Core Viewpoint - The article emphasizes the paradox of modern productivity, where individuals are overwhelmed by tools rather than empowered by them, leading to a need for a new approach to productivity that allows for seamless collaboration and creativity [2][4]. Group 1: Introduction of Fellou CE - Fellou CE was created in response to the limitations of the traditional "Do It Yourself" model, which is inadequate in the increasingly complex digital landscape [4]. - The core philosophy of "Act With Fellou" focuses on collaboration and transforming ideas into reality with enhanced efficiency and freedom [4]. Group 2: Continuous Interaction and Task Management - Fellou CE aims to create a "seamless continuum experience" by enabling natural interactions between humans, AI, and browsers, eliminating the need for manual switching between tools [6]. - The platform allows for goal-oriented automation, where users can set high-level objectives, and Fellou CE autonomously breaks down and executes tasks [7]. - A unified memory layer connects all interaction histories and preferences, optimizing information recommendations and user experience [8]. Group 3: Redefining Human-AI Relationships - Fellou CE redefines the relationship between humans and AI, positioning itself as an intelligent assistant that enhances information management while keeping users in control [9]. - The platform operates in an "Agent" mode, providing end-to-end automation without complex setups, allowing users to visualize and intervene in the process [10]. Group 4: Performance and Competitive Edge - In rigorous benchmarks, Fellou CE demonstrated superior performance in complex tasks, achieving a 72% success rate in high-difficulty "write" tasks, outperforming competitors [25][23]. - The platform supports user intervention and editing, ensuring transparency and control over task processes [27]. Group 5: Future Developments and Community Engagement - The company plans to introduce more features in the coming months, including voice activation and enhanced mobile productivity tools [31]. - There is an open invitation for individuals passionate about breaking tool constraints and reshaping digital productivity to join the team [34].

23 岁时， Harry Qi 已经实现了大多数人一生都难以企及的财务成就——年收入约 100 万美元。 公司发展势头如此强劲，以至于Harry Qi 在 YC 时期的执行教练兼 YC 顾问 Ashutosh Desai 也全职加入了团队。 Motion 专门面向中小型企业（ SMBs ），这些企业没有巨额预算来定制编写和训练自己的 AI Agent。 大学毕业后，他的第一份工作是担任 " 量化分析师 " 。这是对冲基金行业对统计模型驱动的量化基金中股票交易分析师的称呼。但就像许多耗尽精 力追逐更多金钱的人一样，他感到内心空虚。 " 到某个阶段，你会想要对这个世界产生更大的影响， " 现年 29 岁的齐向 TechCrunch 坦言。 在 2019 年，他与高中好友奥米德·鲁霍法达，以及大学同窗（同样任职于对冲基金）的余伊森共同开发了一款 AI 日历与任务管理应用，并向 Y Combinator 提交了申请。他们成功入选 2020 年冬季批次后，立即辞去工作投身创业。此后 Motion 还迎来了第四位联合创始人——早期员工钱德· 拉梅什。 随后的六年里，他们稳步发展 Motion 以专业用户为主的客户群体，直 ...

Z Highlights Aaron Levie ， Box 联合创始人兼 CEO ，长期关注企业级技术与 AI 融合； Steven Sinofsky ，前微软 Windows 部门总裁，现为知名科技评论人与投资者， 深度参与多轮平台变革。本次访谈由 a16z 合作伙伴 Erik Torenberg 和 Martin Casado 于 2025 年 8 月 25 日发起，围绕 Agent 的发展现状、技术挑战与未来 影响展开深度讨论。 Agents 的本质之争：从 Linux 符号到自主执行者 Erik Torenberg ： 我想 用一个非常简单但又很有挑衅性的问题来开启这个范围很广的播客：什么是 Agent ？ Aaron Levie ： 这个问题要问谁？ Steven 。 Steven Sinofsky ： 天哪。我其实对 Agent 有一个非常老派的看法，它就是 Linux 里的 "&" ，就是后台进程。比如你在 O3 里输入一个指令，它会告诉你 " 我正在尝试 " ，然后 " 等等，我需要一个密码，我做不到 " 。这就像世界上最糟糕的助手。归根结底，它只是为了在长时间计算时给你找点乐子。所以我 对 ...

Core Insights - Supabase, a startup providing backend services for "ambient programming" companies, is in talks for funding that could value the company at least $5 billion, with an annual revenue reaching $70 million [1][2] - The company has raised nearly $400 million to date, with existing investors including Accel, Felicis, Coatue Management, and Craft Ventures [1] - If the current funding round is successful, it would more than double the company's valuation from $2 billion in April, when Accel led the investment [1] Group 1 - Supabase's annual revenue has increased from $20 million a year ago to $70 million currently [2] - The company is exploring acquisitions to drive growth, capitalizing on the AI programming assistant trend [2] - The development tool Lovable allows developers to design applications and configure Supabase databases without leaving the platform, with its annual revenue reaching $100 million recently [2]

Core Insights - The article highlights the top AI tools of the week, focusing on their unique features and target audiences, showcasing the growing trend of AI integration across various industries [1][2]. Group 1: Ada - Ada is the world's first AI data analyst, designed to automate data processing and report generation without coding skills, significantly saving time on manual data cleaning and SQL writing [4][5]. - Key features include automated workflows, integration of analytical and visual intelligence, and support for multi-source data fusion [4]. - Ada received 797 Upvotes and 141 comments, indicating strong user interest and engagement [6]. Group 2: CapCut AI Suite - CapCut AI Suite is an intelligent video editing tool aimed at content creators, simplifying the video production process through AI [10][11]. - It offers features like automatic editing, smart segment selection, and AI-generated animations, catering to various creative needs [11]. - The tool garnered 704 Upvotes and 64 comments, reflecting its popularity among users [12]. Group 3: xpander.ai - xpander.ai is a full-stack platform for developing and deploying AI agents, providing a comprehensive environment for AI engineers [13][15]. - It supports both no-code visual development and deep code control, enhancing flexibility and rapid iteration [16]. - The platform achieved 691 Upvotes and 50 comments, showcasing its appeal to developers [17]. Group 4: Sidekick - Sidekick is an AI assistant that automates multi-application workflows through natural language chat, making automation accessible to non-technical users [18][22]. - It simplifies the automation process by eliminating complex interfaces and allowing users to describe workflows conversationally [22]. - Sidekick received 592 Upvotes and 102 comments, indicating a positive reception [24]. Group 5: Incerto - Incerto is an AI co-pilot for database developers, enhancing efficiency by converting natural language commands into database tasks [25][27]. - It automates the detection and resolution of potential issues in production environments, significantly reducing manual operations [27][28]. - The tool achieved 508 Upvotes and 273 comments, highlighting its effectiveness in the database management space [29]. Group 6: ClearCRM - ClearCRM is a comprehensive CRM platform for small businesses, integrating sales, marketing, project management, and billing functionalities [30][32]. - It aims to streamline operations and enhance team efficiency by providing a single platform for managing customer relationships [32][33]. - The platform received 481 Upvotes and 76 comments, reflecting its utility for small teams [35]. Group 7: Dodo Payments - Dodo Payments is a global payment and subscription management platform designed for digital product and SaaS companies [36][38]. - It simplifies cross-border payment processes and automates billing management, catering to the needs of digital entrepreneurs [38][39]. - The platform garnered 444 Upvotes and 63 comments, indicating strong interest from its target audience [40]. Group 8: JoggAI - JoggAI is an AI video generation platform centered around realistic AI avatars, aimed at content creators and marketers [41][44]. - It enhances video content creation by providing expressive digital avatars that improve user engagement [44][45]. - The platform achieved 438 Upvotes and 81 comments, showcasing its appeal in the content creation market [46]. Group 9: Beatoven.ai - Beatoven.ai is an AI music composition tool designed for content creators, generating background music based on text prompts [47][49]. - It offers features for multi-track editing and emotional transitions, making it accessible for users without musical expertise [49][50]. - The tool received 434 Upvotes and 58 comments, reflecting its popularity among content creators [51]. Group 10: Astra API Security Platform - Astra API Security Platform focuses on discovering, scanning, and securing APIs, addressing the needs of security-sensitive teams [52][54]. - It provides real-time traffic analysis and extensive testing coverage to ensure API security and integrity [54][55]. - The platform garnered 418 Upvotes and 83 comments, indicating strong interest in API security solutions [56].

图片来源： Cognition AI Cognition AI 获新一轮融资估值达 100 亿美元人工智能初创公司 Cognition AI ， 在一轮约 4 亿美元的融资中估值达到 102 亿美元，这笔交易凸显出投资者 对 AI 驱动软件开发持续高涨的热情。 本轮融资由彼得 ·蒂尔支持的创投基金 Founders Fund 领投，包括 Lux Capital 、乔·朗斯代尔的 8VC 、埃拉德·吉尔、 Definition Capital 和 Swish Ventures 等现有投资方跟投。 这笔交易标志着 Cognition 估值的大幅跃升，该公司今年早些时候融资时的估值仅为 40 亿美元 。与此同时，公司业务也实现了大幅增长。 Cognition 旗下编程工具 Devin （号称 " 全球首位 AI 软件工程师 " ）的年经常性收入，从 2024 年 9 月的约 100 万美元攀升至 6 月的约 7300 万美元。公司 表示，自 7 月收购编程初创公司 Windsurf 以来，其总收入已翻番。 这家成立仅两年的初创公司称，其净现金消耗始终控制在 2000 万美元以内。 该公司还向部分 Windsurf ...

Z Highlights 图片来源： sequoiacap Sequoia Capital 是全球顶级风投机构红杉资本发布的思想领导力内容，专注科技投资洞察、创业指导和行业趋势分析。本文发布于 2025 年 8 月。 零售业是一个庞大的市场，在美国年产值超过 7 万亿美元。纵观历史， 重大技术进步一直推动着零售业的重大创新。我们相信 AI 时代为创造下一个亚马 逊提供了机遇 。本文将分析零售业的八次历史性技术变革，并预测 AI 革命可能在未来以五种方式改变零售业。 我们为那些对研究过去和理解未来可能面貌感兴趣的创业者、运营者、技术专家和投资者撰写了这篇文章。我们希望分享这一论述将有助于激发创业精 神，甚至可能帮助启动下一家万亿美元级零售公司。 古代零售 每一次重大技术进步都为购物引入新维度。电子商务革命了信息获取，移动设备创新了配送速度，而 AI 则革命性地改变了信息分析方式。 AI 的超能 力是分析和回答复杂问题， 未来零售巨头将通过专注于需要专家问答的咨询式购买而崛起，其独特之处将是集成化咨询服务 。 传统零售如沃尔玛基于明确的产品购买意图，而咨询式零售如家得宝则解决客户问题。顾客带着 " 马桶漏水 " ...

图片来源： Sierra OpenAI 近期向股东们传递了喜忧参半的消息。 ChatGPT 带来的收入增长比半年前公司预期更为迅猛。坏消息是，支撑这款聊天机器人的人工智能研发算 力成本，以及其他数据中心相关支出，将以更快的速度攀升。 • OpenAI预计从2025年到2029年累计现金消耗高达1150亿美元，较半年前预测激增800亿美元。2024年消耗预计超80亿，2028年峰值达450亿。主因是AI训 练（如GPT-5开发试错）、推理计算（1500亿支出）、自建服务器设施（近1000亿计划）及人才竞争（额外200亿股权激励） •OpenAI将2030年收入预测上调15%至2000亿美元。核心驱动是ChatGPT（今年收入近100亿，2030年达900亿），并计划通过免费用户变现。 • 但API和智能体（Agents）业务，收入预期分别下调50亿和260亿美元。 这一史无前例的预计现金消耗。 在之前两年已烧掉约 20 亿美元的基础上进一步增加——有助于解释为何该公司正在筹集比史上任何私营企业都多的资金。 CEO 萨姆·奥尔特曼此前曾告诉员工，他们的公司可能是有史以来"资本最密集"的初创企业。 OpenAI 巨 ...

图片来源： a16z Z Highlights ： a16z 是硅谷最具影响力的风险投资机构之一，它长期深耕人工智能、企业软件等多个领域，支持了包括 OpenAI 、 Databricks 在内的全球领先科技企业。 作为新一代技术与产业变革的推动者， a16z 不仅在资本层面提供支持，更通过研究、行业洞察与政策对话推动技术生态的成熟与落地。本篇文章由 a16z 团队发表于 2025 年 8 月 28 日，系统性地勾勒了 Agents 成为未来数字同事的趋势，并进一步展现了他们在生成式 AI 和未来工作模式探索中的前瞻视野。 自主型智能的临界点：从 RPA 到真正的 Agent 化转变 想象一下，你被要求为公司寻找新的办公空间 —— 这是个你宁愿避免的任务。现在再想象，你能把整个流程都交给一个 AI 来完成：从确定需求、调研地 点、安排看房、谈判租约，到处理保险和突发问题，全程无需你的参与或明确指令。 这样的自主型、任务导向的 "AI Agents" ，长期以来一直是这一领域 的终极目标。 然而，尽管已经投入了大量关注和努力，如今的 Agents 仍然未能实现这一愿景。目前市面上的 Agents 更接近于先进的 ...

随着大型语言模型（ LLMs ）的发展，应用端对其推理效率和性能的要求不断提高，研究人员持续探索更有效的推理方法。因此 Deep Think with Confidence （ DeepConf ）方法应运而生，它克服现有方法的不足，通过利用模型内部置信度信号，在推理过程中或生成后 动态过滤低质量推理轨迹 ，实 现 推理效率和性能的双重提升 。 本次访谈， Z Potentials 特邀 Meta FAIR 研究科学家、加州理工学院博士 Jiawei Zhao 于北京时间9月9日9:00am 直播 talk ，深度解析其团队颠覆性工 作 DeepConf ，揭开了 LLMs 推理中 " 置信度信号驱动低质轨迹动态过滤 " 的核心逻辑。 Yichao FU 和 Jiawei Zhao 为 DeepConf 的共同贡献者。 图片来源： https://jiaweizzhao.github.io/deepconf/ Jiawei Zhao 是 Meta FAIR （ Meta 的基础 AI 研究部门）研究科学家。拥有加州理工学院（ Caltech ）的博士学位。研究成果主要集中在机器学习领域，特 别是针对大型语言模 ...