Model resilience
A real operation needs to degrade with intention when the AI layer cannot complete the turn.
Short answer
If the model fails, the system should fall back to containment behavior: no fake answer, no lost state, and a safe path to continue.
There may be a transient outage, a timeout, or an inability to complete the turn through the primary AI layer.
If the system stays silent or replies with something random, the experience breaks at the exact moment the customer needs more confidence.
The correct behavior is to acknowledge the interruption, preserve useful context, and route toward fallback, retry, or help depending on what the operation allows.
The technical detail is not necessary, but it does help to communicate that there was a problem and how the flow will continue.
It avoids dead-end conversations, protects trust, and preserves continuity even when the main layer is unavailable.
Dialo.ar is designed to sustain sales and support even when the conversation stops following the ideal path.