王峰 AI × 经营 约一次诊断

总目 / 文章 / WF-0481

文章 · 存档

在AI时代“造东西”,让我不再相信“能力主义”

——GPT-5、系统封装与速度,正彻底重构我们对“价值”与“建功立业”的理

解

作者:Hui Jin

过去一年,AI 成了我用过最强大的生产力工具。

写作、编程、研究、逻辑推理、界面测试,甚至是头脑风暴,像GPT-4o、

Claude 3.5、Cursor 和 Lovable 这些工具,把“从想法到执行、从执行到成

果”的距离压缩到了极限。

这就是第一阶段:AI 作为“人类生产力的倍增器”初步登场。

那接下来是什么?

我们正进入第二阶段:端到端整合。

产品正在AI 原生的轨道上被重构:一个界面就能打通从文档解析到决策自动化

的所有流程;一次OpenAI API 调用,就能取代几千行代码。

这非常解放人类。人类的创造力和输出,或许即将迎来历史新高峰。

但问题是:这场革命,可能只能再给我们多争取几年时间。

因为,我们今天在模型之上搭建的这些东西 —— 仪表盘、流程、助手 ——

正在被AI“悄悄吞噬”。

AI 很快就会根据个人情况进行定制,把我们刚搭好的这些系统,再次抽象化、

自动化,并彻底取代它们。

只要看看GPT-5 的传闻能力,趋势就非常明显。

很快,你会看到这样的近未来:

- 你的用户界面,不再需要前端开发团队 —— 它就是一个有记忆的语言模型

- 用户引导流程?不再是Figma 画的页面 —— 它是一段对话

- 你的“产品”?其实就是包在一个智能后端上的一层提示词壳而已

如果你真的想搭上这波浪潮,就必须从“幕布背后”开始构建:

- 模型级的调度和编排

- 智能体设计与提示词原理

- 状态管理、记忆机制、反馈循环

- 那些真正塑造模型本体的工具与协议 —— 而不是仅仅操纵它的输出

但请注意,这第二阶段也只能再给我们多争取几年时间 —— 直到这些底层也

被下一代“模型原生基础设施”所抽象。

之后会发生什么?

我们将进入第三阶段:AI 开始承担大部分实际工作。

这个未来,可能仅仅还需要5 到10 年。

而它提出了一个比技术更深刻的问题:

当智力、技能,甚至创造力都不再稀缺,我们还如何定义“能力”本身?

对我来说,这个认知极度不适,也极度迫切。

我是一个从小将“能力主义”视为和“自由”同等神圣理念的人。

但AI 正在重写关于“生产力、价值、权力”的游戏规则。

这带来一个冷酷的现实:我们必须准备迎接一个“后能力主义”的世界。

当智能与执行力开始泛滥、自动化,并逐渐与“人类努力”脱钩,

“能力”在社会资源配置中的重要性正一点点地——被一层层、一系统一系统

地削弱。

如果你还试图“保住旧游戏”,那不只是自欺欺人,更是一种误导。

真正该问的问题是:

当技能不再稀缺、努力不再决定价值时,我们要建立一种什么样的生存环境?

Building with AI Broke My Faith in Meritocracy

How (rumored) GPT-5, system encapsulation, and speed rewired what it

means to build — and to earn our place

Hui Jin

Over the last year, AI has become the most powerful productivity tool

I've ever used. For writing, coding, research, reasoning, UI testing

— even brainstorming — tools like GPT-4o, Claude 3.5, Cursor, and

Lovable have collapsed the distance between idea and execution,

between execution and results.

That's phase one: the human-AI interface as a personal productivity

multiplier. So what's next?

We're now entering phase two: end-to-end consolidation. Products are

being rebuilt on AI-native rails — one UI wrapping everything from

document parsing to decision automation. One OpenAI API call now

replaces thousands of lines of code.

It's freeing. Human creativity and output may soon reach an all-time

peak. But here's the catch: it probably only buys us a few years.

Everything we're scaffolding on top of today's models — dashboards,

workflows, copilots — is being quietly subsumed under the facade. AI

will soon personalize, abstract, and automate away much of what we

just built.

Just look at GPT-5's rumored capabilities. The trend is clear. Here's

what this near-future looks like:

- Your UI is now an LLM with memory — not a frontend team

- Your onboarding flow is a conversation — not a Figma prototype

- Your “product” is just a thin prompt layer over a cognitive

backend

If you're serious about riding this wave, you have to build under the

facade:

- Model-level orchestration

- Agent design and prompt theory

- State, memory, and feedback loops

- Tools and protocols that shape the LLM itself — not just its

output

But then this phase 2 only buys us another few years — until those

layers are abstracted by the next generation of model-native

infrastructure.

After that? We enter phase three: AI increasingly does most of the

work.

That future may well be mere 5–10 years away. And it raises a

question that cuts deeper than tech:

When intelligence, skill, and even creative output are no longer

scarce, what do we still mean by “merit”?

As someone who grew up seeing meritocracy as fundamental as freedom,

this realization is uncomfortable — and urgent.

AI is rewriting the rules of productivity, value, and power. This

leaves us with a hard truth: We must prepare for a post-meritocracy

world.

The relevance of merit is fading — layer by layer, system by system

— as intelligence and execution become abundant, automated, and

increasingly detached from human effort.

Trying to preserve the old game is worse than denial. It's

misdirection.

The real question is:

What kind of society are we building when skill is no longer scarce,

and effort no longer defines worth?

That's what we should be debating — not quarterly metrics or chatbot

benchmarks, and not the squabbles on the Capital Hill