What is an AI agent?
An AI agent is an AI system given a goal, a set of tools, and the authority to act — choosing what to do at each step, not following a fixed script.
An AI agent is an AI system that's given a goal, a set of tools, and the authority to act — and then chooses what to do at each step to reach the goal. The defining trait is agency: it isn't following a fixed script, it's deciding.
The three ingredients
- A goal — an outcome to pursue ("qualify this lead," "resolve this ticket"), not a single instruction.
- Tools — the actions it can take: send a message, query a database, take a payment, call an API. An agent's power and its risk both come from its tools, which is why scoping them matters.
- A reasoning loop — usually a large language model that, given the goal and what's happened so far, picks the next action, observes the result, and continues until the job is done.
Agent vs. model vs. assistant
A language model on its own only produces text. An agent wraps a model with tools, memory of the task, and a goal — so it acts. An assistant keeps a human in the loop on each step; an agent is trusted to run the task and report back. For why an agent is also different from a scripted automation, see AI agent vs workflow.
Why tool scope is the safety question
Because an agent acts, the important question is what it's allowed to do. A well-governed agent is granted exactly the tools its role needs and no more, enforced by the runtime rather than asked for in a prompt. Compose several scoped agents and you get an agentic app that runs a whole process.
For how this plays out in production — isolation, least-privilege access, and audit — see the enterprise AI agents guide.
Frequently asked questions
What's the difference between an AI agent and a large language model?
An LLM produces text. An AI agent uses an LLM to decide, but adds tools to act, memory of the task, and a goal to work toward — so it does things in the world, not just generate words.
What is 'agency' in an AI agent?
Agency means the system chooses its own next step toward a goal rather than following a pre-defined sequence. Given 'book a meeting with this prospect,' it decides whether to check a calendar, send an email, or ask a question.
Are AI agents and AI assistants the same thing?
Not quite. An assistant helps a person who stays in the loop on each step. An agent is trusted to run a task on its own and report back, escalating only when it needs a human.