An AI agent is AI software that is given tools and access, and acts on its own to reach a goal. A chatbot only answers questions. An agent also does things: it logs in to systems, looks up records, fills in forms and makes payments.
Here is how an agent works. A person or a company gives the agent a goal. The agent runs on a language model, the kind of AI model behind chatbots, which decides each next step. The agent then uses one of its tools, for example to look up a record or send an email. It reads the result and decides the next step again. It repeats this until the goal is reached, or it hands the task to a person. Developers build agents with an AI agent framework.
Agents are used in many industries, and each kind has its own page. AI customer service agents answer customers by chat or phone. Agentic commerce means an agent that searches, books and pays for someone.
Take, for example, a software team with a bug to fix. An AI coding agent writes a fix, runs the tests and changes the code again until the tests pass.
Consider an accounts team that receives hundreds of supplier invoices a week. With agentic automation, an agent reads each invoice, checks it against the purchase order and writes to the supplier when something does not match, work that office staff used to do by hand.
Imagine a warehouse. With physical AI, an agent decides which robot picks which shelf, and in what order.
An agent is only as safe as the access it is given. If it can send refunds, a trick message could make it send a refund that was never owed. AI agent guardrails explains how companies limit what an agent can do. Like an employee, an agent also needs its own login, which is a kind of non-human identity.
How is an AI agent different from a chatbot? Both may run on the same language model. A chatbot gives an answer and stops, and a person must then act on it. An agent takes the action itself, so what it is allowed to touch matters much more.