The term “vibe coding” has emerged as a significant topic in software development, referring to a methodology where large language models (LLMs) generate all or part of a project’s code. Coined in February 2025 by former Tesla Autopilot Vision lead Andrej Karpathy, the concept encourages developers to “fully give in to the vibes” by relying on increasingly capable AI tools like Cursor Composer.
As LLM capabilities have advanced, so too has the popularity of this approach. Proponents highlight the democratization of software creation, citing examples such as a veterinary technician who built a custom application to manage her senior cat’s insulin treatment. However, the trend has sparked considerable backlash regarding code quality and security.
Critics argue that vibe coding produces vulnerable software and creates maintainability challenges. Because many practitioners lack deep coding knowledge, they may be unable to fix errors when AI assistance falls short. Research from Georgia Tech’s School of Cybersecurity and Privacy supports these concerns; in April, analysts identified 74 vulnerabilities in 43,000 security advisories traced to AI-generated code, including 14 classified as critical. Researchers estimated the actual number of vulnerabilities could be five to ten times higher due to underreporting.
Despite professional reservations, adoption among developers is growing. A SonarSource survey of 1,100 programmers found that 72% used AI coding tools daily, with approximately 42% of their codebase being AI-generated or assisted. That same cohort predicts AI-generated code will exceed half of their workflows by next year.
There remains a distinct difference between fully AI-generated code and AI-assisted development. Stack Overflow’s 2025 survey indicated that while 47.1% of respondents use AI tools daily, 72% stated vibe coding is not part of their workflow, with an additional 5% strongly disagreeing. Most professionals continue to utilize AI primarily for autocomplete, code review, and troubleshooting.
The shift also carries economic implications for the industry. Many tasks seniors now automate with AI were historically delegated to junior developers, representing a traditional entry point into the profession. As companies hire fewer junior coders, the pathway into the middle class via coding continues to narrow.
Leave a Reply