"Vibe coding" refers to software development where a large language model generates some or all of a project's code, letting the developer focus on describing intent rather than writing code line by line. The term, coined by AI researcher Andrej Karpathy in February 2025, has become increasingly popular and divisive: supporters say it democratises app-making for non-programmers, while critics warn it produces insecure, hard-to-maintain software and is eroding entry-level coding jobs traditionally used by junior developers to build careers.

Research lends some weight to the criticism. Georgia Tech researchers scanned 43,000 security advisories from early 2025 and traced 74 vulnerabilities directly to AI-generated code, 14 of them critical, with the true figure estimated at five to ten times higher. Professional adoption is also rising sharply: a survey of 1,100 programmers found 72% use AI tools daily and around 42% of their codebase is AI-generated or assisted, a share expected to exceed half within a year, though Stack Overflow's 2025 survey found most professionals still use AI mainly for autocomplete, review and troubleshooting rather than full "vibe coding".