The development style of issuing natural language instructions to an AI to generate executable programs has been dubbed "vibe coding" by Andrej Karpathy (former OpenAI researcher), sparking widespread debate in the tech community. To scientifically investigate the human background factors that influence the success of this "vibes-based" development, a research team at ETH Zurich conducted a detailed study. Their findings, accepted for the premier Human-Computer Interaction conference CHI 2026, provide critical insights into the evolving skill set required for the AI era.

The study involved 100 university students tasked with replicating and improving a meal-planning application using only AI chat interfaces. Participants were prohibited from touching the source code directly, forced instead to rely entirely on dialogue with the AI for debugging and feature additions. Statistical analysis of the results revealed a strong correlation between successful outcomes and two specific areas: academic performance in Computer Science (CS) and the ability to explain concepts clearly in writing.

Notably, CS expertise functioned as an independent predictor of success, separate from general cognitive ability or raw intelligence. This suggests that even when AI handles the granular details of code generation, a fundamental grasp of algorithms and system design remains indispensable. Without these core concepts, developers cannot effectively course-correct the AI's logic when it deviates from the intended path. What may look like simple prompting actually requires sophisticated technical judgment beneath the surface.

Furthermore, writing proficiency has emerged as a vital new development skill in its own right. The ability to logically organize and communicate the current state of a system and the desired ideal state to an AI is essentially replacing the traditional act of manual coding. This research highlights that CS knowledge and writing ability—two seemingly disparate skills—are becoming the dual engines of expertise in modern software development, presenting a significant shift for both education and the industry.

The development style of issuing natural language instructions to an AI to generate executable programs has been dubbed "vibe coding" by Andrej Karpathy (former OpenAI researcher), sparking widespread debate in the tech community. To scientifically investigate the human background factors that influence the success of this "vibes-based" development, a research team at ETH Zurich conducted a detailed study. Their findings, accepted for the premier Human-Computer Interaction conference CHI 2026, provide critical insights into the evolving skill set required for the AI era.

The study involved 100 university students tasked with replicating and improving a meal-planning application using only AI chat interfaces. Participants were prohibited from touching the source code directly, forced instead to rely entirely on dialogue with the AI for debugging and feature additions. Statistical analysis of the results revealed a strong correlation between successful outcomes and two specific areas: academic performance in Computer Science (CS) and the ability to explain concepts clearly in writing.

Notably, CS expertise functioned as an independent predictor of success, separate from general cognitive ability or raw intelligence. This suggests that even when AI handles the granular details of code generation, a fundamental grasp of algorithms and system design remains indispensable. Without these core concepts, developers cannot effectively course-correct the AI's logic when it deviates from the intended path. What may look like simple prompting actually requires sophisticated technical judgment beneath the surface.

Furthermore, writing proficiency has emerged as a vital new development skill in its own right. The ability to logically organize and communicate the current state of a system and the desired ideal state to an AI is essentially replacing the traditional act of manual coding. This research highlights that CS knowledge and writing ability—two seemingly disparate skills—are becoming the dual engines of expertise in modern software development, presenting a significant shift for both education and the industry.