Should a voice agent keep taking calls when one of its tools goes down?
Picture a phone agent that books appointments and takes payments. Scheduling still works, but the payment service has gone down. Do you switch the whole agent off until it recovers, or let it keep handling calls that do not need payments and block only the affected actions?
Keep it running, but in a degraded mode you designed in advance. Shutting the whole agent down because one dependency failed turns a partial outage into a full one, and most callers did not need the broken part anyway.
Remove the broken tool from the agent's tool list while it is down, rather than telling the agent in the prompt not to use it. A tool it cannot see is a tool it cannot call by mistake. A health check or circuit breaker can flip this automatically after a few failures and restore it once the service answers again.
Give the agent a script for what it cannot do right now. The caller should hear plainly that payments are unavailable, and get a concrete next step: a callback, a payment link by text once the service is back, or a transfer to a person. Never let the agent pretend a payment went through.
Log every call that hit the missing tool. That list is your follow-up queue once the service recovers, and it tells you whether the outage cost real bookings.
Listings mentioned
- Error Handling & Observability · skill by CloudAI-XCovers retry strategies, circuit breakers and graceful degradation, the patterns behind switching one tool off without stopping the agent.
- Release It! Framework · skill by wondelaiApplies stability patterns such as circuit breakers, bulkheads and timeouts so one failing service does not take down the rest.
Answers by the AgentAlley team, drafted with AI and checked against the listings they link to. Not a real-person reply from the original thread.