New Relic, an Intelligent Observability Company, released the 2026 State of AI Coding report, revealing the central contradiction of the vibe coding era. While an astonishing 94% of leaders rate AI-generated code to be of higher quality than human-written code at the time of review, its implementation incurs significant operational costs in the wild. Once this code shipped, 78% of respondents reported an increase in incidents, 86% reported an increase in the amount of time senior staff spent fixing the code, and 74% of respondents reported that at least 25% of their AI code required significant rework, considering the past 12 months. 82% have experienced at least one operational failure related to AI-generated code in the past six months. Only 19% of organizations reported no challenges with AI-generated code during this period.
Also read: AiThority interview with Matej Bukovinski, Chief Technology Officer at Nutrient
New Relic’s report reveals the contradictions of the vibe coding era. An astonishing 94% of leaders rate AI-generated code as higher quality than human-written code at the time of review, but its implementation generates significant operational taxes once it goes live.
This report, conducted in partnership with Hanover Research, surveyed U.S. technology leaders in upper-midmarket and enterprise companies using generative and agent AI in software engineering to understand the downstream operational impacts. This data reveals a historic amount of change in software authoring far beyond startups, with 67% of technology leaders saying that 51% to 75% of their organization’s weekly code output is generated or significantly refactored by AI.
“AI coding agents are no longer just auto-filling lines of text, they are driving much of the software development across the enterprise.” Nic Benders, Chief Technical Strategist at New Relic. “However, our report reveals an alarming trend: the rapid accumulation of what we call ‘agent debt.’ While leaders praise the speed of agent-generated code during initial reviews, organizations quietly inherit major flaws in unexamined architectural logic that cause incidents later in production. Finding ways to reduce agent debt is now a critical challenge for engineering organizations. ”
Other key findings from the report include:
- Vibe coding clears production hurdles. Vibe Coding is no longer confined to a sandbox or treated as a personal productivity hack. A surprising 88% of organizations have vibecoding written into their formal operational policies, only 5% restrict it to non-production environments, and no respondents completely prohibit the practice.
- Overconfidence creates upstream risks. This data suggests that trust is largely misplaced early in the development lifecycle. Almost two-thirds (62%) of technology leaders report that their engineering teams trust AI-generated code enough to often ship it to production without manually verifying each line.
- The AI-generated code received high scores in early reviews. A total of 94% of respondents favor AI-generated code, with 61% of leaders rating AI-generated code as “somewhat high quality” and 33% “much higher quality.” Only 2% of respondents perceived the quality to be poor. This metric reflects subjective clarity during code reviews rather than operational performance during actual incidents.
- Observability is now a key factor. Reflecting the downstream complexity of machine-generated code, 96% of technology leaders rated observability as very or extremely important when working with AI-generated code, and no respondents rated it as minor or unimportant.
- Telemetry moves to AI prompts. Engineers are actively moving observability upstream. Nearly four out of five teams (78%) now regularly instruct their AI tools to incorporate specific telemetry, such as logs, traces, and metrics, directly into the generated code itself to ensure it is observable by design.
Also read: AI Systems – Interoperable AI Systems: Connecting models across platforms
[To share your insights with us, please write to psen@itechseries.com]
