Why should you care?

If you only need FAQ answers, a chatbot is enough. If you need “check inventory then create a draft PO”, you need an agent.

See it live — copy this example

Use .NET 8+. Run Semantic Kernel samples in a console or Web API. Keep API keys in user secrets or environment variables — never in source.

public interface IAgentStep
{
    Task<AgentObservation> ActAsync(AgentState state, CancellationToken ct);
}

// Chatbot: one CompleteAsync(prompt)
// Agent: while (!state.Done) await step.ActAsync(state, ct);

What happened?

• IAgentStep is a tiny contract for one action.
• The loop is what makes behavior agentic.

Practice next

1. Draw chatbot vs agent on paper.
2. build a fake step that only logs.
3. Add a Done flag after 2 steps.
4. Add a timeout CancellationToken.
5. Return a user-visible progress message each step.

Remember

Chat = one shot; agent = loop. State carries memory and tool results. Stop conditions prevent runaway cost.

Order status helper

User asks “where is my order?”

Outcome: Agent calls Orders API; chatbot would only guess.