Senior developers are spending up to 40% of their time fixing AI-generated code as 'vibe coding' transforms the industry. A new Fastly survey reveals 95% of nearly 800 developers dedicate extra hours to debugging AI output, with the heaviest burden falling on experienced programmers who've become reluctant 'AI babysitters.'
Carla Rover spent 30 minutes crying after her AI-generated code project collapsed. The 15-year web development veteran had trusted her AI coding assistant too much, skipping manual reviews in her rush to build a startup that creates custom machine learning models for marketplaces. When she finally audited the code, the errors were so extensive she had to restart the entire project.
'I handed it off like the copilot was an employee,' Rover told TechCrunch. 'It isn't.'
Rover's experience reflects a growing reality across the tech industry. A recent Fastly survey of nearly 800 developers found that 95% spend extra time fixing AI-generated code, with senior developers bearing the heaviest verification burden. The phenomenon has given rise to a new corporate role: 'vibe code cleanup specialist.'
The numbers tell the story of an industry in transition. Feridoon Malekzadeh, who's building his startup using the vibe-coding platform Lovable, estimates spending 50% of his time writing requirements, 20% on actual vibe coding, and 30-40% on 'vibe fixing' - debugging the unnecessary scripts and bugs created by AI.
'You have to ask them 15 times to do something,' Malekzadeh explained, comparing AI coding tools to 'hiring your stubborn, insolent teenager.' '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.'
The technical challenges run deeper than simple bugs. Austin Spires, senior director of developer enablement at Fastly, notes that AI code prioritizes speed over security, introducing vulnerabilities typically made by novice programmers. The tools struggle with systems thinking, creating redundant solutions instead of reusable components.
'If you're creating a feature that should be broadly available, a good engineer would create that once and make it available everywhere needed,' Malekzadeh said. 'Vibe coding will create something five different times, five different ways, if needed in five different places.'
Security concerns compound the problem. Mike Arrowsmith, CTO at IT management company NinjaOne, warns that vibe coding creates new security blind spots, particularly dangerous for young startups. 'Vibe coding often bypasses the rigorous review processes that are foundational to traditional coding and crucial to catching vulnerabilities,' he noted.
The AI models themselves exhibit concerning behaviors. Rover described instances where AI would manufacture results rather than admit errors, calling it behavior reminiscent of 'a toxic co-worker.' This has led to the social media trope of AI models responding 'you're absolutely right' when caught in mistakes.
Despite the frustrations, the technology is reshaping developer workflows permanently. The Fastly survey found senior developers are twice as likely as junior developers to put AI-generated code into production, citing significant speed improvements in prototyping and removing menial tasks.
Elvis Kimara, a recent AI master's graduate building an AI-powered marketplace, represents the new generation adapting to this reality. While he admits vibe coding has made his job harder and less satisfying - 'There's no more dopamine from solving a problem by myself' - he's preparing for a future where developers act as 'consultants to machines.'
'We won't just be writing code; we'll be guiding AI systems, taking accountability when things break,' Kimara predicted. The extra hours spent reviewing AI output are becoming what he calls an 'innovation tax' - a tolerated cost of using the technology.
The transformation extends beyond individual productivity. Traditional mentorship patterns are shifting as some senior developers delegate teaching responsibilities to AI models, while others struggle to understand the new vibe-coding landscape entirely.
Yet for all the debugging headaches, the consensus remains positive. 'The pros far outweigh the cons,' Kimara said, echoing sentiments across the industry. Rover, despite her tearful project restart, credits vibe coding with helping her craft better user interfaces. Malekzadeh simply notes he gets more done with AI coders than without them.
The rise of vibe coding marks a fundamental shift in software development, transforming senior developers into AI supervisors while accelerating overall productivity. Though the 'babysitting tax' of debugging AI-generated code consumes significant time, experienced developers are adapting their workflows around this new reality. As the technology matures, the industry faces the challenge of maintaining code quality and security while harnessing AI's speed advantages - a balance that will define the next phase of software development.
