Can Vibe Coding Build Production Software Without an Engineer?

An engineer's role in software development typically splits into two parts: deciding what is in scope and what gets built, and ensuring the code is of high quality and meets all necessary criteria. While AI models can write code, this alone is not sufficient to make software production-ready. Experienced engineers provide crucial oversight, setting priorities, testing, cost monitoring, and ensuring the final product meets expectations.

Vibe Coding's approach attempts to automate this process entirely with AI coding sessions, but this alone is insufficient. The key aspects that an AI cannot provide include setting project scope, determining project priorities, performing thorough testing, and monitoring costs. Without these human elements, AI-generated code can lead to unintended outcomes, unexpected costs, and ultimately a product that fails to meet user needs.

To improve Vibe Coding's output, a more structured approach is recommended. One potential solution is to implement a system called "Code Desks," a small open-source tool that simulates an office environment for AI coding sessions. Each session would involve a job description, with distinct roles such as product manager, developer, reviewer, or launch coordinator. Each role would have its own budget and a set of gates to ensure that all necessary checks are performed before moving on to the next stage.

In this system, the product manager would outline the project scope, the developer would build it, a separate reviewer model would check it, and a launch coordinator would handle the final aspects. This approach allows the AI to focus solely on code generation, while human oversight ensures quality, cost control, and proper decision-making. By combining AI's coding capabilities with human expertise, the resulting software is more likely to be production-ready and align with business goals.