Why is deskilling suddenly a leadership topic?
Because the evidence started arriving. BCG reports that about half of surveyed executives already see skill erosion as AI absorbs tasks, and cites MIT findings that people who wrote with an LLM later could not recall their own text (BCG, 24 September 2026, read 19 October 2026). The industrial version of the concept is a century old; what is new is that the skills eroding are judgment skills, the analysis, writing and decision-making that sit directly under every AI adoption push.
Why does deskilling undermine AI oversight?
Because every safety story ends with a capable human. A human-in-the-loop control assumes the human can still tell good output from bad; BCG’s blunt phrase for review without expertise is “procedural reassurance”. Oversight by someone who can no longer do the work is a signature, not a check. The riskiest version compounds quietly: the expert keeps the title while the expertise fades, and the gap surfaces only when the model meets a case outside its range, exactly when the human was supposed to catch it.
Which skills erode first, and which grow?
What fades is what stops being practiced: drafting from scratch, working a problem without a scaffold, the pattern recognition that comes from doing rather than reviewing. What grows, in teams that manage it, is the ability to direct and evaluate AI, the competence AI fluency names. The difference between erosion and upgrade is whether evaluation stays a practiced skill, and research on AI and skills finds outcomes depend on how the work is designed, not on the technology itself.
How do leaders keep skills alive while automating?
Design the practice in. Rotate people through the manual path on a sample of cases, so the skill that audits the model keeps its edge. Make unaided performance part of training, the way pilots still fly by hand. Score AI-assisted work on outcomes the person must defend, not on volume. And fund real AI literacy, because the alternative to skilled use is not non-use, it is unskilled use. Treat expertise as infrastructure with a maintenance budget, not a stock you can draw down forever.