I don’t know if Vibecoding can be considered a mainstream phenomenon now. A couple of days ago, I shared something about Antigravity on social media. Antigravity is a newly released AI IDE from Google that is completely free to use. A friend who had never touched code before came to me, saying he was extremely interested and wanted to try it out. A few days later, he came back to me saying he had tried it but couldn’t figure it out after tinkering for a long time…
The key to this isn’t whether he could ultimately figure out the code. It’s that Vibecoding has completely lowered the barrier to programming on a cognitive level. Think about it: in the past, for non-programmers, “solving a problem by writing code yourself” was never even an option. It wasn’t a question of whether they could do it, but whether they even thought of it—after all, who would casually do an Ironman triathlon just to exercise? Even if they thought of it, writing a program meant going through installing environments, learning languages, figuring out architectures… These were massive hurdles, both psychologically and practically.
But now, with vibecoding as this “super GUI,” things are different. It’s just like how the graphical user interface drastically lowered the barrier to using computers: you can write programs using natural language. Programming is no longer a life-and-death struggle with tools and environments; it has returned to being a dialogue with requirements and logic.
To be honest, as a programmer, I did feel a sense of crisis that my “rice bowl” might be snatched away. It all happened so suddenly that I couldn’t react at first.
But calming down and thinking carefully, if a programmer were just a “code monkey,” they would have been replaced by AI long ago. Because when it comes to writing code, there is probably no programmer in the world who knows more languages, types faster, and never gets bored than the various LLMs today—as long as there’s power and internet, they can work day and night. But if it were really true that programmer === code typist, then the time utilization rates of all major tech companies should have multiplied by several times. After all, it saves a lot of meeting time.
In the end, programmers are just people who use the skill of writing code to translate requirements into code; they are the implementers of requirements. As long as humans are still the majority in this world, anyone trying to replace the people who implement requirements will have to replace humans first.
Take language learning, for example. For AI, it’s not called learning a language; it’s called devouring a language. Feed it thousands of Japanese videos, articles, and dictionaries, and it quickly becomes a linguistics expert.
But we carbon-based lifeforms have physical limitations, and combined with the characteristics of the human brain, learning a language is destined to be a long process of struggling with forgetting. During this long and unpredictable process, countless pain points based on human experience arise: you can’t remember words because of a lack of episodic memory, you can’t hear clearly because of a lack of cultural context, and you can’t understand articles because of a lack of cognitive transfer. Hitting a bottleneck or using the wrong method—the resulting sense of boredom and frustration is truly torturous.
I was recently preparing for the Japanese JLPT exam. Besides the stress and frustration of exam prep, I also came up with a lot of weird and wonderful ideas. And the existence of AI and vibecoding allowed me to quickly realize these ideas and develop tools to assist my learning. This process can be abstracted as: “A human does something, generates a series of emotions, and uses a series of means to satisfy those emotions.”
AI can indeed complete tasks efficiently, but extremes breed reversals. AI’s “abundance, speed, and tirelessness” means it lacks the unquantifiable value unique to us humans. It cannot define the importance of a problem, much less feel the driving force behind solving it. It lacks the creative leap from zero to one and the understanding of imperfect or irrational needs. Code is a tool, but the pursuit of beauty and the exploration of meaning are value anchors unique to humans that cannot be quantified by data. What we create is not code, but digital bridges connecting people to people and people to the world. Through how many products can we feel the warmth of their creators?
In fact, when looking specifically at the User Experience (UX) aspect of product development, it’s easy to see that humans are the true measure. AI can generate code, but it lacks the subjective experience of an actual user. There are subtle and essential differences in the interaction methods of products tailored for different fields, pain points, and user groups. This is AI’s blind spot: it cannot perceive, let alone understand, the indescribable joy of “smoothness” and fluidity that flows from our fingertips when we experience a very instinctive interaction design.
In the era where AI and Vibecoding are all the rage, we are like passengers in a self-driving car. It handles the calculation and execution, but the decisions of where we go, where we stop, and which road we want to take—the desire for the journey and the ultimate decision-making power—will always be in human hands, always belonging to those carbon-based lifeforms with pain points and emotions. As long as humans are still the majority in this world, anyone trying to replace the people who implement requirements will have to replace humans first.
Finally, attaching a recent post by Vercel CEO Guillermo Rauch for mutual encouragement:
There are no limits anymore. Anyone can do anything. The only limiting factors are agency and ambition.
Never has a college degree, work experience, network, even the accumulation of knowledge been worth less.
You can just ship things.
https://twitter.com/rauchg/status/1999898954427961611?s=46&t=C46q-66ZOD0hggY5a-an4Q