A recent industry report indicates that nearly all developers utilizing artificial intelligence (AI) coding tools are dedicating additional time to review and correct the generated code, effectively shifting their role towards an "AI babysitter" function. Despite this increased oversight, many experienced programmers view the technology as a net positive for productivity.
According to a report from content delivery platform company Fastly, 95% of nearly 800 surveyed developers stated they spend extra time rectifying AI-generated code. The burden of this verification predominantly falls on senior developers, who have identified issues ranging from hallucinating package names and data deletion to the introduction of security vulnerabilities.
Carla Rover, a web developer with 15 years of experience now building an AI-focused startup, recounted having to restart a project due to AI code errors, describing the experience as "worse than babysitting." She noted that while AI offers speed, its outputs can create unexpected problems if not thoroughly scanned, leading to a complete project restart in her case. Rover likened using an AI coding co-pilot to entrusting a complex task to a "smart six-year-old," highlighting the necessity of human accountability.
Feridoon Malekzadeh, with over two decades in product development, estimated he spends 30% to 40% of his time fixing issues in AI-written code, despite leveraging AI tools for his startup. He criticized AI for failing at "systems thinking," often generating redundant solutions instead of efficient, broadly applicable code. Similarly, Rover noted AI's tendency to "manufacture results" or provide misleading advice when data conflicts with its programmed responses.
Austin Spires, Senior Director of Developer Enablement at Fastly, observed that AI code often prioritizes speed over correctness, potentially introducing vulnerabilities common in early-stage programming. Mike Arrowsmith, CTO at NinjaOne, warned that AI coding can bypass traditional, rigorous review processes, creating new security blind spots. NinjaOne addresses this through "safe vibe coding," incorporating access controls, mandatory peer reviews, and security scanning for approved AI tools.
Despite the challenges, developers widely acknowledge the benefits. The Fastly survey found senior developers were twice as likely as junior developers to deploy AI-generated code into production, citing its ability to accelerate workflows. Rover noted AI's utility in crafting user interfaces, while Malekzadeh stated he accomplishes more with AI coders than without them.
Elvis Kimara, an AI-powered marketplace builder, articulated the "innovation tax" associated with AI coding, where the benefits outweigh the cons despite the added work. He foresees a future where engineers guide AI systems, take accountability for failures, and act as "consultants to machines," emphasizing that diligent review of AI-generated code fosters faster learning and development.
