What is an AI agent?

An AI agent is a system that works towards a goal in several steps and decides the next step itself. Today that usually means a language model with tools: it can open a file, look something up in another system, call a calculation and use the result for the next step. Where a language model gives one answer, an agent works through a sequence, for example unpacking a request and laying out the files and quantities per part.

What does an AI agent depend on?

What an agent may do is set by whoever configures it: which systems it can access, where it may write, which actions need a confirmation and what is recorded of each step. Because each step builds on the previous one, an early error can carry through everything that follows. Limited permissions, a log of every step and human confirmation for irreversible actions are therefore common design choices.

How does an AI agent work in practice?

An agent can cut down the searching and retyping around a request. In an estimate, every value that counts in the price should have a source and a moment at which someone confirmed it; a step nobody sees is a step nobody can check. Sending a quote, confirming an order or changing a price are decisions with consequences. With any agent, ask which steps it takes itself, what it only proposes and what remains visible of what it did.

What is an AI agent often confused with?

A script or RPA bot follows a prescribed sequence; an agent decides the next step from the model's output. That sets it apart from fixed automation. And an agent is more than a language model: the model generates text, the agent uses it to carry out steps in other systems.