九月十日

Anthropic's Claude account announces CrowdStrike, Cursor, FactoryAI, Gamma, and Vercel as new additions to Claude Marketplace, saying enterprises can now use their Anthropic spend commitment to buy more Claude-powered products and agents. The substantive change is the purchasing route: these products can draw on a customer's committed Anthropic spending, connecting third-party software procurement to an existing commitment. The announcement describes an expanded purchasing channel, without providing transaction volumes, revenue shares, or detailed eligibility terms.

Anthropic 的 Claude 账号宣布，CrowdStrike、Cursor、FactoryAI、Gamma 和 Vercel 新加入 Claude Marketplace，企业现在可以使用对 Anthropic 承诺的支出额度，购买更多由 Claude 驱动的产品和 agent。实质变化在采购渠道：这些产品可以使用客户已有的 Anthropic 承诺支出额度，把第三方软件采购与现有支出承诺连接起来。这项公告描述了采购渠道的扩展，没有提供交易规模、收入分成或具体准入条款。

OpenAI's Thibault Sottiaux, who works on Codex and ChatGPT, reports that some banked resets did not fully apply when used in ChatGPT Work and Codex during an affected window. He says everyone who used one in that window will receive another reset and an apology email. This is a specific usage-reset problem with a stated remedy tied to having used a reset in the affected period. His post does not specify the window's boundaries or explain the underlying failure.

负责 Codex 和 ChatGPT 的 OpenAI 成员 Thibault Sottiaux 表示，在一段受影响的时间内，部分用户在 ChatGPT Work 和 Codex 中使用已积攒的重置次数时，重置没有完全生效。他说，所有在这段时间内使用过重置的用户都会再获得一次重置，并收到致歉邮件。这是一项具体的用量重置问题，已公布的补偿对象是在受影响时段使用过重置的用户。帖子没有给出该时段的起止时间，也没有解释故障原因。

Vercel CEO Guillermo Rauch reports eight price cuts, fee removals, and pricing-structure improvements, alongside 16 model discounts, over the past six months. The first eight changes span several types of adjustment, so that count should not be read as eight reductions to a single price. He signals that more changes are coming, without providing their timing or terms. The post establishes the breadth of pricing work he reports, but contains no rate card from which to calculate savings for a particular workload.

Vercel CEO Guillermo Rauch 表示，过去六个月内，团队进行了八项降价、费用取消和定价结构改进，此外还有 16 项模型折扣。前八项涵盖多种类型的调整，因此不能把这个数字理解为某一个价格连续下调八次。他预告还会有更多变化，但没有给出时间或具体条款。这条动态呈现了他所报告的定价调整范围，却没有提供价目表，无法据此计算某项具体工作负载能节省多少费用。

OpenClaw and OpenAI builder Peter Steinberger reports getting cloud sessions working fast in OpenClaw, naming Remote Terminal, WebVNC, and CUA for computer use. Separately, he says the dashboards and mini-apps he pushed two months ago have replaced much of the custom tooling his team built around OpenClaw: everything on their team server is now a sidebar entry, dashboard, or plugin. Taken together, his reports describe both access to remote execution and the consolidation of surrounding tools into the team's OpenClaw workspace. The speed assessment and tooling replacement are his reported experience, without latency measurements or a broader deployment evaluation in the supplied posts.

参与 OpenClaw 和 OpenAI 的开发者 Peter Steinberger 表示，他终于让 OpenClaw 的云端会话实现了快速运行，并列出 Remote Terminal、WebVNC 和用于电脑操作的 CUA。另外，他说，自己两个月前推动的 dashboards 和 mini-apps，已经替代团队围绕 OpenClaw 自建的大量工具：团队服务器上的各项功能如今都是侧边栏入口、dashboard 或插件。这两条动态分别呈现了远程执行入口，以及外围工具向团队 OpenClaw 工作空间的整合。速度评价和工具替代情况均来自他的使用经历，所提供的帖子没有延迟测量或更大范围的部署评估。

Box CEO Aaron Levie predicts that lowering the cost of code will expand software use enough to increase demand for engineers, pointing to companies that previously could not afford custom software, cyber defense, life-sciences research, complex data workflows, and legacy-system upgrades. In a second post, he argues that even extremely capable AI remains constrained by data preparation, pipeline setup, process redesign, change management, and agreement on how a workflow should operate. Beyond those organizational constraints sit real-world delays such as customer responses, project permits, and the years required to bring a drug to consumers, while some useful daily AI activity may be neutral for GDP in the near term. His argument combines expanding demand with slow diffusion: he expects more uses for software, but sees the next decade's opportunity in connecting AI capabilities to actual workflows, without assuming that technical capability immediately translates into measurable economic output.

Box CEO Aaron Levie 预测，代码成本下降将扩大软件的使用范围，进而增加对工程师的需求；他举出的方向包括过去无力负担定制软件的企业、网络防御、生命科学研究、复杂数据流程和老旧系统升级。在另一条帖子中，他认为，即使 AI 能力极强，也仍然受到数据准备、管道搭建、流程重构、变革管理，以及就工作流程如何运作达成共识等因素的制约。组织层面的限制之外，还有等待客户回复、项目许可，以及药物最终交付消费者所需的多年时间；与此同时，一些有用的日常 AI 应用在短期内可能并不改变 GDP。他的论点把需求扩张与缓慢普及放在一起：软件的用途会增多，但未来十年的机会在于把 AI 能力接入真实流程，不能假定技术能力会立即转化为可衡量的经济产出。

Every CEO Dan Shipper challenges two assumptions in a report he is responding to: that jobs can be decomposed into tasks, and that human tasks created by automation can always be modeled as a fraction of the tasks automated away. He argues from experience that automation can generate multiples more human work, so a model that limits new work by construction can miss an important source of demand. He accepts task lists as a useful analytical lens, but says a job also involves a way of noticing and caring about the world: editors and nurses notice different problems, and those concerns can generate entirely new tasks as tools and circumstances change. His substantive challenge is to explain where new tasks come from rather than assume a fixed inventory of work; the supplied material contains his critique and linked argument, not the underlying report or a quantitative estimate of net job creation.

Every CEO Dan Shipper 质疑所回应报告中的两个假设：工作可以被拆解成任务，以及自动化带来的人类新任务，总能被建模为已自动化任务的一定比例。根据自己的经历，他认为自动化可能带来数倍的人类新工作，因此，如果模型从设定上就限制了新增工作量，就可能漏掉重要的需求来源。他承认任务清单是一种有用的分析视角，但也指出，职业还包含观察和关心世界的方式：编辑与护士会注意到不同的问题，而当工具和环境变化时，这些关注点会生成全新的任务。他提出的实质问题是解释新任务从何而来，而非预设一份固定的工作清单；所提供的材料包含他的批评及延伸论述链接，没有底层报告正文，也没有净就业增长的量化估计。

通过 Follow Builders 技能生成：https://github.com/zarazhangrui/follow-builders