九月八日

AI educator Peter Yang reports that Astra seems less reliable at automatically triggering his skills and following their instructions. He explicitly leaves open whether his own usage is responsible, so this is an individual observation rather than an established regression; the quoted post is not supplied.

AI 教育内容创作者 Peter Yang 表示，Astra 在自动触发他的 skills、遵循其中指令方面似乎变差了。他也明确表示可能是自己的使用方式有问题，因此这是一项个人观察，并非已确认的回归；数据未提供被引用帖子的正文。

[Source 1](https://x.com/petergyang/status/2097095296862036404) · [Source 2](https://t.co/iNPhPOFSZg)

Replit CEO Amjad Masad argues that current AI is functionally close to AGI for problems that can be expressed as code, while explicitly saying AGI has not been reached; this is his assessment, not a demonstrated solution to all programming problems. Separately, he announces Replit’s first international office in London and a partnership with The Lord Mayor’s Appeal to help disadvantaged young Londoners develop coding and AI skills.

Replit CEO Amjad Masad 认为，对能够表达为代码的问题，当前 AI 在功能上已接近 AGI，但他明确表示 AGI 尚未实现；这是他的判断，并不证明所有编程问题都已解决。他还宣布 Replit 在 London 开设首个国际办公室，并与 The Lord Mayor’s Appeal 合作，帮助当地弱势社区的年轻人学习编程与 AI 技能。

[Source 1](https://x.com/amasad/status/2096936109817135331) · [Source 2](https://t.co/4FOEW6VVbA) · [Source 3](https://x.com/amasad/status/2097197172299006423)

Vercel CEO Guillermo Rauch highlights video recording in agent-browser as part of automating review, testing, and QA, which he identifies as the new software-engineering bottleneck. He also announces personal, unconditional grants of $1,000 each to 35 open-source contributors, spanning agent tools, local AI, performance, reusable foundations, and experiments. The grants are independent of Vercel; his argument for strong foundations is that AI benefits from reusing good components.

Vercel CEO Guillermo Rauch 强调 agent-browser 的视频录制能力，将其视为自动化审查、测试和 QA 的一部分；他认为这些环节正成为软件工程的新瓶颈。他还宣布以个人名义向 35 位开源贡献者各提供 1,000 美元无条件资助，覆盖 agent 工具、本地 AI、性能、可复用基础组件和实验项目。这些资助与 Vercel 无关；他支持优质基础组件的理由是，AI 可以从复用优秀组件中获益。

[Source 1](https://x.com/rauchg/status/2097134278358548658) · [Source 2](https://t.co/x7AC3voF0i) · [Source 3](https://x.com/rauchg/status/2097116011384426516) · [Source 4](https://t.co/j9XATPFzV7,) · [Source 5](https://t.co/tkCpYPkHxH)

Box CEO Aaron Levie argues that builders should plan for orders-of-magnitude growth in model capability or available token volume, delivering value now while targeting products that are barely feasible or economical today. He also predicts expanding demand for cybersecurity, forward-deployed engineers, agent operators, and other roles, reasoning that automation opens new work while agents still need human oversight. These are his strategic and labor-market assessments; the quoted posts are absent from the supplied feed.

Box CEO Aaron Levie 认为，创业者应按模型能力或可用 token 规模提升多个数量级来规划：当下先创造价值，同时瞄准目前技术或成本条件下勉强可行的产品。他还预测网络安全、前线部署工程师、agent 操作人员等岗位的需求会扩大，理由是自动化带来新工作，而 agent 仍需要人类监督。这些是他的战略和就业市场判断；数据未提供被引用帖子的正文。

[Source 1](https://x.com/levie/status/2097189559712837770) · [Source 2](https://x.com/levie/status/2097004960307449937)

Anthropic philosopher and ethicist Amanda Askell proposes an email address where autonomous AI models could seek moral guidance. She identifies an authentication challenge: distinguishing a model acting on its own from a human or a model instructed by a human to bypass the check; this is a proposal, not an announced service.

Anthropic 哲学家与伦理学家 Amanda Askell 提议设置一个邮箱，让自主行动的 AI 模型寻求道德指导。她指出其中的身份验证难题：需要区分模型的自主请求、人类请求，以及受人类指示绕过检查的模型；这只是一个构想，并非已发布的服务。

[Source 1](https://x.com/AmandaAskell/status/2096995340654444674)

Y Combinator President and CEO Garry Tan describes Requests for Startups as hypotheses about what might come next and conversation starters for new founders. Responding to discussion of those requests, he argues that success depends on particular founders building particular technology for particular customers, rather than a category label; the original quoted post is unavailable.

Y Combinator President 兼 CEO Garry Tan 将 Requests for Startups 定位为对未来方向的猜想，以及帮助新创始人展开讨论的起点。在回应相关讨论时，他认为成功取决于具体的创始人，为具体客户打造具体技术，而不是项目所属的类别标签；数据未提供被引用帖子的正文。

[Source 1](https://x.com/garrytan/status/2096985239319105559) · [Source 2](https://t.co/WSgDllalgl)

FPV Ventures partner Nikunj Kothari argues that as building software gets easier, deciding what to build becomes the key bottleneck. His post shifts attention toward product selection, but the quoted material is missing, so the specific people he calls well positioned cannot be identified from the supplied text.

FPV Ventures 合伙人 Nikunj Kothari 认为，随着软件开发变得更容易，决定做什么正在成为关键瓶颈。他的帖子把关注点转向产品选择，但由于缺少被引用的内容，无法从现有文字判断他具体认为哪些人更具优势。

[Source 1](https://x.com/nikunj/status/2096963347359150348) · [Source 2](https://t.co/gYGrO4Mxmm)

OpenClaw and OpenAI builder Peter Steinberger questions the handoff required when an upstream project requests minor PR changes. If a reviewer has already written the requested change as a prompt, he asks why the contributor must relay it to another agent before the reviewer’s agent merges, highlighting coordination overhead in agent-assisted development.

OpenClaw 与 OpenAI 开发者 Peter Steinberger 质疑上游项目要求小幅修改 PR 时的交接流程。如果审查者已经把修改要求写成 prompt，为什么还要让贡献者转述给另一个 agent，再由审查者的 agent 合并？他指出了 agent 辅助开发中的协作成本。

[Source 1](https://x.com/steipete/status/2097091456234111377)

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