Tool calling is the feature that turned chatbots into Agents

What happened

Model providers converged on the same capability: you describe a set of tools (a name, what it does, what inputs it takes), and the model can answer with "call this tool with these inputs" instead of prose. Your code or workflow tool runs the call and hands the result back. The model continues.

Different providers name it differently (tool use, function calling), but the shape is the same, and the common workflow tools expose it as a checkbox or a "tools" panel on the model step.

Why it matters

This is the mechanism behind every Agent on this site. Before it, "read my calendar and summarize" meant you fetched the calendar yourself and pasted it into the prompt. After it, the model asks for the calendar when it decides it needs it, and can ask for more (a specific day, an attendee's other meetings) without you predicting every path.

Two practical consequences:

• Prompts get shorter. You describe tools well once, and the brief only states the goal and the rules.
• Permissions move to the tool definition. A model can only call what you listed. Not listing "send email" is a stronger guarantee than a rule in the prompt saying "do not send email."

What you can try

We tested this with two workflow tools and two providers; the behaviour was the same in all four combinations. The tool definition's description text matters more than you would expect; a vague description gets the tool called at the wrong moments.