What is an AI agent?
An AI agent is an AI system that carries out a multi-step task on its own, for example by searching, using other systems and making small decisions on the way.
An AI assistant answers what you ask. An agent is given a goal and plans the way there itself. It can fetch information, fill in a form and pass on the result.
Take a buyer who asks an agent to compare several quotes. The agent reads the documents, compiles the terms and suggests a choice. The buyer makes the decision.
The more an agent is allowed to do, the more control matters. It needs clear permissions and a log of what it has done. Decisions that affect people or money should be approved by a person.
How it works
An agent is often built on a language model that has been given tools, such as search, calendars or other systems. The model breaks the goal into steps, picks a tool, reads the result and decides on the next step. It repeats this until the task is done or the agent needs help.
Why it matters
Agents can take on work made up of many small steps across different systems. That might mean gathering information, comparing it and preparing a case for a decision. They are most useful where the steps are clear and the outcome can be checked.
What to keep in mind
Any step can go wrong, and a mistake can carry into the next one. An agent that reads outside content can also be steered by instructions someone has hidden in a text.
Start with agents that suggest and prepare. Let them act on their own only once you know how they behave, and keep a log so you can trace what happened.
What it means for you
Give an agent as little access as the task needs, and decide in advance which steps a person must approve.
Common questions
How is an AI agent different from an AI assistant?
An AI assistant answers and leaves the decision to you. An agent can carry out several steps on its own, such as searching, filling in a form or passing something on.
What are the risks with agents?
An agent that can act in other systems can also make mistakes there. It needs clear limits, restricted access and a log of what it does.
How do we stay in control?
Let the agent propose and a person approve any step with consequences. That is the basis of human oversight.
Related terms
Read more
All terms Updated