人工智能总带有"高大上"的滤镜，然而一个好技术，最动人的能力，恰恰在于它能编织出丰富的人间烟火。 时下，AI正在改造中国路边摊。从AI调香、AI手串、AI象棋、AI台球到AI理发，一系列接地气的应用正在涌现。这些AI路边摊能否走远尚不可知，但它 们无疑已为街头，增添了一抹别样的烟火气。 AI席卷路边摊 "想知道最适合你的香水是什么味道吗，来，扫个码。" 北京国家会议中心，一个很小的集市柜台前，却热闹排起了长长的队伍。人们都在等待着队伍尽头的小哥，从十几个瓶瓶罐罐中，挤出一滴一滴的香水， 进行调香。 队伍另一侧，人们纷纷拿出手机，扫码进入小程序。只需要输入自己的名字和MBTI人格，就会生成一个专属于你自己的"香水配方"。一旁设置着一个大 大的看板：AI调香。 有一位年轻人就扫了码，得出自己的香水配方。一边排队等制作，一边她觉得好奇，"专属于自己的香水，究竟是什么味儿。" 集市老板就坐在摊位边。她介绍，自己的传统香水生意在江浙沪做得挺大了，但也一直在想如何让自己的香水，能更加"个性化"，"感觉AI刚好能发挥作 用。"老板说。 其实，大众扫码的这个小程序背后，没啥高深的东西，老板在小程序里嵌入了一个文心大模型，人们 ...

Core Insights - The Chinese AI agent industry is expected to experience explosive growth with a compound annual growth rate (CAGR) of 72.7%, reaching a market size of 852 billion yuan by 2028, and potentially exceeding 2.1 trillion yuan by 2030, driven by technological breakthroughs and deepening application scenarios [1][13][15] Industry Development Trends - The evolution of AI agents in China is characterized by a transition from "model monopoly" to "universal Agent," with advancements in foundational models, architectural innovation, and efficiency optimization driving the industry [1][2] - The breakthrough in foundational models is propelled by the rise of large model capabilities and the trend towards open-source, facilitating a shift from monopolistic control to widespread accessibility [1][2] - Multi-modal fusion technology is expanding the boundaries of models, enabling AI agents to evolve from single-text interactions to multi-sensory perceptions [1][2] Architectural Innovations - The Mixture-of-Agents (MoA) architecture has become an industry standard, integrating general models, specialized scene models, toolchain platforms, and data flywheels, achieving a 15% higher accuracy in specific tasks compared to general models [2] - The Mixture-of-Experts (MoE) architecture reduces computing power consumption by 60%, enhancing system performance through distributed expert networks [2] Product Trends - The AI agent product matrix in China is forming a collaborative development of "general-purpose + vertical" products, catering to diverse market demands [4] - General-purpose products focus on broad scene coverage and the ability to execute complex tasks, while vertical products emphasize deep exploration of specific fields [4] Market Segmentation - The B-end market prioritizes customization capabilities, with AI agent platforms supporting low-code/no-code development and private customization [6] - The C-end market emphasizes standardized experiences, with products aimed at enhancing user efficiency and emotional satisfaction [6] Application Trends - AI agents are penetrating multiple industries, with high application maturity and value release in finance, healthcare, and government sectors [7] - In finance, AI agents have significantly improved efficiency in credit approval processes, reducing processing times from 48 hours to 15 minutes and increasing accuracy to 95% [8] Policy and Governance - The governance framework for AI agents in China aims to balance development and safety, establishing a multi-level legal governance system to mitigate potential risks [9][12] - Challenges in the governance system include traditional governance adaptability, responsibility identification, data governance issues, and compliance challenges for enterprises operating internationally [10][12] Market Growth Drivers - The continuous decline in computing costs is a key driver for the AI agent market, with predictions indicating a reduction to one-tenth of 2024 costs by 2028 [13] - Support from policies for intelligent computing infrastructure is further accelerating technology deployment and market penetration [13]