今天继续这个话题，事实上这个系列不太好写，写深入了容易将现在正在做的项目技术路径泄露，写浅了又有点隔靴搔痒，但其实现在很多公司都有类似 的问题： 1. AI聊得不像人，最常见案例就是生硬，就算上RAG或知识库也不好使； 2. AI准确率不高，最常见就是AI能覆盖80%的场景，但业务的及格线是95%； 这些问题都与我们探讨的问题相关，其中准确率不高这个是专家系统需要解决的任务；而聊得不像人这就比较麻烦了，策略层面涉及了Cot，技术层面暂 时多与RAG相关。 只不过就RAG这个技术，要用好的公司也不多，为避免泄露当前项目技术机密，今天我就借Douwe Kiela（RAG 技术的最初开创者之一）提出的10个宝贵 经验，来聊聊如何做好RAG这件事。 上下文悖论 The Context Paradox，莫拉维克悖论指出：对计算机而言，执行人类觉得困难的任务（如下棋）比执行人类觉得容易的任务（如行走、感知）更容易。 其实这一观点与RL 之父 Rich Sutton某一观点十分类似：依靠纯粹算力的通用方法，最终总能以压倒性优势胜出。 他特别提出：AlphaGo/GPT-3的成功并非源于复杂规则，而是大规模算力支撑的简单算法 ...

Core Insights - The year 2025 is referred to as the "Year of Intelligent Agents," marking a paradigm shift in AI development from "I say AI responds" to "I say AI acts" [1] - The report aims to address whether safety and compliance are ready as intelligent agents rapidly evolve, focusing on their latest developments, compliance awareness, and actual compliance cases [1] Group 1: Definition and Classification - The concept of intelligent agents is currently hot in the market, but definitions are often confused, leading to varied interpretations [2] - OpenAI categorizes AI development into five stages, with L3 representing intelligent agents capable of autonomous planning and execution of complex tasks, along with dialogue, reasoning, long-term memory, and tool invocation capabilities [2] - Intelligent agents' autonomy and interaction capabilities create a core contradiction between utility and risk, necessitating a value ecosystem based on "tolerance" and "autonomy" [2] Group 2: Types of Intelligent Agents - Intelligent agents are divided into general and vertical types, each with significant differences in technology stack, optimization goals, and application scope [4] - General intelligent agents can operate across multiple domains, while vertical intelligent agents focus on specific fields, integrating specialized knowledge and industry data for more precise training outcomes [4] - Vertical intelligent agents are gaining traction in sensitive and regulated industries like finance and law, where compliance and data security are paramount [4] Group 3: Market Dynamics - The intelligent agent market is characterized by a complex "co-opetition" relationship among tech giants, startups, and terminal manufacturers, with players intersecting across various industry segments [5][8] - Major tech companies are building comprehensive "intelligent agent factories" by leveraging large models, funding, data, and cloud infrastructure to attract developers [8] - Startups are innovating in core intelligent agent capabilities while simultaneously competing with tech giants, creating a dynamic competitive landscape [8] Group 4: Industry Applications - Intelligent agents are increasingly being integrated into hardware, with smartphone manufacturers upgrading their devices to feature AI capabilities [12] - AI smartphones are projected to penetrate the market significantly, with an expected penetration rate of 34% by 2025, driven by advancements in edge computing and chip capabilities [12] - AI browsers are also emerging, incorporating intelligent agents to enhance user interaction and streamline web navigation [13] Group 5: Value Ecosystem - A comprehensive understanding of intelligent agents requires a model based on "tolerance" and "autonomy," which can help position various intelligent agent products within a value ecosystem [14] - The X-axis represents "tolerance," indicating the severity of consequences from errors, while the Y-axis represents "autonomy," measuring the agent's decision-making capabilities without human intervention [14]

近日， 未可知人工智能研究院讲师 Jeffrey 为 爱依瑞斯 举办了一场主题为 "AI赋能新零售，助力家居业务增长 "的培训 ，深入探讨了AI技术 在家居销售的革命性应用。AI正在深刻改变家居零售行业的游戏规则。 Jeffrey用一个翻译案例开启了课程。传统人工翻译15万字的书籍需要1.8万元和20天时间，而使用AI工具仅需5小时，成本几乎可以忽略不 计。这个对比让学员们直观感受到AI带来的效率革命。他借用二战时期日军因固守步枪而轻视冲锋枪的历史教训，形象地比喻当下企业如果不 及时拥抱AI技术，就会在竞争中处于劣势。 课程最后，Jeffrey引用英伟达CEO黄仁勋的话作为结语："未来每一份工作都会受到AI的影响，有的岗位会消失，新的会出现，而剩下的则会 被重新定义。"这引发了学员们的深思。随着AI技术的快速进化，家居行业的增长密码就在于选择合适的工具、找到落地场景、建立有效的人机 协作机制。 合作联系微信：duyuaigc 针对家居行业的具体痛点，Jeffrey给出了切实可行的AI解决方案。当产品展示遇到困难时，可以使用即梦AI一键生成3D效果图；面对高昂的 营销成本，可以通过DeepSeek结合飞书多维表 ...

Core Insights - The main narrative in the industry revolves around the adoption of AI Agents, with MiniMax being one of the key players launching new products in this space [1][2][5] - There is a significant gap between the current capabilities of AI Agents and their potential for commercialization, as many companies face challenges in practical implementation [9][10] Product Launches - MiniMax recently introduced the Hailuo Video Agent, which generates professional-quality videos based on user inputs, aiming to democratize video creation [2] - Another product launched is a Long Horizon general-purpose Agent capable of multi-step planning and executing complex tasks, which has seen adoption by over 50% of MiniMax's internal employees [5] Market Position and IPO Plans - MiniMax is reportedly considering an IPO, with a current valuation of approximately $3 billion (around 21.6 billion RMB), and has engaged financial advisors for this process [7][8] - The company previously raised $600 million in Series A funding in March 2024, with a valuation of about $2.5 billion (around 18 billion RMB) [7] Industry Trends - The AI Agent market is rapidly evolving, with many startups and established companies like ByteDance and Alibaba entering the space [6] - The year 2023 is being referred to as the "year of AI Agents," with various companies launching their own products to capture market share [5][6] Challenges and Limitations - Despite advancements, the technology behind AI Agents is still not fully mature, leading to challenges in real-world applications and user satisfaction [9][10] - Industry experts, including Andrej Karpathy, express skepticism about the readiness of AI Agents, comparing their development to the ongoing challenges in autonomous driving technology [10]

金融工程 证券研究报告 |深度研究报告 2025/06/13 AI投研应用系列之二： 从大模型到智能体，扣子Coze在金融投研中的应用 马自妍 S1190519070001 证券分析师： 分析师登记编号： 刘晓锋 S1190522090001 证券分析师： 分析师登记编号： P2 目录 请务必阅读正文之后的免责条款部分 守正 出奇 宁静 致远 1. AI Agent赋能投研应用 2. Coze核心功能解析 3. Coze投研应用实践 4. Coze投研应用相关插件 5. Coze在投研领域的应用前景展望 1、AI Agent赋能投研应用 1.1 AI Agent助力智能投研落地 2025年大语言模型（LLM）在技术层面迎来爆发式发展，但在实际应用落地中仍面临一定的局限：复 杂任务拆解能力不足、多工具协同效率低、专业场景适配成本高等问题，制约了其从技术能力向生产 力的转化。 AI Agent通过整合LLM的核心认知能力与外部工具、自动化工作流及领域知识库，助力大模型在智能 投研场景应用中高效落地。 字节跳动于2024年2月推出Coze扣子平台，2025年4月发布"扣子空间"协同办公系统，并于2025年5 月全面 ...

优于大市·维持 021-61761067 021-60875168 证券分析师：熊莉 证券分析师：库宏垚 证券研究报告 | 2025年06月13日 计算机行业 2025 年 6 月暨中期投资策略 优于大市 AI 产业快速迭代，持续看好 Agent 和算力租赁 核心观点 行业研究·行业投资策略 计算机 证券分析师：艾宪 0755-22941051 aixian@guosen.com.cn S0980524090001 市场走势 xiongli1@guosen.com.cn kuhongyao@guosen.com.cn S0980519030002 S0980520010001 资料来源：Wind、国信证券经济研究所整理 相关研究报告 《稳定币香港政策落地，关注板块投资机会》 ——2025-06-04 《人工智能专题报告：国内大厂扩张资本开支，算力租赁订单持 续落地》 ——2025-05-21 《计算机行业 2025 年 5 月投资策略暨财报总结-大厂布局 Agent 产品，AI 应用快速落地》 ——2025-05-08 《人工智能行业专题：2025Q1 海外大厂 CapEx 和 ROIC 总结梳理 -2025 ...

Investment Rating - The report maintains a "Recommended" investment rating for the AI industry, highlighting the continuous evolution of domestic agents and the promotion of universal protocols for systematic applications [2]. Core Insights - The AI agent field is experiencing rapid advancements, with capabilities doubling approximately every seven months since the release of ChatGPT in 2022, leading to an exponential increase in the tasks that AI agents can complete [5]. - Major internet companies are competing to capture the AI internet era's new interaction entry point, with AI agents inheriting the core attributes of the app era and showing advantages in service delivery, network effects, and developer ecosystems [9]. Summary by Sections Section 1: Domestic Intelligent Agent Technology Innovation - The report discusses the launch of ByteDance's Coze Space, a universal agent product designed to facilitate efficient collaboration between users and AI agents, enabling the completion of complex tasks [13][14]. - Coze Space features two collaboration modes: an exploration mode for lightweight tasks and a planning mode for complex needs, allowing for real-time user input during task execution [17]. Section 2: MCP Open Protocol Continues to Expand - The introduction of the Agent2Agent (A2A) open protocol by Google aims to enable communication and secure information exchange between different AI agents, enhancing collaborative task execution across platforms [69]. - A2A facilitates interactions between client and remote agents, allowing for capability discovery, task management, and secure collaboration, thereby increasing operational efficiency [72]. - Microsoft announced plans to leverage the MCP to create an open "agentic web," where AI agents can autonomously initiate tasks and collaborate with minimal human oversight [80]. Section 3: Related Targets - The report identifies several companies involved in AI agent development, including Zhuoyi Information, which is focused on low-code integrated development environments (IDEs) that incorporate AI capabilities to enhance coding efficiency [84]. - Puyuan Information is integrating advanced large model capabilities into its low-code development platform, enhancing developer productivity through intelligent assistance [89]. - Hehe Information has launched an MCP server service for document processing, significantly improving the efficiency of intelligent document handling tasks [90].

计算机行业 投资评级 推荐 维持 国产Agent不断演进，通用协议推进系统性应用 ——AI行业专题报告 证券研究报告 2025年6月9日 分析师：王朗 邮箱：wanglang2@gyzq.com.cn SAC执业资格证书编码：S0020525020001 目录 • 第一部分：国产智能体技术创新，C端产品百花齐放 请务必阅读正文之后的免责条款部分 2 • 第二部分：MCP开放协议持续扩圈，跨平台协作提升Agent商业价值 • 第三部分：相关标的 • 风险提示 1 国产智能体技术创新，C端产品百花齐放 AI Agent领域的新摩尔定律，显示智能 体的能力急剧快速提升。 2022年ChatGPT发布以来，大模型的能 力持续快速提升，带动AI智能体的能力 提升，进而促进人们竞逐研发更强大的 AI模型。 根据AI研究网站AI Digest发布的研究， AI智能体能够完成的任务时长呈现指数 增长的趋势，其中，任务长度指的是专 业人士完成这些任务需要的时间，从不 到30秒到超过8小时不等。目前，智能体 可以自主完成人类需要一小时才能完成 的编程任务，顶尖的AI系统可以完成的 任务长度正在呈指数级增长——每7个月 翻一番。 ...

Core Insights - The article discusses the competitive landscape of Agent platforms, highlighting the importance of factors such as traffic, data privacy, tool ecosystem, and addressing hallucination issues in vertical domains [1][2]. Group 1: Agent Platforms Overview - Dify has established an early presence in the open-source community, but faces competition from platforms like FastGPT and N8N [3]. - FastGPT, along with Dify and Coze, emphasizes core functionalities such as visual workflow orchestration, a no-code platform, and a toolchain that includes model selection and knowledge bases [4][11]. - FastGPT's tool ecosystem is noted to be weaker compared to Coze and Dify, lacking depth in vertical tools and general life/efficiency tools [7][8]. Group 2: Platform Comparisons - Coze is designed for rapid deployment and ease of use, making it suitable for business departments with tight timelines [26]. - Dify offers a comprehensive LLMOps capability, balancing flexibility and control, ideal for medium to large teams that require private and cloud service options [26]. - N8N is positioned as a workflow automation engine, providing over 500 nodes and script mixing for efficient cross-system integration, catering to development teams [26]. Group 3: User Preferences and Use Cases - Developer preferences for Agent platforms focus on freedom, extensibility, and privatization, while product/operations teams prioritize no-code solutions, visualization, and quick validation [19]. - For quick deployment of a Q&A bot with minimal coding, Coze is the preferred choice, while N8N is favored for complex integrations and custom logic [23][24]. - The article emphasizes that no single platform can meet all needs, suggesting common combinations of platforms for different tasks [28].

证券研究报告/行业研究 MCP 协议加速 AI Agent 生态繁荣 ——人工智能行业专题研究 投资要点 ➢ AI Agent 是AI 发展的第三阶段 2024 年 11 月，Anthropic 发布 Model Context Protocol（MCP），自推 出以来，MCP 迅速成为 AI 原生应用的重要基础设施。MCP 协议如同 AI 应 用的 USB-C 端口，其最关键的设计理念是将"工具调用"与"上下文感知" 统一纳入一个协议框架，使得模型与外部世界之间的交互不仅更自然、更精 准，还可以跨模型平台共用。MCP 协议正在成为 AI 领域连接大模型与外部 世界的核心基础设施，提升了 AI 模型与外部服务的兼容性。预计未来 MCP 协议+Agentic-based 决策路径或将成为主流。 ➢ 科技巨头积极布局AI Agent 产品 从字节和阿里等科技公司近期的动向来看，AI Agent 或成为今年科技公司 布局 AI 的重要主线。整体来看，AI Agent 产业在 2024 年第四季度至 2025 年初呈现快速迭代态势，并逐渐从技术竞争转向生态价值重构。AI Agent 领 域的发展还呈现出明显的结构性分 ...