June 19, 2026

Anthropic's official Claude account formally introduced Artifacts in Claude Code: interactive pages built from your session, like a PR walkthrough or a living project dashboard, shared with your team at a private link. Artifacts draw on the full context of a session (codebase, plugins, skills, connected tools), stay private until you share them, and refresh automatically as the session keeps working so everyone is always looking at the latest version. Available in beta on Team and Enterprise plans.

Anthropic 官方 Claude 账号正式发布了 Claude Code 里的 Artifacts：由你的会话直接生成的交互式页面，比如 PR 走查或者一个会自动更新的项目 dashboard，通过私有链接分享给团队。Artifacts 会调用整个会话的上下文（codebase、plugins、skills、已连接的工具），在你主动分享之前都是私有的，而且会随着会话继续工作自动刷新，确保所有人看到的永远是最新版本。目前在 Team 和 Enterprise 套餐里开放 beta。

Link: https://x.com/claudeai/status/2067671912038240487

Anthropic's Boris Cherny, who works on Claude Code, says Artifacts have become part of his daily workflow: visual explanations of tricky code, system diagrams, quick previews of a few animation options, and data analyses and dashboards he shares with the team. He calls them "a game changer for how I work with Claude."

来自 Anthropic、负责 Claude Code 的 Boris Cherny 说，Artifacts 已经成了他日常工作流的一部分：用来做复杂代码的可视化讲解、系统架构图、几种动画方案的快速预览，以及他分享给团队的数据分析和 dashboard。他形容这是"彻底改变了我和 Claude 协作方式的东西"。

Link: https://x.com/bcherny/status/2067700226669060207

Anthropic's Cat Wu announced that on Claude Team and Enterprise you can now use Claude Code to deploy HTML sites and share them with teammates. She says it has already changed how her team works internally, calling Artifacts a great format for communicating architecture changes, data analyses, and new prototypes.

来自 Anthropic、负责 Claude Code 和 Cowork 的 Cat Wu 宣布，在 Claude Team 和 Enterprise 上，你现在可以用 Claude Code 部署 HTML 网站并分享给同事。她说这已经改变了团队内部的协作方式，认为 Artifacts 是沟通架构改动、数据分析和新原型的绝佳载体。

Link: https://x.com/_catwu/status/2067674836726694200

Anthropic's Thariq shared that Claude Code can now upload and edit HTML artifacts to share with your team or with other Claude instances, starting with teams and coming to Pro and Max plans soon. He will also give the Day 2 opening keynote at the AI Engineer World's Fair on July 1.

来自 Anthropic、负责 Claude Code 的 Thariq 透露，Claude Code 现在可以上传并编辑 HTML artifacts，分享给你的团队或者其他 Claude 实例，先在 team 套餐上线，很快会扩展到 Pro 和 Max 套餐。他还将在 7 月 1 日的 AI Engineer World's Fair 上做第二天的开场 keynote。

Link: https://x.com/trq212/status/2067682475611242546 · https://x.com/trq212/status/2067737883545596368

Linear's Head of Product Nan Yu shared a hard-won product lesson: their AI project-update feature first tried to one-shot updates with little user input, which produced "slop" because users disengaged their brains from the writing. Switching to a multi-turn interactive mode, where the agent asks what to emphasize and what context is missing, produced much better updates. He frames it as "a very good example of introducing friction for a net benefit," and separately likens good agent judgment to a chess player intuiting a move without calculating, which is really exercising taste.

Linear 的产品负责人 Nan Yu 分享了一条踩坑得来的产品经验：他们的 AI 项目更新功能一开始想在用户几乎不参与的情况下一次性生成更新，结果产出了一堆"slop"，因为用户的脑子完全脱离了写作过程。改成多轮交互模式后，agent 会主动问你想强调什么、缺了哪些上下文，更新质量明显提升。他把这总结为"一个'刻意制造摩擦反而带来净收益'的好例子"，并另外把好的 agent 判断力比作棋手不做计算就直觉走出一步棋，本质上是在运用 taste。

Link: https://x.com/thenanyu/status/2067703108344369306

Box CEO Aaron Levie argues that open weights models reaching frontier capability should be "a huge update for many": it ensures you can always have sovereign AI, post-train for your specific workflows, cost-optimize across workloads, and afford to do much more with AI, which he calls a huge win for the applied AI layer. On regulation, he reads the emerging framework gating future model releases past certain capability or compute thresholds as "practically an inevitability," and hopes it leads to bigger, batched model jumps rather than killing the rapid back-and-forth progress everyone has gotten used to.

Box CEO Aaron Levie 认为，open weights 模型达到 frontier 水平的能力，对很多人来说"应该是一次重大认知更新"：它意味着你永远可以拥有 sovereign AI、能针对自己的具体工作流做 post-training、能针对不同负载做成本优化，也能负担得起用 AI 做更多事情，他称这是应用层 AI 的一次巨大胜利。在监管方面，他把那套对超过一定能力或算力门槛的未来模型发布设卡的框架解读为"基本上是必然会发生的事"，并希望它带来的是更大、更集中的模型版本跃迁，而不是扼杀大家已经习惯了的那种快速迭代。

Link: https://x.com/levie/status/2067821985342878180 · https://x.com/levie/status/2067802697324212562

Y Combinator's Garry Tan sharply criticized a bill introduced by Bernie Sanders that would seize 50% of any AI startup crossing $200M in revenue, calling it "a war on building startups in America" and tying it to earlier efforts to ban startup acquisitions, the only exit 85% of founders ever get. He also offered a more philosophical take: "For people who don't have a clear sense of the future they want, AI is just another mechanism of control. But in the hands of someone with agency, AI is the breaker of chains."

Y Combinator 总裁兼 CEO Garry Tan 强烈批评了 Bernie Sanders 提出的一项法案，该法案要没收任何营收超过 2 亿美元的 AI 创业公司 50% 的股权。他称之为"一场针对在美国创业的战争"，并将其与此前试图禁止创业公司被收购的做法联系起来，而被收购正是 85% 的创始人唯一能拿到的退出方式。他还给出了一个更具哲学意味的观点："对那些对自己想要的未来没有清晰想法的人来说，AI 不过是又一种控制手段。但在一个有 agency 的人手里，AI 是挣脱枷锁的工具。"

Link: https://x.com/garrytan/status/2067636692794875923 · https://x.com/garrytan/status/2067606805459714229

FPV Ventures partner Nikunj Kothari ran a quiet experiment to see how far pure SEO/GEO optimization could go when you let Claude Code and Codex "optimize the hell out of" a website. With no backlinks, no mentions on his socials, and no Reddit hacking, he got it to roughly 16,000 impressions and 94 clicks over the last 28 days through old-school optimization alone, with an official launch coming soon.

FPV Ventures 合伙人 Nikunj Kothari 做了个低调的实验，想看看在让 Claude Code 和 Codex "往死里优化"一个网站之后，纯靠 SEO/GEO 优化能走多远。没有任何外链、社交媒体上只字未提、也没在 Reddit 上做手脚，他单靠老派的优化手段，在过去 28 天里把它做到了大约 16,000 次曝光和 94 次点击。正式发布很快就会到来。

Link: https://x.com/nikunj/status/2067830061009633294

Builder Zara Zhang showed off a polished HTML slide deck built with her Frontend Slides skill, packed with easter eggs like click-to-enlarge images and nested interactive content and hyperlinks, and shared both the skill itself and a recording of her talk.

Builder Zara Zhang 展示了一套用她自己的 Frontend Slides skill 做出来的精致 HTML 幻灯片，里面塞满了彩蛋，比如点击任意图片即可放大、层层嵌套的交互内容和超链接。她把这个 skill 本身和她演讲的录像都分享了出来。

Link: https://x.com/zarazhangrui/status/2067850383758901669 · https://x.com/zarazhangrui/status/2067851144664342725

Every CEO Dan Shipper described "extreme time deflation happening right now," predicting a step change in productivity once Fable is unlocked and only half-joking that you might as well go on vacation until then.

Every CEO Dan Shipper 形容"现在正在发生极端的时间通缩"，他预测一旦 Fable 解锁，生产力会迎来一次阶跃式的提升，还半开玩笑地说不如先去度个假等它上线。

Link: https://x.com/danshipper/status/2067630124795662471

Claude Code can now capture a session's work as an artifact: live, shareable web pages, including PR walkthroughs, system explainers, dashboards, and release checklists, that update themselves as the session runs. Built from the full context of your session (your codebase, connectors, and the conversation), a single incident page can pull together a failing test, the function behind it, an error spike from a connected monitoring tool, and the root-cause reasoning, with no data sources to wire up or infrastructure to stand up: "You ask for a page, and Claude Code builds it from what already exists." Every publish becomes a new version at the same link, with version history and restore, plus a gallery to manage everything you've made. When Claude updates an artifact, the open page refreshes in place so teammates always see the latest. Anthropic says debugging has been the most common internal use case: an engineer kicks off an incident investigation before standup, and by the time standup begins Claude has republished the page twice as the investigation progressed, so team members "don't have to walk us through what the agent found" because everyone shares the same view. Every artifact is private to its author by default and can only be shared with authenticated members of your org (never made public), with admin controls for access, retention, and compliance. The post lists role-specific prompts spanning legal license audits, privacy data-flow maps, security findings linked to exact lines, FinOps cost mapping, and PR walkthroughs. Artifacts is in beta for Claude Team and Enterprise orgs, from the Claude Code CLI and desktop app, viewable in any browser.

Claude Code 现在可以把一次会话的工作成果保存成 artifact：可实时更新、可分享的网页，包括 PR 走查、系统讲解、dashboard 和发布清单，并会随着会话运行自动更新。它基于整个会话的完整上下文构建（你的 codebase、connectors 和对话本身），一个事故页面就能把一个失败的测试、它背后的函数、来自已连接监控工具的错误激增、以及会话里得出的根因分析全部汇到一起，不需要你去对接数据源或搭建任何基础设施："你只要开口要一个页面，Claude Code 就会用现有的一切把它搭出来。"每次发布都会在同一个链接上生成一个新版本，带版本历史和恢复功能，还有一个 gallery 用来管理你做过的所有 artifact。当 Claude 更新一个 artifact 时，已打开的页面会原地刷新，团队成员看到的永远是最新版本。Anthropic 说内部最常见的用例是 debugging：一名工程师在站会前启动事故排查，等站会开始时，Claude 已经随着排查进展把页面重新发布了两次，于是大家"不用再让人复述 agent 到底查到了什么"，因为所有人看的是同一个视图。每个 artifact 默认对作者私有，只能分享给组织内经过身份验证的成员（永远无法公开），管理员可以控制访问权限、保留策略和合规。文章还列出了一系列按角色定制的 prompt，涵盖法务的依赖许可证审计、隐私的数据流向图、安全团队精确定位到具体代码行的发现、FinOps 的成本映射，以及 PR 走查。Artifacts 目前面向 Claude Team 和 Enterprise 组织开放 beta，可在 Claude Code CLI 和桌面端使用，页面在任意浏览器中都能打开。

Link: https://claude.com/blog/artifacts-in-claude-code

The Takeaway: GPU compute was never a commodity, and the people betting today's chips will be worthless in five years are, in Stephen Balaban's words, "completely wrong, and they've been wrong the entire time."

Balaban is the cofounder and CTO of Lambda, one of the top "neoclouds" powering the AI boom, with a cloud business now running near a billion dollars in revenue. His core argument: a cloud built for AI is a deeply vertically integrated business spanning land entitlement, power, data center construction, networking, virtualization, and cloud software, which is exactly why the world's biggest companies all compete in cloud. The rental price of a single GPU drifting down barely matters, because it's only one layer of a very complex cake.

His most counterintuitive claims are about depreciation and demand. Lambda leases 2023-era H100s at a higher rate today than when they were bought, because usable economic life far exceeds the six-year accounting schedule most firms assume. He sees "no end to the scaling laws": more compute keeps buying more intelligence, and as models get more capable the cone of addressable work, from customer support to augmenting or replacing software engineering, keeps widening. Even a 10x efficiency gain, he argues, just means everyone processes 10x more tokens against the same fixed pool of compute.

He is refreshingly specific on the physics, tracing the chain from photons and natural gas, to joules per second, to FLOPS, to tokens, and on why NVIDIA's real moat is the cuDNN and NCCL software stack, not just CUDA. His through-line, from a facial-recognition startup and a camera-in-a-baseball-cap to a near-billion-dollar cloud: "we have an amazing system that can take in money and output software."

核心要点：GPU 算力从来就不是大宗商品，而那些押注"今天的芯片五年后就会一文不值"的人，用 Stephen Balaban 的话说，"完全错了，而且他们从头到尾一直都是错的"。

Balaban 是 Lambda 的联合创始人兼 CTO，Lambda 是支撑这轮 AI 浪潮的顶级"neocloud"之一，如今云业务的营收已逼近十亿美元。他的核心观点是：为 AI 打造的云是一门高度垂直整合的生意，横跨土地审批、电力、数据中心建设、网络、虚拟化和云软件，这正是为什么全世界最大的那些公司都在抢做云。单颗 GPU 的租赁价格往下走其实无关紧要，因为它只是这块极其复杂的蛋糕中的一层。

他最反直觉的论断关于折旧和需求。Lambda 今天把 2023 年的 H100 租出去的价格，比当初买进来时还要高，因为芯片可用的经济寿命远远超过大多数公司假设的六年会计折旧期。他认为"scaling laws 还看不到尽头"：投入更多算力就能买到更多智能，而随着模型越来越强，可被 AI 处理的工作范围（从客服一路扩展到增强甚至替代软件工程）也在不断变宽。他还认为，哪怕效率提升 10 倍，也只是意味着所有人都在用同一池固定算力去处理多 10 倍的 token。

他对底层物理讲得格外具体，把整条链路从光子和天然气，到每秒焦耳，到 FLOPS，再到 token 一路拆开，并指出 NVIDIA 真正的护城河是 cuDNN 和 NCCL 这套软件栈，而不只是 CUDA。从一家人脸识别创业公司、一顶帽檐里嵌着摄像头的棒球帽，到一门近十亿美元的云生意，他始终贯穿着一句话："我们已经有了一套了不起的系统，能把钱放进去、把软件吐出来。"

Link: https://www.youtube.com/watch?v=0NttU4CbyVs