Experts split on “vibe coding” as developers trust AI without oversight

“Vibe coding,” a term coined by AI researcher Andrej Karpathy four months ago, describes a hands-off approach to AI-assisted programming where developers let AI generate code without constant oversight or manual tweaking. The concept has sparked debate among software engineering experts about the balance between AI autonomy and human understanding in code development, with implications for how developers will work alongside increasingly capable AI systems.

What you should know: Vibe coding represents a fundamental shift from traditional AI-assisted programming by encouraging developers to step back and trust AI systems to handle code generation independently.

• Unlike standard AI-assisted coding, vibe coding requires developers to “let the AI fully take control and refrain from checking and directly tweaking the code it generates as you go along—surrendering to the vibes,” according to MIT Technology Review’s Rhiannon Williams.
• The approach often means developers don’t read or try to understand the generated code, relying instead on the AI’s output and their intuitive sense of whether it feels right.

What experts are saying: Industry professionals at a recent Imagination in Action panel offered mixed perspectives on vibe coding’s current viability and future potential.

• “You don’t even have to read the code that it produces. You … just basically teleprompt to the system, it generates a part of the code for you, and then, quite often, because developers are notorious for being lazy, you don’t even read the codes,” explained Artem Lukoianov of Hyperskill, an educational technology platform.
• Zach Lloyd warned about the double-edged nature of this approach: “So in the terminal, we see it goes beyond producing code, to this whole feeling of: ‘let me see if the AI can just fix this thing for me, and maybe I won’t have to understand exactly what it’s doing.'”

Current limitations: Several experts highlighted significant challenges with fully autonomous AI coding systems.

• Heena Purohit argued that AI systems struggle with “distance thinking,” or understanding how various components of a system interact with each other.
• “Sure, you can have millions of lines of code be generated in minutes or seconds, but you still need to understand what the code is actually doing, so that you can troubleshoot it and debug it when you need to,” Purohit noted, suggesting scaling could become problematic.

The optimistic view: Some experts believe the technology is nearly ready for widespread adoption, with the main challenge being proper system engineering rather than AI capability.

• “(Vibe coding capability) is already good enough for us to stop coding anything,” Lukoianov said. “To me, it’s more the question of how we engineer the system around this… how do we … provide the correct information to the LLM, how do we summarize our code base, and how do we provide the right tools to the LLM to actually perform … better?”
• He compared the evolution to how developers no longer need to think deeply about hardware or operating systems unless working in specialized fields.

Why human expertise still matters: Despite AI’s growing capabilities, panelists emphasized that full-stack developers remain valuable for managing system complexity and constraints.

• “If you’re using (tools) to synchronize different processes, and you have a million processes, your system is going to break,” Aldo Pareja explained. “You need to understand these constraints.”
• The consensus suggests that while AI might handle 80% of coding tasks autonomously, the remaining 20% requires human oversight for context, debugging, and system-level understanding.