What it means
An AI agent is a system that uses an AI model to pursue a goal by planning steps, using tools such as search, databases or other software, and adjusting its actions based on the results. Unlike a chatbot that only answers, an agent can take actions, within limits set by its designers.
Real-world example
A university's research office uses an agent to prepare funding application checklists. Given a funding call, the agent reads the call document, searches the internal database for the university's standard documents, checks which are outdated, and drafts a checklist with deadlines for the research team. It cannot submit anything; a staff member reviews and sends the checklist.
Why it matters for AI adoption
Agents can take on multi-step work, but every action they can take is a potential risk if the agent misunderstands the goal, is given misleading input or makes errors along the way. Clear limits on tools and permissions, human approval for consequential steps and full logging are essential.
Often confused with
Ordinary automation follows fixed, predefined steps and behaves the same way every time. An AI agent decides some of its steps itself based on the situation, which makes it more flexible but less predictable and harder to test.
Questions to ask
- What actions can the agent take, and which require human approval?
- How can we see and review every step the agent took?