This project investigates the role of personality traits in large language models (LLMs) and their downstream behavior when embedded in agentic systems. While LLMs are increasingly used as autonomous agents or decision-support systems, their responses often exhibit stylistic or behavioral consistencies that resemble personality dimensions. The emergence of such traits—whether as artifacts of pretraining data, model architecture, or fine-tuning—raises critical questions for alignment, robustness, and human interaction. We aim to systematically characterize personality expression in current state-of-the-art LLMs, using controlled experimental prompts and validated psycholinguistic metrics (e.g., Big Five dimensions). The project will then evaluate how these traits persist or change in agentic deployments, where models interact with structured environments and maintain memory or goal-oriented behavior over time. We also explore how explicitly manipulating personality traits—through conditioning, prompt engineering, or fine-tuning—affects performance, consistency, and user trust in multi-agent settings. Comparative experiments will assess whether personality-aligned LLMs improve interpretability or coordination in collaborative tasks, or whether certain personality profiles correlate with undesirable emergent behaviors.