The Cognitive Cloud Continuum envisions a computing ecosystem where edge, fog, and cloud resources dynamically adapt to changing conditions, enabling intelligent, decentralized decision-making. However, current orchestration models, rely on hierarchical control and static policies, which constrain scalability, adaptability, and trust-aware coordination. In this work, we revisit the original principles of the Cognitive Cloud Continuum and propose that software agents represent a natural evolution of this paradigm. Agents act as intelligent, self-aware interfaces capable of interpreting local context, evaluating available resources, and forwarding tasks to the most suitable execution points, coordinating dynamically without a centralized control. We argue that embedding agents as first-class runtime entities transforms orchestration into negotiation, shifting from externally governed scheduling to self-organizing coordination. This vision of agent-based orchestration enables resilient, decentralized behavior and opens new design spaces for adaptive offloading and cognitive autonomy across the continuum. We conclude with a research roadmap identifying orchestration, offloading, and coordination as intertwined challenges for future agentic systems, to be explored jointly through the lenses of autonomy, context-awareness, and local reasoning.