unknown

Vibe coding, coined by Andrej Karpathy in early 2025, became a cultural phenomenon where developers fully delegate code generation to LLMs without reviewing the output. While YC's Winter 2025 batch showed 25% of startups had 95% AI-generated codebases with record growth, a backlash emerged by mid-2025 as security vulnerabilities (45% of AI code introduces flaws), technical debt, and accountability gaps surfaced. The response was 'context engineering' — systematically managing what context AI agents receive. 'Vibe engineering' is proposed as a synthesis: preserving developer flow and AI-generated speed while enforcing quality gates, spec-driven development, security scanning, code review, and human understanding of critical components. It draws clear boundaries (AI handles boilerplate; humans own auth, crypto, architecture), adapts pair programming to include AI as a third collaborator, and treats context as an engineering surface. The argument is that neither pure vibe coding nor traditional engineering alone is optimal — their deliberate synthesis produces better outcomes for security, maintainability, and developer satisfaction.

Vibe Engineering, Disciplined AI Software Development

Exploring the intersections of artificial intelligence, decentralised cognition, posthuman ethics, society and culture.