Vibe Coding Has a Quality Problem — And the Numbers Are Stark

A December 2025 analysis of 470 GitHub PRs found AI-generated code was 1.7× more likely to have major issues and 2.74× more prone to security vulnerabilities than human-written code. LLMs generate vulnerable code at rates between 9.8% and 42.1% across benchmarks. The practitioner consensus: robust CI (automated tests on every commit, code style enforcement, staging deployments) is non-negotiable when AI is contributing significant code volume. Teams that ship well are orchestrators, not prompters.

---

Sources: Hexaware Vibe Coding Best Practices · Softr Guide