Take one simple everyday problem and show how a regular chat would solve it and how an AI agent would solve it. Return JSON {"chatExample":"...","agentExample":"...","keyDifference":"..."}: chatExample - what a regular chat will do, agentExample - what an agent will do (sequence of actions), keyDifference - the main difference is in one short phrase.
A regular chat works according to the question-answer scheme: you write a request, the model responds once with text and stops. The AI agent is designed differently. He has a cycle inside: he looks at the task, plans a step, acts (looks for information, opens a file, calculates), looks at the result and decides what to do next - and so on until the goal is achieved. That is, the agent does not just talk, but does and checks himself. Key lever: The agent must have a clear goal and a clear “done” sign. Without it, it either stops too early or spins in circles. Insider trick: formulate not the topic, but the result - not “help with the report”, but “compile a sales table for June and highlight three conclusions.” Then the agent has something to strive for. A typical rookie mistake is to expect magic from the agent and give a vague task like “do it well.” The agent is strong not in his “mind”, but in the fact that he brings a specific result to the end, step by step.