MUMBAI: AI may be acing the code review, but once it hits production, the bugs are having the last laugh. Enterprise technology leaders are increasingly embracing AI-generated code, but a new study suggests the productivity gains may be coming with an expensive hidden cost.
According to New Relic’s 2026 State of AI Coding Report, 94 per cent of technology leaders rate AI-generated code more favourably than code written by humans during the review process. Yet beneath those glowing assessments lies a growing concern: what happens after deployment.
The survey found that 78 per cent of organisations have seen an increase in incidents linked to AI-generated code, while 86 per cent reported that senior engineers are spending more time fixing issues that emerge in production. Meanwhile, 74 per cent said at least a quarter of AI-generated code required substantial rework over the past year.
The findings highlight a widening gap between development speed and software reliability as AI becomes deeply embedded in engineering workflows.
AI is no longer acting as a coding assistant sitting quietly in the corner. According to the report, 67 per cent of technology leaders said AI now generates or significantly refactors between 51 per cent and 75 per cent of their organisation’s weekly code output, signalling a dramatic shift in how enterprise software is built.
The report introduces a new concern dubbed “agent debt,” the accumulation of AI-generated logic that appears sound during review but later creates operational and architectural problems.
The rise of so-called “vibe coding” is further accelerating the trend. The practice, which relies heavily on AI-generated code and rapid iteration, is now formally included in production policies at 88 per cent of organisations surveyed. Just 5 per cent restrict it to non-production environments, while none reported banning the practice altogether.
That confidence may be contributing to risky deployment habits. Nearly 62 per cent of technology leaders admitted their teams often push AI-generated code into production without conducting line-by-line manual verification.
The consequences are already becoming visible. More than 82 per cent of organisations experienced at least one production failure linked to AI-generated code in the past six months. Only 19 per cent said they had avoided any AI-related coding challenges during that period.
As engineering teams wrestle with these issues, observability tools are becoming increasingly critical. The study found that 96 per cent of technology leaders consider observability either very or extremely important when managing AI-generated software. In response, 78 per cent of teams now regularly instruct AI tools to embed telemetry features such as logs, traces and metrics directly into generated code.
The research, conducted by Hanover Research on behalf of New Relic, surveyed 200 US-based technology decision-makers working in IT and engineering roles at upper mid-market and enterprise organisations using generative and agentic AI.
The message from the report is clear: AI may be writing more code than ever before, but the real challenge is no longer generating software quickly. It is making sure that what is written at machine speed does not create human-sized problems later.