What is an AI agent? Explained with a school bag
An agent is not a smarter chatbot. It is something that starts on its own and does several steps in a row. A school bag, packed the night before, turns out to be the clearest way to explain it.
We asked a group of eleven-year-olds to write down, in exact order, everything that has to happen for a school bag to be packed correctly the night before. Then we had a parent follow only what was written, with no common sense allowed to fill in gaps. The parent packed two left shoes because "pack the shoes" didn't specify which pair, and forgot the water bottle because it was never mentioned at all. The list everyone thought was obvious turned out to have holes in it the moment something followed it literally, without a person there to notice what it meant.
That exercise is the clearest door we have found into what an agent actually is, because a school bag packed the night before already has the shape of one: a trigger, several steps done in order, and no human standing over it while it happens.
What is an AI agent?
Everything before this level in our pathway has needed a human to type a message before anything happened. An agent is different: it starts on a trigger - a time, an event, a condition - and then chains several steps together on its own, the way a school bag gets packed once, in order, without anyone standing over each item.
Waits for a message. Answers it. Waits again. Nothing happens unless a person starts it.
Starts itself at 7pm every evening. Checks what's left to do. Drafts tomorrow's step. Stops, and waits for approval before anything leaves the house.
Why the trigger is the whole difference
A trigger can be a clock, an email arriving, a form being filled in, a sensor moving, a date getting close. The specific trigger matters less than the fact that one exists at all - it is what separates something that runs itself from something that only ever waits to be asked.
The part every family session gets to eventually
Once children understand that an agent runs itself, the very next question they ask - almost without exception - is where it should stop and check with a person before going further. That question is worth taking seriously rather than answering quickly, which is exactly why it is the entire subject of the next article in this level.
One honest limit worth stating plainly: instructions are always vaguer than they feel while you're writing them. The two-left-shoes problem is not a one-off funny mistake - it is the default outcome of any set of steps followed literally, by something with no shared common sense to fall back on. Designing an agent well means assuming your instructions have a gap in them somewhere, not hoping they don't.
Questions we get asked
What is an AI agent?
It is a system that starts itself on a trigger - a time, an event, a condition being met - and then carries out several steps in a row without a person typing a new message for each one. This is the key difference from a chatbot, which only ever responds to a message someone sends it.
What is the difference between an AI agent and a chatbot?
A chatbot waits for a person to send a message, answers it, and then waits again - nothing happens without a human starting each step. An agent starts on its own from a trigger like a time of day or an incoming email, and then chains multiple steps together in sequence before it needs a person again.
How do you explain AI agents to a child?
A school bag packed the night before works well: there's a trigger (bedtime), a sequence of steps done in order (books, then lunch, then shoes), and no one standing over each item as it happens. An agent works the same way, just with digital steps instead of physical ones.