Question

A pro-code team builds a customer-service triage agent using the Microsoft Foundry Agents service. Their ALM pipeline promotes agent definitions from development to a staging environment and then to production. Leadership requires that no agent version reaches production unless it meets a minimum groundedness and relevance score against a curated evaluation dataset. The team wants this enforcement to be automated and to block promotion when scores fall below the threshold. Which approach best implements this requirement in the ALM process?

Accepted Answer

Add an automated evaluation step to the CI/CD pipeline that runs the agent against the curated dataset, computes groundedness and relevance metrics, and fails the deployment gate if scores are below the defined thresholds.

Answer

Have QA analysts manually chat with the staging agent for a fixed number of hours before each release and sign off in a shared spreadsheet.

Answer

Enable Application Insights on the production agent and configure alerts so the team is notified when live groundedness telemetry drops below the threshold.

Answer

Pin the agent to a specific model version in production so that output quality remains stable across releases without needing evaluation.