Overview
A persistent agent is an operational system, not a one-time prompt. Uptime, runaway usage, authentication failures, restarts, and broken integrations are normal parts of operating it. The recovery design should therefore exist before the system is loaded with important workflows.
Cost architecture matters for the same reason. An agent that can schedule jobs and loop toward goals can consume far more model usage than an ordinary chat. Flat subscription usage gives a beginner a safer starting point. A fallback model reduces interruption. A repair agent reduces the chance that a nontechnical user becomes blocked by an installation problem.