当地时间2月25日，三星在旧金山举行Galaxy Unpacked 2026发布会，正式推出Galaxy S26系列旗舰手 机。但真正引爆科技圈讨论的，并非手机硬件本身，而是谷歌安卓生态系统总裁萨米尔·萨马特 （Sameer Samat）登台展示的一项新功能——Gemini智能体，能够在后台自动完成订餐、叫车、购物等 多步骤复杂任务。 这一幕，对于关注国内AI动态的观众来说，可谓"似曾相识"。 就在三个月前的2025年12月1日，字节跳动豆包团队发布了豆包手机助手技术预览版，展示了几乎完全 一致的能力：系统级唤醒、跨应用自动操作、后台任务执行。搭载该助手的努比亚M153工程样机首批3 万台一夜售罄，在科技圈引发现象级关注。 如今，全球科技巨头谷歌携手三星推出同样能力的产品，这是否意味着，跟随这一创新，谷歌的"豆包 手机"也来了？ 智能体AI：手机助手帮你办事 "安卓正从传统操作系统演变为一个真正理解并为你服务的智能系统。"萨马特在发布会上如此定义这次 升级的意义。他将这项能力称为Gemini的"下一次进化"，让手机助手从"回答问题"升级为"帮你办事"。 在现场演示环节，萨马特展示了一个贴近日常的场景：一个家庭 ...

《21世纪经济报道》商业秩序工作室报道 没有哪款AI产品，像手机智能体这样，还在测试阶段就引发如此多争议。2025年年底，字节跳动和中兴努比亚发布搭载"豆包手机助手"的手机，这款3499元 的样机一夜售罄，但马上寸步难行——帮用户点外卖、发消息的智能体，被微信、支付宝、淘宝等App，用安全弹窗拦在门外。 "两方谁有问题？其实做法都有点问题。"在多家手机厂商和互联网公司都曾工作的业内人士说，App 不该彻底拒绝互联，但 Agent 也没有理由要求强制开 放。 事实上，字节对端侧AI的试探开始于更早。2024年，字节开始接触中兴等手机厂商，不只是提供基模，而是想让豆包完全承包手机AI助手和所有流量。置 换条件是免除托管费，自己承担 token 消耗成本。 谈判持续数轮，但最终只谈拢了国内市场份额不足 1% 的中兴手机。参与当时谈判的业内人士回忆道，成本减负固然诱人，但没有主流手机肯放弃自己的AI 战略。而且，"手机整个产线都是成本，没有办法拿整个身家赌一个AI的前途。" 直到豆包手机助手引起激烈讨论，态势有了变化。我们近期了解到，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 ...

Core Insights - The year 2025 marked a significant shift in the AI landscape, with DeepSeek gaining prominence and the launch of the "Doubao" AI mobile assistant, which surpassed 159 million monthly active users, indicating a widespread adoption of AI technology in daily life [1][3][12] - The competitive dynamics in the food delivery sector intensified, with major players like Alibaba, Meituan, and JD.com engaging in aggressive subsidy wars, raising regulatory concerns about market practices [21][22][31] - The regulatory environment for AI and digital platforms is evolving, with a focus on compliance and governance, as seen in the tightening of rules around content moderation and advertising practices [2][18][30] AI Landscape - DeepSeek's open-source model significantly reduced training costs and disrupted the Silicon Valley-dominated AI narrative, being hailed as a "national-level technological achievement" [2][12] - The rise of AI products has led to increased risks, including misinformation and privacy concerns, as users engage with AI in various contexts [4][5][10] - The term "Slop" was coined to describe low-quality AI-generated content, highlighting the challenges of maintaining content integrity in the AI era [4] Advertising and Marketing - The concept of GEO emerged, where brands aim to influence AI-generated content to feature their products without direct advertising, reminiscent of early SEO practices [7][8] - Major AI platforms are grappling with the balance between monetization through advertising and maintaining user trust, with some exploring paid advertising models [9][10] Compliance and Regulation - A notable trend in 2025 was the emergence of AI misuse across various industries, prompting a reevaluation of platform governance capabilities [2][3] - The legal landscape for AI copyright disputes is becoming clearer, with significant cases involving major companies like Disney and OpenAI, indicating a shift towards recognizing AI-generated content as a potential copyright issue [12][13][14] Food Delivery Sector - The food delivery market saw unprecedented subsidy wars, with major companies committing billions to capture market share, leading to regulatory scrutiny [21][22] - Regulatory bodies have begun to intervene more actively, emphasizing the need for fair competition and consumer protection in the face of aggressive marketing strategies [22][31] User Engagement and AI Interaction - The relationship between users and AI is evolving, with increasing instances of users relying on AI for emotional support and companionship, raising ethical and privacy concerns [5][6] - The proliferation of AI tools has lowered the barriers for misuse, leading to new forms of fraud and misinformation in various sectors, including education and e-commerce [10][11] Content Governance - The enforcement of stricter content moderation policies has led to the suspension of accounts that engage in misleading practices, indicating a shift in platform governance strategies [30][31] - Platforms are reassessing their business models in light of the need for authentic content, as the erosion of trust could undermine their commercial viability [31]