Survey Finds AI-Generated Code Increases Debugging and Failure Rates and Creates a Comprehension Gap

A survey by independent research firm Coleman Parkes, commissioned by Undo, a company focused on scaling AI-powered root-cause analysis, examined how AI coding agents affect work on mission-critical codebases. It polled 300 senior engineering leaders responsible for delivering such software, most of whom work with C/C++.

The report found the primary bottleneck has shifted from writing code to debugging, comprehension and maintenance. Respondents said the most demanding task is understanding what code does, how it affects existing codebases, and debugging it when an application misbehaves. Teams in these environments spend an average of 9.8 hours per week producing code but 16.9 hours per week debugging issues found during development or reported by customers in production, which accounts for 42% of the average working week.

Code comprehension emerged as a related challenge. With AI generating most code, engineers no longer have the inherent understanding they once had, making it easier for defects to escape and harder to trace failures to root causes. The survey found 35% of generated code reaches production before the team fully understands it, and 80% of respondents said coding agents struggle to solve difficult problems in complex codebases. About one-third of teams use AI agents for comprehension and debugging only in straightforward codebases, relying on other techniques for more complex systems.

Respondents also reported operational problems. Production incidents or outages affecting internal users or customers were experienced at least once in the previous six months by 81%, with 14% seeing them multiple times per month. Incorrect root-cause or issue diagnosis due to hallucination was reported by 93% at least once, and 18% multiple times per month. Test escapes, serious defects or poorly optimized code entering production were reported by 91% at least once, with 8% experiencing them multiple times monthly.

Overall, 79% of engineering leaders said AI agents can generate code significantly faster, but the resulting shift of effort toward debugging and unpicking AI-generated code means the overall release cycle is no faster than before. Greg Law, founder and CEO of Undo, said engineers lose days trying to unravel what went wrong with code that is almost, but not quite right, and that while agents write reams of code quickly, they are less capable at debugging it.

Why it matters: for developers and product teams working on mission-critical systems, faster code generation may not shorten delivery timelines, since debugging and comprehension consume much of the week, and a large share of generated code reaches production before teams fully understand it.