An agent iteration limit caps how many rounds an agent can spend attempting, reviewing, or revising a result. The limit controls resource use and prevents a workflow from continuing indefinitely when it is not converging.
Reaching the limit does not mean the task succeeded or failed by itself. The workflow should report the current result honestly, including unmet quality requirements and any further work that would be needed.
ELI5
An agent iteration limit is like giving an AI helper a fixed number of tries. When the tries are used up, it must stop and show what it achieved so far.
For example, if the limit is three rounds, the helper can create a scene and revise it twice. If an object is still misplaced, the task should be reported as unfinished.
Why use an agent iteration limit?
It bounds time and cost, prevents endless revision loops, and creates a clear point for reporting or human intervention.
Should the iteration limit be the same for every task?
No. The limit should reflect task complexity, cost, risk, expected quality, and the value of additional attempts.
