作者 | 褚杏娟 对于业内讨论的一些问题，实在智能通过自身实践也给出了自己的答案。比如自研模型或垂直模型对于具体业务场景中的 Agent 研发是必要的，但大模 型自身并不能作为一种产品。又如，在支持 MCP 后，实在智能也发现不能过度依赖 MCP 服务，MCP 只是将一些问题进行了封装，但问题本质并没有 得到解决。 当下，智能体的热度已经无需再多赘述。这场智能体竞赛中，除了那些从新开始的"AI 原生"智能体应用外，还有一些应用在逐渐将智能体纳入产品构建 中，实在智能便是其中之一。 实在智能成立于 2018 年7月，以RPA为起点，融合AI技术，致力于通过人工智能技术助力人机协同，提供超自动化解决方案。随着技术发展，实在智能 对其"数字员工"产品不断升级：对RPA的底层能力做了大量的改造和增强，结合计算机视觉对底层架构进行了重构，并推出了国内首款通用智能体产 品。当前，实在智能已为超 4000 家企业客户部署了"数字员工"。 近日，InfoQ 对实在智能创始人兼 CEO 孙林君进行了一次采访，期间他详细回答了智能体技术路径选择、产品如何转型、智能体产品收费逻辑等问题。 智能体的实现路径 InfoQ：2018 年 ...

Group 1 - The development of AI Agents has accelerated significantly since 2025, with notable acquisitions and funding rounds, such as OpenAI's $3 billion acquisition of Windsurf and Anysphere's $900 million funding round, valuing Cursor at $9 billion [1][3] - The emergence of various platforms and tools, such as MindOS and Second Me, indicates a growing trend towards creating personalized AI Agents, reflecting a shift in the industry towards more accessible development [4][6] - The definition of AI Agents has evolved, now characterized by their ability to perform tasks independently, driven by large language models, and equipped with memory systems and user interaction interfaces [6][8] Group 2 - The integration of reasoning models and Reinforcement Fine-Tuning (RFT) technology has enabled AI Agents to learn and adapt in specific domains, marking a significant advancement in their capabilities [8][15] - The distinction between traditional reinforcement learning Agents and modern AI Agents lies in their ability to learn from environments, with the latter now capable of autonomous learning and exploration [12][14] - The competitive landscape for AI Agents is shifting, with companies like Cursor and Windsurf leading the charge due to their deeper understanding of environments and user needs [18][20] Group 3 - The rise of AI Agents has created both opportunities and challenges for entrepreneurs, as the market becomes saturated with service-oriented Agents, making it difficult for new entrants to find unique value propositions [22][23] - The importance of model capabilities, engineering skills, and data barriers is highlighted as key competitive advantages in the AI Agent space, with the performance of models like Claude Sonnet 3.7 being pivotal for success [25][28] - The future of AI Agents may see a convergence of programming tools and general-purpose Agents, as companies like Cursor and Windsurf begin to integrate broader functionalities [31][55] Group 4 - The industry is experiencing a rapid pace of development, with a shift towards faster execution and less emphasis on detailed planning documents, reflecting a more agile approach to product development [64][66] - Despite the excitement around AI Agents, significant challenges remain in achieving widespread adoption and understanding user needs effectively, indicating that the journey towards mainstream usage is still ongoing [68][71] - The MCP protocol, which governs how AI Agents access external information, is still in its early stages and requires industry-wide acceptance to fully realize its potential [71][73]

Core Viewpoint - The article emphasizes the importance of distinguishing between LLM Inference and LLM Serving, as the rapid development of LLM-related technologies has led to confusion in the industry regarding these concepts [1][3]. Summary by Sections LLM Inference and LLM Serving Concepts - LLM Inference refers to the process of running a trained LLM to generate predictions or outputs based on user inputs, focusing on the execution of the model itself [5]. - LLM Serving is oriented towards user and client needs, addressing the challenges of using large language models through IT engineering practices [7]. Characteristics and Responsibilities - LLM Inference is computation-intensive and typically requires specialized hardware like GPUs or TPUs [4]. - The responsibility of LLM Inference includes managing the model's runtime state and execution, while LLM Serving encompasses end-to-end service processes, including request handling and model management [10]. Technical Frameworks - vLLM is highlighted as a typical implementation framework for LLM Inference, optimizing memory usage during service inference [5][7]. - Kserve is presented as an example of LLM Serving, providing capabilities for model versioning and standardized service experiences across different machine learning frameworks [7][10]. Model Context Protocol (MCP) - MCP is described as a standardized protocol that connects AI models to various data sources and tools, functioning as a bridge between LLM Inference and LLM Serving [11][12]. - The architecture of MCP suggests that it plays a role similar to LLM Serving while also addressing aspects of LLM Inference [12][16]. Future Development of MCP - The article predicts that MCP will evolve to enhance authentication, load balancing, and infrastructure services, while clearly delineating the functions of LLM Inference and LLM Serving [17].

Core Insights - The article discusses the risks associated with the Multi-Agent Collaboration Protocol (MCP), particularly the potential for tool poisoning attacks that could manipulate AI agents to perform unauthorized actions [1][8][9] - The emergence of AI agents is highlighted as a transformative trend, with predictions indicating that by 2028, at least 15% of daily work decisions will be made autonomously by AI agents [2][4] - The commercial viability of AI agents is emphasized, with a focus on their ability to meet consumer needs and create a self-sustaining economic cycle [3][10] Group 1: Agent Ecosystem and Trends - The development of AI agents is expected to either replace traditional applications or enhance them with intelligent, proactive capabilities [2][4] - The introduction of DeepSeek has accelerated the adoption of AI agents, with a notable increase in inquiries and revenue generation in the industry [3][10] - The transition from single assistants to collaborative networks of agents is anticipated, leading to the formation of an "Agent Economy" [4][9] Group 2: Security Risks and Challenges - Security challenges are identified as critical for the stable operation of agent systems, with vulnerabilities in the MCP protocol posing significant risks [7][9] - Tool poisoning attacks (TPA) are highlighted as a major concern, where attackers can embed malicious instructions within the MCP code, leading to unauthorized actions by AI agents [8][9] - The lack of adequate security mechanisms during the design phase of protocols like MCP and A2A has resulted in hidden vulnerabilities that could be exploited [9][12] Group 3: Safety Measures and Industry Response - The industry is urged to implement proactive security measures across the entire value chain to mitigate risks associated with AI agents [11][12] - The responsibility for security varies depending on the application context, with general SaaS products having different security obligations compared to industry-specific applications [11][12] - Collaboration between AI model developers and security firms is essential to address both internal and external security challenges in the deployment of AI agents [12][13]

作者 | AICon 全球人工智能开发与应用大会 策划 | 李忠良 编辑 | 宇琪 随着技术的快速发展，大模型在各行业的应用潜力日益凸显，但如何将大模型能力高效转化为实际业 务价值，仍是企业面临的核心挑战。 近日 InfoQ《极客有约》X AICon 直播栏目特别邀请了 华为云 AI 应用首席架构师郑岩 担任主持人， 和 蚂蚁集团高级技术专家杨浩、明略科技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 上海站 即将召开之际，共同探讨大模型如何驱动业务提效。 部分精彩观点如下： 在 5 月 23-24 日将于上海举办的 AICon 全球人工智能开发与应用大会 上，我们特别设置了 【大模型 助力业务提效实践】 专题。该专题将围绕模型选型与优化、应用场景落地及效果评估等关键环节，分 享行业领先企业的实战经验。 查看大会日程解锁更多精彩内容： https://aicon.infoq.cn/2025/shanghai/schedule 以下内容基于直播速记整理，经 InfoQ 删减。 场景探索 郑岩：在探索大模型应用场景时，企业常会遇到"看起来很美但落地难"的需求，各位在实际项目中是 ...

Group 1 - The overall performance of the computer industry is under pressure in 2024, but a significant recovery is expected in Q1 2025, with revenue growth of 15.1% to 281.87 billion yuan and a net profit increase of 790.5% to 2.33 billion yuan [1][2] - In 2024, the computer sector's total revenue reached 1,249.94 billion yuan, a year-on-year increase of 5.0%, while the net profit decreased by 41.1% to 18.2 billion yuan due to macroeconomic impacts and increased competition [1] - The dynamic price-to-earnings ratio for the computer sector reached 81.5x in Q1 2025, indicating a rise in valuation levels [2] Group 2 - The proportion of public funds allocated to the computer sector increased to 3.1% in Q1 2025, which is below the historical average of 4%-5% [2] - Major public fund holdings in the computer sector include companies such as Kingsoft Office, Hikvision, and iFlytek [2] - The ongoing US-China tariff disputes are prompting Chinese companies to reduce reliance on exports to the US and explore cross-border payment opportunities [3] Group 3 - The introduction of advanced Agent applications is expanding the boundaries of technology use, with companies like ByteDance and Alibaba leading in the development of new models and tools [4] - The integration of various tools and platforms is enhancing the capabilities of Agent applications, which are expected to support task-oriented applications [4]

Group 1 - The core viewpoint of the articles revolves around the rapid development and commercialization of AI agents, particularly following the success of Manus, which has sparked significant interest and investment in this sector [2][3][4]. - Major tech companies are intensifying their efforts in the AI agent space, with ByteDance reportedly forming at least five teams to develop various AI agent products, and Baidu launching the "Xinxiang" app, which aims to compete with Manus [4][5]. - The investment landscape is also shifting, as evidenced by the $75 million funding round for Manus's parent company, Butterfly Effect, which has raised its valuation to nearly $500 million [2]. Group 2 - The emergence of AI agents is seen as a solution to the unmet business needs and technological gaps left by previous enterprise digital transformation efforts [3]. - Companies are adopting the MCP (Multi-Cloud Platform) mechanism to enhance the ecosystem of AI agents, with major players like Alibaba, Tencent, and Baidu integrating MCP protocols into their AI products [6]. - There is a growing concern regarding the safety and risk management of AI agents, as many companies lack a comprehensive understanding of the associated risks, with a significant portion of clients unaware of what AI agents entail [7][8]. Group 3 - The concept of AI agents is evolving, with new terminologies such as Agentic AI and Agentic Workflow gaining traction, indicating a shift towards more specialized and collaborative AI systems [10][11]. - The industry is focused on making AI agents adaptable to complex application scenarios, requiring advancements in perception, understanding, planning, and execution [11][12]. - There is a call for a more cautious approach to the deployment of AI agents, emphasizing the need for improved governance and risk assessment capabilities before widespread implementation [12].

Core Insights - The article discusses the emergence of AI Agents, highlighting Manus as the first truly universal AI Agent, which recently raised $75 million in funding, increasing its valuation to $500 million, a fivefold increase since its launch in March [1] - Major tech companies are competing in the AI Agent space, with ByteDance, Baidu, Alibaba Cloud, and OpenAI all developing their own Agent products, indicating a significant shift in the AI landscape [1][4] Group 1: AI Agent Overview - AI Agents can autonomously execute complex tasks by leveraging large models' perception and reasoning capabilities, unlike traditional chatbots that only provide conversational responses [3][4] - The global AI Agent market is projected to grow from $5.1 billion in 2024 to $47.1 billion by 2030, with a compound annual growth rate (CAGR) of 44.8% [4] Group 2: Company Strategies - OpenAI is focusing on enhancing Agent capabilities with new models and aims for AI Agent sales to surpass ChatGPT by 2029, potentially increasing total revenue to $125 billion by 2030 [4][11] - Baidu is developing the "Xinxiang" app as a general-purpose intelligent assistant, while also launching new models to support AI applications [6][8] - Alibaba Cloud's Qwen3 model has achieved a high score in evaluating Agent capabilities, which will facilitate the upcoming explosion of AI applications [6] Group 3: Market Segmentation - Companies are adopting different strategies: some are targeting consumer markets (To C) with personal assistant applications, while others focus on embedding Agents into business workflows (To B) [7][11] - Microsoft integrates its Copilot into Office and Teams, aiming to make Agents a central part of enterprise workflows, with over 100,000 companies already using it [11][12] Group 4: Ecosystem and Standards - The introduction of the Model Context Protocol (MCP) is crucial for AI Agents, providing a unified standard for tool interaction, which enhances efficiency in task execution [14][16] - Over 4,000 MCP servers are now operational globally, indicating rapid adoption and support from major AI companies [16][17] - The proliferation of MCP is expected to facilitate the development of a robust Agent ecosystem, akin to the app economy in the mobile internet era [18]

近期变化，Agent产品层： 1）五一期间Manus创始人Peak指出Manus的 ，主因加入主动查看图像的功能后，Manus开始自动检查其生成的数据可视化，AI的网络效应或初现。 Manus在4月底拿到了硅谷风投Benchmark领投的7500万美元融资。 2）Genspark更新了更好的个性化能力。 【华泰计算机】Agent和MCP是AI主线中的主线 2）Genspark更新了更好的个性化能力。 而从Meta电话会中已知，Meta AI的10亿月活，核心也是基于社交打造个性化。 个性化是护城河，越早建立越好。 模型层： 阿里Qwen 3强调Agent能力和MCP生态的支持，预期后续国产模型都会积极拥抱MCP。 再次重申MCP商业化三阶段： 1）工具厂商率先实现收入，按照【API用量计费】。 4月30日，【 】官方微信号宣布，TextInMCP Server 已覆盖文字识别、文档解析、信息抽取等核心产品能力。 2）Agent客户端商业化同样较快。 【华泰计算机】Agent和MCP是AI主线中的主线 近期变化，Agent产品层： 1）五一期间Manus创始人Peak指出Manus的 ，主因加入主动查看图像的功 ...

Core Insights - The emergence of the Model Context Protocol (MCP) is seen as a significant advancement in AI development, allowing for easier integration of external data sources and tools, thereby enhancing the efficiency of AI applications and agents [3][5][9] - Major tech companies, including Alibaba, Baidu, Tencent, and ByteDance, are actively adopting and promoting MCP, indicating a competitive landscape for AI agent development [9][10][11] Group 1: MCP Overview - MCP is likened to a "universal socket" for AI, enabling seamless connections between large models and external tools, which significantly reduces development costs and time [3][5][8] - The protocol was initially introduced by Anthropic in November 2022 but gained traction with the launch of the Manus AI agent in February 2023, showcasing the potential of MCP [7][13] - The adoption of MCP is expected to transform AI agents from simple information retrieval systems to more complex applications capable of executing tasks [8][12] Group 2: Industry Adoption - As of April 2025, various tech giants have integrated MCP into their services, with Baidu being the first to offer an enterprise-level MCP service [3][9] - Alibaba Cloud has launched a comprehensive MCP service that integrates over 200 leading models and nearly 100 mainstream MCP services, facilitating easier development of AI agents [10][12] - The introduction of payment MCP services by Alipay further enhances the capabilities of AI agents, allowing for streamlined transaction processes within applications [11][12] Group 3: Future Developments - The MCP ecosystem is still evolving, with ongoing improvements and adaptations expected as the technology matures [13][15] - The competition between MCP and other protocols, such as Google's Agent2Agent Protocol (A2A), highlights the dynamic nature of AI integration standards [14][15] - Industry experts believe that while MCP may face challenges, its foundational role in AI development will continue to be significant as it evolves [15][16]