A fire/EMS dispatch assistant invokes tools for CAD lookups, while a parks FAQ bot only returns text. What architectural distinction should the diagram emphasize?
Select an answer to reveal the explanation.
Short Explanation
A plain FAQ bot talks; an agent reaches for tools—like a dispatcher who can actually query CAD. That extra handhold is why agent diagrams and attack surfaces look different from chat-only apps.
Full Explanation
A fire/EMS dispatch assistant that invokes CAD lookup tools is an agentic system: the language model can trigger side-effecting or privileged actions against external systems. A parks FAQ bot that only returns text is a non-agent LLM app without tool-driven side effects. Architecture diagrams and offensive foundations must emphasize tool trust boundaries, authorization, and human-in-the-loop design for the dispatch assistant that the FAQ bot may not require to the same degree.
Calling both identical agents because any answering LLM is an agent erases the tool-use distinction that drives blast radius. Claiming that holding a multi-turn conversation alone makes a system agentic erases the actual distinction—agency requires tool-driven side effects, and a system that only chats without ever invoking a tool is not agentic no matter how extended the dialogue. Claiming the parks FAQ bot is the agentic system while the dispatch assistant is just a deterministic CAD database interface gets the architecture backward: the dispatch assistant is the one invoking CAD tools with side effects, making it agentic, whereas the FAQ bot returns text only.
Exam caveat: agentic = tool-using side effects; chat-only FAQ ≠ agent; the dispatch assistant is the agentic system because it invokes CAD tools, while the parks FAQ bot is non-agentic despite public engagement, and CAD lookups are LLM tool calls rather than a deterministic database interface. Operational check: diagram which municipal assistants call tools; for fire/EMS, list CAD permissions, authn/z, and HITL gates; include tool-steering tests in authorized purple-team while scoping FAQ bots primarily to content/prompt risks.