Agentic loops are not the same as the workflow-based LLM applications most of us started with. An agent takes multiple steps, calls external tools, reasons across retrieved context, and arrives at a conclusion through a path that is itself part of what needs evaluating.
