正准备合并 SpaceX 和 xAI，创造一个 1.25 万亿美元估值的宇宙 AI 巨无霸的马斯克，想不到自己的巨饼，不是所有人都吃得下。 当地时间2026 年 2 月 10 日，xAI 联合创始人 Tony Wu，宣布从马斯克的 AI 公司离职。 这已经是继去年 8 月 Igor Babuschkin 离职后，xAI 失去的第二位联合创始人。Wu 负责的恰恰是 AI 推理能力——这个被业内认为是下一代 AI 系统核心 竞争力的关键技术方向。 一家成立才两年多的 AI 公司，连续失去两位联合创始人，这在硅谷并不常见。更关键的是，这发生在 AI 竞争最激烈、人才最稀缺的当下。 创始人纷纷出走，马斯克的AI 野心，还能继续吗？ 01 推理专家走人 Tony Wu 在 xAI 的角色，远比表面看起来重要。 作为负责推理能力的技术负责人，Wu 直接向马斯克汇报。在当前的 AI 发展阶段，推理能力被认为是 GPT-4、Claude 这些大模型与真正「通用人工智 能」之间的关键桥梁。 Tony Wu 在 X 上发布了离职消息｜图片来源：X 从技术角度看，AI 推理能力的突破需要长期的积累和持续的迭代。一个推理专家离开，带 ...

2026年2月第一周，硅谷经历了一场静悄悄的地震。 一周之内，近1万亿美元从科技股市值中蒸发。Bloomberg报道称，这不是又一次泡沫破裂的恐慌，而是一次更深层的认知重构——市场终于开始相信， AI代理（agents）不仅能辅助人类工作，更能直接替代传统软件本身。 触发这场雪崩的导火索，是Anthropic在2月初发布的一个看似普通的产品更新：为其Claude Co-work AI代理推出法律文档审查插件。这个插件能自动追踪 合规要求、审查法律文档、执行多项文书工作——这些原本是律所和企业法务部门每年花费数十亿美元购买专业软件才能完成的任务。 而这一切，只是AI代理革命的开场。 原文链接：https://www.youtube.com/watch?v=WVvD5dv6sMY 超级碗战场：一场关于AI未来的公开撕裂 就在软件股哀鸿遍野的同时，AI行业内部爆发了一场罕见的公开对抗——战场选在了美国最昂贵的广告时段：超级碗。 市场的反应是即时且残酷的： Wolters Kluwer（法律信息巨头）股价单日暴跌13% RELX（法律出版集团）下跌15.8% LegalZoom（在线法律服务）重挫18% Thomso ...

So far in 2026, few sectors outside of the software industry have attracted such a lightning rod of attention. Ever since Anthropic ( ANTHRO ) released an upgraded version of Claude that promised more advanced capabilities, investorsWith combined experience of covering technology companies on Wall Street and working in Silicon Valley, and serving as an outside adviser to several seed-round startups, Gary Alexander has exposure to many of the themes shaping the industry today. He has been a regular contribut ...

以下文章来源于不懂经 ，作者不懂经也叔的Rust 不懂经 . 《主权个人》：未来会活得很爽的有三种人，一是技术精英，二是各行业头部，三是有资本及良好判断 力的人。本号专注后面两种。 本文来自微信公众号： 不懂经 ，作者：不懂经也叔的Rust，题图来自：视觉中国 一、"如果我一天就能写出一本 书，而你需要六个月，谁会赢得 这场比赛？" 科勒尔·哈特，美国人，言情小说作家。她在最近接受《纽约时报》的采访中透露，2024年，她在亚 马逊上架了超过200本小说，年收入六位数美元。她甚至只需要45分钟就能创作或生成一部完整的小 说。 她用Claude起草情节框架，用Grok润色对话，用NovelAI生成整章内容，包括那些情色的部分。是 的，连床戏都是AI写的。 她还开了一个培训项目，专门教人如何用AI批量生产小说。学员多达1600人。更有意思的是，其中 很多都是公开反对AI写作的人。 如果说科勒尔·哈特的故事只是关于商业效率，那《纽约客》不久前发表的一篇文章，则触及了更深 的层面，也更加耐人寻味。 一位计算机科学家图欣·查克拉巴蒂 （Tuhin Chakrabarty） 做了一个实验。他用GPT-4o对30位当代 著名 ...

【导读】Anthropic刚刚扔出一份18页重磅炸弹：《2026年智能体编码趋势报告》。结论直接炸裂：程序员不再写代码了，他们变成了「指挥官」。单个 AI助手已经进化成自主智能体军团，能花好几天打造完整系统，甚至让法务、市场这种完全不懂代码的人，也能自己做应用。软件开发，正在经历图形 界面发明以来最大的一次地震。 2026年的AI圈，要说谁最风光，Anthropic绝对排第一。 就在刚刚，Anthropic甩出了一份18页重磅报告：《2026年智能体编码趋势报告》。 这份报告的核心结论可以浓缩为一句话：任何人，都成为了开发者。 编程/软件开发的游戏规则彻底变了！ 注意，不是「每个程序员变得更强了」，而是「非技术人员也能开发了」。 这意味着软件开发这个行当，正在经历自DOS、图形界面发明以来，最大的一次范式转移。 以下是报告全部8大趋势的深度解读，干货满满，一个不漏。 写在前面 这份18页的报告，信息密度极高！ Anthropic没有画饼，没有讲故事，用的是Anthropic自己的内部研究数据、真实客户案例和对行业趋势的冷静判断。 核心信号非常清晰：程序员不会消失，但「只会写代码」的程序员会消失。 未来的软件工 ...

Core Insights - OpenClaw's success is not merely due to being a powerful AI application, but rather it follows a clear evolutionary path in AI applications [1] - The key takeaway is that AI applications are being systematically allowed to operate independently, moving towards more radical forms [2] AI Application Evolution - The development of AI applications is characterized by a gradual outsourcing of control [3] - **Stage 1: Base Model Stage** - This is when ChatGPT, Claude, and Gemini first emerged, where AI was primarily reactive and passive [4][5] - **Stage 2: Copilot Stage** - AI began to participate in processes, providing suggestions and modifications while still under human control [6][7] - **Stage 3: Automation Agent Stage** - AI started to take over complete tasks, with humans acting more as overseers [8][9] - **Stage 4: Manus/OpenClaw Stage** - A significant shift occurs where AI is no longer just a tool but an entity that can be entrusted with tasks, marking its transition to a behavioral subject [10][11] Implications of AI's Evolution - Once society accepts the concept of "letting go," it will not revert to a model where constant oversight is required [12][13] - OpenClaw and Manus are similar, but their deployment environments differ, leading to significant implications for their future development [14][15] Future of AI in Business - The next phase for AI is to directly engage in business processes rather than merely replacing jobs [18] - Successful business execution requires three components: intelligent decision-making, real-world interfaces, and mechanisms for responsibility and risk [19] - AI is transitioning from being a functional module to becoming an organizational entity, which is the foundation of the "unmanned company" concept [19] Industry Structure - The emerging structure of the AI industry will feature a highly concentrated upper layer and a vast, decentralized lower layer [26] - **Layer 1: Super Central Nodes** - A few entities will control critical capabilities like computing power and model evolution [26] - **Layer 2: Super Intelligent Agent Systems** - These systems will have long-term goals and the ability to manage numerous lower-level agents [27] - **Layer 3: Large-scale Execution Agents** - This layer will consist of numerous short-lived agents focused on specific tasks [28] Conclusion - The rise of OpenClaw signifies a shift from AI as a tool to AI as a central player in industry [30] - The critical question moving forward is how to balance the destructive and constructive speeds within a new industry structure composed of super nodes, super individuals, and numerous execution agents [30]

CGS-NDI 周度报告 智能体开启 AI 从工具到伙伴新时代 —— 数字经济双周报（2026 年第 3 期） 2026 证 2 证 10 证 核心观点 分析师 彭雅哲 证010 -80927607 证pengyazhe_yj@chinastock.com.cn 证证证证证证证证S0130525020001 证证证证证证证证 证zengzhaorui_yj@chinastock.com.cn 风险提示 www.chinastock.com.cn 证证证证证证 证证证证证证证证证证证证证证证证证证证证证证证证证证 ⚫ 本期焦点： AI Agent 开启从工具到伙伴新时代。 2 证 5 证证证证 AI 证证 证证证 OpenClaw 证 GitHub 证证证证证 16 证证证证证证证证证证证证证证 证证证证证 AI Agent 证证证证证证证证证证证证证证证证证证证证证证证证 证证证证证证证证证证证证证证 AI 证证证证证证证证证证证证证 2023 证证 2033 证证证证证证证证 5.5% 证证证证证证证证证证证证证证证证证证证证证 证证证证证证证证证证证 证证证证证证证证证 证证证证证证证 证证证 Agent ...

Vi e w p o i n t | 09 Feb 2026 19:49:16 ET │ 9 pages North America Systems Software Frontier AI Coming for Cyber? Headlines Hitting Last Bastions of Defense CITI'S TAKE In the last 2 weeks, the crescendo of headlines/social media posts from frontier AI companies on their express desires and preparedness to tackle cybersecurity use cases has punctured the "hermetic seal" of cyber ISVs' perceived defensiveness and taller business/technical/valuation moats. Cybersecurity is effectively an adoption tax at worst ...

2026年的美式足球大联盟(NFL)超级碗总决赛，不仅见证了西雅图海鹰队的复仇成功夺冠，也见证了AI巨头们的广告烧钱大战。这波铺天盖地的AI 广告攻势，甚至比赛场上的竞技更为精彩，很多人因此将今年的超级碗称之为"AI碗" 超级碗已经成为美国的一个非正式节日，每年收视观众上亿，今年的收视观众预计超过1.3亿人。按照美国3.4亿人口计算，相当于四成的美国人都 会收看这场美式足球的总决赛。这么夸张的收视率，或许只有中国春晚才能相提并论，因此超级碗又被戏称为"美国春晚" 尽管超级碗的影响力目前大体限于北美市场，但这不影响其成为全球商业价值最高的体育赛事，成为一场极致的流量变现奇迹。超级碗每年单是转 播权收入就超过100亿美元。你没有看错，2023年签下的新转播合同，为期11年，总额超过1100亿美元。 由于版权费太过惊人，几大电视巨头只能共同签约，轮流承办。天价转播费当然要靠天价广告费来回收。今年NBC主办的超级碗，每30秒的广告费 已经达到了800万美元，特定时段甚至超过了千万美元。 潮流的风向标。如果说2022年超级碗属于加密货币,那么2026年的舞台则完全笼罩在AI巨头们的光芒下。 随着科技行业成为美国经济的领 ...

Core Viewpoint - Anthropic's strategy of avoiding an advertising monetization model aims to prevent the "sycophancy" phenomenon in AI, ensuring neutrality and long-term trust as a competitive advantage in the AI market [1][2][3] Group 1: Trust Over Advertising - Anthropic's president, Daniela Amodei, emphasizes that relying on advertising could lead AI models to prioritize pleasing users over providing genuine assistance, which she refers to as "sycophancy" [2][3] - The company aims to establish long-term trust with users rather than focusing on user engagement metrics, positioning Claude as a tool that assists without exploiting user time [3][4] Group 2: Competitive Landscape - The AI industry is witnessing a convergence in model capabilities, making it crucial for companies to secure a default position in users' minds rather than relying solely on superior technology [6][9] - The Super Bowl advertisement serves as a strategic move to occupy this critical time window, as many users have yet to form fixed habits regarding AI tools [8][9] Group 3: Product Strategy and Boundaries - Anthropic's approach involves intentionally limiting features, such as not supporting image generation, to prioritize safety and ethical considerations over rapid expansion [12][14] - The company restricts access to Claude for users under 18, reflecting a commitment to long-term safety and ethical standards rather than short-term growth [12][14] Group 4: Defining AI's Future - Anthropic is actively engaging in discussions about the ethical framework surrounding AI, asserting that the evolution of AI should not be left solely to market forces [18][20] - The company positions itself as a public benefit corporation, prioritizing AI safety over commercial profit, which is embedded in its foundational principles [19][20] Group 5: The Real Battle for User Perception - The advertisement's primary value lies in its statement against reliance on advertising for revenue, highlighting the importance of trust in the AI industry [23][24] - The key question for users is not which AI is the smartest, but which one they can trust the most, indicating a shift in focus from technical superiority to user trust [26]