What is Vibe Coding? A New Trend Changing the Future of Programming
In recent years, a major shift has been occurring in the world of software development. One of the symbolic keywords for this is "Vibe Coding." With the evolution of AI-powered code generation tools, a new style is spreading where engineers focus not on "writing" code, but on "giving instructions" to write it. Vibe Coding, proposed by Andrej Karpathy, is based on the idea of "forgetting the existence of code and proceeding with development while simply instructing the AI." Major venture capital firm Y Combinator (YC) has already taken notice, and many startup founders are incorporating this method into their daily routines.
In this article, we will explain what Vibe Coding is, how founders currently in YC's batch are perceiving it, and what impact it has on engineering, hiring criteria, and organizational structure.
1. What is Vibe Coding?
1-1. Background and Definition
The term Vibe Coding gained attention following a social media post by Andrej Karpathy. Karpathy describes the "Vibe Coding" approach as one where, instead of the traditional process of "engineers deeply designing structures and implementing them by hand," one "conveys the 'vibe' to the AI and then checks and corrects the code output by the AI."
Behind this lies the progress of code generation support tools utilizing LLMs (Large Language Models). By using advanced tools such as ChatGPT, OpenAI Codex, Cursor, and WindSurf, engineers are being freed from the task of inputting code and are increasingly able to focus on "thinking about the essential value of the product."
1-2. How Vibe Coding Came to Be
Since 2023, the explosive spread of ChatGPT and various AI coding assistants has dramatically increased software development productivity. Because AI can now instantly take over tasks that previously required high programming skills and long hours, cases have emerged where development speed has increased tenfold, or in some cases, a hundredfold. The foundation of Vibe Coding is that "how you prompt the AI and scrutinize its output" has become more directly linked to results than spending time on the "code itself."
2. AI Utilization and the Evolution of Engineering
2-1. From "Writing Code Fast" to "Creating Products Fast"
One YC partner mentioned in an interview that "the value of the 'skill to write code fast,' which was once considered important in software development, is changing significantly due to AI."
The paradigm that "being able to code quickly" equals "being a talented engineer" is now beginning to collapse, and instead, "the ability to discern what kind of product should be created" and "the ability to debug AI output and optimize systems" are becoming more important.
2-2. From AI as an "Assistant" to AI as an "Implementer"
In the past, AI coding support tools were mainly focused on completion features and template output, requiring engineers to lead the process to completion themselves. However, current LLM-generation tools are reaching a level where they can "effectively generate almost the entire amount of code at once." Some YC founders have noted that "my main job has become thinking and reviewing rather than writing code myself," marking a major change in the workflow.
3. YC Founders on Vibe Coding
When we conducted surveys and interviews with dozens of startup founders currently in the YC batch regarding "how they use AI" and "whether Vibe Coding will truly become mainstream," we gathered some very interesting feedback. Below are some excerpts.
3-1. Explosive Improvement in Development Speed
Founder A (Founder of Outlet)
"By using AI coding tools, the capabilities of engineers will increase tenfold, and we are entering an era where everyone becomes a '10x engineer.' That is precisely why I think the human 'eye for products' will become even more important in the future."Founder B (Founder of TrainLoop)
"I felt it was 10 times faster from six months ago to a month ago, but from a month ago to now, I've experienced a sense of speed closer to 100 times faster. I myself feel like I'm working more as a 'product manager' than an engineer now."
3-2. Changes in "Attachment" to Code
Founder C (RB of Copycat)
"My attachment to my own code has faded. It's faster to rewrite everything if there's a problem than to fix a bug. Since the cost of code regeneration is overwhelmingly low, I feel that even the concept of refactoring is changing."Founder D (Avi from Asra)
"I hardly write code anymore. My role is just to think, review the outputted code, and provide instructions for corrections."
3-3. A New Style of Parallel Work
Founder E (Yoav from cix)
"I often launch two Cursors and send prompts for two different features in parallel. I might work on the frontend in one and the backend in the other, generating and modifying them simultaneously. I even think three might be fine."
In this way, many founders are practicing Vibe Coding and experiencing a level of efficiency and freedom that is completely different from existing software development.
4. Challenges in Debugging and System Construction
4-1. AI's Weak Point: Debugging Ability
No matter how powerful Vibe Coding is, there are areas where current LLM tools struggle. The prime example is debugging. Many founders agree that "humans still have to handle bug fixes" and "even if you try to leave debugging to AI, it is not good at correctly identifying the location of bugs."
However, some have a new perspective: "It is faster in the end to regenerate the code from scratch than to intentionally fix bugs." Because code can be rewritten at lightning speed, the workflow is to perform a rolling reset and leave it to the AI to handle in bulk, rather than painstakingly debugging in the traditional way.
4-2. Problems with Building Scalable Systems
There is also the question, "Vibe Coding is strong for zero-to-one, but what about after it scales?" In fact, as seen in Twitter's scaling issues and the case where Facebook had to create a special compiler (like HipHop) from PHP, challenges that cannot be easily overcome will inevitably arise as systems grow larger.
Current LLMs cannot "understand" an entire massive codebase, and while they are incredibly convenient locally, building a system from the perspective of overall optimization is still a major role for humans.
5. Changes in Hiring and Evaluation Criteria
5-1. Evaluating "Talent That Can Collaborate with AI"
Traditional companies have emphasized "algorithmic understanding" and "coding speed." However, in the Vibe Coding era, the key is "how well you can prompt AI and appropriately manage quality."
YC partners feel that past interview methods (such as having candidates write algorithms on a whiteboard) have become hollow, and they say that evaluating "how they actually solve given problems using AI tools" is more practical.
5-2. The Importance of "Taste" and "Systems Thinking"
Survey results also show that multiple founders emphasize the importance of "what to build" and "the ability to discern what users truly want." In other words, an understanding of product design, user experience, and requirements definition will become a new evaluation axis for software engineers.
At the same time, classical computer science thinking—such as "how to build and optimize scalable large-scale systems and how to control technical debt"—remains indispensable. The fact that Twitter and Facebook ultimately needed engineers with advanced systems thinking to support intense loads is a lesson that remains unshakable even in the AI era.
The new era of software development brought about by Vibe Coding is by no means a temporary boom; it is already beginning to take root as a major coding method in the field. As multiple YC founders show, comments like "I hardly write code anymore," "I have AI implement two features at the same time," and "regeneration is faster than bug fixing" are now part of the daily landscape.
On the other hand, not all challenges can be automated, and the following points are expected to remain important in the future.
Debugging Skills and Systems Thinking
Advanced human expertise is essential for performing bug fixes and optimizing large-scale infrastructure, which AI is not good at.Product Development Skills to Accurately Grasp User Needs
If machines generate code, what is required of engineers is "the ability to discern what features are truly necessary."Redefining Hiring and Evaluation Axes
Skills to operate AI, design correct prompts, and manage quality will become core indicators in engineering teams.
What is required in the Vibe Coding era is not the "speed" of writing code, but the "sense" to oversee the entire product and "systematic thinking." As YC partners agree that "it would be more surprising if Vibe Coding did not become mainstream,"this trend will continue to accelerate, and the gap between those who are left behind and those who are ahead will continue to widen further.
While development productivity and speed are increasing dramatically, it is the talent that can deeply understand systems and compensate for AI's weaknesses that will create long-term value. Keeping this duality in mind, we need to rethink how we update engineering for the new era. Isn't now the very turning point for that?
Related Articles