In the constantly evolving world of technology, a new fad has developed that goes by “vibe coding.” It’s revolutionizing the way developers are approaching their projects. That’s why Fastly recently commissioned a survey. They discovered that most of the time, 95% of the time for nearly 800 developers, goes to re-fixing AI generated code. Even veteran developers are learning to navigate the pitfalls brought on by AI-assisted programming. Though it may come as a surprise, this shift has them worried too.
Feridoon Malekzadeh, a developer who is in the process of developing his own startup, offered interesting perspectives based on his experience with vibe coding. He estimates that he allocates around 50% of his time to writing project requirements, while dedicating 10% to 20% to actual vibe coding. That last 30-40% goes to fixing bugs and dealing with the “junk yard script” formed by AI produced code. Navigating this new change is the challenge. Developers no longer just need to know how to write code, but how to handle the potentially harmful output of intelligent systems.
These challenges with vibe coding are hardly new to Malekzadeh. Carla Rover, another developer going through the process, explained how emotionally draining the process was. She thought back to when she collapsed in tears for an hour after needing to reset a project because her vibe-coding attempts encountered roadblocks. Austin Spires, senior director of developer enablement at Fastly, likened the experience to “hiring your stubborn, insolent teenager to help you do something.”
The AI Dilemma
With AI models like Anthropic Claude continuing to permeate the overall coding process, developers are presented with new and unique challenges. These models are often in concordance when asked to predict their own mistakes. This makes it difficult for engineers to know what’s an acceptable output and what’s unacceptable.
“You’re absolutely right.” – Multiple AI models
To Spires, the experience working with AI tools has been that it forces engineers to take a step back and look at the code that has been produced. “What often happens is the engineer needs to review the code, correct the agent, and tell the agent that they made a mistake,” he stated. This new burden disrupts the typical development cycle, which can overwhelm developers who are always operating under tight deadlines.
Malekzadeh highlighted another drawback of vibe coding: its propensity to create confusion. He noted that these models are reducing the prevalence of boilerplate code, or code created for similar tasks across varying contexts.
“Vibe coding will create something five different times, five different ways, if it’s needed in five different places. It leads to a lot of confusion, not only for the user, but for the model.” – Feridoon Malekzadeh
This redundancy only adds to the already complex coding environment. It further places additional burden on developers who must woefully interpret the AI’s results.
Navigating Mentorship in a Changing Landscape
Since the introduction of vibe coding, senior developers such as Malekzadeh have witnessed a change in mentorship dynamics on teams. In the past, accomplished coders led younger co-workers through difficult procedures. Even senior developers are rapidly adopting new, AI-powered approaches. Instead of the usual route of mentorship by experienced professionals, they’ve opted for AI models to take up that role.
Malekzadeh explained that most senior developers do not know how to vibe-code. He further echoed concerns that many are hesitant to publicly give their endorsement. He noted that this transition means less interaction, mentorship, and day-to-day direction for junior engineers.
Just ask young engineer Elvis Kimara, who has lived through this tumultuous scene. He noted that senior devs can be less helpful with vibes coding these days than they used to be. This pivot in support is sharply evidenced. Perhaps most alarming, this disconnect threatens the skills- and knowledge-transfer imperative to cultivate the next generation of talent that will keep the field growing and vibrant.
To combat these threats, Mike Arrowsmith, chief technology officer at NinjaOne, calls for “safe vibe coding.” Their approach is to require access controls on all approved AI tools, as well as requiring mandatory peer reviews and security scanning processes. By creating these outreach protocols, businesses will be better equipped to mitigate risks associated with AI-generated outputs. Meanwhile, they should focus on using these tools in a more efficient way.
“Vibe coding often bypasses the rigorous review processes that are foundational to traditional coding and crucial to catching vulnerabilities.” – Mike Arrowsmith
The Future of Development
As developers learn how to create in this new world of vibe coding, there’s a scary and exciting future ahead – full of promise and pitfalls. These tools powered by AI almost double their productivity. Now, engineers are held to an even higher standard of accountability as they tiptoe through the deadly outputs.
Malekzadeh stressed that developers will more often be in the role of shepherds of AI systems as opposed to the ones creating the code. “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,” Kimara added.
Feridoon Malekzadeh referenced French theorist Paul Virilio when discussing this duality of progress and its consequences: “Every technology carries its own negativity, which is invented at the same time as technical progress.” This view highlights the need to prioritize both innovation and prudent oversight in the development of software.