What is an AI agent?

An AI agent is a software system that uses an AI model to pursue a goal with limited human direction. An AI agent plans the steps, uses tools such as business software, databases and search, checks its own results and adjusts, instead of following a fixed script or answering a single question.

Why it matters for your business

Most AI tools answer a question and stop. An agent carries a piece of work through: it can read an incoming request, look up the customer record, draft the reply, update the system and pass the unusual cases to a person. That moves AI from helping one individual to taking on steps of a business process, where the larger time savings sit. It also raises the stakes, because software that can act can also act wrongly.

How it works

A person sets the goal and the limits: what the agent may access and do. The agent breaks the goal into steps. It uses tools to gather information and take actions. It checks the result, then continues, corrects itself or hands over to a person.

Costs, limits and when not to use it

Agents are slower and cost more to run than a single AI request, and errors can compound across steps. Anthropic's guidance on building agents is to use the simplest approach that works and add autonomy only when the steps cannot be predicted in advance. If a task follows the same path every time, a fixed workflow is cheaper and more predictable. Actions that are hard to reverse, such as payments, need human approval.

AI agent vs chatbot

A chatbot answers within a conversation. An agent decides what to do next and uses other systems to do it.