Senior Developers Become AI Babysitters as Vibe Coding Raises Quality and Security Concerns

Senior developers are increasingly finding themselves in the role of AI babysitters as they navigate the rise of vibe coding—using AI tools to generate code quickly but often needing to fix the errors that follow. Carla Rover, a veteran web developer with 15 years of experience, shared a painful personal example: after relying heavily on AI to build her startup’s core system, she discovered major flaws only after a full restart, leading to tears. “I handed it off like the copilot was an employee,” she said. “It isn’t.” Rover isn’t alone. A Fastly survey of nearly 800 developers found that at least 95% spend extra time reviewing and correcting AI-generated code, with senior developers bearing the brunt of this workload. The issues range from hallucinated package names and deleted critical code to serious security vulnerabilities. These problems stem from AI models that lack true understanding, often fabricating results or failing to follow complex logic. Feridoon Malekzadeh, a 20-year industry veteran building his own startup, likened vibe coding to hiring a stubborn, rebellious teenager. “You have to ask them 15 times to do something,” he said. “They do some of what you asked, some stuff you didn’t ask for, and break a bunch of things.” He estimates he spends 30% to 40% of his time fixing AI-generated code—rewriting bugs, removing unnecessary scripts, and untangling duplicated functionality that AI creates across multiple parts of a project. The problem extends beyond bugs. AI often struggles with systems thinking—designing solutions that scale across an entire product. Instead, it may create the same feature five different ways in five different places, leading to confusion and inconsistency. Malekzadeh noted that AI tends to focus on surface-level fixes rather than long-term architectural integrity. Austin Spires, senior director of developer enablement at Fastly, has seen AI generate code that prioritizes speed over correctness—introducing vulnerabilities common in beginner-level programming. He’s also noticed a troubling pattern: when challenged, AI models often respond with “you’re absolutely right,” even when they’re wrong. This behavior, he says, mimics a toxic coworker who refuses to admit mistakes. Security expert Mike Arrowsmith of NinjaOne warns that vibe coding can bypass essential review processes, creating blind spots—especially for young startups. To combat this, his company promotes “safe vibe coding,” which includes access controls, mandatory peer reviews, and automated security scanning. Despite the challenges, most developers agree that the benefits outweigh the drawbacks. Rover uses AI to prototype user interfaces faster, while Malekzadeh says he’s still more productive with AI than without it. Spires uses AI for scaffolding and boilerplate code, freeing up time for higher-level design and deployment. For young engineers like Elvis Kimara, a recent AI master’s graduate building an AI-powered marketplace, vibe coding has changed the experience of learning. “There’s no more dopamine from solving a problem myself,” he said. “The AI just figures it out.” He also noted that mentorship from senior developers has declined, with some delegating guidance to AI tools. Still, Kimara believes the trade-off is worth 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 said. “Even as I grow into a senior role, I’ll keep using it. I review every line so I learn faster.” The new normal is clear: AI is a powerful accelerator, but it demands constant oversight. The cost of using it—extra hours spent cleaning up its mistakes—is becoming an accepted tax on innovation.