Simulating Quiet Quitting: A Computational Experiment on Crisis Management Strategies using Generative Agents

<title>Abstract</title> <p>The post-pandemic phenomenon of Quiet Quitting—employees maintaining contractual compliance while psychologically disengaging—poses a detection challenge that traditional surveys cannot resolve. We address this observability gap through a Generative Computational Experiment using Large Language Models (LLMs) to simulate a heterogeneous digital workforce under “Crunch Time” pressure. A 2 × 6 factorial design (Control vs. Commitment strategy × Big Five archetypes; N = 180 dyadic interactions) reveals three findings. First, a dual-output protocol successfully decouples behavioral compliance from internal turnover intention, making Quiet Quitting observable and quantifiable. Second, the Commitment strategy reduces turnover intention from 67.8% to 7.8% (χ2 = 68.4, p &lt; 0.001), but its efficacy is bounded by personality: Cautious agents exhibit hidden friction (high compliance coupled with saturated turnover intention), while Emotional agents remain volatile (46.7%) even under supportive leadership. Third, our Personality-Stripping Protocol filters anthropomorphic noise inherent in RLHF-aligned LLMs, isolating traitdriven psychological signals. The framework provides a computational testbed for predictive retention modeling and supports a shift toward trait-aware, precision-targeted management.</p>