I am writing this article late.
Not because I failed to notice what was happening. Quite the opposite: I have been talking about it since the end of November 2025 with anyone willing to listen. But I only opened this blog a few days ago. And some things need time to settle before they can be clearly articulated.
I believe that moment—late November 2025—will one day be remembered in the history of humanity. Not as a loud revolution with headlines and celebrations, but as one of those quiet inflection points where the direction of history subtly changes.
Many of the most consequential transformations begin that way: almost unnoticed.
When I Stopped Writing Code
Technically, I stopped writing code in March 2025.
Not because software stopped being necessary. Quite the opposite: I probably produce more software today than ever before. But starting in March, the way I worked changed. I was no longer writing code directly. Instead, I spent my time interacting with several LLMs—asking them to write code, fix it, improve it. I compared outputs, had them rewrite parts, refined the results.
This was what many people began calling vibe coding.
A kind of dance between human and machine.
The human expressed the intention, the machine attempted the execution.
But in late November 2025, something changed.
With the release of Claude Opus 2025, and shortly after OpenAI Codex 5.3, the dance ended. Not because the machine simply became a slightly better dancer, but because it began to dance on its own.
We moved from vibe coding to ready-to-run code.
The End of Part of My Job
I am a software architect. I have written code for more than thirty years.
A significant part of my work involved translating ideas into working software. Today, that part is effectively over.
Not because software is no longer needed.
But because machines are better than I am at writing it.
This is not a moral judgment. It is simply an empirical observation.
If I ask a modern model to generate a service, an architecture, or even an entire system from a well-written description, the result is faster, more consistent, and often better than what I would produce by writing it manually.
Continuing to write code by hand would be somewhat like continuing to use an abacus after the invention of the calculator.
I Have Seen This Before
Curiously, I have already lived through a similar moment.
At the beginning of my career, people wrote a lot of C, but critical parts of systems were still written in assembly.
There were programmers who took pride in their ability to optimize every clock cycle. And they often said, with great confidence:
“Compilers will never be as good as we are.”
They were wrong.
Not because they were incompetent—many of them were brilliant—but because they misunderstood the direction of history.
Compilers did not need to be perfect immediately. They only needed to improve steadily. Once they became good enough, there was simply no longer any reason for most developers to write assembly by hand.
The same thing is happening today.
AI will become what compilers once were for machine code: an inevitable layer of abstraction.
Today I no longer worry about the code being written.
I focus on what I want the system to do.
The Prompt as the New Source Code
If this is true, an important consequence follows.
The prompt is the new source code.
If I have a precise and well-written description—usually in English—of a software system, I can generate it whenever I want. I can regenerate it tomorrow with a better AI model. I can modify it, expand it, evolve it.
The code becomes a temporary artifact.
The real asset is the semantic specification of the system.
It is almost as if code itself has become a compiled file, while the prompt has become the true project source.
Why Code Came First
It is not an accident that the first major domain in which AI surpassed humans was code generation.
There are two reasons.
The first is technical.
For those building models, code is an ideal domain: it is structured, verifiable, and supported by vast public datasets.
It is one of the easiest domains in which to train machines effectively.
But the second reason is far more interesting.
Code allows models to help write themselves.
The Moment Machines Begin Programming Machines
This idea may sound like science fiction. But it is not.
OpenAI itself has stated that GPT-5.3 Codex was “instrumental in creating itself.”
In other words, the system contributed to writing the code required for its own development.
This fundamentally changes the dynamics of technological progress.
For thousands of years, humans built tools.
Then we built machines that helped us build tools.
Now we are building machines that participate in building themselves.
We are not yet in a world of uncontrolled self-improvement. But we have crossed an important threshold.
The Real Inflection Point
Many people assume the decisive moment will come when AI surpasses humans on some test of general intelligence.
I disagree.
The real inflection point occurs when a technology becomes the standard tool used to build the next version of itself.
When this happened with compilers, software development accelerated dramatically.
When it happened with machine tools, the Industrial Revolution exploded.
At the end of November 2025 we witnessed the same phenomenon in artificial intelligence.
Machines began to become the primary programmers of machines.
Looking Back
In twenty years, this moment may appear obvious.
It may even seem inevitable.
But those who live through such transitions know something important: the world rarely changes with a single explosion. It changes through a series of small shifts that eventually become irreversible.
At the end of November 2025, one of those shifts became visible.
We did not simply create more intelligent tools.
We created tools that write the next generation of tools.
And when a species invents machines capable of designing the machines that follow them, history enters a new phase.
We do not yet know what that phase will look like.
But we know that it has already begun. 🚀
A blog of AI, tech, and mixed stuffs 
Leave a comment