AWS Launches CloudWatch Coding Agent Insights

AWS documentation says the dashboards appear automatically under GenAI Observability after an agent begins sending metrics in the expected shape. The supported telemetry includes token consumption, per-turn latency, tool calls, API requests and approvals. Teams can segment the data by organization, department, team, cost center and user, export it as CSV, or query the metrics with PromQL.

The schema requirement is central to the release. Native agent metrics provide dimensions such as model and token_type, while fields including user.email, team.id, department, cost_center and organization must be sent as OpenTelemetry resource attributes. AWS warns that changing metric names or moving identity fields off the resource can prevent the dashboards from populating.

AWS documents a self-hosted Claude apps gateway for enterprise Claude Code deployments. The gateway can authenticate developers through a corporate identity provider, hold upstream credentials, enforce group-based model access and managed settings, and route identity-stamped telemetry to CloudWatch or other collectors. AWS lists Claude Code 2.1.195 or later, an OpenID Connect provider, PostgreSQL 14 or later, TLS and a private-network address among the prerequisites.

For data and platform teams, the practical work extends beyond enabling a dashboard. User identities, team labels and cost-center fields need consistent definitions, controlled access and appropriate retention. Measures such as token volume, commits or pull requests can describe activity and cost, but they do not by themselves establish productivity or business value. That distinction is LDS interpretation; AWS documents the available metrics and proposed analyses, not a universal productivity methodology.