Vibe coding: the AI-assisted development trend and its critics
Vibe coding refers to using large language models to generate code for software projects, a practice popularized by AI researcher Andrej Karpathy. Critics argue it produces vulnerable and hard-to-maintain code, with a Georgia Tech study linking dozens of security advisories to AI-generated code. Proponents counter that it democratizes app development, enabling non-programmers to build useful tools.
Karpathy introduced the phrase in early 2025, noting that advanced models like Cursor Composer with Sonnet made it feasible to relinquish direct code oversight. A Georgia Tech investigation examined 43,000 security advisories over a three-month window, attributing 74 vulnerabilities to AI-generated code, with 14 deemed critical. The researchers suspect the true count is five to ten times higher because only properly disclosed AI code could be traced.
Professional adoption is accelerating despite skepticism. A survey of 1,100 programmers who use AI found 72% rely on it daily, with AI now accounting for roughly 42% of their codebases and expected to exceed half within a year. However, a separate 2025 Stack Overflow poll showed most developers use AI for assistance like autocomplete and review, not full vibe coding, while senior staff increasingly automate tasks once assigned to junior hires.
The rise of vibe coding could reshape software development's labor market and security landscape. While it may empower non-professionals to create useful tools, widespread reliance on AI-generated code could introduce systemic vulnerabilities that are difficult to patch. Junior developers might find fewer entry-level opportunities as senior staff automate routine tasks. Conversely, this trend could force a reevaluation of coding education and professional standards, emphasizing oversight and verification over manual syntax mastery. Society may see broader innovation, but could also face trade-offs in software reliability and career pathways.