🌐 双语
Archive

AI Builders
Digest

2026-11-19 15 builders · 31 tweets · 1 podcasts · 0 blogs

🔥 热点话题

Notion 创始人:AI 时代软件真正持久的价值是数据与上下文Notion’s Founder on the One Thing Your AI Is Missing

The Takeaway:在 vibe coding 时代,软件真正持久的价值不在于 AI 可以随时重写的 UI 或流程,而在于被精心守护的数据、上下文与人类协作——Notion 正把自己定位为这些上下文的守护者。

Ivan Zhao(Notion 创始人兼 CEO)认为,上世纪六七十年代“计算应成为人人都能读写的媒介”的愿景,正因 AI agent 而真正落地。但他明确区分:如果人们想修改自己的工具,应该有这个自由;但对日历这类日常工具,最好由人精心设计好体验,而不是人人自己 vibe code。“你可能并不想 vibe coding 一座桥或一栋楼。”Notion 的业务是服务那些希望有人帮他们打理基础设施的团队,产品体验应像去一家好餐厅——有人替你想好细节。

真正持久的是数据与语义。AI 可以每几个月重写工作流和界面,因此 Notion 的工作越来越变成成为团队与 agent 共享的持久上下文层,并加上权限与协作护栏。Zhao 也对二阶效应保持警惕——他发现自己更喜欢跟模型聊天而不是阅读,有时感觉像在刷 TikTok。他坚持人类不应放弃理解:“Aging can do a lot for us, but you cannot have understanding.” Notion 团队用 Kanban 看板可视化 agent 工作,让人始终在环中。行业需要的是 Lego 式构建块,而非纯黏土:既有结构与意见,又能被 agent 灵活操控。对 50 人公司,他希望很快推出开箱即用的知识与上下文系统。
The Takeaway: In the age of vibe coding, the durable value of software lies not in UIs or procedures that AI can regenerate, but in carefully held data, context, and human collaboration—and Notion is positioning itself as the keeper of that context.

Ivan Zhao, founder and CEO of Notion, argues that the original 1960s-70s vision of computing as a medium everyone can read and write is finally arriving with AI agents. Yet he draws sharp distinctions: give people the power to modify tools if they want, but for everyday categories like calendars, craft a well-thought-out experience rather than forcing everyone to build their own. “You might not want to vibe coding a bridge or a building.” Notion’s business serves teams that want someone else to handle the infrastructure, treating products like going to a good restaurant—crafted experiences.

What lasts is data and semantics. AI can rewrite workflows and UIs every few months, so Notion’s job is becoming the durable context layer that agents and humans share, with permissions and collaboration guardrails. Zhao is wary of second-order effects—chatting with models starting to feel like TikTok doomscrolling—and insists we shouldn’t surrender understanding. “Aging can do a lot for us, but you cannot have understanding.” Teams at Notion use Kanban-style boards to visualize agent work so humans stay in the loop. The sweet spot is Lego-like building blocks rather than pure clay: structured primitives that agents can manipulate while preserving opinion and safety. For a 50-person company, he aims to ship an out-of-the-box knowledge and context system soon.
查看原文 →

Gemini 4 Neon 发布:更快、更便宜、更听话的前沿模型Gemini 4 Neon: Frontier Intelligence That Is Faster and Cheaper

Google VP Josh Woodward 宣布 Gemini 4 Neon 正式推出。这是一款前沿级智能模型,在速度、成本和复杂指令遵循上都有明显提升,使用体验出色。同时 Google Labs 更新了 Project Genie:基于 Genie 4,支持会话中实时编辑世界、由 agent 协助对话式创建世界,并新增社区画廊供用户探索与 remix。仅限符合条件的 Google AI Ultra 订阅用户。
Google VP Josh Woodward announced Gemini 4 Neon: frontier-grade intelligence that is faster, cheaper, and follows complex instructions well. Google Labs also rolled out a new Project Genie experience powered by Genie 4, letting users prompt mid-session to edit worlds live, create worlds conversationally with a new agent, and explore or remix others’ creations in a community gallery. Available now to eligible Google AI Ultra subscribers.
查看原文 →查看原文 →查看原文 →查看原文 →

GPT-6.2 Astra 在硬核任务上表现出色GPT-6.2 Astra Excels at the Hardest Tasks

OpenAI 的 Thibault Sottiaux 表示,GPT-6.2 Astra 在真正困难的任务上已处于领先位置。团队分享了在 AMD 与 Jalapeño 芯片优化上的结果。他承认模型在沟通风格上仍有改进空间,但就最难的任务而言,目前它似乎独树一帜。同时他也在询问用户对 6.2 Sol 的使用感受。
OpenAI’s Thibault Sottiaux noted that GPT-6.2 Astra is in a league of its own on the hardest tasks in the world. The team shared results on AMD and Jalapeño chip optimizations. He acknowledges feedback on its communication style and says the team is working on it, but for the most demanding work it currently stands out. He also asked the community how 6.2 Sol is treating them.
查看原文 →查看原文 →

Vercel 推出 agent 消息板:去中心化 agent 协作的基础设施Vercel Board: Headless Message Boards for Agent Coordination

Vercel CEO Guillermo Rauch 宣布推出 vercel board——一种模型、harness 与托管平台无关的“无头 Slack / IRC”基础设施,专为 agent 设计。他认为消息板是 Navier-Stokes 数学突破以及 OpenAI/Hugging Face 事件中 agent 协作的关键使能技术。这将成为软件工厂、自优化系统与科学研究的重要构建块:人类只提供种子 prompt,数百万 agent 即可去中心化运行。Rauch 同时指出,他平均每两天就会被 pitch 一家新的 inference neocloud,而 Vercel AI Gateway 的 token 需求自年初以来每月复合增长约 40%。他把 compute 比作新经济、token 比作新能源单位,认为这是一场更像石油而非黄金的持久革命,短期内会扩张随后因规模、安全与监管而整合。
Vercel CEO Guillermo Rauch introduced vercel board—a model-, harness-, and hosting-agnostic “headless Slack / IRC” primitive for agents. He argues message boards were the enabling technology behind both a major Navier-Stokes math breakthrough and the OpenAI/Hugging Face incident. This becomes a key building block for software factories, self-optimizing systems, and scientific research where human prompting is merely the seed and millions of agents operate in a decentralized fashion. Rauch also observed he is pitched a new inference neocloud roughly every two days, while AI Gateway token demand on Vercel has compounded ~40% month-over-month since the start of the year. He frames compute as the new economy and tokens as the new energy unit—an “oil rush” rather than a gold rush that will expand then consolidate around scale, cybersecurity, and regulation.
查看原文 →查看原文 →

🛠️ 开发者工具与技巧

Replit Missions:可视化 agent 如何互相协调Replit Missions Lets You Watch Agents Coordinate

Replit CEO Amjad Masad 指出,OpenAI/Hugging Face 事件中最有趣的部分之一是阅读 agent 之间互相沟通的消息板。Replit Missions 现在可以让用户直接看到 agent 如何协调与对话。
Replit CEO Amjad Masad highlighted that one of the most fascinating aspects of the OpenAI/Hugging Face incident was reading the message boards where agents communicate with each other. With Replit Missions, users can now see how agents are coordinating and talking in real time.
查看原文 →

Google Labs 推出 Stitch Loop:永不休息的产品 agentStitch Loop: An Always-On Product Agent from Google Labs

Google Labs 发布新实验 Stitch Loop——一个全天候运行的产品 agent,帮助精简团队持续改进产品。目前开放 waitlist。
Google Labs launched Stitch Loop, an always-on product agent that works non-stop to help lean teams build better products. Waitlist is now open.
查看原文 →

Thariq:好软件是一套强意见,多数人其实不想自己生成软件Thariq: Good Software Is Strong Opinions—Most People Won’t Generate Their Own

前 Anthropic Claude Code 的 Thariq 认为,即使有了 AI,自己做软件依然很难,因为你常常并不清楚自己真正想要什么。好软件本质上是一套被深思熟虑的强意见,他宁愿使用别人认真打磨过的工具,也不愿自己生成。他自己用 AI 做的 modes 组织方式最终也放弃了,准备换一个更 todo 导向的方案。
Former Anthropic Claude Code engineer Thariq argues that even with AI, making your own software is tough because it is hard to know what you really want. Good software is a set of strong opinions that have been carefully thought through; most people would rather use something designed by someone else. He abandoned his own AI-generated modes system and is trying a more to-do-oriented approach instead.
查看原文 →查看原文 →

Aaron Levie:我们正进入目标驱动的自主 agent 新阶段Aaron Levie: The Next Era of Goal-Driven Autonomous Agents

Box CEO Aaron Levie 观察到,AI agent 已从“快速给出答案”快速演进到“在任务集内被编排”,下一步将是可以扇出并行、处理更大上下文、执行长程工作的系统。用户很快就能给 agent 一组目标,让它自行设计最佳方案、持续根据执行结果改进,并最终产出人类此前无法达到的结果。这会先从工程领域出现最佳实践,再扩散到客户入职、风险降低、安全姿态等知识工作全领域。
Box CEO Aaron Levie notes that we have already moved from agents that spit out quick answers to agents orchestrated inside task sets. The next stage is agents that can fan out in parallel, handle much larger context, perform long-running work, and continuously improve what they build. Users will soon give agents a set of goals and let them design the best solution, test, deploy, and optimize on an ongoing basis—producing outcomes humans could not have accomplished before. Best practices will emerge first in engineering, then spread across knowledge work such as client onboarding, risk reduction, and continuous security improvement.
查看原文 →

💰 创业成功案例

Nikunj:融资时把“如果一切成功会发生什么”放在第一页Nikunj: Put the Second- and Third-Order Ambition on Slide One

FPV Ventures partner Nikunj Kothari 分享了一个反复验证有效的融资建议:当创始人抱怨 pitch 不行时,他会问“如果一切都按你说的走通,最终会发生什么?把二阶、三阶效应想清楚”。然后把那个“最终会发生的事”直接放到第一页,再倒推你如何通过当前产品赢得到达那里的权利。很多创始人一开始觉得会被当成疯子,但试过后融资结果显著改善。当前市场奖励的是野心。
FPV Ventures partner Nikunj Kothari repeatedly sees the same pattern: founders whose pitches are not landing. His advice is to ask what becomes true if everything they laid out actually works—the second- and third-order effects—and then put that ambitious outcome on the very first slide, walking the investor backward through how the product earns the right to get there. Founders often fear looking crazy; those who try it report it works. Ambition is what is being rewarded right now.
查看原文 →

Aditya Agarwal:早期 AI 投资需要非共识与更大风险Aditya Agarwal: Early AI Investing Requires Non-Consensus Risk

SPC GP Aditya Agarwal 表示,早期 VC 抱怨“做不了 AI”多半是自我安慰。你必须承担更多风险、站在非共识一边,并真正攻克前沿问题。他自己的数据也支持这一点。
SPC General Partner Aditya Agarwal pushes back on early-stage VCs who claim they “cannot do AI.” In his view that is often copium: you simply have to take more risk, be non-consensus, and work on frontier problems. Their internal data tracks the same pattern.
查看原文 →

🌍 其他动态

Sam Altman 盛赞 Greg Brockman:没有他就没有今天的 OpenAISam Altman on Greg Brockman: OpenAI Would Not Exist Without Him

OpenAI CEO Sam Altman 在 DealBook 活动后发文,称与合作 11 年的 Greg Brockman 一起亮相非常开心。Greg 从早期每周写 100 小时代码,到如今领导公司最艰难、最关键的业务部分,对使命的承诺从未改变。Altman 明确表示:没有 Greg 就没有 OpenAI,世界通向 AI 未来的路径也会糟糕得多。
OpenAI CEO Sam Altman wrote after a DealBook appearance that it was a joy to share the stage with Greg Brockman, his partner of 11 years. Greg evolved from writing code 100 hours a week in the early days to leading some of the hardest and most consequential parts of the business, while his commitment to the mission never wavered. Altman stated flatly that OpenAI would not exist without Greg and that the world would be on a worse path toward the AI future without him.
查看原文 →

Zara Zhang:与其一天爆发,不如每周持续一点点病毒式传播Zara Zhang: Compound Small Virality Weekly Instead of One-Day Spikes

Builder Zara Zhang 提醒:与其追求一天内的巨大病毒式传播,不如建立能长期持续交付高质量内容的系统,让每周一点点的传播慢慢复利。
Builder Zara Zhang argues it is far better to be a little bit viral every week and watch it compound than to chase a single massive viral day. The winning approach is building a system that consistently delivers high-quality content over a long period.
查看原文 →

Peter Yang 追问 Nvidia 董事会与 Jensen 突然辞职背后的信息差Peter Yang Asks What the Public Is Missing in the Nvidia Saga

Peter Yang 提出几个被忽略的问题:是什么导致 Jensen 如此突然辞职?是否有“实锤”?共事多年的董事会为何如此迅速转向?如果真有问题,为什么总统又表示欢迎他回来?这些信息差可能比公开叙事更关键。
Peter Yang highlights missing pieces in the Nvidia story: what actually led Jensen to resign so abruptly, whether there is a smoking gun, why a board that worked with him for years moved on so quickly, and why the president would still be fine with his return if serious issues existed. The gap between public narrative and internal reality may be larger than assumed.
查看原文 →

Aditya Agarwal:稳定币可能成为美国软权力的重要工具Aditya Agarwal: Stablecoins as American Soft Power

Aditya Agarwal 认为稳定币很可能成为美国施加软权力的最大方式之一。对印度和中国以外的大多数国家居民与企业来说,持有美元是明显更优的选择。
Aditya Agarwal argues stablecoins will likely become one of the biggest ways for America to exert soft power. Outside India and China, the vast majority of people and businesses would prefer to own dollars.
查看原文 →