机器之心编辑部 智能体（AI Agent）是 AI 演进的关键方向，正逐渐成为人机协同与机器自主决策的核心载体，并有望演进为未来智能社会的基本单元。而基础设施作为智能体落 地的重要支撑，却仍然依赖大量「 胶水代码」来机械拼合各个环节，并时常陷入这样的困境 —— 宝贵的智算资源在静默中闲置，突发的故障让天价训练任务一夜 中断，庞大的运维团队在无尽的告警和排障中疲于奔命 —— 传统的工具链和人力运维模式，已无法应对智能体生产的动态性与复杂性。 我们需要对当下的智能体基础设施进行一次范式上的革新，让基础设施系统本身拥有自主决策、协同和进化的能力 。凭借智能体的主动性和智能性，做出比人类 手动操作更优、更快、更精准的决策，完成原本依赖顶尖专家高级脑力劳动才能完成的复杂工作，实现超越人类经验的卓越运营，以支持更高效、稳定与普惠的 智能体创新。 今天，无问芯穹正式推出基础设施智能体蜂群，它是基于无问芯穹长期的 AI-Native 基础设施建设成果及经验沉淀，深度融合多智能体协同架构与行业场景需求， 所提供的新一代基础设施智能化解决方案 。通过封装 SOTA 模型筛选、基础设施平台管家、资源运营、答疑排障、智算集群运维等多 ...

Core Viewpoint - The recent development of a "multi-scale big data fusion-based mineral exploration prediction system" has been awarded a national invention patent, marking a significant advancement in intelligent mineral exploration prediction technology in China [1][2]. Group 1: System Development - The big data intelligent mineral exploration prediction team has created an innovative system that integrates geological big data management, data mining, multi-factor information extraction, multi-scale grid computing, intelligent prediction evaluation, and decision support [1]. - This system effectively combines geological exploration data, knowledge bases, algorithm libraries, model libraries, and software tools into a collaborative platform, enhancing the efficiency and accuracy of mineral exploration predictions [1]. Group 2: Future Directions - The construction of the big data intelligent mineral exploration prediction system is part of a demonstration initiative by the China Geological Survey to deeply integrate AI technology into geological survey applications [2]. - The team plans to further innovate by integrating cutting-edge technologies such as AI, large models, and intelligent agents into mineral exploration prediction, aiming to establish a new working model of "AI+" in geological exploration [2].

Core Viewpoint - The rapid development of artificial intelligence (AI) is transforming the economy and society, with 2024 marking the year of AI application and 2025 the year of intelligent agents. This shift will lead to a new economic paradigm where humans may become a minority in terms of intelligence compared to machines [1][4]. Group 1: AI and Economic Impact - AI is expected to contribute approximately 0.6% to 1% annual growth to GDP over the next decade, with some experts predicting even higher impacts due to unmeasured efficiencies [4][5]. - The productivity growth rate in the U.S. could rise to an average of 3% annually, driven by reduced reliance on human labor and advancements in AI technologies [5][6]. - The introduction of reasoning models like Deepseek and ChatGPT-5 is expected to enhance AI's capabilities, allowing for autonomous task completion without human intervention [5][6]. Group 2: Global AI Competition Factors - Key factors determining success in the global AI competition include talent, energy, technology, and legislation [8][9]. - China leads in talent availability, with a significant number of AI developers and scientists, many of whom are of Chinese descent [8]. - China also has a competitive edge in energy resources, which are crucial for AI development, while the U.S. faces challenges in this area [8][9]. Group 3: Future Economic Landscape - By 2050, the number of intelligent robots is projected to reach 4 billion, fundamentally altering the economic landscape and human roles within it [4]. - The competition among machines is expected to suppress inflation, leading to increased investment in non-scalable assets like luxury goods and collectibles [6]. - The current economic environment shows signs of a potential bubble, particularly in tech valuations, reminiscent of the internet bubble era [6]. Group 4: China's Economic Outlook - China's economic prospects are viewed positively, with expectations of a recovery from the real estate cycle and a shift towards urbanization driving growth [12][15]. - The valuation of Chinese tech companies is often lower than their Western counterparts, presenting attractive investment opportunities [15][16]. - The government is focused on stabilizing the economy, which is expected to create a favorable market environment for growth [15][16].

Core Insights - The core issue for enterprises regarding AI is its practical application and ability to solve real problems, rather than just being a concept or demonstration [1][3] - The rise of "intelligent agents" in 2025 is driven by business needs for tools that can effectively address challenges [3][4] Group 1: Intelligent Agents Development - The Intelligent Agent Development Platform 3.0 enhances features such as RAG (Retrieval-Augmented Generation), multi-agent collaboration, long-term memory, and plugin ecosystems, allowing agents to work continuously across tasks and scenarios [3][4] - Intelligent agents can now remember user interactions and follow up on needs, moving beyond one-time responses [3][4] Group 2: Real-World Applications - Intelligent agents are being applied in various sectors, such as hotel services where they can interpret guest requests and manage room conditions, and in industrial quality inspection where they identify product defects in real-time [4][5] - The media industry benefits from intelligent agents that streamline the entire content production process, enhancing efficiency [4] Group 3: Open Source and Industry Trends - Tencent's Youtu-Agent framework has been open-sourced, gaining over 2700 stars on GitHub, positioning it as a leading project in the intelligent agent domain [6] - The company emphasizes the importance of combining technology with real industry scenarios, as frameworks without genuine data and context may lack substance [6] Group 4: Future Directions - The focus is on addressing current pain points in the industry rather than rushing to define "super intelligent agents," with an emphasis on improving reliability and reducing erroneous outputs [7] - The industry is still in an exploratory phase, and overcoming obstacles is essential for the widespread adoption of intelligent agents as effective productivity tools [7]

闻乐 发自 凹非寺 量子位 | 公众号 QbitAI 不谈智能体，客户都不见你。 腾讯云副总裁、腾讯云智能负责人、腾讯优图实验室负责人吴运声接受采访时这么说道。 意思就是，现在搞云服务，必须得有点智能体干货。 怎么说？在2025腾讯全球数字生态大会上，腾讯云公布了一份自己的答案： 智能体开发平台3.0（ADP3.0） 面向全球上线，腾讯优图实验室的关键智能体技术也将持续开源。 据说，这次新版本打磨了3个月，完成近600个功能上线，从RAG能力到Workflow，从Multi-Agent协同到应用评测，再到插件生态，看样子 是把所有模块都更新了一遍。 目前插件已经有140多个，全面支持MCP， 真・妈妈再也不用担心我接不上插件了 。 更值得注意的是，腾讯云透露，未来几个月都有关键智能体技术开源。 智能体开发平台3.0 先说清楚这次升级到底升了啥，一个一个看。 首先是 RAG升级 。以前你可以叫它AI资料员，但现在它不仅能管知识库，还能对比文档冲突、自定义切块、接入主流数据库。 官方说法叫"从传统RAG升级至Agentic RAG"，简单说就是，你问个问题，它现在不光能检索，还会自己动脑子组答案。 再看Multi ...

大家早上好！这里是今天的早报，每天早上，我都会在这里跟你聊聊昨夜今晨发生了哪些大事儿。 昨夜今晨 【上海通报多校午餐发臭事件】 上海市教育委员会表示，近期有市民反映上海绿捷实业发展有限公司9月15日供应本市部分学校午餐中的虾仁 炒蛋存在问题。该委会同市场监督管理部门第一时间赴现场调查，相关菜品留样及原料已送上海市质量监督检 验技术研究院有限公司和上海中维检测技术有限公司进行检验。根据相关规程，一周内将出具检验结果。 该委将广泛听取学生和家长意见，严格遵守校园食品安全相关规定，进一步落实学校负责人陪餐制，发挥膳食 监督家长委员会作用，增加菜品选择等，不断提高供餐质量。 同时，会同相关部门，加大对学校供餐企业的监管力度。 【香港警方：黄金盗窃案已有13人被捕】 据财联社9月19日电，香港警方18日就红磡黄金盗窃案召开记者会。西九龙总区重案组第三队主管总督察许仲 恒在记者会上表示，警方于17日至18日在全港多区以盗窃罪及协助罪犯罪共计拘捕13人，包括12男1女，年龄 在29岁至60岁。当天早些时候，香港警方公布，香港红磡一黄金及珠宝首饰加工工场于17日凌晨发生一起盗窃 案，有10名被通缉男子闯入并夺走金条、金砖及金 ...

Investment Rating - The report maintains a "Buy" rating for Tencent Holdings with a target price of HKD 705.0, indicating a potential upside of 6.6% from the current price of HKD 661.50 [3]. Core Insights - Tencent Cloud is focusing on two strategic directions: "intelligentization" and "globalization," with AI-enabled customer solutions and international business growth as key priorities [1][8]. - The company has accelerated the deployment of AI applications across both consumer (C-end) and business (B-end) sectors, showcasing significant growth in user engagement and application usage [7][10]. - Tencent Cloud's international business has seen high double-digit growth over the past year, with the number of overseas customers doubling, positioning it as a preferred choice for over 90% of internet companies and 95% of leading gaming companies going global [1][20]. Financial Summary - For FY23A, Tencent's revenue is reported at RMB 609,015 million, with an adjusted net profit of RMB 157,688 million. Projections for FY24A and FY25E show revenues of RMB 660,257 million and RMB 736,804 million, respectively, with adjusted net profits of RMB 222,703 million and RMB 252,172.5 million [2][27]. - The adjusted earnings per share (EPS) for FY24A is projected at RMB 23.96, increasing to RMB 27.17 in FY25E, reflecting a positive growth trajectory [2][27]. Strategic Developments - Tencent Cloud has launched a comprehensive AI application matrix, with over a hundred applications upgraded to include AI capabilities, enhancing user experience across platforms like WeChat and Tencent Meeting [8][10]. - The company has introduced the "Agent Runtime" solution to support intelligent agents, significantly improving deployment speed and operational efficiency [13][18]. - Tencent Cloud's mixed 3D model has improved modeling accuracy by three times compared to previous versions, facilitating faster production cycles for 3D assets [22][23].

本报（chinatimes.net.cn）记者卢晓 北京报道 进入AI时代，云计算厂商普遍被打了一剂强心针。不过它们还想获得更多增长。 9月16日，腾讯云宣布正式发布智能体战略全景图，全面开放AI能力、C端和B端优势场景。"向智能化 要产业效率，向全球化要收入规模，已经成为企业增长的两大核心动力，也是影响未来行业格局的关 键。"9月16日，腾讯集团高级执行副总裁、云与智慧产业事业群CEO汤道生在腾讯全球数字生态大会上 这样说。 不过所有云计算厂商显然都不想错过这场由AI和出海共同带来的集体普涨，B端市场的竞争已经越发激 烈。 To B营收双位数增长背后 9月16日，腾讯云发布了多项AI技术产品与最新升级，其中腾讯云智能体开发平台（ADP）面向全球发 布3.0版。据悉，面对产业实战中的智能体需求，ADP在3个月内完成了近600项需求的开发，持续迭代 了LLM+RAG、Workflow、Multi-Agent等多种智能体开发框架。 不过吴运声同时也表示，AI智能体真正落地时面临的是一系列系统性问题，需要逐步解决，"智能体及 智能体开发平台不是半年、一年就能做好的短期项目，而是需要三五年甚至十年持续投入的长期事 业 ...

开场环节，腾讯集团高级执行副总裁汤道生以数字分身身份亮相，完成整段演讲。他指出腾讯AI发展始终锚定两个方向：一是内部多元生态 的实践，二是外部产业的真实需求。 他还提到，大模型落地效果较好的场景包括客服、销售话术、知识问答、情感陪伴、文档总结等，需要将语言与逻辑推理能力与企业数据、流 程结合，同时具备联网与系统调用能力，完成任务。 针对这些现实需求，腾讯还在会上披露了智能体开发平台3.0的技术机制：通过接入大模型，融合文档解析、知识检索、流程编排等组件，帮 助企业将原有的数据、规则与任务流程结构化拆解，并交由Agent完成问答、调用或执行。在处理复杂任务时，平台也支持多Agent协同，并可 嵌入企业原有系统中运行。 一、智能体开发平台3.0版本上线，开放全链路Agent开发能力 在本次峰会上，腾讯云副总裁吴运声对昨日发布的腾讯智能体开发平台3.0进行了详细介绍。 据其介绍，腾讯云智能体开发平台面向企业用户提供完整的智能体构建能力，底层能力包括权限体系、数据库、合规机制与运行环境，中间集 成RAG引擎、工作流引擎与Agent引擎，同时配套模型广场、插件工厂、构建模板等模块，支持从开发、调试、评测、发布到运营的全 ...

Group 1 - The core viewpoint of the article emphasizes the transformative impact of artificial intelligence (AI) on global industrialization and the need for new industrialization concepts to integrate emerging technologies [1][2] - The 2025 BRICS New Industrial Revolution Partnership Forum highlighted discussions on how AI can empower new industrialization and how BRICS countries can seize opportunities [1] - The President of the New Development Bank, Dilma Rousseff, stated that AI and other emerging technologies are fundamentally changing production, trade, innovation, and development methods, necessitating an upgrade in industrialization concepts [1][2] Group 2 - China is accelerating the promotion of new industrialization, with AI as a key driver, and has established a complete AI industry system, focusing on innovation and the "AI + manufacturing" initiative [1][2] - Research indicates that the proportion of Chinese industrial enterprises applying large models and intelligent agents is expected to rise from 9.6% in 2024 to 47.5% in 2025 [1] - China has maintained its position as the world's largest industrial robot market for 12 consecutive years, with over 7,000 advanced and more than 230 excellent smart factories established [1] Group 3 - Cross-industry integration driven by AI is becoming a trend, with advancements in embodied intelligence and world models enabling real-time, precise mapping of the real world [2] - The Chairman of China Mobile, Yang Jie, noted that AI is facilitating the deep integration of the digital and real worlds, leading to new business models such as digital twin factories and industrial metaverse laboratories [2] - Challenges remain, particularly for global South countries in key technologies and digital infrastructure, necessitating inclusive AI technology that bridges the digital divide [2] Group 4 - The Minister of Industry and Information Technology, Li Lecheng, emphasized the importance of keeping pace with technological changes and promoting interconnected digital infrastructure [3] - China established the China-BRICS AI Development and Cooperation Center last year, with plans to create an innovation-sharing platform to share mature large model tools and reduce technical barriers [3]