What an agent actually is

Strip away the excitement and an agent is a small, understandable thing.

So far your automations follow steps you wrote. An agent is an automation where a model decides which step to take next. You give it a goal and a set of things it is allowed to do; it chooses among them, looks at the result, and chooses again until it is done or you stop it.

That is the whole idea:

goal + available tools + a loop

A worked example. Goal: answer this student's enquiry.

Available tools: look up a course fee, check batch dates, create a lead, send a reply.

The model reads the enquiry, decides it needs the fee, calls that tool, reads the answer, decides it now has enough, and writes the reply.

Nothing mystical. A loop, a decision each time round, and a list of permitted actions.

Why it is genuinely useful: when the sequence cannot be known in advance. A fixed automation must anticipate every path. An agent handles the enquiry it was not designed for.

Why it is dangerous, and this is the whole course: the thing choosing the steps is the same system from AI Fundamentals — the one that produces plausible output with no signal for 'I am unsure', and that will state something false with complete confidence.

Give that thing a set of tools and it can now act on a wrong conclusion. A confident mistake stops being a bad sentence and becomes a bad action — a message sent, a record changed, money moved.

So the discipline in this course is not how to build one. It is what to let one touch.