Glossary
AI Agent
An AI agent is a system that uses a language model to decide which actions to take toward a goal (calling tools, querying systems, and continuing over several steps) rather than producing a single block of text.
Glossary
An AI agent is a system that uses a language model to decide which actions to take toward a goal (calling tools, querying systems, and continuing over several steps) rather than producing a single block of text.
A chatbot answers. An agent can look up an order, check a policy, draft a reply and, where permitted, complete a step in a process, deciding at each point what to do next.

An agent given broad permissions and vague instructions is a system that will eventually take a confident, incorrect action against real data. The useful question is never how autonomous it can be, but which actions it may take unsupervised and what happens at the edge of that set.
A defined task, a small set of tools, explicit permissions, a clear confidence boundary, and an escalation path that hands a person the full context rather than a bare failure.
Most agent projects that disappoint were scoped by what the technology could theoretically do rather than by which decision a business was willing to delegate.
A chatbot produces replies. An agent can take actions (querying systems, calling tools, completing steps) and decides which to take next. That makes permissions and escalation the central design questions.
Only with defined boundaries. A narrow task, limited permissions, grounded answers and an escalation path to a person when confidence is low. Broad autonomy over real data without those controls is where agent projects go wrong.
An AI agent is a system that uses a language model to decide which actions to take toward a goal (calling tools, querying systems, and continuing over several steps) rather than producing a single block of text.