Vibe-Coding Survey: 167 Engineers Weigh In

A recent survey of 167 software engineers explores the industry's reaction to 'vibe-coding,' a term coined by Andrej Karpathy for using AI to generate code. The results show a divided workforce: 46.9% of respondents feel they are 'keeping up' with AI tools, while smaller groups feel ahead or behind. Notably, 17.5% have opted out entirely, citing the tools' lack of advancement or steep learning curves.

Individual experiences vary widely. Some developers, like Ryan Shah, believe the future lies in reading AI-generated code, while others, like Javanie Campbell, warn that over-reliance on LLMs could endanger careers. Productivity remains a central debate; while some report efficiency gains, a METR study suggests AI assistance may actually increase task time due to review and prompting. Ultimately, the consensus is that human oversight remains essential.

The concept of vibe-coding—creating code using AI—has rapidly gained traction, even being named Collins Dictionary's Word of the Year for 2025. To understand how developers are adapting, a survey was conducted with 167 software engineers regarding their feelings on this shift.

The results indicate a complex landscape of sentiment:

• 46.9% (75 engineers) reported feeling like they are 'keeping up' with AI tools.
• 18% (30 engineers) felt they were ahead of the curve.
• 16% (27 engineers) admitted to feeling behind.
• 17.5% (28 engineers) are opting out of using AI code editing tools entirely.

Those opting out cited two main reasons: the tools were not advanced enough, or they took too long to learn. While the survey is not scientific, it offers a snapshot of an industry in transition. In follow-up conversations, eight engineers described their specific experiences, noting that while the tools are helpful, usage ranges from occasional assistance to being a 'lifesaver.'.

Debate is brewing regarding the long-term effects of AI on software engineering jobs. Some fear workforce shrinkage, while others view the technology as merely a new tool. Ryan Shah, a 23-year-old AI consultant, questions if traditional coding skills are still necessary. He utilizes tools like Cursor and Google's Antigravity, paired with Claude Opus 4.5, which he describes as 'midlevel engineer status.' However, Shah believes his ability to 'read' code keeps him employable.

Javanie Campbell, CEO of DevDaysAtWork, offers a stark warning: 'For people who turn to the LLM as the God or the expert, they will be replaced.' Conversely, Ryan Clinton, a 46-year-old developer, initially feared for his job but no longer does. He argues that experienced engineers focus on 'architecture and design,' while AI handles junior coding tasks, though he stresses that 'human intervention is also still routinely necessary.' Barry Fruitman, a 56-year-old developer, believes the threat is overstated and likely five to ten years away.

The promise of increased productivity is a major driver of AI adoption, yet data suggests the reality is nuanced. Ed Gaile, a principal solutions architect, claims AI tools have doubled or tripled his productivity, specifically praising the decrease in 'context switching.' He noted, 'I wish I had this 15 years ago.' Gus De Souza, a software architect, agreed that coding time is saved but noted that time is reallocated to reviewing AI-generated code, with real gains found in troubleshooting.

However, a study by METR challenges the narrative of immediate efficiency. The study found that developers using AI assistance spent 20% more time reviewing outputs, prompting the AI, or waiting for responses. Ultimately, the study concluded that AI-assisted developers were less productive than those working without AI. Shawn Gay, an R&D manager, expressed the difficulty of adapting, stating, 'I have decades of experience, so I feel like it's a huge effort to try to change the way my brain thinks about software.'

While the term vibe-coding was originally defined by Andrej Karpathy as developers 'fully giving in to the vibes' and forgetting the code exists, practitioners have varying definitions. Lara Fraser, a data analyst, does not consider herself a vibe-coder despite using tools like ChatGPT. She emphasizes that model generations vary significantly—calling GPT 5.1 'great' but 5.2 a 'disaster.'.

For Fraser, the distinction lies in maintenance capability: 'Anyone can create an app, but not everyone can maintain it.' She argues that if a developer cannot fix code when 'inevitably, something's going to break,' they are a vibe-coder. This highlights the distinction between using AI as a crutch versus using it as a tool for skilled professionals who retain the ability to debug and maintain complex systems.