Understanding Vibe Coding in the Age of AI
Regardless of your engagement level with AI chatbots, the term “vibe coding” is likely familiar to you. For those not yet acquainted, vibe coding represents a novel software development technique that leverages large language models (LLMs) to create a significant portion, or even all, of the code needed for a project.
The phrase was introduced by Andrej Karpathy, a renowned AI researcher famed for steering Tesla’s Autopilot Vision initiative.
“There’s a new kind of coding I call ‘vibe coding,’ where you fully give in to the vibes, embrace exponentials, and forget that the code even exists,” Karpathy tweeted in February 2025.
His assertion underscores the remarkable capabilities of advanced LLMs—such as Cursor Composer with Sonnet—continuing to evolve.
Since Karpathy’s pronouncement, the proficiency of LLMs in code generation has improved substantially, resulting in a burgeoning interest in vibe coding. However, with burgeoning trends often comes scrutiny and pushback.
Critiques of Vibe Coding
Advocates of vibe coding contend that LLMs democratize the software development landscape, enabling individuals to create their own applications.
An example comes from my neighbor, a former veterinary technician who crafted an app that simplifies monitoring her elderly cat’s insulin administration. Conversely, detractors argue that vibe coding yields inherently vulnerable software.
The extensive use of such coding practices can render a codebase challenging to maintain, particularly if a vibe coder lacks the requisite knowledge to rectify errors when an LLM falls short.
Preliminary investigations lend credence to these concerns. In April, researchers from the School of Cybersecurity and Privacy at Georgia Tech conducted a survey of 43,000 security advisories and identified 74 vulnerabilities directly attributable to AI-generated code.
Among these, 14 were categorized as critical security threats. While this may appear minimal, it is essential to note that these advisories spanned only a three-month timeframe.
Moreover, the researchers estimated that the actual figure of AI-induced vulnerabilities could be five to ten times greater, given that they only accounted for code disclosed as AI-generated.
This situation would be less alarming if vibe coding were primarily a pastime of amateurs. However, increasing evidence indicates that more professionals are embracing vibe coding, despite reservations expressed by themselves and their colleagues.
A recent study involving 1,100 professional developers revealed that 72 percent utilize AI tools daily, with approximately 42 percent of their code comprising either AI-generated or AI-assisted content.
This demographic anticipates that AI-generated code will constitute over half of their codebase within the coming year.
Differentiating AI-generated Code from AI-assisted Code
When discussing vibe coding, it is essential to distinguish between code that is entirely produced by an LLM and code that has been enhanced or refined with AI assistance.
According to Stack Overflow’s 2025 developer survey, 47.1 percent of respondents reported using AI tools daily.
However, 72 percent mentioned that vibe coding does not form a part of their development methodology, with an additional 5 percent vehemently asserting it is not part of their professional practice.
Although these numbers may have evolved, as of mid-2025, the majority of professionals seem to utilize AI tools for tasks such as code completion, review, and troubleshooting.
Furthermore, it is crucial to recognize the ramifications of this trend: many responsibilities traditionally delegated to junior developers are now being automated by senior programmers employing AI.
Coding, once regarded as a dependable pathway for young individuals to enter the middle class, appears to be losing this potential, as companies are hiring fewer junior coders than ever before.
Source link: Aol.com.
