one framework, providing generative UI support and production-ready infrastructure [3][4] - An example workflow involves a LangGraph agent pulling data via MCP, delegating analysis to a CrewAI agent via A2A, and streaming results to a React app via AG-UI [6] Development Focus - Protocols allow developers to focus on building agent capabilities instead of integration mechanics, as interoperability is handled automatically [3]

MCP Ecosystem & Specification - The Model Context Protocol (MCP) ecosystem is still in its early stages, with significant room for growth and development [2][3] - The industry emphasizes the importance of adopting the full MCP specification to unlock rich, stateful interactions between agents [9] - The industry acknowledges a gap in MCP implementation, with a tendency to treat it as just another API wrapper [5] - Technical barriers, including missing support in clients, SDKs, documentation, and references, contribute to the limited adoption of the full MCP spec [6] - The industry highlights the need for developers to stay updated with the latest MCP specification and provide feedback on draft features [29] Tools & Dynamic Discovery - Tools are the most immediately successful aspect of MCP, but overuse can lead to quality problems and AI confusion [7][11][12] - Dynamic tool discovery allows servers to provide context-aware tools, enhancing the user experience [16][17][18] - VS Code offers user controls like per-chat tool selection and user-defined tool sets to manage tool complexity [13][15] Resources & Sampling - Resources provide a semantic layer for exposing files and data to both the LLM and the user, enabling more dynamic and stateful interactions [19][20] - Sampling allows servers to request LLM completions from the client, enabling progressive enhancement and interesting functionalities [22][23][24] Developer Experience & Community - The industry recognizes the need for improved developer experience when working on MCP servers, including debugging and logging [26] - VS Code offers a dev mode with debugging capabilities for MCP servers, simplifying the development process [26][27][28] - A community registry is being developed to facilitate the discovery of MCP servers [32]