AI Builders Digest - 2026-06-24
X / TWITTER
Swyx
Swyx highlighted a strategically important angle on the AI infrastructure market: the emerging combination of model lab plus NeoCloud. His point is that SpaceX has already recovered roughly half of its Cursor investment through compute deals, and could recover the rest if Composer 3 performs well. The deeper signal: GPU supply is becoming a go-to-market weapon, not just an infrastructure cost.
Swyx 提醒了一个 AI 基础设施市场里的关键结构变化：model lab + NeoCloud 的组合正在变成竞争优势。他认为 SpaceX 已经通过 compute deals 收回了 Cursor 投资的大约一半，如果 Composer 3 表现好，另一半也可能被覆盖。更深层的信号是：GPU supply 不只是成本项，正在变成 go-to-market 武器。
Source: https://x.com/swyx/status/2069301071965741388
Thibault Sottiaux
OpenAI's Thibault Sottiaux amplified the "Patch The Planet" security push around Codex Security and GPT-5.5-Cyber. The notable builder signal is that AI security is moving from detection toward automated repair: finding vulnerabilities is no longer enough, the product direction is to close the loop and patch them.
OpenAI 的 Thibault Sottiaux 推动了围绕 Codex Security 和 GPT-5.5-Cyber 的 "Patch The Planet" 安全叙事。这里真正值得关注的 builder 信号是：AI security 正从发现问题走向自动修复。仅仅发现漏洞已经不够，产品方向正在变成闭环 patch。
Sources:
https://x.com/thsottiaux/status/2069152290326630518
https://x.com/thsottiaux/status/2069267387464274366
Peter Yang
Peter Yang is still looking for the practical edge of coding agents. He asked who is best at using Codex or Claude Code to make pixel or Three.js games, and separately said he still does not understand when to use "dynamic workflows" in Claude Code. That is a useful market signal: even sophisticated AI product people need clearer examples, patterns, and teaching around agent workflows.
Peter Yang 仍在寻找 coding agents 的实操边界。他想找擅长用 Codex / Claude Code 做 pixel 或 Three.js games 的人做一期播客，也坦言自己还没理解 Claude Code 里的 "dynamic workflows" 到底适合什么时候用。这个信号很有价值：即使是成熟的 AI 产品人，也需要更清晰的案例、模式和教学来理解 agent workflows。
Sources:
https://x.com/petergyang/status/2069118077313425840
https://x.com/petergyang/status/2069267139576693028
Guillermo Rauch
Vercel CEO Guillermo Rauch announced two product-facing updates: Claude Design can deploy to Vercel in one click, and Vercel now supports WebSocket and socket.io from CDN to Fluid. The combined signal is clear: Vercel wants to be the default deployment surface for AI-generated interfaces and more stateful real-time applications.
Vercel CEO Guillermo Rauch 发布了两个偏产品化的更新：Claude Design 可以一键部署到 Vercel，Vercel 也开始支持从 CDN 到 Fluid 的 WebSocket 和 socket.io。合起来看，Vercel 想成为 AI 生成界面和更强实时交互应用的默认部署层。
Sources:
https://x.com/rauchg/status/2069219190834127276
https://x.com/rauchg/status/2069109057433035171
Aaron Levie
Box CEO Aaron Levie argued that almost all meaningful AI model and agent progress is downstream from evals. His strongest claim: enterprise automation will depend on whether companies can understand their own workflows and measure how well agents participate in them. Box also added support for previewing, editing, versioning, and securely sharing HTML content, a timely move as agents increasingly produce HTML artifacts.
Box CEO Aaron Levie 的核心判断是：几乎所有 AI model 和 agent 的实质进步都取决于 evals。他最重要的观点是，企业自动化能不能真正落地，取决于企业是否理解自己的 workflow，并能衡量 agents 在其中的表现。Box 也新增了 HTML 内容的预览、编辑、版本管理和安全分享能力，这正好对应 agents 越来越多产出 HTML artifacts 的趋势。
Sources:
https://x.com/levie/status/2069228335255949775
https://x.com/levie/status/2069140445205348432
Ryo Lu
Ryo Lu shared his Cursor Compile talk on building in the age of AI and what does not change. The interesting framing is not just "AI changes how we build", but the inverse: the durable parts of taste, craft, judgment, and product thinking may become more important as raw implementation gets cheaper.
Ryo Lu 分享了他在 Cursor Compile 的演讲，主题是 AI 时代如何 build，以及哪些东西不会变。这里有意思的不只是 "AI 改变了开发方式"，而是反过来：当实现成本下降，taste、craft、judgment 和 product thinking 这些不会被轻易自动化的能力会变得更重要。
Source: https://x.com/ryolu_/status/2069218497272717661
Zara Zhang
Zara Zhang published a walkthrough of her Frontend Slides skill, a GitHub project with more than 22k stars. The tutorial covers creating HTML slides with Claude Code, how the skill was built, how others can make their own skills, and practical details like inserting media and publishing. This is part of a broader pattern: "skills" are becoming a reusable packaging layer for agent workflows.
Zara Zhang 发布了她的 Frontend Slides skill 完整 walkthrough，这个 GitHub 项目已经超过 22k stars。教程覆盖如何用 Claude Code 生成 HTML slides、她如何创建这个 skill、别人如何创建自己的 skill，以及插入图片/视频和发布等实操细节。这代表一个更大的趋势：skills 正在成为 agent workflows 的可复用封装层。
Source: https://x.com/zarazhangrui/status/2069311440692072481
Peter Steinberger
Peter Steinberger echoed the "Patch the Planet" security direction. Paired with OpenAI's GPT-5.5-Cyber messaging, this reinforces that AI coding agents are being positioned not only as builders of new software, but as maintainers and repair systems for existing codebases.
Peter Steinberger 也转发强化了 "Patch the Planet" 的安全方向。结合 OpenAI 的 GPT-5.5-Cyber 叙事，可以看到 AI coding agents 不再只是用来写新软件，而是在被定位成现有代码库的维护和修复系统。
Source: https://x.com/steipete/status/2069132838356840857
Sam Altman
OpenAI CEO Sam Altman framed GPT-5.5-Cyber, Patch The Planet, and Codex Security as part of a broader effort to help companies become secure in collaboration with the U.S. government and the security ecosystem. The product direction is explicit: AI should solve security problems, not merely report them.
OpenAI CEO Sam Altman 将 GPT-5.5-Cyber、Patch The Planet 和 Codex Security 放在一个更大的安全框架里：与美国政府和安全生态协作，帮助企业提升安全能力。产品方向非常明确：AI 不应只是报告安全问题，而应真正解决安全问题。
Source: https://x.com/sama/status/2069121360744550796
PODCASTS
AI & I by Every - How Anthropic Uses Claude Fable 5 With Mike Krieger
The Takeaway: Mike Krieger's real lesson from using Claude Fable 5 is that frontier models are becoming less like autocomplete and more like teammates that can carry long-horizon work.
Mike Krieger, head of Anthropic Labs and cofounder of Instagram, described a shift in how he works with advanced models. The change is not just better answers. It is a different operating rhythm: more architectural planning upfront, clearer intent setting, and then delegation of substantial work across hours or overnight. He says the model can now get stuck, scaffold around missing dependencies, document the workaround, and continue when services return. That changes the human role from constant micromanagement to setting direction, reviewing results, and maintaining taste and verification.
The counterintuitive point is that stronger models may require more skill from users, not less. Krieger says he felt like a beginner again because his old prompting and task decomposition habits were out of date. He now spends more time aligning on architecture, asking for shareable HTML or markdown plans, and deciding when to use long-running sessions versus fast, high-context interaction loops.
For builders, this is the important lesson: the productivity gain does not come from asking the model to do the same small tasks faster. It comes from redesigning the workflow around delegation, review, and parallel execution.
核心结论：Mike Krieger 使用 Claude Fable 5 的真正经验是，frontier models 正在从 autocomplete 变成可以承担长周期任务的 teammate。
Mike Krieger 是 Anthropic Labs 负责人，也是 Instagram 联合创始人。他描述的变化不只是模型回答更好，而是工作节奏变了：前期做更多 architecture planning，更清楚地表达 intent，然后把跨数小时甚至 overnight 的复杂任务委托出去。他提到模型现在可以在遇到 remote service 故障时，先搭一个 scaffolded backend，记录 workaround，等服务恢复后再补齐。这意味着人的角色从持续微管理，转向设定方向、review 结果、保持 taste 和 verification。
反直觉的一点是：模型越强，对使用者的技能要求可能越高，而不是更低。Krieger 说自己一开始像新手，因为旧的 prompting 和 task decomposition 习惯已经过时。他现在会花更多时间和模型对齐架构，让模型输出可分享的 HTML 或 markdown plans，并在 long-running sessions 和 fast high-context loops 之间切换。
对 builders 来说，关键不是让模型更快完成原来的小任务，而是重新设计 workflow：delegation、review、parallel execution 才是真正的生产力增量。
Source: https://www.youtube.com/playlist?list=PLuMcoKK9mKgHtW_o9h5sGO2vXrffKHwJL
Generated through the Follow Builders skill: https://github.com/zarazhangrui/follow-builders