AI点餐时代来了？ 这一年，AI Agent 的热度一路走高。 从大洋彼岸的 OpenAI、谷歌等 AI 巨头，到国内的各类创业公司，大家都在摸索如何让 AI 真正帮人干活。 相比需要学习复杂操作的传统应用，Agent 的最大魅力就在于无界面交互，只需一句自然语言指令，就能触 发一整套自动化行动。 机器之心报道 编辑：杨文 在这股浪潮中，美团也交出了自己的答卷。9 月 12 日，美团 首个 AI Agent 产品「小美」 正式开启公测。 与外卖 App 不同，小美没有繁杂的界面，只需说一句话，它就能帮你点咖啡、找餐厅，甚至为你规划好一 整周的早餐菜单。 更重要的是，它还能持续学习你的口味偏好，为老人、小孩等特定人群提供更贴心的方案。 我们第一时间拿到了内测码，接下来就奉上一手实测，看看美团首个 AI Agent「小美」的体验究竟如何。 一手体验 点餐这事儿，大家都再熟悉不过了。 打开美团 App，进入外卖页面，搜个关键词，浏览商家列表，再点进具体餐厅看菜单，挑菜、加购、确认 配送地址、支付…… 一顿操作下来，至少要跑完六七个步骤。虽然我们对此已习以为常，但整个流程依然 耗费不少精力。 而到了「小美」这里，这个 ...

"不管用工程化的能力，或者说用垂直领域的专有Know-How，甚至通过服务用户的Case形成比较好的 专有Reward奖励函数，让你的Agent变得越来越强，我觉得在这方面我们看见大多数早期创业者遇到比 较大的问题。"黄明明表示。（文猛） 责任编辑：张恒星 "不完美的创始人才能做出相对完美甚至是伟大的产品，比如一个有激情的人，他可能不够理性，但是 你要做出伟大的产品，太理性的人在这个时代做不出伟大的产品来。"黄明明表示。他还进一步指 出，"好的Agent创始团队或创业者，也要能够让自己做的Agent能随着模型能力提升，不断解锁更多能 力的。" 谈及行业发展，黄明明认为，今天大家往往夸大模型对Agent或者是应用层吞噬的能力，但在判断早期 创业者时却往往忽视了是否真正切入到"用户有刚需痛点"的问题上。 在黄明明看来，互联网应用更多是完成信息与信息、信息与人的连接，搭建的是生产关系。而今天的 AI Agent是生产力工具，衡量它最核心的指标是完成任务的复杂程度和完成度，因为它是需要交付结果 的。"今天很多Agent只能做到30-40分，如果有一个Agent能把模型能力充分榨干，做垂直细分领域'最后 一公里'的苦活 ...

Group 1: Economic Indicators - The US Consumer Price Index (CPI) rose by 2.9% year-on-year in August, matching expectations and slightly up from the previous value of 2.7% [2] - The core CPI also increased by 3.1% year-on-year, with a month-on-month rise of 0.3%, consistent with expectations and previous values [2] - The "super core CPI," which excludes housing and energy prices, slowed to 3.52% year-on-year, indicating a potential easing in inflation pressures [2][3] Group 2: Trade Relations - Mexico plans to raise import tariffs on approximately 1,400 products, including automobiles and textiles, to between 10% and 50% for countries without free trade agreements, including China [4] - In 2024, the bilateral trade volume between China and Mexico is projected to reach $109.43 billion, with Chinese exports to Mexico increasing by 10.77% [4] - The tariff increase may lead to higher prices for Chinese goods in Mexico, potentially raising inflation levels in Mexico [4][5] Group 3: Energy Storage Development - China's National Development and Reform Commission aims to achieve a new energy storage capacity of over 180 million kilowatts by 2027, with direct investment expected to reach approximately 250 billion yuan [6] - The focus will remain on lithium-ion battery storage, with plans to enhance various technology routes and application scenarios [6] - The growth in energy storage is crucial for stabilizing power supply from renewable sources like wind and solar [6][7] Group 4: Technology and Consumer Electronics - Apple is working to launch eSIM-enabled iPhones in China, awaiting final regulatory approval [8] - eSIM technology allows for virtual SIM cards, which can enhance user convenience by enabling online activation and operator switching [8][9] - The introduction of eSIM may lead to changes in the competitive landscape among mobile operators in China [9] Group 5: Corporate Developments - Starbucks is in the final stages of negotiations to sell its China business, with potential buyers including several investment firms [10] - The CEO of Starbucks has indicated a desire to retain a significant minority stake in the business to maintain brand integrity [10][11] - The sale reflects a strategic move to adapt to the evolving coffee market in China, where competition has intensified [11] Group 6: AI Applications in Business - Meituan has launched its first AI Agent product, "Xiao Mei," which facilitates local service transactions through natural language processing [12] - The AI application aims to enhance user experience in ordering and recommendations, although it still requires improvements in interaction quality [12][13] - The integration of AI Agents into consumer apps represents a promising avenue for commercializing AI technology [12][13] Group 7: Fund Management Regulations - The China Securities Regulatory Commission has proposed new regulations that significantly reduce sales fees for public funds, encouraging long-term investment [14] - The proposed changes include a reduction in subscription fees for various fund types, with some fees decreasing by up to 67% [14][15] - The regulations may impact the profitability of fund management companies, pushing them to rely more on performance-based fees [14][15]

Group 1 - Meituan's first AI Agent product, "Xiao Mei" App, has officially launched public testing, utilizing the self-developed model LongCat-Flash-Chat for local life services [1][2] - "Xiao Mei" aims to provide a full-chain local life service experience, allowing users to place orders, receive restaurant recommendations, and navigate reservations through simple natural language interactions [1][2] - The product is designed to cater to a wide range of user needs, including those of elderly and children who may find it difficult to operate smartphones, emphasizing ease of use without complex interfaces [1][2] Group 2 - Future developments for "Xiao Mei" include deeper personalization and proactive services, aiming for comprehensive coverage of users' life scenarios [2] - Unlike traditional AI assistants, "Xiao Mei" is an independent C-end intelligent agent rather than a feature embedded within the Meituan App, focusing on serving users directly [2] - The launch of "Xiao Mei" comes at a critical time as competition in the local consumption sector intensifies among internet giants [2][3] Group 3 - Alibaba's release of the "Gaode Street Ranking" directly targets Meituan's "Dazhong Dianping," intensifying competition in the local life service market [3] - Meituan has responded by officially restarting its quality takeaway service, leveraging B-end self-developed models and real review data to enhance user decision-making [3] - The competitive landscape is shifting, with Meituan's ability to utilize AI and other new tools being a focal point for potential recovery in market positioning [3]

Core Insights - Meituan launched its first AI Agent product, "Xiao Mei" App, for public testing on September 12, positioning it as a "small and beautiful AI life assistant" [2] - The app is powered by Meituan's self-developed model LongCat-Flash-Chat, enabling users to interact through natural language for local life services such as food delivery, restaurant recommendations, and navigation for reservations [2] - "Xiao Mei" App is available on major app stores, and users can participate in the public test by obtaining exclusive invitation codes from the official account [2]

编辑：panda、+0 姚顺雨加入腾讯了？已被辟谣。 昨天，有消息称，OpenAI 著名研究者、清华校友、著名博客《AI 下半场》的作者姚顺雨已经加入了腾讯混元大模型团队，并且还传言说他将在这里组建一支自己 领导的研究团队。 这一消息瞬间点燃了 AI 社区，「年薪 1 亿」的传闻更是让话题迅速升温。 相关的内容也是越传越全面，根据小红书用户 @Top华人科创社区 的贴文，有网友透露，姚顺雨上周已经来腾讯开会了。 机器之心报道 不过，今早鹅厂黑板报已正式辟谣，否认了相关说法。但也有网友追问：辟谣的是「没加入」，还是「没有上亿薪酬」？ 不过谣言也并非空穴来风， 据多个信息源表示，姚顺雨确已从 OpenAI 离职。 如今 AI 人才争夺战确实日趋激烈。海外，Meta 的扎克伯格高调抛出「上亿年薪」挖角各个公司的顶尖研究员；国内，各大厂同样全力加码，力求稳住核心团队。 可以预见，类似的风声还会不断出现。 从清华学霸到 OpenAI 姚顺雨，毕业于清华姚班，普林斯顿大学计算机科学博士，2024 年 8 月加入 OpenAI。 在加入 OpenAI 之前，他就已经在语言智能体领域做出了一系列开创性的工作： 此外，他还参与 ...

Core Viewpoint - Meituan has launched its first AI Agent product, "Xiao Mei" App, which aims to enhance local life services through AI technology [1] Group 1: Product Overview - "Xiao Mei" is positioned as a "small and beautiful AI life assistant," emphasizing simplicity and a better quality of life [1] - The app utilizes Meituan's self-developed model, LongCat-Flash-Chat, to provide features such as native food delivery ordering, restaurant recommendations, and reservation navigation [1] - "Xiao Mei" App is now available on major app stores for public testing [1]

Core Insights - Meituan has officially launched the public beta of its first AI Agent product, "Xiao Mei" App, which aims to serve as a "small and beautiful AI life assistant" [1] - The app utilizes Meituan's self-developed model LongCat-Flash-Chat, enabling seamless local life service experiences such as food ordering, restaurant recommendations, and navigation [1] - "Xiao Mei" is designed to simplify daily tasks for users, including ordering coffee and planning meals, while also catering to the needs of elderly and children [1] Product Features - "Xiao Mei" App is available on major app stores, and users can join the public beta through exclusive invitations from the official account [1] - The app focuses on vertical service scenarios in local life, allowing users to interact using natural language [1] - Future updates will aim to enhance personalization and proactive services, covering a wide range of life scenarios including dining, accommodation, travel, entertainment, shopping, and healthcare [1]

Core Insights - The article highlights the transformative impact of AI Agents in office environments, evolving from mere concepts to integral "digital employees" capable of meeting KPIs and integrating into core business processes [1][11]. Group 1: Evolution of AI in Office Settings - AI in the office has progressed from a "show-off" phase to a practical application phase, marked by the introduction of tools like Microsoft Office Copilot and WPS AI 1.0 [2]. - The initial phase, termed the Copilot assistance phase, involved AI acting as a passive tool for text generation and basic data analysis, requiring user initiation for tasks [2]. - By mid-2024, AI is expected to enter the Agent task phase, where it can understand context and automate multi-step tasks, as demonstrated by AI assistants handling 80% of HR inquiries [2][5]. Group 2: Case Studies and Applications - Recent developments at the WAIC showcase AI Agents deeply embedded in business processes, such as EHGO's LuminaSphere, which deploys specialized AI assistants across departments [3]. - Real-world applications include a significant reduction in processing time for financial operations at Hebei Telecom, where AI cut task duration from 2 hours to 10 minutes [3]. - The integration of AI in various companies, like 百丽时尚, has led to improved operational efficiency and sales performance through innovative AI-driven solutions [4]. Group 3: Driving Forces Behind AI Adoption - The rise of AI in office settings is driven by three main factors: increasing labor costs, the need to address high-frequency, high-error, and repetitive tasks [5]. - Technological advancements, particularly the integration of LLM, RPA, and low-code solutions, have overcome previous limitations in task automation [5]. - The ecosystem of platforms like DingTalk and WeChat has facilitated the development and deployment of AI Agents, allowing business personnel to create their own solutions [5][6]. Group 4: Challenges and Limitations - Despite the success of AI Agents, challenges remain, such as the contradiction between development efficiency and implementation depth, often leading to a lengthy and burdensome process [8]. - Data integration issues arise from the fragmentation of enterprise data across various systems, complicating real-time access and decision-making for AI [8][9]. - Many AI systems still struggle with executing final operations, limiting their ability to take full responsibility for tasks [9][10]. Group 5: Future Directions - The future of AI in the workplace is expected to involve a "golden triangle" of MCP, LLM, and Agent technologies, enhancing task management and execution feedback [10]. - Multi-modal interactions, including text, voice, and video, are anticipated to become mainstream, improving user engagement and collaboration [10]. - The vision for AI in organizations includes a shift from being mere tools to becoming integral team members, potentially leading to new operational models like "human directors with AI execution teams" [10][11].

Core Insights - The first L4 level mass-produced Robotaxi from Hello has been unveiled at the 2025 Inclusion·Bund Conference [1] - The Robotaxi features an end-to-end integrated architecture and collaborates with Alibaba's team on model and computing power [1] - The vehicle's cabin includes an AI Agent that provides passengers with a comprehensive range of services, akin to a ride secretary [1] Company Strategy - Hello aims to launch over 50,000 Robotaxis into the market by 2027 [1] - The CEO of Hello, Yang Lei, emphasizes that the current challenge in the Robotaxi sector is technological, while operational challenges will arise in the long term [1] Cost Optimization - Yang Lei highlighted the potential for significant cost savings through operational efficiencies, particularly in reducing air conditioning usage [1] - The estimated annual electricity cost for a single Robotaxi is around 80,000, which translates to 80 billion for 1 million vehicles [1] - The company is focused on optimizing algorithms to manage air conditioning usage effectively, which could lead to substantial savings [1]