七月十二日

Shawn Wang argues that most people have only met Jevons paradox in the narrow context of software demand under agentic engineering, and haven't internalized how far it reaches. Once you have humans who can wield coding agents well plus agents that "break containment" into every other kind of knowledge work, the unit cost of knowledge work falls broadly and total demand for work and for better knowledge rises rather than shrinks. His framing: what happened to coding is not the exception, it's the herald of what's coming for all knowledge work. https://x.com/swyx/status/2076155833428431012

Latent Space 联合创始人、AI Engineer 大会组织者 Shawn Wang（Swyx）认为，大多数人只在 agentic engineering 带来的软件需求这个狭窄语境里接触过 Jevons paradox，却没真正理解它的影响范围有多广。当一批人能熟练驾驭 coding agents，而这些 agent 又"突破边界"渗透进其他各类知识工作时，知识工作的单位成本会全面下降，对工作总量和更优质知识的需求反而会上升，而不是萎缩。他的判断是：coding 领域发生的事情并非特例，而是所有知识工作即将迎来变化的前兆。https://x.com/swyx/status/2076155833428431012

Thibault Sottiaux positions the new GPT 5.6 Sol as OpenAI's answer to the recurring complaints about speed, code slop, and weak frontend output — calling it fast and token-efficient, hardcore at backend work, strong on frontend, and notably "does not use useEffect everywhere." For people not yet ready to install the Codex app, he shared a five-minute workaround (crediting Theo) to point Claude Code at GPT 5.6 Sol using CLIProxyAPI and a `claudex` shell alias that sets the model plus a few env flags. https://x.com/thsottiaux/status/2076145711922696371

OpenAI 负责 Codex 与 ChatGPT 的 Thibault Sottiaux 把全新的 GPT 5.6 Sol 定位为 OpenAI 对速度慢、代码质量差、前端能力弱这些长期抱怨的回应——他称它速度快、token 高效、后端能力硬核、前端表现出色，还特意点出它"不会到处乱用 useEffect"。对于还不想装 Codex app 的人，他分享了一个五分钟的变通方案（并把功劳给了 Theo）：通过 CLIProxyAPI 让 Claude Code 指向 GPT 5.6 Sol，再用一个 `claudex` shell alias 设定模型和几个环境变量即可。https://x.com/thsottiaux/status/2076145711922696371

Aaron Levie makes the counterintuitive case that the job AI was supposed to replace is doing the opposite — software job postings are now outpacing other fields. His logic is classic Jevons paradox applied to knowledge work: when you lower the cost of producing something with many use cases, people want far more of it, so companies light up more software projects than ever. But because the job itself isn't fully automated (and likely won't be for as far out as we can see), you still need people to maintain the code, decide what to build, and run it long-term. His conclusion: "Agents will cause more abundance than replacement." https://x.com/levie/status/2076116544980214164

Box CEO Aaron Levie 提出了一个反直觉的观点：AI 本应取代的那份工作，现实却恰恰相反——软件类岗位的招聘数量如今正超过其他领域。他的逻辑正是把 Jevons paradox 套用到知识工作上：当你降低某个用途广泛的东西的生产成本时，人们会想要更多，于是公司启动的软件项目比以往任何时候都多。但由于这份工作本身并没有被完全自动化（而且在可见的未来大概率也不会），你依然需要人去维护代码、决定要做什么、并长期运行它。他的结论是："Agent 带来的将是更多的富足，而非替代。"https://x.com/levie/status/2076116544980214164

Garry Tan argues that reigniting building in California requires repealing and reforming CEQA (the California Environmental Quality Act). He points out that millions of units of housing are blocked every year using CEQA, and makes the case that housing should be built "by-right" rather than subjected to the level of review it faces today. https://x.com/garrytan/status/2075944103867830352

Y Combinator 总裁兼 CEO Garry Tan 认为，要在加州重新点燃建设的活力，就必须废除并改革 CEQA（加州环境质量法案）。他指出，每年有数百万套住房因为 CEQA 而被阻挡，并主张住房应当"依权建设"（by-right），而不是像今天这样承受如此高强度的审查。https://x.com/garrytan/status/2075944103867830352

Sam Altman claims that a number of benchmarks now suggest GPT 5.6 Sol is the best model in the world — while joking that the most reliable signal isn't the benchmarks at all, but that "elon is obsessed with me again." https://x.com/sama/status/2075983427019612242

OpenAI CEO Sam Altman 声称，多项 benchmark 显示 GPT 5.6 Sol 已是目前全球最强的模型——同时他开玩笑说，最可靠的信号其实不是那些 benchmark，而是"Elon 又开始盯着我了"。https://x.com/sama/status/2075983427019612242

The Takeaway: In the AI era there is no such thing as a moat — durable advantage comes only from relentlessly building better services for both sides of your marketplace.

Glenn Fogel is CEO of Booking Holdings, parent of Priceline and Booking.com. He joined in 2000, months after Priceline's IPO, just as its market cap collapsed from roughly $30 billion to a couple hundred million and the stock flirted with delisting near a dollar a share. Twenty-seven years later it's a $130 billion-plus company — and he sees sharp parallels between the late-'90s internet bubble and today's AI wave, including the certainty that many will lose a lot of money even as real value gets built.

His central discipline: "There is no such thing as a moat. There is no such thing as somewhere you're gonna be protected against innovation." Even Booking's scale advantages — 8.6 million alternative-accommodation listings, roughly three-quarters of Airbnb's size in that category and growing faster for five years — buy no permanent safety. The only defense is to keep building.

On AI specifically, Fogel is a pragmatist, not a doomer or an evangelist. Priceline's agentic assistant "Penny" is seeing adoption double every month, lifting conversion and cutting cancellations, though it's still tiny against $186 billion in annual travel — and token economics (which model, at what cost, for what lifetime value) remain unsolved. AI-handled customer service has already cut cost-per-contact about 10% while raising satisfaction, but he insists some customers will always want a human.

His sharpest worry is jobs — not that automation destroys them (it always has) but the speed of the change and the people who can't retrain. His archetype is the 50-year-old truck driver suddenly automated out of a livelihood. His answer is constant upskilling and AI literacy, and an honest public conversation before fear pushes people to reject technology outright. https://www.youtube.com/watch?v=8nj_0wZkbtA

要点：在 AI 时代，护城河并不存在——真正持久的优势只来自不断为你市场两端的用户打造更好的服务。

Glenn Fogel 是 Booking Holdings 的 CEO，旗下拥有 Priceline 和 Booking.com。他在 2000 年、也就是 Priceline 上市几个月后加入，当时公司市值正从约 300 亿美元崩塌到几亿美元，股价一度逼近一美元、面临退市。二十七年后，它已是一家市值超过 1300 亿美元的公司。他看到 90 年代末互联网泡沫与今天这波 AI 浪潮之间存在鲜明的相似之处——包括一个几乎可以确定的结局：即便真正的价值会被创造出来，也会有很多人亏掉大量的钱。

他最核心的信条是："护城河并不存在，你不可能找到一个地方，让自己永远免受创新的冲击。"哪怕是 Booking 的规模优势——860 万套非标住宿房源、在该品类上体量约为 Airbnb 的四分之三、且连续五年增长更快——也换不来永久的安全。唯一的防御，就是不停地建设。

具体到 AI，Fogel 是个务实派，既不是唱衰者也不是布道者。Priceline 的 agentic 助手"Penny"采用率正每月翻倍，提升了转化、降低了取消率，但相对于每年 1860 亿美元的旅行交易额仍然很小——而且 token 经济学（用哪个模型、成本多少、换来多大的用户终身价值）仍未解决。由 AI 处理的客服已经把每次接触的成本降低了约 10%，同时提升了满意度，但他坚持认为，总会有一部分客户想要真人服务。

他最尖锐的担忧是就业——不是自动化会消灭岗位（它一直都在这么做），而是这次变化的速度，以及那些无法再培训转型的人。他常举的例子是一位 50 岁的卡车司机，突然因自动化而失去谋生手段。他给出的答案是持续的技能升级和 AI 素养，以及在恐惧把人们推向彻底排斥技术之前，进行一场诚实的公共对话。https://www.youtube.com/watch?v=8nj_0wZkbtA