Vibe coding an application or an AI agent sounds too good to be true. A single developer prompts their way through a plan and has the code for a working prototype in minutes instead of days. Fixes and improvements come through iteration until the developer is satisfied with the results. According to the State of Code Developer Survey, 48% of respondents have adopted vibe coding for new projects, and 62% say it’s effective. But 96% don’t fully trust that AI-generated code is functionally correct.
There’s reason for concern. The State of AI Versus Human Code Generation report shows AI pull requests have 1.4 times more critical issues and 1.7 times more major issues than when developers create the pull request. “Developers write bad code, and AI helps them write bad code faster,” says Brett Smith, distinguished software developer at SAS. “The fallacy lies in organizations assuming AI-generated code is inherently correct instead of subjecting it to the same testing, validation, and quality standards as every other release.”
Subjecting AI code generation to the same testing as traditional code may not be sufficient. One report states that 41% of all global code is now AI-generated. The code reviews, security scans, and testing practices deployed for developer-generated code may not support the volume, complexity, and new risks from AI-generated code. These metrics don’t suggest that vibe coding is bad, but they do suggest developers should slow down and avoid costly mistakes. Here are seven bad habits to avoid.
