🌐 双语
Archive

AI Builders
Digest

2026-09-22 10 builders · 21 tweets · 0 podcasts · 1 blogs

🔥 热点话题

个人Agent的商业化潜力与软件使用爆发Personal Agents' Monetization Potential and 100X Software Usage

Box CEO Aaron Levie指出,能代表用户完成端到端交易的个人Agent,其变现潜力巨大。用户会从简单日常任务逐步升级到复杂任务,降低商业摩擦后反而会增加整体消费,为Agent提供商以及Agent交互层(电商、本地服务、B2B)创造多层级机会。他还强调AI Agent将比人类多使用100倍软件,核心原语如CRM、ERP和数据平台在安全护栏、数据管理和工作流编排上变得更关键,给能快速行动的新老平台都带来巨大机会。
Box CEO Aaron Levie highlights the significant monetization potential of personal agents that transact end-to-end on users' behalf. Users will start with simple tasks and escalate to complex ones; lower friction for commerce actually increases overall spend, creating opportunities for agent providers and the layers agents interact with (commerce, local, B2B services). He also notes AI agents will use software 100X more than people ever did. Core primitives like CRM, ERP, and data platforms become even more critical as security layers, data managers, and workflow orchestrators for agents that can take destructive actions, offering huge openings for both startups and existing platforms that move fast.
查看原文 →查看原文 →

Agent浏览将颠覆广告市场Agents Will Disrupt Ad Markets

Practical AI教程作者Peter Yang警告广告市场将迎来猛烈觉醒:如果大量业务依赖向人类展示定向展示广告,当Agent自主浏览网站并完成任务、人类根本看不到广告时,会发生什么?他还分享ChatGPT Finances帮助发现遗漏收费并成功退款的案例,并表示自己已从邮件和短信全面迁移到与Agent的对话中生活。
Practical AI tutorials creator Peter Yang warns of a rude awakening in ad markets: if a huge part of your business relies on showing targeted display ads to humans, what happens when agents browse websites and complete tasks without any human ever seeing the ads? He also shared how ChatGPT Finances spotted missed charges and secured reimbursements, and noted he no longer lives in email or text but in chats with his agent(s).
查看原文 →查看原文 →查看原文 →

AI正在复兴美国梦AI Is Reviving the American Dream

Replit CEO Amjad Masad表示,AI正在复兴美国梦,引用相关讨论强调技术赋能普通人实现雄心的力量。
Replit CEO Amjad Masad states that AI is reviving the American Dream, quoting related discussion on how the technology empowers ordinary people to achieve ambition.
查看原文 →

OpenAI即将迎来重置OpenAI Reset Teased for Tuesday

OpenAI Codex与ChatGPT工程师Thibault Sottiaux高调预告:女士们先生们,启动引擎吧。我们几乎到了周二,我承诺过周二会有重置,还有其他内容。很快见。这条推文获得上万点赞和大量互动。
OpenAI Codex & ChatGPT engineer Thibault Sottiaux teased with high engagement: Ladies and gentlemen... start... your... ENGINES. We are almost Tuesday and I promised a reset for Tuesday. Among some other things. See you soon.
查看原文 →

🛠️ 开发者工具与技巧

Claude in Chrome全面可用Claude in Chrome Generally Available

Claude Blog: Claude in Chrome is generally available。Claude in Chrome现已在所有付费Claude计划全面可用,Claude可在浏览器中自主采取行动,安全分类器会在执行前验证每个动作是否安全且符合请求。它能访问内部仪表盘、遗留系统和供应商门户,使用现有登录查看页面、点击、输入和填表。Anthropic强化了针对prompt injection的防护:模型训练库持续更新、探针扫描网页内容、动作验证分类器。最新评估显示,在探针+分类器加持下,Sonnet 5和Opus 5等模型攻击成功率为0%。从Chrome Web Store安装即可使用,企业版支持管理员域名限制。
Claude Blog: Claude in Chrome is generally available. Claude in Chrome is now generally available on every paid Claude plan. Claude can take actions autonomously in the browser, with a safety classifier validating each action before it runs. It accesses internal dashboards, legacy systems, and vendor portals using existing logins to view pages, click, type, and fill forms. Anthropic strengthened prompt-injection defenses via expanded attack libraries for training, probes that screen web content, and action-verification classifiers. Latest evaluations show 0% attack success against Sonnet 5 and Opus 5 with probes plus classifiers. Install from the Chrome Web Store; Enterprise admins can limit to approved domains.
查看原文 →

Vercel AI SDK新增Jev HTTP支持,Grok 4.7表现惊艳Vercel AI SDK Adds Jev over HTTP; Grok 4.7 Impresses

Vercel CEO Guillermo Rauch宣布,除了类型安全的AI SDK API外,现在可在AI Gateway通过HTTP使用Jev。他还称赞Grok 4.7成功解决了一个涉及逆向工程运行二进制的高难度问题,且速度极快。
Vercel CEO Guillermo Rauch announced that in addition to the type-safe AI SDK API, users can now use Jev over HTTP in AI Gateway. He also praised Grok 4.7 for beautifully solving a hard reverse-engineering problem on a running binary, noting its impressive speed.
查看原文 →查看原文 →

Capydotai成为新宠Agent编码武器,Cluely仍有价值Capydotai as New Agentic Coding Weapon; Cluely Still a Good Idea

Y Combinator总裁兼CEO Garry Tan分享,Capydotai能跟踪多步骤工作流并比Codex或Claude Code更快完成大型PR,成为他过去一周最爱的Agent编码秘密武器。他还展示了其在GBrain上执行雄心勃勃bug修复的清晰任务划分、自动并行化和干净GitHub PR工作流。同时他重申,Cluely作为实时思维助手和半对抗式、持续上下文的助手,依然是个好主意。
Y Combinator President & CEO Garry Tan shared that Capydotai tracks multi-step workflows and executes large PRs faster than Codex or Claude Code alone, calling it his favorite new agentic coding secret weapon. He showed an example of it handling an ambitious bug-fix wave on GBrain with clear task delineation, automatic parallelization, and clean GitHub PR/CI workflow. He also reaffirmed that Cluely as a realtime thought helper and semi-adversarial assistant with ongoing context is still a good idea.
查看原文 →查看原文 →查看原文 →

Codex仍是专业工作首选,少即是多Codex Remains Undefeated for Pro Work; Less Is More

FPV Ventures合伙人Nikunj Kothari表示,自己长期把Codex当作主要Agent,以至于忘记普通人从未用过它。Instinct和Muse在手机浏览器场景很出色,零配置即可体验优秀Agent;但Mac上的Codex凭借Computer Use在复杂工作流上一骑绝尘。他期待其面向大众的Agent发布,并呼吁展示真实工作并降低入门门槛。他还批评tokenmaxxing公司产品体验最差,因为好产品靠策展和园艺,而非把厨房水槽扔给Agent让其自己摸索,少即是多。
FPV Ventures partner Nikunj Kothari noted he has used Codex as his primary agent for so long he forgets the average person has never tried it. Instinct and Muse excel for zero-setup browser use on phones, but Codex on Mac with Computer Use remains undefeated for real workflows. He looks forward to its mass-market agent launch and asks teams to showcase real work and make onboarding easy. He also criticized companies boasting about tokenmaxxing for having the worst product experiences: good products are about curation and gardening, not throwing the kitchen sink at agents. Less is more.
查看原文 →查看原文 →

OpenClaw启发Meta自建Agent,安全与自主并存OpenClaw Inspired Meta’s Agent; Security and Autonomy

OpenClaw创建者Peter Steinberger澄清,Meta是受启发后自建了自己的Agent,而非直接使用OpenClaw,并向Nat团队致意。他自豪地表示自家系统通过了安全审计,未发现关键问题,并强调自己运行Claw的好处:别人无法屏蔽你。
OpenClaw creator Peter Steinberger clarified that Meta built its own agent inspired by OpenClaw rather than using it directly, giving kudos to Nat and team. He proudly shared that their system passed security review with nothing critical found, and highlighted the beauty of running a claw yourself: they cannot block you.
查看原文 →查看原文 →查看原文 →

提示词小技巧:大图少字Prompt Tip: Big Pictures and Few Words

Anthropic Claude Code工程师Thariq分享日常实用技巧:我现在每天都会多次输入“use big pictures and few words”。
Anthropic Claude Code engineer Thariq shared a practical daily habit: I now type "use big pictures and few words" several times a day.
查看原文 →

🌍 其他动态

无其他显著动态No Other Notable Updates

今日其余推文多为播客推广或轻量互动,无额外实质内容。
Remaining posts today were mostly podcast promotions or light engagement with no additional substantive content.