Understanding how AI coding agents interact with software is a growing challenge for developers. Microsoft has developed Scope, an open-source platform built on Kubernetes, to address this. Scope turns agent interactions into actionable evidence for product teams. It supports a continuous learning loop grounded in repeatable experiments, demonstrating its value for product improvement. The Azure Cosmos DB team, for instance, resolved numerous issues and improved documentation through Scope's evaluations. Scope helps identify documentation errors and assess how agents use SDKs and APIs. Its open-source nature allows broader community contribution to address diverse agent behaviors and integration scenarios. The MIT License permits teams to use and extend Scope, with self-managed deployments enabling private workload evaluation. Scope streams agent activity, capturing interactions and snapshots for detailed review. This observable evidence facilitates concrete improvements to documentation, SDKs, and tooling.