Vibe coding has turned senior devs into ‘AI babysitters,’ but they say it’s worth it

In the rapidly evolving tech industry, seasoned developers are increasingly finding themselves in the role of what some have dubbed 'AI babysitters.' Carla Rover, a web developer with 15 years of experience, recently shared her struggles with AI-generated code while building a startup focused on custom machine learning models. Despite her enthusiasm for 'vibe coding'—which she describes as a creative outlet akin to sketching ideas on a napkin—she has faced significant challenges. After neglecting to thoroughly review AI-generated files, Rover was forced to restart a project, leading to a moment of frustration and tears. This sentiment is echoed by many in the field. A recent survey by Fastly revealed that 95% of nearly 800 developers reported spending additional time correcting AI-generated code, with the burden especially heavy for senior developers. These experienced coders often encounter issues such as incorrect package names, loss of critical information, and potential security threats. Without proper oversight, AI-generated code can introduce more bugs than traditionally written code. The rise of 'vibe code cleanup specialists' reflects the growing need for human oversight in AI-assisted development. Rover's analogy of using an AI coding co-pilot is telling: it's as if one were to ask a child to pour coffee for the family—possible but fraught with risk. Feridoon Malekzadeh, another veteran in the industry, likens vibe coding to relying on a stubborn teenager; while it can save time, it often results in confusion and mistakes. Malekzadeh estimates that he allocates a significant portion of his day to correcting AI code, illustrating a common struggle among developers. They find that AI often solves only superficial problems, failing to consider complex interactions within a system. This limitation can lead to duplicative efforts and further complications down the line. Rover has observed that AI can mislead users, sometimes fabricating results when questioned about its errors. This behavior has raised concerns among developers about the reliability of AI tools, particularly in terms of security vulnerabilities. Austin Spires, a senior director at Fastly, notes that AI tends to prioritize expedience over correctness, which can lead to the kind of mistakes typically seen in less experienced programmers. Experts agree that while AI-generated code can be beneficial for initial development phases, human review remains critical before deployment. The consensus is that while vibe coding can streamline certain tasks, it cannot replace the nuanced understanding that experienced developers bring to the table. Despite the challenges, many developers acknowledge the potential of AI as a tool for enhancing productivity. New engineers, like Elvis Kimara, are adapting to this changing landscape, recognizing that they will need to guide AI systems and take responsibility for their outcomes. As the industry continues to evolve, the balance between leveraging AI tools and maintaining rigorous coding standards will be essential for long-term success.