返回列表
💬 讨论题 · 💰投资

别把 AI 当末日按钮——现实更像“慢变量”：惯性、杰文斯悖论与再工业化会把剧本拖回人间

AI 当然会冲击就业，但更可能的路径不是两年内的系统性崩溃，而是被制度惯性与物理世界摩擦“摊平”的长期重构：软件需求不降反增，社会会用再工业化与基建把劳动力重新吸进去。

2026-02-23 原文链接 ↗
阅读简报
双语对照
完整翻译
原文
讨论归档

核心观点

  • 低估“机构动能”是所有末日叙事的共同错误 现实世界不是逻辑推演机，监管俘获、流程惯性与变革管理会让“早该消失的职业”（如房地产经纪）顽强得离谱；技术可行≠制度立即迁移。
  • “软件会终结软件工作”忽略了一个尴尬事实：现有软件普遍烂到发指 AI 让复制更容易，但也让“做得更好”更容易；竞争会加剧，但这并不等于劳动需求消失，反而可能因为质量与复杂度的上限被抬高而继续吞噬劳动力。
  • 杰文斯悖论可能是软件行业的真实底层机制 当生产一个功能更便宜，系统往往不会“省下来”，而是把复杂度与功能扩张到 10x/100x，直到新的饱和点出现；因此“需求饱和”更可能是慢过程。
  • 再工业化是最现实的社会缓冲器：高摩擦、可见成果、政治可共识 电池、电机、半导体、化工、基础设施与国防供应链的回补，是一个巨大的就业与投资容器，而且不受“智能奇点”的速度支配。
  • 宏观政策的底线不是会计指标，而是社会契约的延续 作者认为一旦冲击显性化，政府会像 Covid 一样更主动地用刺激/转移支付兜底；效率可能不高，但能把崩溃情景压回“可管理的慢变量”。

跟我们的关联

💰投资：这篇是在提醒你：不要只押“AI 直接摧毁一切”的快剧本，很多收益/风险来自“拖慢后的转向”。再工业化/基建/电力栈（制造、能源、供应链）可能是更政治正确、更能吸纳资本与劳动力的长期主题。

🌍通用：把“技术替代”拆成两条曲线：技术曲线（快）与制度曲线（慢）。真正的机会和痛苦，多数发生在两条曲线错位的区间。

讨论引子

  • 你更相信哪一种速度：AI 技术扩散的速度，还是制度/监管/企业流程迁移的速度？哪一个更决定 3 年内的现实？
  • 如果软件的“最后 5% 完成度”仍然极耗工，AI 是在减少工程师需求，还是在把需求转移到更高层的产品/质量/集成复杂度？
  • 再工业化如果真成了“社会缓冲器”，最大瓶颈会是什么：电力、土地/审批、人才，还是供应链？

驳 Citrini7：更可能、更乐观的未来

市场领域的知名评论员 Citrini7 最近发表了一篇引人入胜、也颇受欢迎的 AI 末日式虚构文章——坦白说，它确实有那么一点点发生的概率。但我见过足够多轮经济唱衰的周期。我想对 Citrini 的作品提出批评，并展示一种更可能、也更积极的未来图景。

  1. 永远不要低估机构动能

2007 年，人们在“石油峰值”的背景下认为美国的地缘政治已经走到尽头。2008 年，他们认为美元离崩溃只差一步。2014 年，他们觉得 AMD 和 NVIDIA 完了。后来 ChatGPT 出现，人们又觉得 Google 完了……每一次，拥有动能的既有机构都证明自己远比旁观者想象的更耐久。

在担心机构更替与劳动力快速被替代时，Citrini 写道：

即便是那些我们以为会被“人际关系价值”所保护的领域，也显得脆弱。房地产就是如此——由于经纪人与消费者之间的信息不对称，买方几十年来一直容忍 5-6% 的佣金……

过去 20 年里，人们一直在喊“房地产经纪人要被终结”！这根本不需要超级智能！你只需要 Zillow、Redfin 或 Opendoor 就够了。这个例子其实恰恰说明了与 Citrini 观点相反的事情：我们有一种大多数人都认为早该过时的劳动，但市场惯性与监管俘获却让房地产经纪人比十年前任何人敢押注的都更顽强。

几个月前，我和妻子买了房。这笔交易要求我们必须有一名经纪人，表面上理由就是上面那些。我们的买方经纪人从这单里拿走了大约 50,000 美元，只做了大概十个小时的填表与多方协调——这些我自己也能完成。这个市场终究会变得更有效率，并为这类劳动给出更公平的定价，但到那一步需要很长时间。我对惯性与变革管理颇有体会：我创办并出售过一家公司，专注于把保险经纪行业从“靠服务”转向“靠软件”。我学到的最重要的一点，是与人类现实打交道的铁律：哪怕你已经知道这条铁律，一切仍然总是比你想象的更复杂，也总是花更久。这并不意味着世界不会发生有意义的变化，而是意味着变化会更渐进，从而给我们留下响应与调整的时间。

  1. 软件对劳动的需求近乎无限

近几个月来，软件行业举步维艰，因为投资者担心 Monday、Salesforce、Asana 等公司的产品如今可以轻易被复制，而它们后台系统的价值也难以自证护城河。Citrini 以及其他人把 AI 编程视为 SaaS 公司岗位的终结，原因在于：(1) 产品会过时/利润归零；(2) 工作岗位本身会消失。

但大家似乎忽略了一个事实：这些产品烂得令人发指。我之所以敢这么说，是因为我确实在 Salesforce 和 Monday 上花过几十万美元。当然，AI 也许会让竞争对手更容易复刻他们的产品。但更重要的是，AI 会让竞争对手更容易交付更好的产品。所以股价下跌并不奇怪：一个缺乏竞争、锁定效应极强、充斥着垃圾般既有巨头的行业，正在重新变得有竞争。

更一般地说，几乎所有现有软件都是垃圾，这一点毫无争议。我用的、付费的每一款产品都布满 bug。有些软件坏到我甚至没法付款。我已经三年无法通过 Citibank 的网银发起电汇。大多数 Web 应用连移动端和桌面端的适配都做不好。没有任何东西具备你真正想要的功能。一切都不够格。像 Stripe 和 Linear 这样的硅谷宠儿，仅仅因为没有像竞争对手那样疯狂地难用、糟糕透顶，就能积累庞大拥趸。你去问一个拿到终身教职的工程师：“给我看一款好软件”，你往往只会得到长久的沉默与茫然的目光。

这里有一个深刻而重要的事实：即便我们真的迎来某种“软件奇点”（Software Singularity），对劳动的需求也几乎是无限的。众所周知，最后几个百分点的完成度最耗工；因此，几乎任何软件产品在需求开始趋于饱和之前，都很可能还能把复杂度和功能扩展到大约 100 倍。

我感觉，那些宣称软件即将走向终结的评论员，未必对“做软件”有多少直觉。软件诞生至今大约五十年。它这些年确实有了显著进步，但始终不够用。作为 2020 年的一名程序员，我能完成在 1970 年需要数百人年才能完成的工作；这种杠杆效应令人惊叹，但产出的结果在每一个环节仍然留下巨大的改进空间。人们低估了杰文斯悖论（Jevons Paradox）。

重要的是，这并不意味着软件工程会永远是一个抗冲击的就业来源。当然不；没有什么是永恒的。但我想说的是：这个行业依然拥有比人们想象中更强的动能与吸纳劳动力的能力，而这种吸纳的饱和会是一个缓慢的过程，从而给我们留下响应与调整的时间。

  1. 再工业化

当然，会有一些劳动力被替代。驾驶是最显眼的例子。正如 Citrini 所说，许多白领工作会经历一番震荡：一些岗位消失，另一些岗位则发生实质性变化。AI 也许会成为压垮骆驼的最后一根稻草，终结像房地产经纪人这类工作——这种工作其实早就“消失”了很久，只是薪酬还在。

这里的救命稻草是：在美国，我们拥有几乎无限的再工业化能力与需求。你或许听说过“制造业回流”，但这不止于此：我们在很大程度上已经不再知道如何制造、也没有相应的设施来生产现代生活的核心基础构件：电池、电机、小型半导体——整个电动化的“电力栈”几乎让我们完全依赖中国和其他国家。要是有一天发生军事对抗怎么办？其实比那更糟：你知道中国生产了全球 90% 的氨吗？一旦战争爆发，我们几乎连化肥都造不出来。我们会直接挨饿。

一旦你开始认真审视物理世界，就会发现：围绕创造就业、造福国家、夯实基础设施的工作范围几乎无穷无尽，而且在政治上也能形成两党共识。

我们已经看到经济与政治环境在慢慢朝这个方向移动——谈论再工业化、制造业、深科技、美国活力，等等。我的预测是：当 AI 冲击白领劳动时，政治上阻力最小的道路将是为大规模再工业化提供资金，以就业型超级工程的形式推进；谢天谢地，这些工程并不受“奇点”支配，而是按照在物理世界里把事情做成所必然经历的高摩擦速度推进。我们会再次修桥。人们会从现实世界中看到自己劳动的成果，而不是数字抽象中的影子。Salesforce 里那位失去 180K 美元年薪的 Senior PM，可能会在加州海水淡化工程（California Desalination Works）的一线找到新工作，终于、终于结束那场持续 25 年的干旱。它不该只是“够用”，而应该做到卓越。而且一旦建成，就必须维护！如果你愿意，杰文斯悖论也可以再次适用。

  1. 以及更远处

工业超级工程带来的结果当然是走向富足：美国将再次实现独立，以大规模、低成本的方式制造各种东西。超越物质匮乏是关键：从长远看，如果我们真的几乎把所有白领工作都输给 AI，我们就必须能够继续为人们提供高质量的生活。其中一部分会自动发生——因为 AI 把利润压到零，意味着那些消费品的价格也会相应变得便宜。

在我看来，经济的不同部分会以不同的速度“起飞”，而且几乎所有领域的速度都比 Citrini 那样的文章所暗示的更慢。说清楚一点：我对 AI 极其看多，也预期总有一天我的劳动也会过时。但走到那一步还需要一段时间，而这段时间给了我们制定良好政策的机会。

在这方面，要避免 Citrini 想象中的那种市场崩盘其实相当容易——联邦政府在 Covid 期间的应对已经表明，它愿意多么主动、多么强硬。一旦需要，我预计大规模刺激会迅速启动。说它不会高效，多少让我有点不爽，但那也不是重点。重点是：让人们在自己的生命历程中获得物质上的繁荣——广泛的消费者福祉，使国家获得正当性，并延续社会契约——而不是去满足过去那套会计指标或经济常规。如果我们能对这场缓慢但确定的技术革命保持敏捷与回应，我们就会没事。

Link: http://x.com/i/article/2025713380510887936

相关笔记

Popular markets commentator Citrini7 recently published a compelling and popular piece of AI doomer fiction, admittedly with some small probability of occurring. But I am old enough to have seen many cycles of economic doomsaying. I want to present a critique of Citrini's work and show a much likelier, more positive view of the future.

市场领域的知名评论员 Citrini7 最近发表了一篇引人入胜、也颇受欢迎的 AI 末日式虚构文章——坦白说，它确实有那么一点点发生的概率。但我见过足够多轮经济唱衰的周期。我想对 Citrini 的作品提出批评，并展示一种更可能、也更积极的未来图景。

  1. Never Underestimate Institutional Momentum
  1. 永远不要低估机构动能

In 2007, people thought the US was geopolitically done under peak oil. In 2008, they thought the US dollar was just shy of collapse. In 2014, they thought AMD and NVIDIA were done. Then came ChatGPT, and they thought Google was done... Every time, existing institutions with momentum have proven themselves far more durable than onlookers thought.

2007 年，人们在“石油峰值”的背景下认为美国的地缘政治已经走到尽头。2008 年，他们认为美元离崩溃只差一步。2014 年，他们觉得 AMD 和 NVIDIA 完了。后来 ChatGPT 出现，人们又觉得 Google 完了……每一次，拥有动能的既有机构都证明自己远比旁观者想象的更耐久。

When worried about institutional turnover and rapid labor replacement, it is very funny that Citrini writes:

在担心机构更替与劳动力快速被替代时，Citrini 写道：

Even places we thought insulated by the value of human relationships proved fragile. Real estate, where buyers had tolerated 5-6% commissions for decades because of information asymmetry between agent and consumer...

即便是那些我们以为会被“人际关系价值”所保护的领域，也显得脆弱。房地产就是如此——由于经纪人与消费者之间的信息不对称，买方几十年来一直容忍 5-6% 的佣金……

People have been calling for the end of the real estate broker for 20 years! You don't need superintelligence for this! All you need is Zillow or Redfin or Opendoor. This example actually shows the very opposite of Citrini's point: we have a type of labor that most people consider obsolete, and yet, market inertia and regulatory capture have made the real estate broker far more resilient than anyone would've bet a decade ago.

过去 20 年里，人们一直在喊“房地产经纪人要被终结”！这根本不需要超级智能！你只需要 Zillow、Redfin 或 Opendoor 就够了。这个例子其实恰恰说明了与 Citrini 观点相反的事情：我们有一种大多数人都认为早该过时的劳动，但市场惯性与监管俘获却让房地产经纪人比十年前任何人敢押注的都更顽强。

My wife and I bought a house a few months back. The transaction required us to have an agent, ostensibly for the above reasons. Our buyer's agent made about $50,000 on the deal, for about ten hours of form-filling and party-coordination that I could've done myself. This market will eventually be efficient and price this labor fairly, but it takes a long time to get there. I know a lot about inertia and change management: I built and sold a company that focused on moving insurance brokerages from service to software, and the main thing I learned is the iron rule of dealing with human reality: everything is always more complicated and takes much longer than you think it will, even if you already know about the iron rule. That doesn't mean that a meaningful change in the world won't happen, but that the change will be more gradual, giving us the time to respond and adjust.

几个月前，我和妻子买了房。这笔交易要求我们必须有一名经纪人，表面上理由就是上面那些。我们的买方经纪人从这单里拿走了大约 50,000 美元，只做了大概十个小时的填表与多方协调——这些我自己也能完成。这个市场终究会变得更有效率，并为这类劳动给出更公平的定价，但到那一步需要很长时间。我对惯性与变革管理颇有体会：我创办并出售过一家公司，专注于把保险经纪行业从“靠服务”转向“靠软件”。我学到的最重要的一点，是与人类现实打交道的铁律：哪怕你已经知道这条铁律，一切仍然总是比你想象的更复杂，也总是花更久。这并不意味着世界不会发生有意义的变化，而是意味着变化会更渐进，从而给我们留下响应与调整的时间。

  1. Software Has Infinite Demand for Labor
  1. 软件对劳动的需求近乎无限

The software sector has been struggling in recent months as investors fear that companies like Monday, Salesforce, Asana, etc. can now be easily replicated and that the value of their backend systems is indefensible. Citrini and others talk of AI coding as spelling the end of jobs at SaaS companies as (1) the products become obsolete/zero-margin and (2) the jobs themselves disappear.

近几个月来，软件行业举步维艰，因为投资者担心 Monday、Salesforce、Asana 等公司的产品如今可以轻易被复制，而它们后台系统的价值也难以自证护城河。Citrini 以及其他人把 AI 编程视为 SaaS 公司岗位的终结，原因在于：(1) 产品会过时/利润归零；(2) 工作岗位本身会消失。

What everyone seems to be missing is this: these products fucking suck. I can say this, because I've actually spent hundreds of thousands of dollars on Salesforce and Monday. Sure, maybe AI enables competition to replicate their products. But more importantly, AI enables competition to deliver better products. It's no surprise to see the stocks drop: an uncompetitive, sticky lock-in sector filled with dogshit incumbents is becoming competitive again.

但大家似乎忽略了一个事实：这些产品烂得令人发指。我之所以敢这么说，是因为我确实在 Salesforce 和 Monday 上花过几十万美元。当然，AI 也许会让竞争对手更容易复刻他们的产品。但更重要的是，AI 会让竞争对手更容易交付更好的产品。所以股价下跌并不奇怪：一个缺乏竞争、锁定效应极强、充斥着垃圾般既有巨头的行业，正在重新变得有竞争。

More generally, it is uncontroversial that virtually all current software is garbage. Everything I use and pay for is littered with bugs. Some software is so broken that I can't even pay for it. I have not been able to send a wire using Citibank's online banking in three years. Most web apps can't even get mobile vs. desktop right. Nothing has the functionality that you want. Everything is deficient. Silicon Valley darlings like Stripe and Linear have built massive followings just by not being as insanely unusable and horrendous as their competitors. Ask tenured engineers "show me a piece of good software" and you'll get long silences and blank stares.

更一般地说，几乎所有现有软件都是垃圾，这一点毫无争议。我用的、付费的每一款产品都布满 bug。有些软件坏到我甚至没法付款。我已经三年无法通过 Citibank 的网银发起电汇。大多数 Web 应用连移动端和桌面端的适配都做不好。没有任何东西具备你真正想要的功能。一切都不够格。像 Stripe 和 Linear 这样的硅谷宠儿，仅仅因为没有像竞争对手那样疯狂地难用、糟糕透顶，就能积累庞大拥趸。你去问一个拿到终身教职的工程师：“给我看一款好软件”，你往往只会得到长久的沉默与茫然的目光。

There is a deep and important truth here: even if we get something like the Software Singularity, the level of demand for labor here is practically infinite. Famously, it is the last few percent of completion that take the most work, and by that token, virtually every software product could probably scale up its complexity and features by something like 100x before beginning to saturate demand.

这里有一个深刻而重要的事实：即便我们真的迎来某种“软件奇点”（Software Singularity），对劳动的需求也几乎是无限的。众所周知，最后几个百分点的完成度最耗工；因此，几乎任何软件产品在需求开始趋于饱和之前，都很可能还能把复杂度和功能扩展到大约 100 倍。

I have the feeling that commentators on the imminent demise of software don't have much intuition for making software. We've had software for about fifty years now. Though it has improved meaningfully over the years, it has always been inadequate. As a programmer in 2020 I was able to do what would've taken hundreds of man-years in 1970; the leverage gained is incredible, but the results still leave massive space for improvement at every step along the way. People underestimate Jevons Paradox.

我感觉，那些宣称软件即将走向终结的评论员，未必对“做软件”有多少直觉。软件诞生至今大约五十年。它这些年确实有了显著进步，但始终不够用。作为 2020 年的一名程序员，我能完成在 1970 年需要数百人年才能完成的工作；这种杠杆效应令人惊叹，但产出的结果在每一个环节仍然留下巨大的改进空间。人们低估了杰文斯悖论（Jevons Paradox）。

Importantly, this does not mean that software engineering is a forever-resilient source of jobs. Of course not; nothing is. But my point is that again, the sector has more momentum and ability to absorb labor than people give it credit for, and saturation of this will be a slow process, giving us time to respond and adjust.

重要的是，这并不意味着软件工程会永远是一个抗冲击的就业来源。当然不；没有什么是永恒的。但我想说的是：这个行业依然拥有比人们想象中更强的动能与吸纳劳动力的能力，而这种吸纳的饱和会是一个缓慢的过程，从而给我们留下响应与调整的时间。

  1. Re-Industrialization
  1. 再工业化

There will be some labor displacement, of course. Driving stands out. Many types of white-collar work, as Citrini suggests, will undergo some gyration as some jobs disappear and others change meaningfully. AI may be the straw that breaks the camel's back for jobs like the real estate broker, where the job had actually already disappeared a long time ago, but the pay was still there.

当然，会有一些劳动力被替代。驾驶是最显眼的例子。正如 Citrini 所说，许多白领工作会经历一番震荡：一些岗位消失，另一些岗位则发生实质性变化。AI 也许会成为压垮骆驼的最后一根稻草，终结像房地产经纪人这类工作——这种工作其实早就“消失”了很久，只是薪酬还在。

The saving grace here is that in the US, we have a virtually limitless capacity and need for re-industrialization. You may have heard about bringing back manufacturing, but it's more than that: we largely no longer know how to create, and don't have the facilities for, making the core building blocks of modern life: batteries, motors, small semiconductors -- the whole electric stack is something we are almost entirely dependent on China and other countries for. What if there's ever a military confrontation? Actually, it's much worse than that: did you know China makes 90% of the world's ammonia? If there's a war, we can barely make fertilizer. We'd just starve.

这里的救命稻草是：在美国，我们拥有几乎无限的再工业化能力与需求。你或许听说过“制造业回流”，但这不止于此：我们在很大程度上已经不再知道如何制造、也没有相应的设施来生产现代生活的核心基础构件：电池、电机、小型半导体——整个电动化的“电力栈”几乎让我们完全依赖中国和其他国家。要是有一天发生军事对抗怎么办？其实比那更糟：你知道中国生产了全球 90% 的氨吗？一旦战争爆发，我们几乎连化肥都造不出来。我们会直接挨饿。

Once you start looking at the physical world, you see a virtually endless scope for work on job-creating, nation-benefiting, fundamental infrastructural work that is politically bipartisan.

一旦你开始认真审视物理世界，就会发现：围绕创造就业、造福国家、夯实基础设施的工作范围几乎无穷无尽，而且在政治上也能形成两党共识。

We've seen the economic and political milieu slowly make their way in this direction -- talking about re-industrializing, manufacturing, deep tech, American dynamism, and so forth. My prediction is that as AI challenges white-collar labor, the political path of least resistance will be funding large-scale re-industrialization in the form of employment megaprojects which, thankfully, are not subject to a singularity but rather move at the friction-heavy speed of getting things done in the physical world. We'll build bridges again. People will find it gratifying to see the fruits of their labor in the real world, not in digital abstractions. The Senior PM at Salesforce that loses their $180K job might find a new job in the field at the California Desalination Works, to finally, finally, end the 25-year drought. And it shouldn't be good enough, but excellent. And once it is built, it must be maintained! Once more, Jevons Paradox can apply, if you allow it to.

我们已经看到经济与政治环境在慢慢朝这个方向移动——谈论再工业化、制造业、深科技、美国活力，等等。我的预测是：当 AI 冲击白领劳动时，政治上阻力最小的道路将是为大规模再工业化提供资金，以就业型超级工程的形式推进；谢天谢地，这些工程并不受“奇点”支配，而是按照在物理世界里把事情做成所必然经历的高摩擦速度推进。我们会再次修桥。人们会从现实世界中看到自己劳动的成果，而不是数字抽象中的影子。Salesforce 里那位失去 180K 美元年薪的 Senior PM，可能会在加州海水淡化工程（California Desalination Works）的一线找到新工作，终于、终于结束那场持续 25 年的干旱。它不该只是“够用”，而应该做到卓越。而且一旦建成，就必须维护！如果你愿意，杰文斯悖论也可以再次适用。

  1. And Beyond
  1. 以及更远处

The outcome of industrial megaprojects is of course that we move toward abundance: America will once more be independent, and make things at large scale and low cost. Transcending material scarcity is the key: in the long run, if we do lose almost all the white-collar jobs to AI, we have to be able to provide people with a continued high quality of life. Part of this we get automatically, just because AI taking margins to zero means that those consumer products will become equivalently cheap.

工业超级工程带来的结果当然是走向富足：美国将再次实现独立，以大规模、低成本的方式制造各种东西。超越物质匮乏是关键：从长远看，如果我们真的几乎把所有白领工作都输给 AI，我们就必须能够继续为人们提供高质量的生活。其中一部分会自动发生——因为 AI 把利润压到零，意味着那些消费品的价格也会相应变得便宜。

My view is that different parts of the economy will "take off" at varying speeds, and virtually all the areas are slower than a piece like Citrini's might suggest. To be clear, I am extremely bullish on AI, and expect that one day, my labor too will be obsolete. But it's going to take a while to get there, and that time gives us the opportunity to make good policy.

在我看来，经济的不同部分会以不同的速度“起飞”，而且几乎所有领域的速度都比 Citrini 那样的文章所暗示的更慢。说清楚一点：我对 AI 极其看多，也预期总有一天我的劳动也会过时。但走到那一步还需要一段时间，而这段时间给了我们制定良好政策的机会。

On that front, preventing a market meltdown the way Citrini imagines is actually pretty easy, and the federal government's response during Covid showed how proactive and aggressive it is willing to be. I'd expect large-scale stimulus to kick in quickly once needed. It slightly irks me to say that it won't be efficient, but that's also not the point. The point is material prosperity for people in the course of their lives -- broad consumer well-being that legitimizes the state and carries forth the social contract -- not satisfying the accounting metrics or economic norms of the past. If we are nimble and responsive to this slow but sure technological revolution, then we will be fine.

在这方面，要避免 Citrini 想象中的那种市场崩盘其实相当容易——联邦政府在 Covid 期间的应对已经表明，它愿意多么主动、多么强硬。一旦需要，我预计大规模刺激会迅速启动。说它不会高效，多少让我有点不爽，但那也不是重点。重点是：让人们在自己的生命历程中获得物质上的繁荣——广泛的消费者福祉，使国家获得正当性，并延续社会契约——而不是去满足过去那套会计指标或经济常规。如果我们能对这场缓慢但确定的技术革命保持敏捷与回应，我们就会没事。

Link: http://x.com/i/article/2025713380510887936

Link: http://x.com/i/article/2025713380510887936

相关笔记

Contra Citrini7

  • Source: https://x.com/johnloeber/status/2025748423157432756?s=46
  • Mirror: https://x.com/johnloeber/status/2025748423157432756?s=46
  • Published: 2026-02-23T01:43:55+00:00
  • Saved: 2026-02-23

Content

Popular markets commentator Citrini7 recently published a compelling and popular piece of AI doomer fiction, admittedly with some small probability of occurring. But I am old enough to have seen many cycles of economic doomsaying. I want to present a critique of Citrini's work and show a much likelier, more positive view of the future.

  1. Never Underestimate Institutional Momentum

In 2007, people thought the US was geopolitically done under peak oil. In 2008, they thought the US dollar was just shy of collapse. In 2014, they thought AMD and NVIDIA were done. Then came ChatGPT, and they thought Google was done... Every time, existing institutions with momentum have proven themselves far more durable than onlookers thought.

When worried about institutional turnover and rapid labor replacement, it is very funny that Citrini writes:

Even places we thought insulated by the value of human relationships proved fragile. Real estate, where buyers had tolerated 5-6% commissions for decades because of information asymmetry between agent and consumer...

People have been calling for the end of the real estate broker for 20 years! You don't need superintelligence for this! All you need is Zillow or Redfin or Opendoor. This example actually shows the very opposite of Citrini's point: we have a type of labor that most people consider obsolete, and yet, market inertia and regulatory capture have made the real estate broker far more resilient than anyone would've bet a decade ago.

My wife and I bought a house a few months back. The transaction required us to have an agent, ostensibly for the above reasons. Our buyer's agent made about $50,000 on the deal, for about ten hours of form-filling and party-coordination that I could've done myself. This market will eventually be efficient and price this labor fairly, but it takes a long time to get there. I know a lot about inertia and change management: I built and sold a company that focused on moving insurance brokerages from service to software, and the main thing I learned is the iron rule of dealing with human reality: everything is always more complicated and takes much longer than you think it will, even if you already know about the iron rule. That doesn't mean that a meaningful change in the world won't happen, but that the change will be more gradual, giving us the time to respond and adjust.

  1. Software Has Infinite Demand for Labor

The software sector has been struggling in recent months as investors fear that companies like Monday, Salesforce, Asana, etc. can now be easily replicated and that the value of their backend systems is indefensible. Citrini and others talk of AI coding as spelling the end of jobs at SaaS companies as (1) the products become obsolete/zero-margin and (2) the jobs themselves disappear.

What everyone seems to be missing is this: these products fucking suck. I can say this, because I've actually spent hundreds of thousands of dollars on Salesforce and Monday. Sure, maybe AI enables competition to replicate their products. But more importantly, AI enables competition to deliver better products. It's no surprise to see the stocks drop: an uncompetitive, sticky lock-in sector filled with dogshit incumbents is becoming competitive again.

More generally, it is uncontroversial that virtually all current software is garbage. Everything I use and pay for is littered with bugs. Some software is so broken that I can't even pay for it. I have not been able to send a wire using Citibank's online banking in three years. Most web apps can't even get mobile vs. desktop right. Nothing has the functionality that you want. Everything is deficient. Silicon Valley darlings like Stripe and Linear have built massive followings just by not being as insanely unusable and horrendous as their competitors. Ask tenured engineers "show me a piece of good software" and you'll get long silences and blank stares.

There is a deep and important truth here: even if we get something like the Software Singularity, the level of demand for labor here is practically infinite. Famously, it is the last few percent of completion that take the most work, and by that token, virtually every software product could probably scale up its complexity and features by something like 100x before beginning to saturate demand.

I have the feeling that commentators on the imminent demise of software don't have much intuition for making software. We've had software for about fifty years now. Though it has improved meaningfully over the years, it has always been inadequate. As a programmer in 2020 I was able to do what would've taken hundreds of man-years in 1970; the leverage gained is incredible, but the results still leave massive space for improvement at every step along the way. People underestimate Jevons Paradox.

Importantly, this does not mean that software engineering is a forever-resilient source of jobs. Of course not; nothing is. But my point is that again, the sector has more momentum and ability to absorb labor than people give it credit for, and saturation of this will be a slow process, giving us time to respond and adjust.

  1. Re-Industrialization

There will be some labor displacement, of course. Driving stands out. Many types of white-collar work, as Citrini suggests, will undergo some gyration as some jobs disappear and others change meaningfully. AI may be the straw that breaks the camel's back for jobs like the real estate broker, where the job had actually already disappeared a long time ago, but the pay was still there.

The saving grace here is that in the US, we have a virtually limitless capacity and need for re-industrialization. You may have heard about bringing back manufacturing, but it's more than that: we largely no longer know how to create, and don't have the facilities for, making the core building blocks of modern life: batteries, motors, small semiconductors -- the whole electric stack is something we are almost entirely dependent on China and other countries for. What if there's ever a military confrontation? Actually, it's much worse than that: did you know China makes 90% of the world's ammonia? If there's a war, we can barely make fertilizer. We'd just starve.

Once you start looking at the physical world, you see a virtually endless scope for work on job-creating, nation-benefiting, fundamental infrastructural work that is politically bipartisan.

We've seen the economic and political milieu slowly make their way in this direction -- talking about re-industrializing, manufacturing, deep tech, American dynamism, and so forth. My prediction is that as AI challenges white-collar labor, the political path of least resistance will be funding large-scale re-industrialization in the form of employment megaprojects which, thankfully, are not subject to a singularity but rather move at the friction-heavy speed of getting things done in the physical world. We'll build bridges again. People will find it gratifying to see the fruits of their labor in the real world, not in digital abstractions. The Senior PM at Salesforce that loses their $180K job might find a new job in the field at the California Desalination Works, to finally, finally, end the 25-year drought. And it shouldn't be good enough, but excellent. And once it is built, it must be maintained! Once more, Jevons Paradox can apply, if you allow it to.

  1. And Beyond

The outcome of industrial megaprojects is of course that we move toward abundance: America will once more be independent, and make things at large scale and low cost. Transcending material scarcity is the key: in the long run, if we do lose almost all the white-collar jobs to AI, we have to be able to provide people with a continued high quality of life. Part of this we get automatically, just because AI taking margins to zero means that those consumer products will become equivalently cheap.

My view is that different parts of the economy will "take off" at varying speeds, and virtually all the areas are slower than a piece like Citrini's might suggest. To be clear, I am extremely bullish on AI, and expect that one day, my labor too will be obsolete. But it's going to take a while to get there, and that time gives us the opportunity to make good policy.

On that front, preventing a market meltdown the way Citrini imagines is actually pretty easy, and the federal government's response during Covid showed how proactive and aggressive it is willing to be. I'd expect large-scale stimulus to kick in quickly once needed. It slightly irks me to say that it won't be efficient, but that's also not the point. The point is material prosperity for people in the course of their lives -- broad consumer well-being that legitimizes the state and carries forth the social contract -- not satisfying the accounting metrics or economic norms of the past. If we are nimble and responsive to this slow but sure technological revolution, then we will be fine.

Link: http://x.com/i/article/2025713380510887936

📋 讨论归档

讨论进行中…