🌐 双语
Archive

AI Builders
Digest

2026-09-10 14 builders · 28 tweets · 0 podcasts · 0 blogs

🔥 热点话题

Aaron Levie:编程 Agent 将远超预期,工程师需求不减反增Aaron Levie: Coding Agents Will Be Used Far Beyond Expectations

Box CEO Aaron Levie 指出,世界将大规模使用 coding agents,远超此前想象。人们会用它们构建全新品类工具、为以前负担不起软件的公司开发系统、防护网络风险、自动化生命科学研究、处理复杂工作流数据、升级遗留系统等。降低代码成本意味着软件更有用,工程师的杠杆大幅提升,反而需要更多而非更少工程师。

他进一步分析 AI 能力与 GDP 影响的落差:AI 扩散远比人们想象的慢,受限于企业物理定律——数据准备、流程再造、变革管理、对齐新工作流,以及真实世界的速度(客户回复、许可审批、药物研发周期等)。许多日常 AI 使用短期内对 GDP 是中性的。未来十年的主题是 AI 扩散,最大机会在于搭建超级智能与真实工作流之间的桥梁。
Box CEO Aaron Levie argues the world will use coding agents for far more than anyone expected. Teams will build tools in entirely new categories, create software for companies that previously could not afford it, protect systems from cyber risk, automate research in life sciences, process massive complex workflows, and upgrade legacy infrastructure. By lowering the cost of code, software becomes more useful and engineer leverage rises dramatically—meaning we will need more engineers, not fewer.

He also reconciles AI capability with limited near-term GDP impact: diffusion will take much longer than expected and is constrained by corporate physics (data pipelines, process reengineering, change management) plus real-world latency (customer responses, permits, multi-year drug pipelines). Many positive daily AI uses are GDP-neutral in the short run. AI diffusion is the theme of the next decade; the opportunity lies in building the bridges between superintelligence and actual workflows.
查看原文 →查看原文 →

Dan Shipper:自动化未必减少人类工作,反而常创造更多Dan Shipper: Automation Often Creates More Human Work, Not Less

Every CEO Dan Shipper 对一份关于自动化与就业的报告提出尖锐质疑。报告假设工作可分解为任务,且自动化总是减少(或至少不增加)人类劳动。他反驳:在自己的经验里,自动化经常创造数倍于被替代的人类工作。报告还忽略了工作的本质——它是一种看待与关心世界的方式(借用 Wittgenstein 与 Heidegger)。编辑注意到的东西与护士不同;工具与环境变化时,同样的关切会生成全新任务。这个过程本身就是工作的一部分,却几乎无法被任务列表建模。认真对待这一点,会大幅改变对 AI 自动化与就业的看法。
Every CEO Dan Shipper challenges a widely discussed report on automation and jobs. The report assumes jobs decompose cleanly into tasks and that automation always reduces (or at best holds constant) human labor. In his experience the opposite is often true: automation frequently creates multiples more work for humans. A deeper blind spot is treating jobs as mere task lists. Drawing on Wittgenstein and Heidegger, a job is a particular way of seeing and caring about the world—an editor notices different things than a nurse. As tools and circumstances change, the same cares generate entirely new tasks. That generative process is part of the job itself and is almost impossible to model as a static task inventory. Taking it seriously substantially changes the analysis of AI and employment.
查看原文 →

Claude Marketplace 扩展:企业可用 Anthropic 支出承诺直接购买合作产品Claude Marketplace Adds Major Partners; Enterprises Can Spend Commitment on Claude-Powered Products

Anthropic 官方账号宣布 Claude Marketplace 新增 CrowdStrike、Cursor、Factory、Gamma 与 Vercel。企业现在可以用自己的 Anthropic 支出承诺直接购买更多 Claude 驱动的产品与 agent,降低采购摩擦。同时鼓励在 Claude Platform 上构建的开发者上架,方便客户发现与购买。
Anthropic’s Claude account announced new Claude Marketplace listings for CrowdStrike, Cursor, Factory, Gamma and Vercel. Enterprises can now apply their Anthropic spend commitment directly toward Claude-powered products and agents, reducing procurement friction. Builders on the Claude Platform are encouraged to get listed so customers can more easily discover and buy their solutions.
查看原文 →查看原文 →

Sam Altman 欢迎 Paul 回归 AI 安全,同时强调客户服务优先Sam Altman Welcomes Paul Back to AI Safety Work; Prioritizes Customer Service

OpenAI CEO Sam Altman 发帖欢迎 Paul(此前长期从事 AI 安全工作的关键人物)回归,表示感激其过往贡献,并期待再次合作。另有帖子回应服务压力时明确表态:即使短期内体验可能受影响,也会优先保证对客户的优质服务,直到产能重新跟上。
OpenAI CEO Sam Altman publicly welcomed Paul (a long-time AI safety contributor) back, expressing gratitude for prior work and excitement about collaborating again. In a separate post addressing capacity strain, he stated that OpenAI will prioritize great service for customers until it can get fully back on top of demand.
查看原文 →查看原文 →

Thibault Sottiaux:ChatGPT Work 与 Codex 重置额度问题已补偿Thibault Sottiaux: ChatGPT Work & Codex Reset Glitch Fixed with Extra Credits

OpenAI Codex & ChatGPT 团队的 Thibault Sottiaux 说明,早些时候部分 banked resets 在 ChatGPT Work 与 Codex 中未完全生效。所有在受影响时间窗口使用过重置的用户都会额外获得一次,并收到道歉邮件。
OpenAI Codex & ChatGPT team member Thibault Sottiaux reported a morning glitch in which some banked resets failed to fully apply in ChatGPT Work and Codex. Everyone who used a reset during the affected window is receiving another one plus an apology email.
查看原文 →

🛠️ 开发者工具与技巧

Peter Steinberger:OpenClaw 云会话大幅加速,仪表盘与插件成主力Peter Steinberger: OpenClaw Cloud Sessions Now Fast; Dashboards & Plugins Take Over

OpenClaw 相关开发者 Peter Steinberger 宣布终于让云会话变得很快,支持 Remote Terminal、WebVNC 与 computer-use CUA。两个月前他推动的 Dashboards/Mini-Apps 现已取代大量定制工具——团队服务器上几乎所有东西都变成了侧边栏入口、仪表盘或插件。
Peter Steinberger (OpenClaw) reported that cloud sessions are finally working fast, with Remote Terminal, WebVNC and CUA for computer use. Dashboards and mini-apps he pushed two months ago have now replaced most custom tooling; everything on the team server is a sidebar entry, dashboard or plugin.
查看原文 →查看原文 →

Peter Yang:用「残酷诚实反馈」提示词测试新模型思维能力Peter Yang: Best Way to Test a Model’s Thinking—Ask for Brutal Blind-Spot Feedback

实用 AI 内容创作者 Peter Yang 分享测试新模型思维能力的最爱方法:直接把包含完整提示词的图片丢给模型,让它对你当前阻碍进步的盲点给出残酷诚实的反馈。他特别提到 Astra 的反馈尤其直接。完整帖子与提示词可在链接中查看。
Practical AI educator Peter Yang shared his favorite way to test a new model’s thinking ability: paste an image containing a full prompt that asks the model for brutally honest feedback on the blind spots holding you back. Astra’s response was especially blunt. Full post and prompt available via the link.
查看原文 →

Guillermo Rauch:Vercel 过去六个月持续降价与折扣Guillermo Rauch: Vercel Delivered 8 Price Cuts + 16 Model Discounts in Six Months

Vercel CEO Guillermo Rauch 指出,过去六个月公司已完成 8 次价格下调、费用取消与结构优化,外加 16 次模型折扣,并预告还会有更多。同时宣传与 Benchmark 联合举办的 AI Benchmarks 活动,聚焦模型评测这一关键软件品类。
Vercel CEO Guillermo Rauch highlighted that the company has shipped 8 price cuts, fee removals and structural improvements plus 16 model discounts in the last six months alone, with more coming. He also promoted the AI Benchmarks event co-hosted with Benchmark, calling model evaluation one of the most important software categories of the generation.
查看原文 →查看原文 →

🌍 其他动态

Aditya Agarwal:Waymo 联席 CEO Dmitri Dolgov 将做客 SPCAditya Agarwal: Waymo Co-CEO Dmitri Dolgov Visiting SPC

SPC 普通合伙人 Aditya Agarwal 介绍,Waymo 是人类导向技术的典范之一。Dmitri Dolgov 自 2009 年加入 Google 自动驾驶项目,现任 Waymo 联席 CEO,本月将到访 SPC 分享。名额有限,需提前 RSVP。
SPC General Partner Aditya Agarwal noted that Waymo remains one of the greatest examples of human-oriented technology. Dmitri Dolgov joined Google’s self-driving project in 2009 and now co-leads Waymo as co-CEO; he will visit SPC this month. Spots are limited and require RSVP.
查看原文 →查看原文 →

Zara Zhang:想法容易,信念很难Zara Zhang: Idea Is Easy, Conviction Is Hard

Builder Zara Zhang 用一句简洁的观察总结创业核心难点:「Idea is easy. Conviction is hard.」
Builder Zara Zhang distilled a core startup truth into four words: “Idea is easy. Conviction is hard.”
查看原文 →