“Agent” has become one of the most overused terms in AI marketing, applied to everything from a simple chatbot with a system prompt to genuinely autonomous multi-step systems. The actual technical distinction is worth being precise about.
A Chatbot: One Turn, One Response
A traditional chatbot takes a message, generates a response, and stops. Even a well-engineered chatbot with RAG and good prompting is fundamentally reactive — it answers what was asked, once, and waits for the next input.
An Agent: Plans, Acts, Observes, Repeats
An agent operates in a loop: given a goal, it decides what action to take (often via function/tool calling), executes that action, observes the result, and decides whether to take another action or whether the goal is complete — potentially many steps, with no human in between each one.
Chatbot: User asks -> Model answers -> done
Agent: Goal given -> Model plans step 1 -> executes tool -> observes result
-> Model plans step 2 -> executes tool -> observes result
-> ... continues until goal is met or a limit is hit
-> Final result returned to user
A Concrete Example of the Difference
Task: "Find and book me the cheapest flight to Berlin next Friday"
Chatbot: "I can't book flights, but here are some websites you could check."
Agent: 1. Searches flight APIs for the given date/destination
2. Compares results across multiple airlines
3. Checks the user's calendar for conflicts
4. Requests confirmation before the irreversible booking step
5. Completes the booking and reports back
Why the Distinction Matters Practically
Agents carry meaningfully higher risk — more autonomous steps means more opportunities for the model to take an unintended or irreversible action, which is exactly why well-designed agentic systems build in human-in-the-loop confirmation for consequential steps, sandboxed execution environments, and hard limits on how many steps an agent can take unsupervised.