AI has clearly boosted developer output, but new research shows our measurement frameworks are failing to capture the work AI has added, leaving organisations with a misleading view of engineering productivity.
The State of Engineering Excellence 2026 report—based on a survey of 700 engineering practitioners and managers across the US, UK, France, Germany and India—finds that while leaders see clear productivity gains from generative AI, much of the benefit is offset by a growing, invisible validation burden that current metrics do not capture.
The report highlights a core disconnect: 89% of engineering leaders say productivity metrics have improved after deploying AI, yet 81% report that code review time has increased. This points to a paradox where gross output rises—more code, more commits, faster cycle metrics—while the time required to validate, correct and explain AI‑generated work grows, creating an “invisible tax” on developer time.
The study quantifies that invisible work: around 31% of a developer’s day is now consumed by AI‑related tasks that frequently do not appear in traditional engineering dashboards.
Exactly where that time goes is revealing. Developers cite reviewing AI code for accuracy (53%), fixing subtle bugs introduced by AI (52%), explaining AI outputs to colleagues (48%) and context‑switching between tools (45%) as the top sources of friction. Yet only 38% of organisations track AI review time, meaning the activity that most erodes net productivity is routinely unmeasured.
That gap helps explain why 94% of respondents say crucial factors such as technical debt, validation time and burnout are missing from their metrics, and why only 6% believe current frameworks are sufficient for the AI era.
This mismatch has social as well as technical consequences. Perception between managers and practitioners diverges sharply: leaders often report more favourable conditions than the engineers doing the work. Managers are nearly four times as likely to say they have no concerns about AI productivity data being used to evaluate staff (15% of managers vs 4% of practitioners).
The report warns that metrics designed without practitioner involvement risk becoming instruments of mistrust rather than insight. Indeed, 54% of developers fear that AI data could be used in individual performance evaluations.
The report’s lead recommendation is simple but consequential: measure what AI adds. Rather than discard established performance frameworks—DORA metrics, cycle time, survey‑based developer experience—the study urges organisations to instrument new, net‑effort measures alongside them.
Practical steps include tracking AI validation time, debugging overhead, context‑switching costs and agent accuracy; distinguishing generated code volume from shipped value (ship rate); and treating high metric confidence as a risk signal that warrants audit.
Developers’ acceptance of measurement hinges on trust and governance. Engineers said they would accept measurement when data is used for improvement rather than punishment, and when they have transparency and a role in defining metrics: 55% want a clear separation between improvement and performance data, 50% seek transparency about what is measured, and 49% demand developer involvement in metric design.
Third‑party research supports these conclusions. Jellyfish’s 2026 State of Engineering Management report similarly finds rapid AI adoption correlates with perceived productivity gains, but also warns that teams must refine measurement to capture real value and avoid hidden costs.
GitLab and others have documented the “hidden tax” of reviewing AI‑generated code and the increased context switching that reduces developer focus and flow.
For COOs and engineering leaders, the takeaway is operational: instrument validation work, audit trusted metrics, redesign performance systems with practitioner input, and build clear policies on how AI‑derived data will be used. Only by making the invisible visible can organisations ensure AI’s promise translates into sustained, trustworthy productivity gains.
