City IT just deployed a permitting agent and must observe traces, failures, and usage in production. What should they add?

After deploying a permitting agent, city IT should add tracing and monitoring hooks integrated into the newly deployed agent so traces, failures, and usage are observable in production. Instrumentation is how municipal operators detect tool errors, latency spikes, and abnormal usage during launch week and beyond; without it the permitting counter becomes an opaque black box. Integrating hooks at deploy time—not as an afterthought—matches Foundry operational practice for agentic solutions. Diagnostic log streaming configured for the deployed permitting agent host only streams host-level diagnostic logs and misses the structured agent traces, failure correlation, and usage visibility that integrated tracing provides in production. Token usage analytics dashboards connected to the deployed permitting agent show usage and cost metrics but cannot capture the tool-call fidelity or failure traces needed for modern debugging. Scheduled quality evaluation runs scored for the deployed permitting agent assess quality in periodic scored runs but do not observe live production traces, failures, or usage as they happen. On the exam, deploy-time tracing/monitoring is required; distractors host-level diagnostic log streaming, token-usage analytics dashboards, or scheduled quality evaluation runs. Caveat: telemetry may include resident prompts—apply data-minimization, retention limits, and access controls appropriate to permitting data sensitivity. Operational check: confirm traces for a sample permit dialogue show model and tool spans, alert on failure rates, and verify operators can pivot from a resident incident to the correlated trace ID.