In 2020, a Bulgarian developer suddenly came up with the idea of replicating Red Alert 2 on the web? So he spent 5 years using the most extreme human reverse engineering, handwritten code line by line, to simulate and restore every operating logic according to "Red Alert 2". Finally, I wrote a web page red alert, so that everyone born in the 80s and 90s who wants to be nostalgic can go back to their childhood by just opening the browser without having to worry about configuring the environment. However, just a month ago, someone around us did something similar.
He has reproduced three of Xishanju's early works, "New Swordsman Love", "Legend of Moon Shadow", and "Swordsman Love 2", into the web version, and adapted them to the mobile version, so you can click on them to play.
Everyone is familiar with these games.
The bad reviewer still remembers turning on the computer after returning home from elementary school, controlling the Dugu Sword and wandering around Lin'an City, and chatting with a drunkard in a restaurant can actually trigger hidden martial arts. Then in "Legend of Moon Shadow", he controlled Yang Yingfeng to pick up herbs and chase a man in black to the Batcave. As a result, he encountered a group of bandits and was surrounded and beaten, but he couldn't beat them.
Opening this remastered version after more than 20 years, the negative reviewer found that the taste is very pure: the NPC dialogue is still so long-winded, pressing V to meditate to restore the vitality is still there, and there are also many treasure chests hidden behind stones in the corners of the map.
But what is shocking is:
The authors who reproduced these three games did not write a single line of code by hand.
He relied entirely on AI, and in less than a month, he not only reproduced three complete martial arts RPGs, but even built a universal game engine, complete with debugger and editor, so that others could reference and reproduce other games.
How is this done? The bad reviewer found the author of this project, Teacher Chen.
Mr. Chen has been teaching himself programming since he was in the third or fourth grade of elementary school. He has nearly 20 years of coding experience and has open sourced many projects on GitHub early on.
It stands to reason that such a senior geek should have a deep belief in handwritten code.
But Teacher Chen told me that now he is not even willing to type an equal sign.
Because he has tried a fixed process: first let the current most powerful model write a detailed design document, and then execute it in modules strictly according to the document. Each module is independently verified, and then put together after it runs through. The quality of the engineering architecture and code produced in this way is indeed higher than handwriting from scratch in most cases.
In fact, as early as 2020, Mr. Chen had the idea of reprinting these childhood masterpieces, but gave up after calculating the engineering workload of manually reverse-engineering binary files and rewriting the entire game system.
Until recently, he believed that AI’s capabilities had finally crossed that tipping point.
Therefore, this re-enactment is not only a childhood nostalgia, but also an extreme test of AI programming.
The first thing Mr. Chen did was to throw the original resource files of the game client directly to AI and let AI analyze it.
These game files are binary, and all information is packed and compressed into a human-readable format for distribution and installation. After the AI got the file, it started frantically calling various viewing tools, listing the file structure, reading hexadecimal, and writing Python code to try to parse it. If one path doesn't work, immediately try again with another logic.
In the past, programmers had to hold a hex editor, analyze bit by bit based on experience and intuition, and then write scripts and try again and again.
Teacher Chen estimates that AI’s efficiency in this link is 10 to 100 times that of humans.
It makes sense to me. AI is good at violent exhaustion. It doesn’t need to sleep, it won’t be agitated, and it won’t open TikTok in the third hour to get bad reviews.
This is just the beginning, the real test is how to rebuild the entire game system.
When reproducing with VScode Copilot, Mr. Chen mainly uses two models: Claude Opus is responsible for in-depth thinking and acts as an architect, splitting the game into modules and writing technical documents; Claude Sonnet is fast and acts as an executor and writes the code according to the documents.
What makes people's scalp numb is that AI is no longer just immersed in execution, it also has the ability to autonomously correct errors.
Once when I was working on a pathfinding system, Teacher Chen felt that this module was too performance-intensive and TypeScript might not be able to handle it, so he asked Sonnet to try rewriting it in Rust. After Sonnet was written, I ran a round of performance comparisons and found that the overhead of translating and transferring data back and forth between Rust and TypeScript was too high, and the overall calculation was even slower.
Therefore, it took the initiative to feedback this test conclusion to Teacher Chen and suggested falling back to the TypeScript solution.
With such a responsible thinking and responsible action, the project progressed faster than Mr. Chen himself expected.
From the start of work on January 25 to the end of February, excluding the Spring Festival break, it only took about 20 days to fully calculate. During this period, Teacher Chen had four to five thousand conversations with AI.
He usually opens 5 to 6 AI dialogue windows at the same time and processes different modules in parallel. When I work in the company during the day, I also use the remote desktop to let the AI at home work crazily in the background, and I can directly check the results when I go home at night.
Someone must be questioning this after seeing this: 200,000 lines of code are all automatically generated by AI. Isn’t it really a mountain of shit code?
Indeed, the bad reviewer also used AI to write an app. As a result, every time a function was added, there would be one more bug. Even if one bug was fixed, two new ones would pop up.
Teacher Chen said that the biggest misunderstanding when people with zero programming experience use AI to write code is that AI can write a complete APP in one go and lacks architectural design. It was okay at the beginning, but the more functions were added later, the more confusing it became, because AI is best at imitating previous writing methods. Once the logic is confusing in the early stage, it will become more and more chaotic in the later stage.
Therefore, Mr. Chen’s approach is to develop in modules, with each module designed and tested separately, and then try again after it is passed. When a bug is encountered, the running log is sent to AI, and the specific module is located for repair. After each major change is completed, he will open a new window and let another AI scan the newly changed code from scratch.
In short, after no less than 10 rounds of optimization and refactoring, nearly 200,000 lines of code were finally successfully implemented.
So what exactly do these 200,000 lines of code do?
You will know when you open the game: when the character walks around the map, he can find the shortest route around obstacles. The weather will suddenly turn from sunny to rainy, and raindrops will hang on the screen and slowly fall. This effect is not available in the original version, but is newly added in the remastered version. The martial arts system also has 22 flight trajectories and 10 status effects, including spiral, tracking, and fan-shaped diffusion, which can combine hundreds of spells.
Since the original game was written by Xishanju using his own script syntax to string together the entire script, in order to make the game run in the browser, Mr. Chen also rewrote hundreds of plot instructions and added support for the Lua language commonly used in the modern game industry. In the future, if he wants to add new plots and change dialogue branches to the game, he does not need to ponder Xishanju's original syntax, as he can use Lua scripts.
Not only that, he also made a game editor.
Martial arts, NPCs, maps, items, dialogue trees, all 13 modules can be edited directly on the web page, and the effects can be seen in the game after refreshing. In the future, as long as the engine is connected to AI, users can generate game plots and MODs in one sentence.
These three games already have a group of MOD creators on Bilibili. In the past, they were all made by hand, which was time-consuming and labor-intensive. If this function is implemented, everyone can create their own swordsman romance.
As I write this, the old question comes up again: Are programmers really going to be replaced?
Teacher Chen’s answer is no at present, but the definition of the profession of programmer is already being rewritten.
In the past, to develop games independently, you had to devote 70% of your energy to coding, modeling, fixing bugs, and optimizing performance, leaving 30% for gameplay.
But now, AI can take over these coding tasks. You need to use 30% of your energy to make architectural decisions, and the remaining 70% is devoted to gameplay design.
For example, how to determine the item interaction logic? How to draw a numerical growth curve? How to arrange NPC behavior tree? These are decisions at the product and design level, and there are no standard answers.
AI can provide you with a hundred plans, but in the end you still need humans to make the final decision on what kind of game you want to make.
In other words, Mr. Chen’s 20 years of programming experience have not been invalidated, but have been transformed into architectural decisions in the process of reproducing the game.
People like you, who have no programming experience, don’t know how to locate bugs when they encounter them. They have no idea how to dismantle the game system, and even if there is AI, they still stare at it.
but. . .
This chat also led to a deeper inference:
The distribution of programmers' work focus has changed from "7 to 3" to the current "3 to 7". So with the rapid iteration of AI models, will this ratio soon become "1 to 9" or even "0 to 10"?
How long will this stage of requiring experienced people to lead the code last?
Even after tomorrow, will AI be able to think faster and more thoughtfully than most people in creative work such as "how to design a game?"
In this field where the time scale is extremely compressed, no one dares to give a definite conclusion.
After all, DeepSeek was just released more than a year ago, and AI only helps you fill in a few lines of code and write the simplest programs. Even going forward a year, everyone is still talking about AI customer service.
Today, 20 days, 200,000 lines of code, and three games are the result of the cooperation between an old programmer and AI.
No one would have thought that one day programmers would no longer type code with their hands. There was also ridicule in the industry that "Handwritten code with ingenuity is an intangible cultural heritage, ancient programming method", which clearly stated that handwritten code is an ancient method.
These are the two sides of the AI coin.
On the one hand, the skills you are most proficient in are becoming the default function of AI.
AI has gone from asking and answering questions in a chat box to a tool that can read documents, remember context, call tools, actively work, and check for errors.
This change is spreading from the programming field to all mental work. Design has changed from hand-made materials to generating and adjusting AI-generated solutions. Videos have changed from manual editing to creative planning + AI execution.
The focus of work in all walks of life is being redefined on a monthly basis, and the shelf life of human experience and skills also follows the version number of the top model.
On the other side, there is a door you have never opened.
The distance between idea and finished product is shortened extremely quickly.
A person who has never learned to edit can directly turn the picture in his mind into a short film. A grand idea in one’s mind no longer needs to wait for hundreds of thousands of funds and a complete team to be implemented.
Execution ability is depreciating, but people who know "what to do" will be more irreplaceable than before.
No one knows which side the coin will end up on. But at least now, it gives everyone a chance to be dealt their cards again.