The phrase "AI agent" is everywhere and rarely defined. Stripped of the hype, an agent is software that can be given a goal and then take the steps to reach it, rather than just answer a question about it.
The short version
An AI agent is a system built on a language model that can reason about a task, choose actions, use tools to carry them out, and check its own progress. The model supplies the reasoning; the tools let it act; the goal you give it sets the direction. The result is something that does work, not just talks about it.
Agent versus assistant
A chat assistant responds to each message in isolation: you ask, it answers, you act on the answer. An agent holds a goal across many steps and works toward it, deciding what to do next based on what it finds. Ask an assistant to "summarise this" and it does; ask an agent to "close out the month" and it invoices, chases, reconciles, and reports, calling each tool in turn.
What makes an agent useful
Three things. Tools: the actions it can actually take, like creating a record or sending a message. Reasoning: the judgement to pick the right tools in the right order for the goal. And bounded autonomy: the freedom to act, inside limits you set, so it can finish a task without a human approving every step but cannot stray beyond what it is allowed.
Agents at work in a business
Inside software built for them, agents stop being a novelty and become useful. Connected to your workspace, an agent can run real operations: finance, the inbox, sales, stock, the work that used to need a person at a keyboard. It acts within that person's permissions and a budget you control, and routes the decisions that need you back to you.
Where to go next
If you want the precise vocabulary, the glossary defines the terms. If you are weighing the words, chatbot, copilot, or agent draws the distinction. And to see one operate your own business, connect the agent you already use.