Unanticipated Effects of Generative AI on Expertise Pathways and Performance Perception in System Administration

Synthesis: Drawing on 14 semi-structured interviews with IT professionals, this study explores how GenAI integration reshapes professional practice in system administration — troubleshooting, scripting, and system verification — extending concerns about Cognitive Offloading and skill decay from classrooms to professional practice.

Key Findings

Drawing on 14 semi-structured interviews with IT professionals, this study explores how GenAI integration reshapes professional practice in system administration — troubleshooting, scripting, and system verification.

• Compression of traditional expertise pathways: GenAI acts as both mentor-like tutor and "ladder-shortening" tool, accelerating unfamiliar-domain task performance while reducing exposure to the foundational build-fail-debug cycles that historically built expertise.
• Performance perception shift: AI-assisted speed resets organizational and self-expectations, creating a "two-speed culture" within teams and "productivity guilt" — a metacognitive cost of AI-augmented work.
• The findings extend concerns about Cognitive Offloading and skill decay from classroom settings to professional practice, with implications for Workplace Learning programs and workplace learning — and for human-in-the-loop AI-mediated work.

What this means for practice

• Instructors. Preserve the build-fail-debug cycles that build technical intuition: participants reported GenAI handling 60–70% of a task, removing the repetitive training ground on which professional judgment rests.
• Instructors. Train novices to audit AI output line by line rather than accept it, since participants warned that the tool can "very confidently" give a wrong answer and that new users tend to "take it as is."
• Designers. Treat verification as a required competency in Workplace Learning and Lifelong Learning programs rather than an afterthought; experienced participants positioned human expertise around context, verification, and judgment.
• Administrators. Make productivity metrics context-aware by counting the invisible labor of iterative prompting and output "babysitting," so a two-speed culture does not penalize staff without AI access or push them toward shadow AI use.

Limitations

• Fourteen semi-structured 45-minute interviews with IT professionals; the authors frame the findings as exploratory rather than generalizable.
• Participants all had 2–15 years of experience and prior GenAI use, so the study does not follow novices over time and cannot confirm that the predicted "learning debt" materializes.
• Accounts are self-reported; the claimed time savings (for example a "70, 80 percent time saving," or a two-month task compressed to two weeks) are participants' own estimates, not measured.
• The account rests on one interview study, with no control condition and no observation of actual task performance.

Citation

Abou Khamis, R., Assal, H., & Matrawy, A. (2026). Unanticipated effects of generative AI on expertise pathways and performance perception in system administration.