Developers spend more time fixing AI code than writing it, survey finds
Software developers report productivity gains from AI tools, but the benefits mask a growing problem: 81% now spend more time reviewing code than before AI adoption, according to research from Harness.
The 2026 State of Engineering Excellence report reveals developers are caught in a cycle of manual remediation. More than one-quarter spend 30% longer on code reviews alone.
The real issue is invisible. Nearly one-third of this work-roughly 31% of total developer time-goes untracked by organizations. This includes reviewing AI-generated code, fixing bugs, and switching between tools.
Measurement frameworks lag behind reality
Organizations can't see the problem because they're using the wrong metrics. Around 89% of tech leaders rely on productivity measures that don't reflect AI's actual impact on developers or teams.
Even worse: 94% of leaders said their metrics miss critical factors like tech debt and developer burnout.
When asked to identify their biggest AI challenge, leaders pointed to visibility problems. The top three: measuring true productivity impact (26%), maintaining code quality with AI (24%), and proving return on investment to leadership (18%).
What needs to change
Harness recommends treating AI performance as a separate discipline. Organizations should track AI agent accuracy, acceptance rates of AI outputs, and tool costs separately from human developer metrics.
Code validation and overall code quality should factor into how organizations measure the impact of AI tools on workloads.
Developers themselves should have a voice in defining these metrics. Nearly half (49%) said they want involvement in how performance is measured-a crucial point given the gap between manager and developer perspectives.
Managers are nearly four times more likely to report no concerns about how productivity metrics are measured compared to developers on the front lines.
For developers managing AI-generated code and struggling with quality issues, an AI learning path designed for software developers can provide frameworks for handling these workflows. AI coding courses also address the specific challenge of code review and validation in AI-assisted development.
Your membership also unlocks:
