过去两年，智能体（Agent）是AI行业最重要的叙事，现在聚光灯正收束到一个更具体的方向：端侧智能体。 在海外，名为OpenClaw的智能体在硅谷技术圈走红，接管一众开发者的电脑；在国内，字节跳动把豆包嵌入手机，样机价格在二手市场居高 不下。这些智能体运行在手机、电脑和汽车上，能操作本地环境和所有工具，点外卖、打游戏、炒股票，把执行力拉到极致。 手机智能体，体验在退化？ 越来越多智能体从云端落入个人终端。在国内，豆包手机助手是端侧智能体破圈的一个起点，但这条路并不始于此。 智能体还会接管更多个人设备。在发售工程版"豆包手机助手"后，据媒体披露，字节已于去年年底启动正式版手机项目，搭载智能体的新机预 计于今年Q2发布。 我们近期还从多方了解到，包括阿里系在内的多家App与字节跳动达成停火协议，App允许努比亚设备的手动登录，豆包主动限制AI操作场 景，双方回到"井水不犯河水"的状态。 行业正在形成一个共识：未来智能体的壁垒，在于能打通多少个人设备，能互联多少服务。智能体想成为新的能力层，重组我们与设备、与 App的连接方式。 但这种互联互通的技术趋势，也撞上了合规边界。智能体要想操作手机，需要利用高敏感权限进行 ...

Core Insights - The AI industry is shifting focus towards edge-based intelligent agents, with significant developments in both domestic and international markets [1][2][3] - ByteDance's Doubao and other intelligent agents are gaining traction, operating across various personal devices and applications, but face challenges related to privacy and operational capabilities [2][3][5] - The industry consensus is that the future of intelligent agents will depend on their ability to connect multiple personal devices and services, reshaping the ecosystem [2][3] Group 1: Market Developments - OpenClaw has gained popularity in Silicon Valley, while Doubao is embedded in smartphones in China, with high demand reflected in the second-hand market [1] - The user base for intelligent agents among six major Chinese smartphone manufacturers grew by 65 million in one year, reaching a total of 535 million users [3] - OPPO, Vivo, and Honor are leading the charge in the domestic market, with OPPO's Xiaobu Assistant achieving 160 million monthly active users by June 2025 [3] Group 2: Technical Challenges - Current intelligent agents are struggling with task execution, often only able to perform basic functions like opening apps rather than completing complex tasks [7][8] - A recent evaluation of seven intelligent agents revealed a success rate of only 20% for completing tasks, with many tasks failing or being interrupted [5][7] - The technology behind these agents is still in its infancy, with most operating at a basic level and lacking the ability to handle complex, multi-step tasks effectively [8][10] Group 3: Privacy and Security Concerns - Intelligent agents require extensive permissions, often exceeding 100, raising concerns about privacy and potential misuse of sensitive data [17][18] - The use of high-sensitivity permissions, such as accessing location and reading messages, poses significant risks to user privacy [19][20] - The reliance on cloud processing for data handling introduces vulnerabilities, as sensitive information may be exposed during transmission [27][28] Group 4: Industry Dynamics - A ceasefire agreement has been reached between ByteDance and several app developers, allowing for normal login procedures while limiting AI operations [35] - The competitive landscape is marked by tensions between smartphone manufacturers and app developers, with concerns over the implications of intelligent agents on existing business models [30][31] - The future of intelligent agents hinges on establishing a viable revenue-sharing model and addressing the concerns of both hardware manufacturers and app developers [31][32]

智能体是能够感知环境、自主规划、决策并执行任务的AI应用系统，2025年被业内视为"AI智能体元年"。环球时报研究院舆情监测团队对外媒与国 外社交媒体2024年1月-2026年1月关于智能体报道的热度进行了舆情监测，其间外媒与国外社交媒体对智能体的关注度呈现出总体走高的态势。 所谓"端侧智能体"，是指搭载在终端设备（如手机、电脑、IoT设备等）上能够独立完成推理、决策、任务执行的智能体。中国信息通信研究院人工 智能研究所副所长巫彤宁表示，智能体被视为人工智能从感知理解（会说话）向主动服务（会做事）演进的关键形态。 上海交通大学人工智能学院副教授、无问芯穹联合创始人兼首席科学家戴国浩在研讨会上介绍说，AI的终极目标是服务于物理世界中的人。因此智 能必须通过我们身边的终端设备（如手机、汽车、机器人）来实现，这就是"端侧智能"。在端侧如何构建人机交互与处理的"语言"，衍生出不同技 术路线。 据专家介绍，第一种是GUI（图形用户界面）路线，靠视觉识别读取屏幕信息后进行"模拟操作"。戴国浩形容，"这好比另有一人通过观察界面来操 作用户设备，它将物理世界的信息——如视觉、听觉等——转化为AI可理解的表达方式"。GUI的优 ...

文｜邱晓芬 编辑｜苏建勋 1月26日，在腾讯公司年会上，马化腾罕见评价了豆包手机——他直言腾讯反对将用户的手机屏幕传到 云端，因为这"极其不安全、不负责任"。 对此，字节方面暂无回复。 如何避免继续被"围剿"，会是第二代硬件产品的关键课题。 对此评价，豆包手机方面快速做出回应，称将严格遵守用户授权，云端处理遵循"不存储、不训练原 则"。 马化腾的评价，也揭开了豆包手机此前的隐痛。2025年12月，字节旗下豆包手机在发布仅一天后，便经 历了一场严重"围剿"——其AI Agent由于触及了各大超级APP的流量和核心数据，引发了美团、微信、 阿里系在内互联网厂商的联合封锁，AI功能几近瘫痪。 尽管如此，豆包手机的故事尚未落幕。 据《智能涌现》独家获悉，字节已于去2025年底开启豆包手机助手正式版项目，新机预计将于2026年 Q2中晚期发布。 有供应链人士称，字节对新机的预期不低，比第一代的测试版大大提升。在模式方面，豆包二代手机依 旧合作中兴努比亚，由中兴负责硬件，豆包负责AI。 《智能涌现》了解到，豆包手机团队过去正在与多数主流应用厂商谈判，目前已经与部分互联网公司 （打车、外卖、订票等）谈好了部分常用权限。 谈 ...

Core Insights - The emergence of "Physical AI" is accelerating, with significant developments showcased at the 2026 CES and Alibaba Cloud's event in Shenzhen, indicating a shift towards AI integration in physical devices [1][3][4] Group 1: AI Hardware Transformation - AI hardware is evolving from mere marketing gimmicks to genuine intelligent devices capable of perception, understanding, and decision-making, marking a significant step in AI's practical application [3][4] - The Pettichat pet translator, which translates pet sounds into human language with a 94.6% accuracy rate, exemplifies the innovative capabilities of AI hardware [4] - The market for AI hardware is expanding rapidly, with over 1,000 smart hardware products showcased at Alibaba Cloud's event, indicating a growing demand and the entry of numerous brands [6] Group 2: Industry Dynamics - Alibaba Cloud's "Qianwen" model aims to become the Android of the AI hardware sector, facilitating a diverse range of devices to adopt intelligent systems through open-source and adaptable solutions [7][10] - The shift from hardware sales to service-oriented models is becoming prevalent, as companies leverage AI capabilities to enhance user experience and create sustainable profit models [10][11] - The introduction of a licensing fee model for AI services is expected to alleviate cost concerns for hardware manufacturers, encouraging mass production and competitive pricing [10][12] Group 3: Market Outlook - The Chinese AI hardware market is projected to exceed 1.1 trillion yuan by 2025, with a year-on-year growth of 13.4%, indicating a robust growth trajectory for the industry [12]

Core Viewpoint - The article emphasizes the rapid advancement of AI hardware, transitioning from novelty to practical applications, particularly in the realm of "Physical AI" as highlighted at major tech events like CES 2026 and Alibaba's cloud exhibition [1][3][4]. Group 1: AI Hardware Transformation - AI hardware is evolving into intelligent entities capable of perception, understanding, and decision-making, marking a significant shift from mere marketing gimmicks to functional devices [3]. - The demand for AI hardware is increasing, with new applications being rapidly developed, as evidenced by the diverse range of products showcased at the Alibaba cloud exhibition, including over 1,000 smart hardware items from more than 200 brands [8][9]. Group 2: Innovations in AI Hardware - Notable innovations include the Pettichat pet translator, which achieves a 94.6% translation accuracy for pet sounds, demonstrating the potential of AI in everyday applications [6]. - The AI hardware ecosystem is expanding, with various products like smart glasses, AI companions, and training devices becoming mainstream, indicating a robust market for AI-integrated solutions [8]. Group 3: Competitive Landscape - Major players in the AI hardware space, such as Alibaba's Qianwen, are positioning themselves similarly to Android by providing open-source and adaptable solutions that lower barriers for hardware manufacturers [9][14]. - The shift from hardware sales to service-oriented models is becoming prevalent, allowing manufacturers to enhance user experience and establish sustainable profit models [14][15]. Group 4: Future Outlook - The Chinese AI hardware market is projected to exceed 1.1 trillion yuan by 2025, with a year-on-year growth of 13.4%, indicating a strong and ongoing expansion in this sector [16]. - Companies like Alibaba are focused on fostering an ecosystem that encourages innovation and collaboration among hardware manufacturers, aiming to enhance the overall market potential for AI hardware [16].

Group 1 - The article highlights the impressive technological innovations showcased at the CES, particularly emphasizing NVIDIA's CEO Jensen Huang's presentation on the imminent arrival of physical AI [1] - Huang's engaging offline discussions following his main presentation are noted for their personality and interest [2] - The article recommends several insightful readings, including Huang's full speech at CES 2026, which covers three key topics and introduces a "chip monster" [3] Group 2 - The article mentions a bullish consensus among private equity firms regarding a return to 4000 points in the market, focusing on identifying "uncrowded growth" opportunities and capturing "low-hanging fruit" [3] - It references insights from Duan Yongping's 2025 sharing, stressing the importance of investing in good companies and the necessity for individuals to understand and learn about AI [3] - Liu Yuhui's latest commentary on the "G2 competition" era emphasizes the significance of edge intelligence, gold, and commodities as major beta factors [3] - The article notes that key variables include monetary policy, with Ray Dalio's latest review predicting a long-term return of only 4.7% for the US stock market [3]