What is an AI agent?

What an AI agent is

A chatbot answers a question. An AI agent does a job. You give it a goal. It makes a plan, uses a tool, checks what came back, and decides what to do next. It repeats this loop until the job is done, then gives you the answer.

That loop is the whole idea. Anthropic describes agents as language models that use tools in a loop, guided by what each step sends back.

How an AI agent is made

Most agents have four parts.

• A brain: the language model. It reads the goal and decides what to do next.
• Hands: the tools. A web search, a file, a calendar, an app. Tools let it act, not just talk.
• A rulebook: the instructions and guardrails. What the job is, how to do it, and what it must never do.
• A notebook: memory. What it has seen and done so far in this job, and sometimes things from before.

The model is not the agent. It is one part inside it. The agent is the model, plus the tools, plus the loop.

What an AI agent needs

• A clear goal. Vague asks give vague results.
• The right tools, each with a plain description, so it knows when to use which.
• A check after every step. It should look at what a tool returned before it moves on.
• A stop rule. A limit on how many steps it may take, so it cannot run forever.
• A human for big decisions. It should pause and ask before it sends, buys, or deletes anything.

Agents that skip these make confident mistakes. The good ones are tested in a safe sandbox first.

Try one

The quickest way to understand an agent is to run one.

Try a free agent Watch the video guide

Sources: Anthropic, Building effective agents · OpenAI, A practical guide to building agents