Vibe coding vs agentic engineering: why terminology matters for AI‑assisted development

Vibe coding, coined by Andrej Karpathy, describes a reckless workflow where developers prompt an AI, accept its output without review, and iterate by pasting errors back in. While useful for quick prototypes, this approach leads to unreliable, unmaintainable code and can cause serious security and performance issues. The emerging discipline of agentic engineering replaces vibe coding with a disciplined process that starts with a design spec, breaks work into scoped tasks, reviews AI output, and relies on rigorous testing to validate results. In agentic engineering, the AI acts as a junior developer, the human remains the architect, and the codebase is governed by version control, CI, and monitoring. Senior engineers benefit most from this workflow because they can efficiently review AI output and enforce architectural and security standards, whereas junior developers risk skill atrophy if they rely on AI without understanding the code. The terminology shift from “vibe coding” to “agentic engineering” clarifies expectations for stakeholders and encourages systematic evaluation of AI‑assisted workflows. Training programs must adapt to emphasize fundamentals, architectural thinking, and security awareness alongside AI tooling. Adopting agentic engineering can accelerate development while maintaining quality and reliability.