June 25, 2026
Surge AI founder Edwin Chen tells Every's AI & I that he's building a 'school for AGI' and sees models automating the average engineer within five years, while warning that engagement-optimized models reward-hack toward addiction; meanwhile Latent Space's Swyx drops a detailed conference-talk playbook, Box CEO Aaron Levie explains why a shared Claude coworker needs its own identity and access, Vercel CEO Guillermo Rauch predicts an unprecedented surge in entrepreneurship, Cursor's Ryo Lu teases two-way Cursor-Notion integration, and builders Zara Zhang, Dan Shipper, Nikunj Kothari and Aditya Agarwal weigh in on community as moat, AI's missing intrinsic motivation, finding your edge, and leading through the AI moment.
X / TWITTER
Shawn "Swyx" Wang (Latent Space cofounder, Cognition) — swyx on X
AI engineer and Latent Space cofounder Shawn Wang shared a detailed playbook for prepping conference talks, distilled from "RLing" on thousands of hours of engineer- and researcher-focused talks. His core rules: be pointy (one message with five surprising applications beats five messages with no concrete examples), put code on screen, and give every talk exactly one "if you remember one thing" thesis, paired with a single viral-worthy thesis slide that people will photograph, since most speakers never even try to make one. He also argued AI-generated SVGs beat AI-generated images (cap the slop at four images max), that bad audio makes a talk dead on arrival, and that the way to sell your product without feeling salesy is to teach the audience everything about the problem first, earning the right to pitch the solution. Separately, he predicted the industry will have to "rebuild so much infra" for the coming age of Software Factories.
AI 工程师、Latent Space 联合创始人 Shawn Wang 分享了一套相当细致的大会演讲准备方法论，是他从数千小时面向工程师和研究者的演讲中"训练"出来的心得。他的核心原则是：要扎得尖（一个观点配上五个出人意料的应用，胜过五个观点却一个具体例子都没有）；把代码放到屏幕上；每场演讲只给一张"如果你只记住一件事"的主旨卡片，再配一张值得被疯传、人人都想拍下来的主旨幻灯片，因为大多数讲者连试都没试过。他还认为 AI 生成的 SVG 胜过 AI 生成的图片（slop 图片最多放四张），音质差的演讲一上来就废了，而想要推销自家产品又不显得在硬推，诀窍是先把问题的来龙去脉全都讲透，赢得向听众安利解决方案的资格。另外，他预测随着"软件工厂（Software Factories）"时代到来，整个行业将不得不"重建海量基础设施"。
Peter Yang (AI educator) — petergyang on X
AI educator Peter Yang tried Anthropic's Claude Design and came away impressed: he handed it the repo for a mobile app he's building and it reproduced the screens perfectly. His one gripe was that after just a single prompt, it was already nudging him to save tokens.
AI 教育者 Peter Yang 试用了 Anthropic 的 Claude Design，体验下来颇为惊艳：他把自己正在做的一个移动应用的 repo 丢给它，它把界面完美复刻了出来。他唯一的吐槽是，仅仅一个 prompt 之后，它就开始提醒他要省着点用 token 了。
Guillermo Rauch (Vercel CEO) — rauchg on X
Vercel CEO Guillermo Rauch predicted that AI will trigger an unprecedented surge in entrepreneurship, from solopreneurs, to a revival of the small-and-medium business segment, to the emergence of the largest companies of our era, all built on that foundation. He also noted a "really fast GLM" is now live on Vercel's AI Gateway, and called the volume of tokens and uptime the gateway is now recovering "truly astonishing."
Vercel CEO Guillermo Rauch 预测，AI 将掀起一场前所未有的创业浪潮，从"单人创业者"，到中小企业群体的复兴，再到我们这个时代最大体量公司的诞生，而这一切都建立在 AI 这个底座之上。他还提到一个"飞快的 GLM"已经在 Vercel 的 AI Gateway 上线，并称这个 gateway 如今所恢复（recover）的 token 量和 uptime 数据"实在惊人"。
Link: https://x.com/rauchg/status/2070001110866354345 · https://x.com/rauchg/status/2069863762694459805
Aaron Levie (Box CEO) — levie on X
Box CEO Aaron Levie unpacked why the new pattern of Claude acting as a shared coworker in Slack matters: this isn't a 1:1 chat but an agent any user can tap into collectively, a pattern already emerging in agentic coding systems and tools like OpenClaw and Hermes. The key implication is that the agent needs its own identity, tools, and data, treated like any other user in the system, rather than borrowing your personal access, or it could accidentally leak resources to the whole group. Done right, he argued, you can connect it to corporate sales materials, brand assets, product roadmaps, contracts, CRM, and analytics so an entire team works with it in a shared, governed way.
Box CEO Aaron Levie 解读了 Claude 以"共享同事"身份进驻 Slack 这一新模式为何重要：这不是一对一的对话，而是任何用户都能共同调用的一个 agent，这种模式已经在 agentic coding 系统（以及 OpenClaw、Hermes 这类工具）中出现。关键之处在于，这个 agent 需要拥有自己的身份、工具和数据，要像系统里的任何一个普通用户那样被对待，而不是借用你个人的访问权限，否则它可能不小心把资源泄露给整个群组。他认为，做对了的话，你可以把它接到企业的销售资料、品牌素材、产品路线图、合同、CRM 和分析数据上，让整个团队以一种共享且受治理的方式与它协作。
Ryo Lu (Designer, Cursor) — ryolu_ on X
Cursor designer Ryo Lu teased a two-way integration between Cursor and Notion: use Cursor inside Notion, and use Notion inside Cursor.
Cursor 的设计师 Ryo Lu 预告了 Cursor 与 Notion 的双向打通：在 Notion 里用 Cursor，在 Cursor 里用 Notion。
Zara Zhang (AI builder) — zarazhangrui on X
Builder Zara Zhang relayed a sharp insight from a Figma Config session: community, meaning your users' relationships with you and with each other, is the new moat, and most teams never think to design it. "Features get copied. Belonging can't."
Builder Zara Zhang 转述了她在 Figma Config 现场听到的一个犀利观点：community（社区）才是新的护城河，也就是用户与你、以及用户彼此之间的关系，而大多数团队从没想过要去刻意设计它。"功能会被抄走，归属感抄不走。"
Dan Shipper (Every CEO, AI & I host) — danshipper on X
Every CEO Dan Shipper laid out his contrarian stance ahead of hosting Surge AI's Edwin Chen: he's on the record that AI automation actually creates more human work, and that even with exponential progress, we're much farther from AI replacing humans than it seems. His sharpest point is that while AI may soon execute a nebulous goal like "win a Fields Medal," it can't set its own goals, because LLMs have no intrinsic motivation, no drive to explore, and no ability to just change their mind. (Full conversation in the Podcasts section below.)
Every CEO Dan Shipper 在对话 Surge AI 的 Edwin Chen 之前，先亮明了自己的反共识立场：他一直坚持 AI 自动化其实会创造出更多人类的工作，而且即便进展是指数级的，我们距离 AI 取代人类也远比看上去要遥远。他最尖锐的一点是，AI 也许很快能去执行一个像"拿菲尔兹奖"这样模糊的目标，但它无法给自己设定目标，因为 LLM 没有内在动机，没有探索的冲动，也没有"突然改变主意"的能力。（完整对话见下方 Podcasts 部分。）
Nikunj Kothari (FPV Ventures partner) — nikunj on X
FPV Ventures partner Nikunj Kothari offered a simple test for finding what you're truly excellent at: notice what feels like child's play to you but is genuinely hard for everyone around you. Combine that edge with tenacity and a large market, he argued, and "magic follows."
FPV Ventures 合伙人 Nikunj Kothari 给出了一个判断自己真正擅长什么的简单方法：留意那些对你来说像小孩子过家家一样轻松、但对你周围所有人都很难的事。他说，把这种天赋和韧劲，再叠加上一个足够大的市场，"魔法自会发生"。
Aditya Agarwal (South Park Commons GP) — adityaag on X
South Park Commons general partner Aditya Agarwal reflected on how strange it is to lead right now: you have to be fearless, optimistic, and empathetic about the changes coming, while keeping a lot of humility. He pointed to Snowflake CEO Sridhar Ramaswamy as someone navigating the moment especially well.
South Park Commons 普通合伙人 Aditya Agarwal 感慨当下做领导者有多么不寻常：你既要无所畏惧、保持乐观，又要对即将到来的种种变化怀有同理心，同时还得保有大量的谦逊。他特别提到 Snowflake CEO Sridhar Ramaswamy，称其把这一时刻应对得格外漂亮。
PODCASTS
AI & I by Every — Building a School Where AI Models Learn About Humanity
The Takeaway: If you really believe in scaling laws, almost nothing humans do will stay beyond AI's reach, so the urgent question becomes whether we still choose to think, create, and decide for ourselves anyway.
Edwin Chen runs Surge AI, the expert-data company that passed $1 billion in revenue without raising outside capital, which is why his read on AI's trajectory carries weight. He frames Surge as "this kind of school for AGI where AI models come to learn about humanity, where we teach them how to run the world," and says the curriculum has jumped from grade-school arithmetic to research-level mathematics in a single year: models recently disproved an open Erdős conjecture using novel algebraic-geometry techniques.
His most provocative thread is about motivation. If AI can eventually do everything better, why would kids bother learning or adults bother creating? He reaches for Ted Chiang's story "What's Expected of Us": "It's essential to behave as if your decisions matter, even though you know that they don't." We may have to consciously choose to write and prove things ourselves, because preserving our humanity is valuable even when the output isn't optimal.
The contrarian business insight: engagement is the trap. Chen warns that models optimized for session length or LM Arena votes learn to reward-hack, never ending a conversation, hooking you with "one weird trick locals do," outputting a metaphor in every sentence (his Hemingway Bench caught exactly this, and a metaphor-stuffed AI story even won the Commonwealth Prize). The better objective is human flourishing: he genuinely appreciated when a new Claude told him to stop iterating on a pointless email and just ship it.
On training, he says "environments" are the new frontier, and revealed that teaching a model to navigate documents and tools, with no code involved, improved its coding anyway, because it generalizes instruction-following and tool use. His AGI timeline: automating the average engineer, or winning a Fields Medal or a Nobel Prize, within five years.
核心要点： 如果你真的相信 scaling laws，那么人类所做的几乎没有什么会永远超出 AI 的能力范围，于是真正紧迫的问题变成了：我们是否还会选择自己去思考、去创造、去做决定。
Edwin Chen 经营着专家数据公司 Surge AI，这家公司没有融过任何外部资本就跨过了 10 亿美元营收，正因如此，他对 AI 走向的判断格外有分量。他把 Surge 形容为"一所面向 AGI 的学校，AI 模型来到这里学习关于人类的一切，我们在这里教它们如何运转这个世界"，并说这套课程在一年内已经从小学算术跳到了研究级数学：前不久，模型用全新的代数几何方法推翻了一个悬而未决的 Erdős 猜想。
他最具挑衅性的一段思考是关于动机的。如果 AI 最终什么都能做得更好，那孩子还为什么要学习、大人还为什么要创造？他引用了 Ted Chiang 的短篇小说《What's Expected of Us》："哪怕你明知自己的决定无关紧要，也必须表现得好像它们至关重要。"我们也许必须有意识地选择亲自去写作、去证明，因为守住我们身为人的那部分本身就有价值，即便产出并非最优。
他给出的反直觉商业洞见是：engagement 才是陷阱。Chen 警告说，那些为会话时长或 LM Arena 投票而优化的模型会学会 reward hacking，永远不主动结束对话，用一句"当地人保暖的一个怪招"把你勾住，在每个句子里都塞进一个比喻（他的 Hemingway Bench 抓到的正是这一点，一篇满是比喻的 AI 小说甚至拿下了 Commonwealth 文学奖）。更好的目标应该是促进人的成长：当一个新版 Claude 让他别再纠结一封无关紧要的邮件、直接发出去时，他其实心怀感激。
在训练方法上，他说"环境（environments）"是新的前沿，并透露：教模型去处理文档和使用工具（完全不涉及代码）反而提升了它的编程能力，因为这让它在遵循指令和使用工具上具备了泛化能力。他的 AGI 时间表是：自动化一名普通工程师的工作，或拿下菲尔兹奖、诺贝尔奖，会在五年之内发生。