New Relic Report: 78% More Incidents From AI-Generated Code in Production
AI-generated code gets higher marks than human code during review, but 82% of orgs hit production failures from it, 86% of senior devs spend more time fixing it, and a new term is born: agent debt.
New Relic's 2026 State of AI Coding report surveyed 200 tech decision-makers at mid-market and enterprise companies. The headline finding: 94% of leaders rate AI-generated code as higher quality than human code during review. Only 2% see it as worse. Then production happens.
By the numbers
| Metric | Value |
|---|---|
| More incidents from AI code | 78% |
| Orgs with AI-code production failures (6mo) | 82% |
| Senior staff spending more time fixing | 86% |
| AI code needing significant rework | 74% (at least 25% of output) |
| Code generated or refactored by AI | 67% (51-75% of weekly output) |
The report coins a term for the pattern: agent debt. AI tools generate code that passes review but hasn't been deeply understood or tested. The debt accumulates silently until production surfaces it.
The trust paradox
62% of teams routinely ship AI-generated code without line-by-line review. Velocity goes up. Review scores go up. Then senior engineers spend their time on incident response instead of feature work. The operational cost shifts from writing code to debugging it.
Observability as the counterweight
96% of leaders now rate observability as very or extremely important when working with AI code. 78% of teams prompt AI tools to include logs, traces, and metrics directly in generated output. The idea: instrument the code at generation time since you can't trust it to be instrumented later.
The bottom line
Agent debt is the technical debt of the AI era. AI writes code faster than humans can review it. The code passes review more often than human code. Then it breaks more in production. The fix isn't less AI. It's more observability, earlier in the pipeline, and a review process that doesn't confuse velocity with quality.