It is only when various apps and websites start pushing their year-end summaries that I realize 2025 is drawing to a close.
If you were to ask me what left the deepest impression this year, I’d think back to the beginning of the year when Deepseek was released. Over the course of the year, Large Language Models (LLMs) have become increasingly powerful while consuming fewer and fewer tokens. Last year, we might have still been debating what AI could or could not do. This year, we realize how trivial that discussion was. Whether you like it or not, you and I are no longer spectators watching the AI wave; we are the surfers riding it.
I remember when I first learned to program, I would feel a sense of smug satisfaction about how many languages I could write "Hello World" in, and how excited I was to master a new one. Later, with the sudden emergence of GPT-3.5, the internet was flooded with prompt tutorials, claiming that everyone would need to be a "prompt engineer" in the future. We scrambled to imitate other people's prompts. Looking back now, that seems laughable. Today, what we need to learn is far more than just a language or how to mimic a prompt to generate an image or a webpage.
To get a satisfactory image now, I no longer need to type obscure prompts into Stable Diffusion. Instead, I find a reference image, toss it to the AI, ask it to generate a prompt for me, and then feed that prompt into Nano Banana Pro. If I want to generate code, I simply type my ideas into the input box of Cursor or Claude Code, and the AI generates code ready for a production environment.
Everything has become simple, perhaps even boring. The curriculum has shifted from learning a language or a prompt to learning how to learn—learning how to be a conductor, orchestrating these agents to complete work that originally required humans. Everyone can be a team leader; only now, we aren't leading carbon-based humans, but silicon-based agents built on chips and electrical power.
In the past, closing my laptop after a day's work, I would think about how much code I wrote, how many features I implemented through various classes, and secretly delight in cleverly solving a problem—sometimes even clapping my hands, immersed in that "Aha!" moment. Nowadays, after a day of work, my hands seem to be on the keyboard constantly, yet the actual code I write is countable on one hand. My work has shifted from execution to thinking. Every day feels like a conversation with another engineer: I constantly test how much work he can handle, tell him my ideas, and he instantly finishes a volume of code that might have taken a week to submit before (and by "before," I might just mean last year).
I saw a saying that Go (Weiqi) and programming were once considered "high-intellect professions." But with the advent of AlphaGo and LLMs, these two jobs have, in a sense, become "assembly line" work. A Go player's job shifted from freely seeking the optimal move to memorizing the optimal solutions found by AI via Monte Carlo tree search. The competition between players became about who could memorize more AI game records under time pressure and play more like an AI. Similarly, a programmer's job is no longer about applying learned languages and theories to production, but about discovering and identifying problems, and then letting AI solve them. Although we are still seeking the optimal solution, the search space has expanded exponentially compared to before.
This year, from developing casual tools to diving into completely new product fields I knew nothing about, collaborating with AI has become my daily bread. Work that used to take weeks or even months can now be compressed into days or hours. The original workflow of tapping the keyboard and compiling has been completely replaced. I feel like I've become a typist: entering requirements into a box and waiting for the result. If I choose the right tools and methods, the result is not only much faster than I could produce, but often more professional than what even a senior engineer could deliver.
This inevitably raises a worry: Will I be replaced by AI? As the "beasts of burden" (corporate drones) of the modern age, will we be replaced by AI just as real oxen and horses were replaced by machines?
The mid-life crisis of the past was often about physical stamina hitting a bottleneck. Because AI can now handle "digital manual labor" cheaply, stamina is no longer the constraint—thinking is the new bottleneck. Therefore, even when you stop writing code directly, never stop thinking. AI can replace execution and implementation, but at least for now, AI cannot replace critical thinking and architecture. We must use the time AI saves us to enrich our arsenal, understand more concepts—forcing ourselves to understand deeper, underlying logic—and keep our brains capable of continuous thought.
I don't know if AI will be recorded in history books as the N-th Industrial Revolution, but for us ordinary people in various industries living through it, all we can do is adapt to this change. We must change ourselves, step out of our inertia, and try to use this ability that was once viewed as a superpower. The old order has crumbled, but the new order has not yet formed. All we can do is strive to find our own place in this new era.
I believe humanity can reach the stars. I hope we can all meet there among them.