A product team is building a customer support tool and has been using "chatbot" and "AI agent" interchangeably. The lead engineer pushes back: "These are different things — and we need to decide which we're building before we pick architecture."
What's the operational distinction?
Why did you pick that answer? Two or three sentences. The act of articulating it is what builds the judgment — not the click that follows.
**The chatbot/agent distinction is about action, not size or interface.** A chatbot generates text. An agent has tools, makes decisions about which to call, observes the results, and takes further action — LLM + Tools + Reasoning Loop + Memory. This isn't pedantry; it determines architecture, cost, latency, error handling, and risk surface. Agents need approval gates, tool descriptions, error recovery, audit logging — concerns that don't apply to chatbots. Picking the wrong category at design time produces the wrong system.
Voice and text are both interfaces; both can be backed by either a chatbot or an agent. The interface is a presentation choice; the architecture distinction is whether the system can take actions.
Size isn't the distinction — capability surface is. A small agent with three tools is still architecturally different from a large chatbot. The relevant axis is whether the system can affect the world beyond producing text.
This is the genuinely tempting wrong answer because agents do often have reasoning loops while chatbots often have single-turn responses. But loop count is a consequence of the architectural difference, not the difference itself. A chatbot can have a multi-turn conversation with several model calls and still be a chatbot — it isn't taking action in the world. An agent with one tool call is still an agent — it can affect the world beyond producing text. The distinguishing axis is action capability, not iteration count.