This paper investigates cybersecurity decision-making challenges by developing an adaptive network model that integrates AI coaching. The performed simulations reveal the dynamics between cybersecurity analysts and AI coaching during real-time security monitoring. In the first scenario, analyst fatigue and resulting weak mental model connections lead to systematic decision-making errors and security vulnerabilities. In the second scenario, AI coaching results in effective error correction by improvements at the level of the knowledge underlying the decision making and improved consistency. The findings underscore the importance of context-sensitive AI coaching and adaptive coaching mechanisms in maintaining cybersecurity effectiveness while preserving human autonomy.