Tool calling is what lets an AI do something instead of just describing it. Given access to a set of tools — a web search, a calculator, your calendar, a database query — the model can decide to call one, pass it the right inputs, read the result, and continue. It is the hands attached to the model's brain.
This is the capability that turns a chatbot into an agent. A model that can only generate text can advise you to check today's prices; a model with tool calling can actually go look them up and come back with the number. The same mechanism, standardised, is what the Model Context Protocol exposes to many tools at once.
Why it matters at your desk. For an engineer, tool calling is why Cursor can run your tests and read your files rather than guess; for a freelancer, it is why workspace agents in ChatGPT and the new Notion workspace agents can act across your documents instead of living in a separate chat box.
What to watch for: a model decides when and how to call a tool, and it can call the wrong one or feed it bad inputs. Tools that take real actions — sending, deleting, paying — deserve a confirmation step, because the convenience of "it just did it" cuts both ways.