Tool use, also called function calling, is the ability of an AI model to request that a specific piece of software be run on its behalf. Instead of guessing an answer, the model asks to call a defined function such as "check stock for this SKU" or "create a booking", and then uses the result.
The developer describes each available tool to the model, including its name, what it does and what information it needs. During a conversation the model decides when a tool is useful and produces a structured request. The surrounding software runs it, often through an API, and passes the result back so the model can reply.
This is how an AI assistant moves from chatting to doing. A retail assistant can look up a live price, a hospitality assistant can check table availability, and a service agent can find an order status. It is the link between the language model and systems like POS, bookings and CRM.
Watch the permissions. A tool that can only read data is low risk, while one that can refund, delete or send messages needs limits, validation and sometimes human approval. Confirm that the AI cannot call tools beyond what it needs, and that every call is logged so mistakes can be traced.
It matters because answers based on live data are far more accurate than those from a model's general knowledge, and because it is the foundation of every useful AI agent.