Glossary
Tool use
Tool use is an AI agent calling an external function — an API, a database query, a browser action — to get information or make something happen outside the conversation.
What it changes
Tool use is the line between describing and doing. Without tools a model can tell you how to update a record; with them it updates the record. Every meaningful capability and every meaningful risk enters here.
How it works
The agent is given a description of each available tool and what it expects. When a step needs one, it emits a structured call, the platform executes it, and the result comes back for the agent to react to.
The governance point
Tools are where the approval boundary belongs. A read-only tool can usually run freely; one that sends, pays or publishes is where a human gate earns its place. Which tools an agent holds is the most consequential configuration decision you make.
Related terms
FAQ
What is tool use in AI agents?
An agent calling an external function — an API, a database query, a browser action — to get information or make something happen outside the conversation. It is the line between an agent describing an action and performing it.
How do you control what an agent's tools can do?
By scoping which tools it holds and gating the consequential ones. Read-only tools can usually run freely; tools that send, pay or publish are where a human approval gate belongs. Which tools an agent holds is the most consequential configuration decision available.