An AI agent is a system with a goal, access to a defined set of tools, and a loop: it decides which tool a task needs, calls it, reads back what happened, and decides the next step — repeating until the goal is met or it hands off to a person.
The mechanics are described plainly in vendor documentation for the tool-calling feature that makes agent behaviour possible. Anthropic’s own developer docs put it this way, in one continuous passage: “Tool use (also called function calling) lets Claude call functions that you define or that Anthropic provides. Claude determines when to call a tool based on the user’s request and the tool’s description. It then returns a structured call that your application executes (client tools) or that Anthropic executes (server tools)”. That is a vendor describing its own product’s capability, not a Canadian standard — but it is the clearest plain description of the decide-call-read-back loop that makes something an agent rather than a chat window.
What separates that loop from ordinary software is that the tool being called, and when, is decided by the model reading the request — not hard-coded by a developer for every case in advance. A support form that always emails the same address on submit is not an agent making a choice; a system that reads the message and decides whether to email support, open a ticket, or reply directly is.
An agent built to triage IT tickets doesn’t only answer “how do I reset my password”: given a new ticket, it decides whether the fix is already in the knowledge base, whether it needs to open a request in the ticketing system, or whether it should escalate to a person — then acts on that decision itself, and checks the ticket’s status again before marking it closed.
See also: what is agentic AI, what is tool use in AI, what are AI guardrails.
An agent is only as useful as what it can reach — ai-integration-automation covers connecting one to the systems it needs to act on.
