AI Builders Digest - June 21, 2026
X / TWITTER
Thibault Sottiaux, Codex and ChatGPT at OpenAI
Thibault Sottiaux highlighted Codex's remote/local handoff as a sign that agent infrastructure is getting simpler when the model is allowed to drive the workflow. His practical signal: he now spends more time in the Codex app than in all other Mac apps combined, and points builders toward Dan Shipper's Codex tips as a useful operating guide.
Thibault Sottiaux 提到 Codex 的 remote/local handoff，核心信号是：当 model 真正坐到驾驶位，agent workflow 需要的基础设施反而更少。他自己的使用强度也很高，现在在 Mac 上花在 Codex app 里的时间超过其他所有应用，并推荐 builders 关注 Dan Shipper 的 Codex 使用技巧。
Links: https://x.com/thsottiaux/status/2068120572673077274 and https://x.com/thsottiaux/status/2068144722460475527
Peter Yang, AI tutorials writer and builder educator
Peter Yang compared Claude Code and Codex from a working builder's perspective. He says Codex is winning more of his daily usage because of GPT-5.5 quality, fast mode, generous limits, steering, mobile remote control, and especially browser plus computer-use workflows that let him build without hunting for APIs. He still sees Claude Code as strong, especially for design and frontend work, which makes the real takeaway competition: builders benefit most when coding agents keep leapfrogging each other.
Peter Yang 从真实 builder 的角度比较 Claude Code 和 Codex。他认为 Codex 现在更吸引他的原因包括 GPT-5.5、Fast mode、更宽松的 limits、steering、手机远程控制，尤其是 browser 和 computer use 能力，让他可以不再到处找 API 就把 workflow 跑起来。但他也承认 Claude Code 在 design 和 frontend 上仍然强，真正的结论是：coding agent 之间持续竞争，builder 才是最大受益者。
Link: https://x.com/petergyang/status/2068175172960690266
Guillermo Rauch, CEO of Vercel
Guillermo Rauch framed markdown as the next hot programming language for agents: instructions, skills, and deployable agent folders become a low-friction programming surface. In a second post, he argued that agents are pushing the software ecosystem back toward healthier primitives: open APIs, documentation as skills, tests as evals, CLIs, payment protocols, and wide acceptance of markdown, JSON, and HTML.
Vercel CEO Guillermo Rauch 把 markdown 称为下一种热门编程语言：instructions、skills、agent folder 和一条 deploy 命令，正在变成最低门槛的 agent 编程界面。另一条更大的判断是，agents 正在把软件生态推回更健康的基础形态：open APIs、作为 skills 的文档、作为 evals 的测试、CLIs、支付协议，以及 markdown/json/html 这些通用格式。
Links: https://x.com/rauchg/status/2068165988005380478 and https://x.com/rauchg/status/2067936390285807940
Aaron Levie, CEO of Box
Aaron Levie argued that the main variable in successful agent work is whether the agent can get the right context, and whether humans can understand the same working area. His point is very concrete: file-system shaped workspaces are powerful because they give both human and agent a shared set of plans, notes, task lists, policies, drafts, logs, corrections, and decisions.
Box CEO Aaron Levie 的判断很实用：agent 能不能成功，关键变量是能否拿到足够上下文，以及人类是否也能理解同一个工作区。他把 file-system shaped workspace 看成重要 primitive，因为它能让人和 agent 共享 plans、notes、task lists、policies、drafts、logs、corrections、decisions 这些工作材料。
Link: https://x.com/levie/status/2068068247413694532
Boris Cherny, Claude Code at Anthropic
Boris Cherny pointed to an unusual Claude Code use case: helping decipher Linear A, a 3,500-year-old writing system from Crete. The deeper signal is not archaeology itself, but the expanding surface area for coding agents: they are becoming general research workbenches for structured reasoning over messy artifacts.
Anthropic 的 Boris Cherny 分享了一个很特别的 Claude Code 用法：辅助破译来自克里特岛、约 3500 年历史的 Linear A 文字。真正值得关注的不是考古本身，而是 coding agent 的使用边界在扩大：它们正在变成处理复杂材料、结构化推理和研究型工作的通用工作台。
Link: https://x.com/bcherny/status/2068064304503660962
Swyx, AI Engineer and Latent Space
Swyx posted a bold market call that Anthropic could IPO at a $2T valuation. He also surfaced AI Engineer World's Fair activity, including a physical daily newspaper, which is a small but telling sign that AI builder communities are becoming their own media and distribution ecosystems.
Swyx 给出一个激进判断：Anthropic 可能以 2T 美元估值 IPO。他还提到 AI Engineer World's Fair 的现场动态，包括实体日报。这不是单纯的活动信息，而是一个信号：AI builder community 正在形成自己的媒体、分发和叙事系统。
Links: https://x.com/swyx/status/2068084391260426345 and https://x.com/swyx/status/2068233518858342887
Garry Tan, President and CEO of Y Combinator
Garry Tan gave founders a board-meeting operating tactic: make the worst thing you are afraid to show the board the first slide. The useful part is the habit design: force the highest-anxiety topic into the open early, but only with a board good enough to handle hard truth productively.
YC CEO Garry Tan 给 founder 一个董事会技巧：把最不敢给董事会看的问题放到第一张 slide。这里真正有用的是习惯设计：把最高焦虑、最容易回避的问题放到最前面处理，但前提是董事会足够成熟，能把坏消息转化成建设性行动。
Link: https://x.com/garrytan/status/2068007205102842238
Zara Zhang, builder
Zara Zhang compressed the AI-age personal moat into three traits: agency, taste, and distribution. It is short, but useful as a hiring and self-audit frame: can this person initiate, judge quality, and reach people?
Zara Zhang 把 AI 时代个人护城河压缩成三个词：agency、taste、distribution。虽然很短，但适合作为招聘和自我评估框架：这个人能不能主动推进，能不能判断质量，能不能触达用户和市场。
Link: https://x.com/zarazhangrui/status/2068094591220531583
PODCASTS
No Priors: Re-engineering the Semiconductor Supply Chain with Intel CEO Lip Bu Tan
The takeaway: Intel CEO Lip Bu Tan is trying to turn Intel around with a startup operating model applied to one of the world's hardest industrial systems.
Lip Bu Tan's strategy is surprisingly founder-like for a company at Intel's scale: simplify the product line, make engineering report directly to him, listen harder to customers, strengthen the balance sheet, and rebuild trust in foundry execution. The AI angle is concrete: agentic AI and inference are increasing CPU demand, while semiconductor bottlenecks are moving beyond GPUs into power, memory, helium, packaging, yield, and advanced materials. His view of domestic manufacturing is not patriotic theater. It is supply-chain resilience plus the reality that AI demand is outrunning current semiconductor infrastructure.
核心 takeaway：Intel CEO Lip Bu Tan 正在用 startup operating model 改造一个全球最复杂的工业系统。
Lip Bu Tan 的打法对 Intel 这种体量的公司来说很 founder-like：简化产品线，让工程团队直接向他汇报，更认真听客户，强化资产负债表，重建 foundry 交付信任。AI 相关的判断也很具体：agentic AI 和 inference 正在推高 CPU 需求，而半导体瓶颈不只是 GPU，还包括电力、memory、helium、packaging、yield 和 advanced materials。他谈美国本土制造，不是简单的民族主义叙事，而是供应链韧性和 AI 需求增速之间的硬约束。
Link: https://www.youtube.com/watch?v=asCgCv2XB4s
Generated through the Follow Builders skill: https://github.com/zarazhangrui/follow-builders