One prompt vs a loop

You already know the shape of prompting: you write a clear request, the model writes its best answer, and you are done. That is a single round trip — one prompt in, one reply out. It is powerful, and it is not going away. But it has one built-in limit that matters enormously: the model never finds out whether its answer actually worked. It hands you the answer and moves on. If the answer was wrong, you discover that, not the model.

Looping removes that limit. Instead of stopping after one reply, the model runs in a cycle: it decides what to do, does it, looks at the result, and then decides again based on what it just saw. It keeps cycling until the job is done. The one-sentence version worth memorizing: prompt engineering gets one response right; looping gets a whole sequence of actions to reach a goal.

Here is the difference made concrete. Imagine you ask an AI to find the cheapest flight for a trip.

• Single-shot prompting gives you its best guess from memory — which may be outdated or invented, and it has no way to know.
• A looping agent searches a travel site (action), reads the actual results (observe), notices the dates were wrong (think), searches again with corrected dates (action), and repeats until it has a real, current answer it can defend.

The looping version can be right in ways the single-shot version simply cannot, because it checks reality along the way instead of guessing once.

Why this is the idea behind every AI agent you keep hearing about. When people say agent, they almost always mean prompting plus a loop plus some tools plus a goal to keep working toward. The loop is the engine. Everything else in this course builds on it.

Recap. Normal prompting is one round trip and never learns whether it succeeded; looping runs a repeating cycle where the AI acts, sees the result, and adjusts until the goal is met. Prompting gets one answer right; looping drives a sequence of actions to a goal. That single change — checking reality between steps instead of guessing once — is what turns an AI that answers into an agent that acts, and it is the foundation of everything called agentic AI.