AWS tracks Codex adoption on Bedrock with new CloudWatch OTel dashboard

As coding agents like OpenAI's Codex move from pilot experiments to company-wide rollouts, engineering leaders need more than anecdotal evidence of usefulness. AWS's new guidance shows how to build exactly that visibility with native tools: Codex can emit OpenTelemetry (OTel) metrics, and when developers authenticate via AWS IAM Identity Center, a local OTel collector on each workstation enriches those metrics with user, team, department, and cost center attributes before forwarding them to Amazon CloudWatch. Crucially, the architecture avoids a centralized proxy—developers keep using Codex locally, and the collector only listens on 127.0.0.1, signing requests to CloudWatch with AWS SigV4. This keeps the model-request path untouched while giving administrators a unified view.

The bundled CodexOnBedrock dashboard provides rolling 24-hour totals for active users, conversation turns, API requests, and token usage, with drill-downs by model, token type, user, department, team, and session source. That telemetry helps answer practical questions: Is adoption spreading across teams or concentrated in a few power users? Where is token consumption highest for showback? Are agentic workflows (tool calls) taking hold? Are API status and duration metrics pointing to degraded experiences? AWS stresses that CloudWatch metrics are not a billing ledger—for actual spend, use AWS Cost and Usage Reports (CUR) 2.0 with IAM principal allocation. The reference implementation, available in the Codex on AWS guidance repository, gives enterprises a template to scale Codex responsibly.