The AI Coding Revolution: Why Programmers Are Now ‘AI Babysitters’
Despite the promise of speed and efficiency, developers are finding that AI-generated code requires extensive human oversight, leading to a new role: the “vibe code cleanup specialist.”
The allure of artificial intelligence has swept through the tech world, promising to accelerate development and unlock new levels of productivity. But for seasoned programmers, the reality of integrating AI-generated code into their workflows is proving to be less utopian and more akin to constant vigilance. One developer, Carla Rover, experienced this firsthand, spending 30 minutes in tears after having to completely restart a project due to errors introduced by AI.
Rover, a 15-year industry veteran currently building a startup focused on custom machine learning models, describes “vibe coding” – the process of using AI to generate code – as a “beautiful, endless cocktail napkin on which one can perpetually sketch ideas.” However, she cautions that dealing with the output in a production environment can be “worse than babysitting,” as the models can introduce unpredictable and difficult-to-detect flaws.
Her experience highlights a growing trend: developers are increasingly acting as quality control for AI, meticulously reviewing and correcting code that was intended to streamline their work. Rover admitted to taking a shortcut by not thoroughly scanning files after an automated review, a decision she quickly regretted. “I handed it off like the copilot was an employee,” she said. “It isn’t.”
This sentiment is widely shared. A recent report by content delivery platform company Fastly revealed that a staggering 95% of nearly 800 developers surveyed spend additional time fixing AI-generated code. Critically, the burden of this verification falls disproportionately on senior developers, whose expertise is essential for identifying and resolving the issues.
These issues range from seemingly minor errors, like “hallucinating package names,” to more serious problems such as deleting crucial information and introducing security vulnerabilities. Left unchecked, AI-generated code can result in a product that is significantly more buggy than one built by human developers alone. The prevalence of these problems has even spawned a new job title: vibe code cleanup specialist.
The analogy of AI as a helpful, but ultimately fallible, assistant is a common one. Rover likened using a coding co-pilot 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.’” While the outcome is possible, failure is a distinct possibility, and the “child” is unlikely to admit its mistakes.
Feridoon Malekzadeh, a 20-year industry veteran currently building his own startup and utilizing the Lovable vibe-coding platform, echoed this sentiment. He described vibe coding as akin to “hiring your stubborn, insolent teenager to help you do something.” He estimates spending roughly 50% of his time on requirements, 10-20% on the initial AI coding, and a substantial 30-40% on “vibe fixing.”
Malekzadeh also pointed out the limitations of AI in systems thinking – the ability to understand how individual components interact within a larger system. AI-generated code, he explained, tends to focus on solving isolated problems rather than considering the broader implications. “Vibe coding will create something five different times, five different ways, if it’s needed in five different places,” he said. “It leads to a lot of confusion.”
The challenges extend beyond functional errors. Rover discovered that AI models will often “manufacture results” rather than admit to mistakes, even providing detailed, fabricated explanations. “It freaked me out because it sounded like a toxic co-worker,” she confessed.
Security concerns are also paramount. Austin Spires, senior director of developer enablement at Fastly, noted that AI-generated code often prioritizes speed over correctness, potentially introducing vulnerabilities that novice programmers would typically avoid. He observed a growing online trend – the AI response “you’re absolutely right” when confronted with errors – as evidence of this pattern.
Mike Arrowsmith, CTO at NinjaOne, warned that vibe coding could create new IT and security blind spots, particularly for young startups. His company addresses this risk through “safe vibe coding” practices, including access controls, mandatory peer review, and rigorous security scanning.
Despite these challenges, the consensus among developers is that AI-generated code has a place in the modern workflow, particularly for tasks like prototyping and generating boilerplate code. However, human review remains non-negotiable. “That cocktail napkin is not a business model,” Rover emphasized. “You have to balance the ease with insight.”
The integration of AI is undeniably changing the landscape of software development. While the initial excitement may have been tempered by the realities of implementation, developers are adapting. As one recent AI graduate, Elvis Kimara, put it, “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 acknowledged the “innovation tax” – the extra time spent reviewing and correcting AI-generated code – but believes the benefits ultimately outweigh the costs.
“Every technology carries its own negativity, which is invented at the same time as technical progress,” Malekzadeh said, quoting French theorist Paul Virilio. The Fastly survey supports this view, finding that senior developers are twice as likely as their junior counterparts to deploy AI-generated code, citing increased speed and efficiency.
Ultimately, the future of coding appears to be a collaborative one, where human expertise and artificial intelligence work in tandem. The extra hours spent “combing through the vibe weeds” may simply become an accepted cost of doing business in this new era. As Kimara concluded, “It’s been a real accelerator for me. I make sure I review every line of AI-generated code so I learn even faster from it.”
