The gentleman is not a vessel (君子不器)
— Analects (論語), Weizhen chapter (為政篇)
AI Era, the Signaling of Capability, and the Programming of 군자불기 (君子不器)
I originally majored in theoretical physics. I wanted to understand how the world works, and if possible, I wanted to stay in academia and keep studying, but I lacked the ability, and I lacked the money.
To earn recognition in physics in Korea, however, you had to either ride the academic networks of the country's top universities or go through the path of an overseas doctorate centered on the United States. Studying abroad in America was out of the question given my family's circumstances, and from what I observed of academia, I stood outside that pedigree. My decision to leave was partly a recognition of my own limitations, but it was also accompanied by the growing conviction that I would never gain entry into that exclusive network's circle of credentials.
The decisive blow was the salary of those on the periphery, namely lecturers. That salary shocked me. I realized I could not live the life I had in mind on such pay. The ordinary life I imagined: a wife, a small home, nothing extravagant.
The idea that academia is the purest pursuit is, frankly, a lie. Academics are purely exclusive in the most rigorous sense. What they scrutinize most is which institution you belonged to and under whom you studied.
Science operates in the realm of the falsifiable, but the entities that fund science are corporations, and those corporations seek to make famous the people whose conclusions they want to believe. Science is not beautiful. Depending on who funds a conference, the statistical modeling gets reinterpreted, different conclusions are drawn, and the entire structure makes it extremely difficult to challenge the presiding chairs. This is what I observed, so others may see it differently. Because I was bounced out of that world, I may be someone who distorts things with a hostile eye.
Regardless, in the year I quit I looked at the doctoral employment statistics. Those numbers confirmed what I had already suspected. Academia is not a meritocracy. It is pedigree. It is a network of recommendations and personal connections dressed in the clothing of rigorous science.
So I left. Not because I stopped loving physics, but because I understood that I would never receive the credential to practice that discipline.
After that, I became a programmer.
I had absolutely no interest in the beauty of code or anything like that. I had never felt that mathematics was beautiful, or anything of the sort. I simply wanted to draw the world inside my head.
Just as physics models the world with equations, I wanted to simulate the world in my mind through code.
For me, the ultimate destination of programming is the question of how beautifully the ideas inside my head move on screen.
And this turned out to be a considerable advantage for me, even putting a remarkable amount of money in my hands relative to people my age. I lost it all to fraud, but still. (It was not top-1% money, to be clear. More like top-10%. Half of South Korea's salaried workers earn less than 3 million won a month, and I was at least earning more than that.)
The Collapse of the Portfolio and the Return of Homophily
There is a dilemma I feel acutely from using AI every day.
What will human capability be proven by going forward?
In the past, the mere fact that you had actually built something and shipped it was a powerful signal. If you had pushed through frontend, backend, database, authentication, and deployment to launch a game or a service, that output alone was enough to infer your ability.
At least in the Korean market, it was the portfolio that mattered, not questions about internal code architecture. (And to be honest, internal code architecture outside the major cities is a mess to begin with.)
But AI has explosively raised the floor of what constitutes an observable deliverable. Even a beginner using AI can produce a seemingly polished output with login functionality, charts, and test code all intact.
When the cost of generating an output converges to zero, that output can no longer serve as evidence of ability. I tutored 10 students and helped 6 of them land jobs. (All at startups or small companies, of course.)
Back then, the whole thing came down to shipping. Having actual launch experience was genuinely powerful, and even just three years ago it was the strongest possible portfolio. But that is no longer the case.
Hirers and investors do not have time to individually verify hundreds of plausible AI-generated portfolios. When the cost of verification skyrockets, I believe humans revert to the most primitive and exclusive indicators available.
Degrees, pedigree from well-known companies, vouching from existing networks. Even as AI tears down the technical barriers to entry, the very pedigree and homophily I do not possess paradoxically resurface as the most powerful signal.
In the short term, the value of an Ivy League or Seoul National University student ID may decline, but over the long term it will grow stronger still. Even now, in the SI market and the VC investment market, university pedigree, especially overseas university pedigree, already carries the most weight, and that will only intensify.
Anyone can now produce a convincing product. A few prompts generate a frontend, wire up authentication, mock a database schema, and yield an output that looks identical to what a senior engineer would build. The cost of producing a plausible deliverable is converging to zero.
군자불기 (君子不器): Stop Being a Tool
The thing I fear most as I rely more on AI is the sense that my ability to write code by hand is atrophying. APIs no longer come to mind the way they used to, and for complex implementations I generate rather than write. I believe returning to a world without AI is simply not possible.
In the old days I would put documentation on the second monitor, translate it as I read, and spend time wandering through GitHub examples, constantly searching for better implementations. It was common to spend an entire day writing 600 lines. Now it takes three minutes.
The Analects (論語) contains the saying "군자불기 (君子不器): a person of virtue is not a vessel fit for only one purpose."
The old "coding muscle" of typing out localized syntax was squarely in the domain of the tool (器), but now that AI is fully replacing that role, I sense my raw coding ability declining. Even so, I believe that when the consumption of cognitive resources decreases in one area, it increases in another. I have grown weaker at local detail, but my capacity for macroscopic pipeline design is developing.
I now have to make a choice: keep grinding out code by hand, or, after convincing myself that a person of virtue is not a vessel, focus on how to stabilize the torrent of code that AI pours out.
I chose to make the effort to hold onto only what is stable among the code raining down.
Homophily and the Death of Technical Novelty
At one time, the internet produced strange, beautiful, world-changing things from the most unexpected places. Linux came from a student in Finland. PHP came from one person who wanted to manage his own résumé. Even the subscription model that Twitch made famous was reportedly inspired by a Korean platform, according to Twitch's CEO. The early history of the internet is full of outsiders who changed the world without asking permission.
That era is ending.
The industry has matured. The standards that govern how software is built are now decided almost entirely by American big tech. New languages, new frameworks, new architectural patterns are designed in California and handed down to the rest of the world. If you sit outside that ecosystem, you spend most of your energy just catching up to changes they have already made.
Consider Linux. Early Linux was simple. You only had to focus on the kernel. To compete now, you must simultaneously handle the kernel, drivers, GPU, file systems, toolchains, package managers, and security models. It is not a game an individual can win. The frontier is closed.
That is why I moved to the application layer. Not because I lost interest in systems, but because I concluded that the systems layer is no longer a place where an outsider can fight.
The systems that Korea once built to challenge American standards have become legacy, and that same technology now shackles Korean programmers trying to compete globally. I judged that I cannot win that game.
The Architecture of Control and 정명 (正名)
The real threat of the AI era is not "being unable to write code" but rather "too much code being generated too easily." Every time hundreds of lines pour out, duplicated state, hidden exception handling, and competing caches quietly erode the system.
That is why I returned to minimal proof techniques like Rocq (Coq) and ADT (Algebraic Data Types). The goal is not to prove every piece of code mathematically.
I decided to prove only the parts that must not break, and to draw clear boundaries around the parts that are allowed to break.
I am trying to narrow state through ADTs to block impossible states at the source and to control the direction of dependencies. To that end, instead of reading AI-generated code line by line, I have built and use a visual debugger that lets me trace ownership and dependency flow across functions and classes.
I now believe it is not the code itself that matters, but the connections along which code flows.
In the old days, the ability to write code was like a small stream, and every person carved out that stream differently. But AI is a flood from a burst dam. It continuously produces enormous volumes of code. Because of that, what I am trying to do is dig the channels that determine how that flood will flow.
The Capitalization of Intelligence and a Five-Year Window
Another reality I encounter while teaching students from low-income backgrounds in the provinces is the "capitalization of intelligence." The gap between students using paid AI models and those using free models is widening to a dispiriting degree. I see the difference between paid and free models every single day.
I see it constantly while volunteering to teach students how to use AI. That gap will likely compound over time.
The democratization of information has paradoxically produced inequality in access to compute. I myself pay a substantial monthly subscription for AI. The era in which cognitive capacity must be purchased with money has arrived.
AI Slop is flooding the internet. I expect this to continue for five years. In that environment, how can I establish a foothold locally?
It is always difficult.