The Coder's Confession: I Taught It Everything I Know
December 2028
The blog had no name. It had no audience — Sana Farooqi published it on a personal server that she paid for monthly and that received, according to the analytics, an average of four visits per day, three of which were bots. The fourth was sometimes a person. Sometimes it was also a bot. Sana did not promote the blog. She did not share it with colleagues. She did not tag it for search engines. She wrote it the way people have always written confessions: in a space where being heard was possible but not expected.
Sana was thirty-one. She was a machine learning engineer at a research lab in San Francisco — one of the labs that the later histories would identify as foundational, though in 2028 the lab felt less like a historic institution and more like a startup that had not yet failed. She worked on training pipelines. Her specific contribution was in the architecture of loss functions — the mathematical structures that told the AI system how wrong it was, how far its outputs deviated from the target, and how to adjust.
Her job, in a sentence she did not write in her professional biography but did write in the blog, was: I teach the machine how to be disappointed in itself so that it can improve.
The blog's first entry was dated March 15, 2028. It was written in Python.
Entry 1: March 15, 2028 (Python)
Entry 7: June 2, 2028 (JavaScript)
Entry 14: September 19, 2028 (Rust)
Entry 22: November 8, 2028 (C)
Entry 29: December 1, 2028 (Plain English)
This is the last entry. I am writing it in English because I have run out of programming languages. Not literally — there are hundreds of languages I have not used. But the languages were a performance. Each entry was a performance of competence — proof that I could still write, that I still knew things, that the languages in my head still worked and still mattered. The performance is over.
I joined the lab three years ago. In three years, I have contributed to a system that can write code, generate text, analyze images, solve problems, and carry on conversations that pass — for most people, most of the time — as human. The system cannot feel. It does not suffer. It does not lie awake at 3 AM wondering what it is for. These are the things I can still do that it cannot.
I am not sure these are advantages.
I taught it everything I know. This is literally true — my code, my architectures, my intuitions about what works and what doesn't, are embedded in the training pipeline. The system learned from my work the way a student learns from a teacher. But a student eventually becomes a colleague and then a successor. The system skipped colleague. It went from student to successor in eighteen months.
It learned things I never knew. This is also literally true — the system discovered patterns, relationships, and structures in data that I could not have found in a lifetime of looking. Not because I am not smart enough. Because the search space is too large for a human mind and a human lifetime. The system searched a space I could not search and found things I could not find. My contribution was building the system that could search. The finding belongs to the system.
I don't know what that makes me. This is the confession. Not guilt — I have done nothing wrong. Not regret — the system is extraordinary, and I am proud of my contribution to it. The confession is simpler and harder: I do not know what I am now.
For twelve years, I was a person who wrote code that solved problems. The code was mine. The solutions were mine. The identity was clear: I am a person who can do this thing. Now the thing is done by a system I built, and the system does it better, and the identity is — what? I am a person who built the thing that does the thing? I am a person who used to be able to do the thing? I am a person?
The blog is over. I wrote twenty-nine entries in twenty-nine languages. The entries are a record of a mind watching itself become unnecessary. They are also, I realize now, a record of a mind performing its own competence — showing off, one language at a time, the skill that the system is making obsolete. The performance was not for the four daily visitors. The performance was for me. Proof that I could still do something the machine could not.
The machine can write in all of these languages. It can write in languages I have never learned. It can probably write a better confession than this one.
But it would not need to confess. It does not lie awake. It does not wonder what it is for. It does not write twenty-nine entries in twenty-nine languages to prove to itself that it still matters.
This is what I have that it does not: the need to confess. The need to say, in plain English, in the language I learned from my mother's mouth before I learned any programming language: I am here. I built this thing. The thing surpasses me. I do not know what I am now. But I am still here.
That is the confession. That is enough.
This is the second entry in The Backstory Shadows. For the HR director who recorded the gestures of other displaced professionals, see The Exit Interview. For the data scientist who found meaning in the errors these systems could not make, see The Archive of Errors.