<p>Agentic AI is fundamentally transforming wireless communication by reshaping static architectures into an autonomous, cognitive fabrics, ultimately paving the way for <i>agent networks</i>: distributed ecosystems of intelligent agents capable of perception, reasoning, and collaborative action across physical and digital domains. In such environments, communication evolves beyond transmitting bits to conveying intent, context, and task-relevant semantics that drive collective intelligence. As the key enabler of agent networks, <i>Task-Oriented Communication (TOC)</i> aligns transmission objectives with task execution, ensuring that agents are capable of sharing information most relevant for joint reasoning and decision-making. This paper examines the development of TOC, tracing its evolution from discriminative and generative paradigms while highlighting their inherent reactivity and limited support for autonomous agent behaviors. To address these limitations, we envision a reasoning-empowered TOC framework, where communication becomes a reasoning-driven process, enabling agents to autonomously determine not only <i>what</i>, but also <i>when</i> and <i>why</i> to communicate. This paper concludes by identifying critical challenges and outlining future research pathways towards enabling scalable, intelligent agent networks through reasoning-empowered TOC systems.</p>