Skip to course content
Free agentic AI fundamentals course

Introduction to Agentic AI and Workflow Automation

Unit 07.00: Loops, and the two detectors that catch them

Loops are the most expensive agent failure and the easiest to detect.

A ceiling and a repeat counter

The ceiling bounds total steps. The repeat counter watches for the same tool with the same arguments.

The example below runs both over a five-step sequence.

1. lookup   INV-1187    
2. lookup   INV-1187    
3. lookup   INV-1187      <- REPEAT DETECTOR: ('lookup', 'INV-1187') seen 3x
4. search   contract    
5. lookup   INV-1187      <- REPEAT DETECTOR: ('lookup', 'INV-1187') seen 4x

repeat detector fired at step 3; the ceiling would not fire until 5

The repeat detector fires at step three; the ceiling would not fire until step five. On a tight loop that difference is two paid steps every time it happens.

The ceiling still earns its place for a different shape: an agent that keeps doing slightly different things without converging, never repeating an exact call. The repeat detector never fires on that one.

The mistake this prevents

The mistake is a ceiling alone because it is simpler to implement. A ceiling of eight lets a two-step loop run four times, which is eight paid calls to learn something detectable at the third.

Takeaway

Run both detectors. The repeat counter catches tight loops early; the ceiling catches drift that never repeats exactly.