A new study of 700 engineering practitioners and managers across the U.S., U.K., France, Germany, and India reveals a fundamental shift in the software development landscape. While generative AI has accelerated code production, it has introduced a massive “invisible” workload that traditional productivity metrics fail to capture.
For decades, technological shifts like the internet, the cloud, and DevOps changed how software was distributed and deployed, but the core cognitive act of development remained largely the same. Generative AI has broken this pattern, moving the transformation to the cognitive layer. Developers have shifted from being the primary authors of code to becoming validators of machine-generated output.
According to the 2026 State of Engineering Excellence report from Harness, 31% of a developer’s day is now consumed by AI-related invisible work. This includes deeper scrutiny of code quality, increased accountability for downstream outcomes, and complex judgment calls regarding when to trust or override AI. Despite this, established frameworks like DORA metrics and cycle time were not designed to measure these new requirements.
The data highlights a significant “productivity offset.” While AI improves gross output volume and shortens cycle times, 81% of engineering leaders report that code review time—often viewed as overhead or “toil”—has risen sharply since deploying AI. This rise in validation effort often exists outside the measurement process, leading to systemic friction.
Developers identified the top sources of this AI-driven friction as reviewing AI code for accuracy (53%), fixing subtle bugs (52%), and explaining AI-generated code to teammates (48%). Ironically, only 38% of organizations actually track the time spent reviewing AI-generated code.
There is also a stark disconnect between leadership and practitioners. While 94% of respondents agree that tech debt, validation time, and burnout are missing from current metrics, managers generally report more favorable conditions than those doing the work. Furthermore, 54% of developers fear that AI productivity data will be used against them in individual performance evaluations.
To bridge this gap, the report suggests five key starting points for organizations in 2026:
- Measure validation work: Track debugging overhead and context-switching alongside output.
- Prioritize ship rate: Distinguish between generating code volume and shipping actual value.
- Audit frameworks: Treat high confidence in incomplete measurement systems as a risk signal.
- Plan for complexity: Anticipate increased needs for governance and security reviews as AI scales.
- Build trust: Establish clear policy guardrails about data usage to encourage developer partnership.
As AI tools consume a larger share of engineering budgets, the industry must evolve its productivity frameworks to account for the true shift in effort.
