Build visibility for Codex on Amazon Bedrock with OpenTelemetry and Amazon CloudWatch

As organizations transition from experimenting with coding agents like Codex to deploying them across engineering teams, understanding adoption, consumption, and reliability becomes crucial. Codex can emit OpenTelemetry (OTel) metrics about its activity, which, when routed through a local collector to Amazon CloudWatch, provides an AWS-native view of usage. This setup allows visibility by user, team, department, or cost center without adding centralized proxies to the model request path.

The telemetry data includes metrics such as API requests, token usage, conversation turns, and tool calls, which can be visualized in CloudWatch dashboards. These insights help technology leaders distinguish broad adoption from isolated experimentation, identify workflow patterns, and investigate degraded experiences. The approach also supports governance by enabling access controls and privacy considerations.

The architecture involves a developer signing in via IAM Identity Center, running Codex with Amazon Bedrock, and sending metrics to a local collector on the same host. The collector enriches metrics with organizational attributes and signs requests to CloudWatch using SigV4. The deployment process includes enabling CloudWatch OTel capabilities, deploying dashboards and collectors, generating per-developer configurations, and validating the complete telemetry flow.

This pattern supports informed decision-making around scaling Codex, balancing developer enablement with operational governance and cost management. It offers a scalable, non-intrusive way to monitor AI tool usage and operational health, facilitating a measured path from pilot to enterprise-wide adoption.