A housing work-order agent sometimes fails when calling maintenance APIs. What should operations dashboards track?

A housing work-order agent that sometimes fails when calling maintenance APIs needs operations dashboards that track tool-call failure rates and related errors. Agents that look fine in chat can still leave residents stranded if tools fail. Tool-call success/failure rates, error codes, and related exception trends show whether the maintenance integration is healthy. That is the operational signal for this civic agent—not cosmetic or marketing metrics.

Token usage volumes and related costs for the work-order agent fail because usage and cost metrics track spend and efficiency, not whether maintenance API calls succeed, and cannot show current failure rates, spikes, or error codes. Dialogue completion times and related latencies for the work-order agent fail because latency shows how slow conversations are, not whether live maintenance API calls are failing for residents, and cannot show operators tool-call error rates or failure spikes. Grounding retrieval precision and related recall for the work-order agent fail because retrieval quality shows whether the right information was found, not whether the agent's specific maintenance-API tool calls executed, and can look good while tool calls are still failing underneath it.

Exam caveat: for tool-using municipal agents, ops must monitor tool-call failure rates and related errors—not app cosmetics, marketing CTR, or office printers. Operational check: instrument every maintenance API tool invocation with success/failure and error class, dashboard failure rate and spikes for the work-order agent, alert on sustained error rises, and page on-call when tool failures correlate with open work-order backlogs.