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

The AWS Machine Learning Blog published a post on enabling observability for OpenAI Codex coding agents that use Amazon Bedrock as the model provider. Developers run Codex locally, and it emits OpenTelemetry (OTel) metrics. These metrics are sent to a collector running on each developer workstation, which enriches them with organizational context (user, team, department, cost center) and forwards them to Amazon CloudWatch using AWS Signature Version 4. The reference deployment includes a CloudWatch dashboard with rolling 24-hour totals for active users, conversation turns, API requests, and token usage, plus views by model, user, team, and more. The post describes the architecture, implementation steps (enabling CloudWatch OTel capabilities, deploying the dashboard, generating per-developer configuration, configuring Codex, and validating the flow), and governance considerations such as metric cardinality control and phased rollout. It distinguishes usage telemetry from billing-grade cost data, recommending AWS Cost and Usage Reports (CUR) 2.0 for financial tracking.