Senior Developers Turn into AI Code Babysitters Amid Vibe Coding Surge
Key Points
- Senior developers are spending significant time fixing AI‑generated code.
- Fastly survey: 95% of developers report extra effort on AI code; seniors twice as likely to deploy it.
- Carla Rover had to restart a project after AI errors, describing the process as emotionally taxing.
- Feridoon Malekzadeh spends 30‑40% of his time fixing AI code, noting poor systems‑thinking by the models.
- Security experts warn AI code can introduce vulnerabilities by favoring speed over correctness.
- NinjaOne implements "safe vibe coding" with access controls, peer review, and security scans.
- AI code is useful for prototyping, boiler‑plate, and test scaffolding, but requires human review.
- Developers acknowledge an "innovation tax"—extra time for auditing AI‑generated code.
Developers are increasingly using AI‑generated code, known as vibe coding, to speed up projects. Senior engineers, however, find themselves spending significant time correcting the AI's output, which can include hallucinated packages, deleted information, and security risks. Interviews with developers like Carla Rover and Feridoon Malekzadeh reveal frustrations, costly rewrites, and a new "innovation tax" of extra review work. Companies such as Fastly and NinjaOne acknowledge the productivity boost but stress mandatory human oversight and security scanning to keep AI‑generated code safe for production.
AI‑Assisted Development Becomes a Double‑Edged Sword
Vibe coding, the practice of using AI agents to generate code, is gaining traction across the tech industry. While it promises rapid prototyping, boiler‑plate creation, and faster delivery, developers report that the technology often produces buggy or insecure output that must be meticulously reviewed.
Senior Engineers Acting as Babysitters
Carla Rover, a veteran web developer, described her experience with AI‑generated code as a "beautiful, endless cocktail napkin" that quickly turned into a painful rewrite. After relying on the AI to meet a tight deadline, she discovered numerous errors and was forced to restart her entire project, leading to a distressing moment of tears.
Fastly’s survey of nearly 800 developers found that at least 95% of respondents spend extra time fixing AI‑generated code. Senior developers, in particular, are shouldering the bulk of this effort and are twice as likely as junior developers to place AI‑generated code into production.
Time Spent on Vibe Coding and Fixing
Feridoon Malekzadeh, a seasoned product engineer, breaks down his workflow: roughly half of his day is spent writing requirements, 10‑20% on AI‑generated code, and 30‑40% on fixing the resulting bugs and unnecessary scripts. He notes that AI struggles with systems‑thinking, often creating multiple disparate implementations for the same feature.
Security Concerns and Organizational Responses
Austin Spires of Fastly observes that AI tends to prioritize speed over correctness, introducing vulnerabilities reminiscent of those found in novice coding. Mike Arrowsmith, CTO of NinjaOne, warns that vibe coding can create new security blind spots, especially in startups that may skip rigorous review processes.
NinjaOne mitigates these risks by enforcing "safe vibe coding" policies: approved AI tools, strict access controls, mandatory peer review, and automated security scanning.
Balancing Benefits and the Innovation Tax
Despite the challenges, many developers acknowledge the productivity gains. AI‑generated code excels at mocking up ideas, scaffolding tests, and handling repetitive tasks, freeing engineers to focus on higher‑level design and scaling.
Elvis Kimara, a recent AI master’s graduate, admits that vibe coding can diminish the personal satisfaction of solving problems manually, yet he continues to rely on it, emphasizing thorough line‑by‑line review to learn from the AI’s output.
The consensus across interviews is clear: AI‑generated code is a powerful tool, but human oversight remains essential before deploying it in production. The industry is adapting to an "innovation tax"—the extra time required to audit, correct, and secure AI‑written code—while still leveraging its speed advantages.