Over the past few years, artificial intelligence (AI) has transformed how software is developed. Consequently, developers became impatient with such paradigms and celebrated new paradigms like “vibe coding.” Fastly recently conducted a survey that revealed an interesting and perhaps surprising finding. Almost 800 developers responded to the survey, and a whopping 95% said they’ve wasted additional time debugging AI-generated code. Such a phenomenon raises critical concerns about the effectiveness and correctness of AI generated programming solutions. These are just a few examples of how developers are actively grappling with the complexities that these technologies introduce.
Feridoon Malekzadeh, a physician and developer who is currently the CTO of a small enterprise actively developing his own startup. Besides using the vibe-coding platform Lovable, Jeremy figures he spends over half his work hours simply writing requirements. He personally devotes another 10% to 20% of his time to vibe coding. He invests another 30-40% of the project on bug fighting and removing scripts that AI code creates unnecessarily. With over 35,000 responses, this distribution highlights the top challenges developers face. Their goal is to use AI’s strengths while reducing its weaknesses.
These are not the only challenges vibe coding presents. Countless developers have shared their agony on the restrictive and odd behavior of AI models, such as ChatGPT. Malekzadeh goes so far as to compare vibe coding to parenting, meaning it’s an ongoing process of nudging and redirection. He reflects on the dual nature of technological advancement, quoting French theorist Paul Virilio: “Every technology carries its own negativity, which is invented at the same time as technical progress.” This insight highlights the need for cautious implementation when developing AI to ensure it is designed to enhance rather than replace the development process.
The Reality of AI-Generated Code
Developers are a rush to AI tools, such as Anthropic Claude. Instead, they frequently become trapped in a loop of watch-shame-watch. Austin Spires, senior director of developer enablement at Fastly, notes that many AI models tend to respond with phrases like “you’re absolutely right” when their mistakes are pointed out. This attitude adds to the already difficult task of delivering defect-free and reliable code.
Feridoon Malekzadeh highlights specific issues with vibe coding practices, explaining that “in the end, they do some of what you asked, some stuff you didn’t ask for, and they break a bunch of things along the way.” This level of unpredictability can result in major disruptions in development workflows, requiring labor-intensive reviews by engineers.
Vibe coding may sound harmless, but no one is more aware of its dangers than Mike Arrowsmith, chief technology officer at NinjaOne. He asserts that “vibe coding often bypasses the rigorous review processes that are foundational to traditional coding and crucial to catching vulnerabilities.” To address these issues, NinjaOne has implemented safe vibe coding methodologies. These practices include robust access controls that restrict the use of approved AI tools, a peer-review process, and extensive security scanning.
Balancing Innovation and Oversight
The development community is beginning to understand the importance of finding the right mix between harnessing AI’s potential and upholding robust accountability. Austin Spires suggests that “what often happens is the engineer needs to review the code, correct the agent, and tell the agent that they made a mistake.” This iterative process can be lengthy but is necessary to make sure that the content going out into the world is high quality.
React developer Carla Rover is familiar with the pains of vibe coding. Even today, she recalls with great detail a painful assignment that had her in tears for a full half hour after having to restart due to failures in AI-generated code. She likens using AI in coding to “giving a coffee pot to a smart six-year-old and saying, ‘Please take this into the dining room and pour coffee for the family.’” Whether poetic or prophetic, this analogy underscores the wild and chaotic nature of AI tools that, left unchecked, can lead to significant disruptions.
Elvis Kimara adds an innovative approach to this discussion. He argues that we’re soon going to see developers move from simply writing code to actually training AI systems. “We won’t just be writing code; we’ll be guiding AI systems, taking accountability when things break, and acting more like consultants to machines,” he explains. This change of roles requires a reconsideration of what it means to be a developer in a quickly evolving technological landscape.
The Future of Vibe Coding
Well, developers are addressing the creative chaos of vibe coding with both hands. Unlike some other professions, this ongoing trend is obviously changing their day-to-day work and duties. Feridoon Malekzadeh points out that “if you’re creating a feature that should be broadly available in your product, a good engineer would create that once and make it available everywhere that it’s needed.” With vibe coding generating multiple versions of solutions for the same problem—”five different times, five different ways, if it’s needed in five different places”—the likelihood of confusion increases significantly.
This new landscape creates both opportunities and challenges for developers as they work to produce and utilize AI tools in a way that best fits into their workflows. AI tools can drastically improve productivity, but they must be intentionally governed to avoid the dangers. As Mike Arrowsmith recommends, taking a proactive, standardized approach to vibe coding will allow companies to sidestep these pitfalls.
