Week 3 was about one great prompt. This week the model gets to send many prompts: a loop where it sees a result, thinks, and chooses its own next step. The loop is the part of an agent that you own: you can read it, budget it, audit it, and stop it.
No tools, no verification, no recovery. If the memory is wrong, the answer is wrong, and nobody notices.
Each observation is fed back before the next decision. Wrong turns become visible, and recoverable.
The loop is four arrows and one check. Press Play to watch a travel-cost question answered in three iterations, then watch the same loop get stuck and be rescued by its max-iteration guard. The guard is a safety cap; the stop condition is the goal. They are not the same thing.
Try this: in “Stuck”, name the exact iteration where a human reading the trace would have noticed the loop. What observation repeated?
An explicit loop logs every thought, tool call, and state change. When compliance asks "why did the system do that?", you replay the trace.
Steps are the unit of spend. A max-iteration guard and a token budget turn "AI costs" from a surprise into a line item.
Feeding observations back is what lets an agent notice failure and recover: the difference between confident nonsense and a checked answer.