Not Just One API Call

Building With AI APIs covered a single request and response: you send one input, the model sends back one output, and that's the whole interaction. The perceive-decide-act loop is a genuinely different pattern — it keeps cycling instead of stopping after one exchange.

In the loop, each cycle builds on the last: perceive the result of the previous action, decide the next step based on it, act, then perceive the new result. A single API call has no "previous action" to build on — it's one shot, and whatever happens next is up to whoever reads the response.

That raises an obvious question: if the loop keeps cycling instead of stopping after one response, what makes it stop at all? A loop with no way to end would just keep running — so a real agent needs some condition that tells it the run is over, which is exactly what the rest of this level covers.

How does the perceive-decide-act loop differ from a single API call?

A single API call is one request and one response; the agent loop keeps ____ instead of stopping after one exchange.

A loop with no stopping condition at all would simply keep running indefinitely.