Human–AI Interaction · Longitudinal Study
January 2026 · Work Systems Correspondence

Skill Atrophy Thresholds

A speculative longitudinal exploration of how sustained copilot use shapes error detection, memory, and autonomous problem-solving. speculative longitudinal pre-registered

AI copilots can markedly accelerate complex work while leaving it subjectively demanding, and a team at the Center for Adaptive Work Systems argues that this gap conceals a measurable dynamic: skill atrophy thresholds — points beyond which reliance on assistance improves speed but gradually weakens independent error-checking and knowledge retention.

Core claim: The study does not assert that “AI makes users less capable.” Instead it isolates a narrower effect: when internal checking is consistently offloaded, the habit of checking atrophies — and that erosion compounds over time.
Working Definition
A skill atrophy threshold is the minimum level and duration of copilot-assisted work after which a person’s unaided performance on the same task family declines relative to their pre-assistance baseline.
Study design

Over 28 days, 96 participants were followed across three task categories: code debugging, analytical writing, and spreadsheet-based modeling. Each participant was randomly assigned to one of three copilot exposure conditions: low, moderate, or high.

Condition Assistance exposure Weekly “no-copilot” test
Low Guidance and hints only; no full solutions Yes
Moderate Drafts and suggestions; user must revise and finalize Yes
High Default to auto-complete and auto-solutions Yes
Primary outcomes
−12%
Error-detection rate (high exposure, week 4, unaided)
−9%
Retention score (high exposure, week 4, unaided)
+22%
Task throughput during assisted sessions
+2%
Throughput on weekly “no-copilot” evaluations

Declines were most pronounced for debugging tasks and smallest for spreadsheet work; experienced practitioners showed slower, but still noticeable, degradation in unaided checking.

Why there is no single “magic number”

Rather than a universal tipping point, the authors propose a family of thresholds that vary by domain, task structure, and user experience level. A recurrent pattern, however, is a behavioral shift: atrophy accelerates when workers move from a verification-first habit (“check, then accept”) to an acceptance-first habit (“accept, then occasionally check”).

“The copilot didn’t erase the underlying skill. It erased the requirement to exercise it — and that alone was enough to move the baseline.” — Dr. Rowan Keats, Study Lead

What deteriorates first

The earliest measurable change was not the quality of work produced with the copilot turned on, but self-correction capacity. When the assistant was removed, participants became slower to spot contradictions, boundary conditions, and subtle logic errors in their own output.

Limitations

Contextual references

  1. NIST AI Risk Management Framework — highlights evaluation, monitoring, and human oversight in socio-technical systems.
  2. NIST AI RMF 1.0 (PDF) — lifecycle framing that emphasizes measurement of human factors and downstream impacts of AI tools.