What is an AI agent? A plain-English explanation

An AI agent is a program that uses a language model to work toward a goal, choosing some of its own steps on the way. That's the short version. The makers word it a little differently, and the differences matter when you read their pages.

The older idea behind the word

"Agent" was in use long before chatbots. Wikipedia describes an intelligent agent as an entity that perceives its environment and takes actions on its own to achieve goals¹. Under that wording, even a basic thermostat counts¹.

Software makers use a narrower meaning. They mean a language model that can act, not only talk.

How the makers define it

Anthropic says the word can be defined several ways. Some customers use it for fully autonomous systems that work on their own over extended periods. Others use it for set procedures².

Anthropic groups all of these as "agentic systems" and then draws a line between two kinds²:

• Workflows are systems where models and tools are run along paths that code has laid down in advance.
• Agents are systems where the model directs its own process and its own use of tools, and keeps control over how the task gets done.

OpenAI's documentation puts it more briefly. Agents can plan and complete tasks using tools, work with other agents, and keep context across steps³.

The two makers agree on the core. A model, some tools, and the model choosing what to do next. Agent or fixed workflow looks closer at where the line falls.

Agent or chat window

A plain chat window takes your message and replies with text. An agent setup adds tools to that. With tools switched on, each turn brings a choice for the model: use a tool, or respond⁴.

Anthropic's own example sits between the two. Customer support, it says, combines familiar chatbot interfaces with extra abilities added through tool integration². So the same chat box can be a plain chatbot or the front of an agent. What differs is what's connected behind it.

Exceptions and cautions

There's no single agreed definition. One company's "agent" can be another company's workflow, because the word is used for the whole range from fully autonomous to tightly scripted².

So when a product page says "agent", look for the specifics. What tools can it call? Who decides each step, the model or a fixed script? Can it change anything in your accounts, or only read? The pages on tool calling and on what an agent is allowed to do cover those questions.

This page describes what the makers' documents say. It doesn't say any product is or isn't a true agent.