The difference between a chatbot and an agent
A chatbot takes input and returns a response. It has no memory between sessions, no ability to act on the world, and no capacity to course-correct when something goes wrong. It is a sophisticated autocomplete wrapped in a conversation window. Most of what gets shipped under the label "AI assistant" is exactly this — a well-prompted language model that answers questions and nothing more.
An agent, by contrast, maintains state, can call tools, makes decisions across multiple steps, and knows when to escalate to a human rather than guess. The line is not about the underlying model — it's about architecture: does the system have goals, memory, and the ability to take actions with real-world consequences? If not, you have a chatbot. If yes, and those capabilities are handled carefully, you have the beginning of an agent.
