专业软件和编程语言，会不会变成机器内部的东西？ / Will Professional Software and Programming Languages Become Internal to Machines?
专业软件和编程语言，会不会变成机器内部的东西？
本文包含中文与英文两个版本。
This post includes both Chinese and English versions.
[CN] 中文版
未来像 Photoshop、Unity、Unreal 这样的专业软件，会不会消失？
再往前想，编程语言会不会也消失？以后只要说出想做什么，AI 就能直接给出可以运行的应用。
我觉得这个方向是有可能的。但准确地说，消失的可能不是软件和语言，而是它们作为“人必须亲自操作的工具”这一层。
人真正想要的不是工具
想做一张图，我们学 Photoshop；想做一个游戏，我们学 Unity 或 Unreal；想做一个 App，我们学编程语言、框架和构建工具。
想法要变成结果，中间隔着很多层：
想法 → 工具 → 操作方法 → 工程文件 → 最终结果
这些工具当然有用。它们把复杂能力装进软件里，让人可以一点点控制结果。但很多时候，我们学一套工具，只是因为机器还不能直接听懂我们要什么。
以前想修一张图，就得知道图层、蒙版、曲线在哪；想做一个功能，就得知道代码该写在哪里、项目怎么编译。我们花时间学的，常常是怎样把自己的想法翻译成软件能接受的操作。
AI 改变的可能就是这件事。
从“怎么做”到“做成什么样”
一开始，AI 只是帮我们操作工具：写一段代码、生成一张素材、修改一个材质。工具还在，人也还在工程里，只是有些步骤交给了 AI。
再往后，人可能只说目标：做一个下雨的东京街头；敌人发现玩家后先找掩体，再叫队友包抄；做一个能导入运动记录、生成轨迹回顾的 App。
AI 去改场景、写代码、生成资源，再运行起来看看哪里不对。人看结果，告诉它哪里需要调整。
这时 Unity、Unreal、Photoshop 还在，只是人未必需要一直打开它们。它们可能更像 AI 背后的能力库：需要渲染时调用引擎，需要修图时调用图像工具，需要运行程序时调用构建环境。
工具还在，操作工具的人变了。
编程语言也一样
现在做应用，大致要经过这样的过程：
想法 → 编程语言 → 编译器 → 可执行程序
如果 AI 能把想法变成程序，编程语言是不是就没用了？
对大多数人来说，可能确实不必再亲自写那么多代码。但程序还是得有一种明确的表示方式，机器才能执行，人也才能检查它到底做了什么。
这个表示方式可以是今天的代码，也可能是以后新的中间形式。它甚至可能大部分时间都不显示给人看。
所以，“AI 直接生成二进制”听起来很直接，却不一定是重点。程序能运行，不代表它做对了事。出了问题，总得知道哪里错了；需求变了，也得知道改动会影响什么。
编程语言可能从多数人的日常工具，变成少数人维护系统时才会直接接触的东西。就像汇编语言没有消失，但大多数人写程序时不需要碰它。
会操作工具的人会变少吗？
可能会有一部分操作能力不再值钱。以前要花很久才能做出来的东西，AI 也许几分钟就能给出一个版本。
但这不代表“会判断”就能凭空出现。
AI 做出来的图好不好，游戏能不能玩，App 的数据处理有没有问题，还是要有人判断。只会说“再好一点”，很难让结果真的变好。你得知道哪里不对，什么才算达到要求。
做图、做游戏、做软件，专业知识还是有用，只是用途可能变化了。过去靠它一步步把东西做出来，以后可能更多靠它判断结果、提出修改、确认风险。
而且，亲手操作本身也有价值。人经常是在做的过程中才发现自己真正想要什么。AI 如果只会按一句话交付成品，却不让人中途参与，未必是更好的工具。
以后更重要的是什么
过去，一个人会不会 Photoshop、会不会 UE、会不会写代码，能决定他能做多少事。
以后这些能力可能变得便宜。更重要的会是：你想解决什么问题，什么样的结果算好，做出来以后有没有真的用处。
这不是说人人都不必学工具和技术。懂得越多，越容易看出 AI 在哪里胡说、哪里做得不对。只是学习的目的，可能不再是记住每个按钮和语法，而是知道怎样把事情做好。
专业软件和编程语言大概不会突然消失。它们会从人手里的工具，慢慢变成机器背后的实现。
人越来越少地告诉机器“怎么做”，越来越多地告诉它“要做成什么”。
真正难的，最后可能不是让 AI 做出东西，而是知道自己要什么，也知道做出来的东西值不值得留下。
[EN] English Version
Will Professional Software and Programming Languages Become Internal to Machines?
Will professional tools like Photoshop, Unity, and Unreal disappear in the future?
If we take the idea further, will programming languages disappear too? Will we simply describe what we want, and have AI deliver an application that runs?
I think that direction is possible. More precisely, what may disappear is not the software or the languages themselves, but their role as tools people must operate by hand.
What People Really Want Is Not a Tool
To make an image, we learn Photoshop. To make a game, we learn Unity or Unreal. To build an app, we learn programming languages, frameworks, and build tools.
An idea has to pass through several layers before it becomes a result:
Idea → tool → operating method → project files → final result
These tools are useful. They package complex capabilities into software and let people control the outcome step by step. But often, we learn a tool because machines cannot yet understand directly what we want.
To edit an image, we had to know where to find layers, masks, and curves. To build a feature, we had to know where to write the code and how to compile the project. Much of what we learned was how to translate an idea into actions a tool could accept.
AI may change that.
From “How to Do It” to “What to Make”
At first, AI helps us operate tools: write a piece of code, generate an asset, or adjust a material. The tools remain, and people still work inside the project. Some steps are simply delegated to AI.
Later, people may describe the goal: make a rainy Tokyo street; have an enemy find cover and call for backup after spotting the player; build an app that imports exercise records and creates a route recap.
AI edits the scene, writes code, creates assets, and runs the result to see what is wrong. People review it and say what needs to change.
Unity, Unreal, and Photoshop would still exist, but people might not need to keep opening them. They could become capabilities behind AI: an engine for rendering, an image tool for editing, and a build environment for running software.
The tool remains. The person operating it changes.
The Same May Be True of Programming Languages
Today, building an app roughly follows this path:
Idea → programming language → compiler → executable program
If AI can turn an idea into a program, do we still need programming languages?
Most people may no longer need to write as much code by hand. But a program still needs some precise representation so that a machine can execute it and people can inspect what it does.
That representation might be source code, or a different intermediate form in the future. It may not even be shown to people most of the time.
So “AI generates a binary directly” sounds straightforward, but it may miss the point. A program that runs has not necessarily done the right thing. When something breaks, we need to find out why. When requirements change, we need to know what else will be affected.
Programming languages may move from everyday tools for most people to something specialists use when maintaining systems. Assembly still exists, but most people do not need to touch it when writing software.
Will Fewer People Need to Operate Tools?
Some operating skills may become less valuable. Work that once took a long time might get a first version from AI in minutes.
But that does not make good judgment automatic.
Someone still has to decide whether an image looks right, whether a game is fun, and whether an app handles data correctly. It is hard to improve a result by saying only “make it better.” You need to know what is wrong and what would count as good enough.
Professional knowledge in art, games, and software will still matter, but its use may change. In the past, it helped people make things step by step. In the future, it may help them evaluate results, request changes, and identify risks.
And hands-on work has value of its own. People often discover what they really want while making something. An AI that delivers a finished product from one sentence, but leaves no room for people to take part along the way, may not be a better tool.
What Will Matter More?
In the past, knowing Photoshop, Unreal, or how to code could determine how much a person was able to make.
Those skills may become cheaper to access. More important will be knowing what problem to solve, what a good result looks like, and whether the result is useful in practice.
That does not mean everyone can skip technical knowledge. The more you understand, the easier it is to spot when AI is making things up or getting something wrong. The purpose of learning may shift: less memorizing every button and syntax rule, more understanding how to make the work succeed.
Professional software and programming languages probably will not vanish overnight. They may move from tools in people’s hands to implementations behind the machine.
People may spend less time telling machines how to do something, and more time telling them what to make.
The hardest part may not be getting AI to produce something. It may be knowing what we want, and whether the result deserves to stay.