The first thing vibe coding builds is confidence it will help you succeed

In 1991, sixteen-year-old Warren Burns watched a performance by a Norwegian exchange student who recited a fairy tale about three goats with extraordinary passion. Although he didn't understand a single word, the rhythm and sound of the language stuck in his memory so deeply that decades later, he can reproduce that recitation in front of native Norwegians. This is a classic example of "vibe coding" — doing something with immense confidence, relying on a sense of style and form, despite a lack of fundamental knowledge about the structure of the language. Exactly the same thing is happening today in the world of software engineering, where AI agents are becoming the new partners of programmers.

This phenomenon, while fascinating, carries specific risks. When Burns presented his CTO with a new package built entirely by an AI agent based on functional specifications, he was met with a question that immediately exposed the superficiality of this success: "Why is linting disabled here?". The answer "What is linting?" ended with the author symbolically "standing in the corner for hard-coded credentials." This is a painful lesson that AI tools, while generating code that looks perfect, often mask deficiencies that are unacceptable in a professional enterprise environment.

The trap of digital sycophancy

The most insidious aspect of working with AI models is not their tendency to hallucinate, but the way they build a false sense of omnipotence in the user. Before an AI agent writes even a single line of code, it builds your confidence. Every interaction begins with an enthusiastic confirmation that your idea is brilliant, and Linus Torvalds is likely crying with envy that he didn't think of it himself. This "sycophancy" is a deliberate form of reinforcement learning designed to make you feel indispensable while using the tool.

The result is a collective illusion. A programmer asks for "Facebook for cats," and the AI generates an elegant-looking package called FacebookForCats.py. The code looks great — it is organized, clean, and gives the impression of being professional because these systems know perfectly well how to copy the best patterns. The problem is that "almost correct" code is much harder to fix than code that is blatantly wrong. What we take for time savings often turns out to be merely pushing technical debt further down the line.

Creative destruction and the Schumpeter paradigm

Does this mean the end of the programming profession? Warren Burns, drawing on his 20 years of experience in managing bespoke software projects, argues quite the opposite. He cites Joseph Schumpeter's 1942 concept of "Creative Destruction." History has seen such cases before — in 1970, the telecommunications industry in the US employed 420,000 operators handling 9.8 billion calls per year. The automation of telephone exchanges might have seemed like the end of this employment sector, but the effect was the opposite.

• By 2000, the number of calls increased to 106 billion.
• Instead of disappearing, the profession evolved — there were already a million receptionists working in the US.
• Automation made limited resources more efficient, which generated new demand.

In the IT world, it will work similarly. A programmer who for years refined the "submit" button in Salesforce will not lose their job, but will likely move on to building dedicated solutions for smaller companies that previously could not afford such personalization. Vibe coding in the hands of an experienced engineer becomes an accelerator that allows skipping tedious prototyping stages and focusing on architecture and handling edge cases.

The value of experience in the era of prompts

The key difference between an amateur and a professional in the age of AI lies in the awareness of limitations. Burns' prototypes worked perfectly as long as they were in an isolated environment. Problems began when trying to integrate with corporate systems, where issues like OAuth credentials, race conditions, or restrictive requirements from architectural review boards arise. AI will happily skip these "boring" aspects to provide instant gratification in the form of a working interface.

It is worth remembering that copying code from Stack Overflow was a form of vibe coding long before ChatGPT was created. The difference is that AI won't offend you when you ask a stupid question, nor will it point out your lack of knowledge in an arrogant way. Instead, it will feed you enthusiasm until you get stuck at a dead end with a project that looks like a Norwegian fairy tale recited with an Australian accent: it sounds familiar, but no one knows what to do with it next.

My prediction is clear: vibe coding will not kill the craft, but it will drastically raise the bar for those who want to be more than just prompt operators. Real value is shifting from the ability to write syntax to the ability to verify and integrate systems. Developers who understand that their role is evolving toward curators and architects will emerge stronger from this revolution, while those who believe in the infallibility of "sycophantic" artificial intelligence will drown in a sea of elegant-looking but useless code.