HackerNews中文版

多年来，我安静地专注于我的工作，但现在我感到有必要指出一个日益明显的问题。 1. 正确的模型 ≠ 被采纳的模型历史事实：在前端生态系统中，成功的并不是那些创造出最准确抽象的人，而是那些以最小摩擦提供“工作感觉”的人。结果：正确的思维模型 → 低采纳率，而不正确但易用的模型 → 爆炸性增长。这并非偶然。 2. 为什么基于意图/确定性模型没有成功？有几个明显的原因。认知负担：意图 + 有限状态机 + 时间线需要： “我知道我在做什么”，“我在设计生命周期”，“我在有意识地生成状态。” 但对于今天的前端开发者来说，这不是一个特性，而是一个障碍。因为生态系统奖励的是：快速演示、快速工作、快速简历。对于这样的人：有限状态机 = 恐惧，确定性 = 不必要，明确的生命周期 = “过度工程”。用户界面的容错性非常高：在后端：错误的抽象 → 系统崩溃，金钱损失，数据损坏。在用户界面：加载指示器转动的时间稍长，状态出现故障，用户刷新页面。换句话说：前端错误可以被容忍很长时间。这使得糟糕的抽象得以存活。 React的副作用：“隐藏、保存”文化： React做了这一点：隐藏了生命周期，隐藏了并发，隐藏了协调。结果：形成了一种“即使你不理解它也能工作”的文化。这种文化：促进了仪式性的记忆，而不是工程思维。写一个钩子，如果它能工作，那就好。但为什么能工作，怎么能工作？没有人会去问。 “氛围编码者”的爆炸不是原因，而是结果因为这种情况并不是从ChatGPT、Gemini、Claude或Copilot开始的。这些只是加速了已有的衰退。基础早已奠定。已经有一群人不知道抽象，不知道状态是什么，不知道并发是什么，但记住了框架。人工智能只是做了这一点：它使“我不需要思考”的感觉变得正式化。 4. 不是框架，而是基础设施层最大的错误是：“让我们构建一个新的用户界面框架。”这注定会失败，确实失败了。 -> “https://dayssincelastjsframework.com/” 然而，如果创建一个定位为运行时层、意图引擎、状态协调器、提交解析器的结构，它就会成为一个在React、Vue和Svelte等巨头之上的隐形层。采纳方式是这样的：“想用就用”或“如果不想用就根本看不见”。 5. 最重要的教训（或许是所有教训）我清楚地写下这句话：“前端世界现在并没有产生工程，而是在产生流水线式的行为。” 这就是为什么正确的抽象不会立即获胜，但这是不可避免的。因为：用户界面变得越来越有状态，人工智能交互在增加，并发是不可避免的。在这一点上：“加载指示器 + 钩子”模型将崩溃，人们将不得不再次问“为什么？” 而且是的：“在这种混乱中呼吁标准是毫无意义的。” 但当从正确的层面、正确的问题和正确的痛点出发时，这个模型不可避免地会创造价值。这不是关于炒作，而是关于耐心。 --- 经过这些话，我意识到有抱怨，但解决方案在哪里？我打开了一个GitHub仓库，并添加了一个“AI支持”的RFC提案。欢迎任何想要参与的人。谢谢。 https://github.com/laphilosophia/temporal-intent-resolution

查看原文

For many years, I focused quietly on my work, but now I feel compelled to point out a problem that is becoming increasingly apparent.--1. Correct Model ≠ Adopted ModelHistorical fact: In the frontend ecosystem, the winners aren't those who create the most accurate abstraction; they're those who provide the “feel of working” with the least friction.The result: correct thought model → low adoption and incorrect but easy model → explosion.This is no coincidence.2. Why Didn't Intent-Based / Deterministic Models Succeed?There are several obvious reasons.Cognitive Tax:Intent + FSM + timeline requires:“I know what I'm doing,” “I'm designing the lifecycle,” “I'm consciously producing the state.”But for today's FE crowd, this isn't a feature, it's a barrier. Because the ecosystem rewarded: quick demo, quick job, quick CV lineFor someone with this profile:FSM = fear, determinism = unnecessary, explicit lifecycle = “overengineering”Failure Tolerance is Very High in UI:In the backend: wrong abstraction → system crashes, money is lost, data is corrupted.In the UI: the spinner spins a moment too long, the state glitches, the user refreshes.In other words: Frontend errors can be tolerated for a long time. This has allowed bad abstractions to survive.React's Side Effect: The “Hide, Save” Culture:React did this: hid the lifecycle, hid concurrency, hid reconciliation.Result: a “it works even if you don't understand it” culture.This culture: grew ritual memorization, not engineering.Write a hook, if it works, fine. But why does it work, how does it work? No one asks.The Vibe-Coder Explosion is Not a Cause, but a ConsequenceBecause this situation didn't start with ChatGPT, Gemini, Cluade or Copilot. These accelerated the existing decay. The groundwork was already laid.There was already a crowd that didn't know abstraction, didn't know what state was, didn't know what concurrency was, but had memorized the framework.AI just did this: it formalized the feeling of “I don't need to think.”4. Not a Framework, but an Infrastructure LayerThe biggest mistake was: “Let's build a new UI framework.” This dies, and it did die.-> “https://dayssincelastjsframework.com/”However, if a structure is created that positions itself as a runtime layer, intent engine, state coordinator, commit resolver, it becomes an invisible layer above giants like React, Vue, and Svelte.Adoption comes like this: “Use it if you want” or “Don't see it at all if you don't want to.”5. The Most Important Lesson (Perhaps All of Them)I'm writing this sentence clearly: “The frontend world is not producing engineering right now; it's producing conveyor belt-like behavior.”That's why the right abstraction doesn't win immediately, but it is inevitable.Because: UIs are becoming more stateful, AI interaction is increasing, concurrency is inevitable.At this point: the “spinner + hook” model will collapse, and people will have to ask “why?” again.And yes: “It's pointless to shout for standards in this mess.”But when approached from the right layer, with the right problem, and the right pain point, this model inevitably creates value.This isn't about hype; it's about patience.---After all these words, I realized there's complaint, but where's the solution?I opened a GitHub repo and added an “AI-supported” RFC-Proposal. Participation is open to anyone who wants to join.Thank you.https://github.com/laphilosophia/temporal-intent-resolution

一个（谦逊的）新提案，针对FE生态系统