An agent becomes useful when it can call tools. The simplest way to give it one is a connector: a small piece of code that turns the model's request, such as search("ferry times"), into the real call the tool expects, then turns the answer back into text.
With one agent and one tool this is cheap. You write the connector once, test it, and move on. The trouble starts when the numbers grow.
💡Good to know
A connector is glue: it knows the tool's format on one side and the agent's on the other.