Recently, tech influencers have been hyping up Vibe Coding, which sounds pretty cool. The general idea is that you don’t need to know rigid programming syntax—as long as you have a feeling or an idea, AI can help you write the code.
Many people claim that this completely removes the barrier to programming, making everyone a programmer. But based on my personal experience, this has been somewhat overblown.
In fact, over the past few years, programming tools have consistently been at the forefront of AI applications. Before general-purpose agents like Manus entered the public eye, tools like Cursor, Winsurf, and Copilot were already widely used by engineers.
Every time a new tool emerges, a wave of influencers proclaims that programming is dead and that soon, ordinary people will be able to develop software as easily as ordering takeout. But is that really the case?
As Long as Code Exists, the Barrier Remains
If you open any AI-powered programming tool, such as Cursor, Antigravity, Trea, or Tongyi Lingma, you’ll still see lines of code filling the screen.
This is actually a very important signal. The reason code is still displayed is that AI-generated output isn’t yet 100% reliable. In professional software development, code is the only standard for verification. It serves the functions of validation, debugging, accountability, and maintenance. Once a system becomes complex, there must be an intermediate form that can be inspected and corrected. At this stage, code is precisely that intermediate form. If the AI makes a mistake or the functionality doesn’t meet expectations, you still have to go back and figure out where the code went wrong.
Any user-facing tool product will not expose its core business logic. If AI editors were truly fully reliable, there would be no need to show you the code—just like when you use Meitu to edit photos, it doesn’t show you the algorithmic instructions for image processing.
So, current Vibe Coding is actually serving those with basic engineering judgment. AI helps developers save the effort of memorizing syntax and writing boilerplate, but it can’t replace professional developers in thinking through the logic of an entire system.
Therefore, a more accurate positioning for Vibe Coding is to reduce the operational burden on professional developers, not to allow non-professional users to bypass engineering complexity. It reduces boilerplate labor and compresses implementation time but still requires users to have the ability to judge right from wrong.
The Steering Wheel in Autonomous Driving—You Still Have to Hold It
To understand this, we can draw an analogy to a familiar scenario: autonomous driving.
In closed, simple, and standardized conditions like highways, autonomous driving performs exceptionally well—AI can help you stay steadily in your lane. But once you enter complex urban traffic, it’s prone to errors.
That’s why even cars equipped with the most advanced systems and running the most sophisticated autonomous driving technology still retain steering wheels, accelerator pedals, and brake pedals. This design ensures that when the AI encounters situations it can’t handle, the driver can quickly take over to prevent accidents.
Similarly, with Vibe Coding, AI excels at writing small tools or standalone webpage features. But when it comes to developing complex software with multiple integrated modules, issues like AI hallucinations and attention drift emerge. This complexity isn’t just about the volume of code but involves expanding states, increasing implicit constraints, and early decisions being magnified later on. Current models struggle to maintain global consistency over the long term and lack reliable risk perception and self-degradation capabilities. Errors generated early on can accumulate over time, eventually causing the program to fail.
AI editors expose code because humans are still seen as the fallback, requiring developers to take over or verify at any time. Just like cars with autonomous driving still have steering wheels and pedals, drivers need to be ready to take control.
Should Ordinary People Learn Vibe Coding?
Despite this, it doesn’t mean Vibe Coding is overhyped to the point of being useless.
I believe Coding Agents are likely to become key middleware for various AI Agents in the future. Code is a universal, composable, and verifiable way to express actions. Translating high-level intent into executable structures is an unavoidable part of intelligent systems.
However, for a considerable time, it’s better suited as a constrained execution module rather than an independent entity handling complex goals.
From a product experience perspective, ordinary people don’t need to learn how to use Coding Agents. Any truly mature product aimed at the general public should aim for zero learning costs. If users need to understand context management, failure rollbacks, or handling bugs caused by model biases, it means the product hasn’t been fully encapsulated.
If you’re learning complex tool operations just out of FOMO (fear of missing out), I think it’s completely unnecessary. Tools are constantly evolving, and simplicity is an inevitable trend. If a tool still requires you to deliberately learn how to use it, it means it’s not yet good enough.
However, if you want to use these tools to understand how programs work, that’s highly valuable.
The future is a world driven by AI and data. While you may not need to write code yourself, understanding how programs break down problems and what logical constraints are is like learning to drive in the age of autonomous driving. It’s not about becoming a professional driver but about understanding the rules of how the world operates, enabling better judgment when problems arise.
From the Decline of Prompt Engineering to the Rise of Vibe Coding
Looking back at the hype around prompt engineering in 2023, which has since noticeably cooled, we can see a similar pattern.
So-called techniques are essentially temporary patches when a system isn’t yet mature. Once model capabilities improve, these techniques are quickly absorbed. What remains isn’t the method of writing prompts but more fundamental abilities: the ability to express needs clearly and the ability to evaluate results.
Whether it’s prompt engineering or Vibe Coding, what truly determines effectiveness is never the technique itself but whether you know what you want and what results are acceptable.
Vibe Coding does make development faster, but it doesn’t make programming itself disappear. It’s more like giving professional chefs a set of more modern cooking tools, but these tools can’t turn someone who’s never been in a kitchen into a Michelin-star chef overnight.
AI hasn’t created entirely new barriers; it amplifies the differences that already exist between people. Those who express themselves clearly will be more efficient, while those lacking judgment will only get unusable results faster.
AI is just a tool. In the end, the person responsible for setting goals, making judgments, and verifying results is always you.