This study introduces an innovative methodological framework for AI-driven talent chain management, addressing key challenges in workforce optimization, collaboration dynamics, and innovation assessment within complex and uncertain environments. Traditional talent management methods often lack the adaptability needed to handle dynamic demands and stochastic task variations. To overcome these limitations, the framework incorporates the Adaptive Talent Dynamics Planner, composed of three modules: the Constraint-driven Workforce Optimizer, the Agent-based Collaboration Forecaster, and the Uncertainty-aware Innovation Evaluator. These components optimize workforce allocation, predict collaboration patterns, and evaluate innovation potential under uncertainty. The framework further integrates policy-grounded reasoning and uncertainty-aware refinement to ensure alignment with organizational objectives and robustness to fluctuations in talent and task parameters. By formalizing the problem mathematically and leveraging a multi-agent system architecture, this research provides an adaptive solution for talent chain management. Experimental results demonstrate improvements in allocation efficiency, collaboration prediction accuracy, and innovation assessment, underscoring its potential to support sustainable growth and organizational innovation.