Stop Conditions
An agent needs explicit rules for when it is done, when it is stuck and when it must stop for safety.
An agent needs explicit rules for when it is done, when it is stuck and when it must stop for safety.
Define the completed state.
Cap steps, time or cost.
Stop and ask a person when needed.
Understand the working model
Without stop conditions, an agent can loop, repeat actions or keep spending resources because “try again” always appears possible. Bound the loop before deployment.
Use both positive and negative stops: finish when the objective is verified; stop when step limit is reached, required data is missing or a protected action is requested.
Practice
Apply the pattern above to one real task you already do. Keep the first version small enough that you can inspect every input and output.
Do not use “the model decides when it is done” as the only stop rule.
Bound agents with explicit success criteria, resource limits and escalation rules.