The difference from a chatbot is the loop: the model decides what to do next — search, call an API, write a file — sees the result, and continues until the task is done or the budget runs out. 'Summarise this email' is a call; 'triage the inbox and draft replies' is an agent.
What makes one production-grade is the same thing that makes any automation production-grade, squared: permissioned tools, a step limit, logging of every action, and a human checkpoint before anything irreversible happens. An agent that can send an email needs the same review an intern would get.
Related terms
AI agent
A language model given tools it can call and a goal to pursue, so it decides the steps rather than following a fixed script.
Tool calling
The mechanism by which a model asks the surrounding application to run a named function, and gets the result back as text.
AI guardrails
The checks around a model that constrain what it can say and do — input filters, output validation, permissioned tools — so a bad response fails safely.
The bench this belongs to
AI automationThe repetitive half of your week, handed to software that does not get bored. Inbox triage, follow-ups, reporting, data entry between tools that were never meant to talk.
