以下文章来源于硬AI ，作者专注科技产研的 系统级记忆能力： 主流OEM厂商能够访问和存储用户全面的行为和偏好数据，这是训练和优化AI代理的关键。第三方AI在这方面处于天然劣势。 高盛12月4日发布的最新研报中分析认为，"豆包AI手机助手"标志着AI竞争进入到操作系统层面，然而，第三方AI代理在获取系统级权限、用户数据和跨应用连 接方面面临重重挑战，反而更加验证了小米等现有手机巨头长期竞争力的稳固性。 豆包AI来袭，但系统级集成面临三大挑战 根据报告，字节跳动于12月1日发布的"豆包AI手机助手"是一个系统级的图形用户界面（GUI）代理，它将豆包大模型深度集成到移动操作系统中，旨在实现屏 幕内容的视觉解读和跨应用的复杂任务操作，例如演示中的跨平台比价和自动领取优惠券。 报告显示，豆包已成为多家智能手机原始设备制造商（例如OPPO、vivo、荣耀）部署的热门大语言模型，5月份硬件AI助手在豆包上日均生成的token数达到 1.3万亿（占豆包总token消耗的8%）。 尽管技术上展示了潜力，但高盛指出，这类第三方AI代理要实现大规模应用，面临三大核心挑战： 系统级操作权限： AI代理需要获得读取屏幕内容、模拟用户 ...

硬·AI 高盛认为，豆包AI手机虽然"火爆"，但其在系统权限、数据获取和应用联通性上面临重重挑战，反而凸显了小米等手机巨 头的结构性优势。小米凭借其"人车家"生态、近10亿的AIoT设备连接以及渗透率高达71%的AI助手"超级小爱同学"，已构 建起强大护城河。小爱同学已经能发布微博笔记、筛选淘宝商品等。 作者 | 龙 玥 编辑 | 硬 AI 近期，字节跳动高调发布"豆包AI手机助手"预览版，意图通过大语言模型（LLM）重塑手机交互体验，引 发市场高度关注。 高盛12月4日发布的最新研报中分析认为，"豆包AI手机助手"标志着AI竞争进入到操作系统层面，然而，第 三方AI代理在获取系统级权限、用户数据和跨应用连接方面面临重重挑战，反而更加验证了小米等现有手 机巨头长期竞争力的稳固性。 01 豆包AI来袭 但系统级集成面临三大挑战 根据报告，字节跳动于12月1日发布的"豆包AI手机助手"是一个系统级的图形用户界面（GUI）代理，它将 豆包大模型深度集成到移动操作系统中，旨在实现屏幕内容的视觉解读和跨应用的复杂任务操作，例如演 示中的跨平台比价和自动领取优惠券。 报告显示，豆包已成为多家智能手机原始设备制造商（例如O ...

Core Insights - The article highlights the competition for the "super entry point" in the AI era, emphasizing that the ability to control data and traffic is shifting from traditional apps to system-level AI agents [2][17] - The recent developments surrounding the Doubao AI assistant reveal the challenges faced by AI companies in navigating existing app ecosystems and data security regulations [5][19] Group 1: Doubao AI Assistant Launch and Challenges - Doubao AI assistant was launched on December 1, enabling cross-application operations, which generated significant market excitement [7][4] - Shortly after its launch, Doubao announced a temporary suspension of its ability to operate financial apps due to security concerns, highlighting the need for clear AI operation guidelines [3][4] - Major apps like WeChat, Taobao, and banks implemented measures to prevent Doubao AI from functioning properly, indicating a defensive response to the new technology [8][10] Group 2: Industry Dynamics and Competitive Landscape - The conflict between AI companies, hardware manufacturers, and app developers illustrates a complex interplay of interests and poses challenges to existing data security frameworks [5][12] - Goldman Sachs identified three core obstacles for third-party AI agents: system-level operation permissions, memory capabilities, and cross-application interface connections [12][14] - The dominance of major smartphone manufacturers in the Chinese market, which hold over 90% market share, makes it difficult for new players to disrupt the ecosystem [14][15] Group 3: Future of AI and App Ecosystem - The article suggests that as voice interaction becomes the primary entry point, traditional app operations may collapse, making it crucial to establish a default system entry point [17][20] - The ongoing battle for AI agents signifies a broader struggle involving major tech companies and highlights the potential for significant shifts in the internet landscape over the next decade [19][20]

Core Insights - The "Doubao Phone" has decided to temporarily withdraw its capability to operate financial apps, citing the need for "prudent adjustments" to align technology development with industry ecology [1][2] - The launch of the Doubao AI Assistant, which allows cross-application operations, has faced significant backlash from major apps like WeChat and Taobao, indicating a conflict between AI developers and existing app ecosystems [3][4] Group 1: Doubao Phone and AI Assistant - Doubao's announcement on December 5 highlights a shift in strategy to ensure safe AI operations, particularly in financial applications [1][2] - The Doubao AI Assistant, launched on December 1, was initially well-received, leading to a surge in demand and price inflation for the associated smartphone [2][3] - Users reported issues with the AI Assistant causing abnormal logouts from WeChat and other apps, raising concerns about security and operational integrity [3][4] Group 2: Industry Response and Implications - Major apps like WeChat and Taobao have implemented measures to block the Doubao AI Assistant, reflecting a defensive stance against third-party AI operations [4][5] - The situation underscores the challenges faced by third-party AI agents in gaining system-level permissions and accessing user data, which are tightly controlled by smartphone manufacturers [5][6] - High-profile firms like Xiaomi are highlighted as having a competitive advantage due to their integrated AI systems and established ecosystems, which are difficult for new entrants to disrupt [6][7] Group 3: Future of AI and App Ecosystems - The conflict initiated by Doubao reveals a broader struggle for control over AI as a new entry point in the digital landscape, potentially reshaping user interactions with applications [7][8] - The need for third-party app authorization poses significant barriers for AI agents, as existing apps are likely to maintain strict security measures to protect user data [9][10] - The ongoing evolution of AI agents is seen as a pivotal moment in the tech industry, with major players vying for dominance in a rapidly changing environment [10]

摘要：高盛认为，豆包AI手机虽然"火爆"，但其在系统权限、数据获取和应用联通性上面临重重挑战， 反而凸显了小米等手机巨头的结构性优势。小米凭借其"人车家"生态、近10亿的AIoT设备连接以及渗 透率高达71%的AI助手"超级小爱同学"，已构建起强大护城河。小爱同学已经能发布微博笔记、筛选淘 宝商品等。 近期，字节跳动高调发布"豆包AI手机助手"预览版，意图通过大语言模型（LLM）重塑手机交互体 验，引发市场高度关注。 高盛12月4日发布的最新研报中分析认为，"豆包AI手机助手"标志着AI竞争进入到操作系统层面，然 而，第三方AI代理在获取系统级权限、用户数据和跨应用连接方面面临重重挑战，反而更加验证了小 米等现有手机巨头长期竞争力的稳固性。 豆包AI来袭，但系统级集成面临三大挑战 根据报告，字节跳动于12月1日发布的"豆包AI手机助手"是一个系统级的图形用户界面（GUI）代理， 它将豆包大模型深度集成到移动操作系统中，旨在实现屏幕内容的视觉解读和跨应用的复杂任务操 作，例如演示中的跨平台比价和自动领取优惠券。 这与中国新能源汽车（NEV）市场形成了鲜明对比。NEV市场更为分散，为具备软件差异化（如华为 HIM ...