Image: AWS / aws.amazon.com AWS Launches CloudWatch Coding Agent Insights for Claude Code, Codex, and Copilot
Amazon CloudWatch now surfaces OpenTelemetry metrics from AI coding tools in a ready-made dashboard, letting engineering leaders track usage, cost, latency, and tool approvals across Claude Code, OpenAI Codex, and GitHub Copilot.
Amazon CloudWatch added Coding Agent Insights on July 20, giving engineering organizations a prebuilt dashboard for the telemetry their AI coding tools emit. It works with Claude Code, OpenAI Codex, and GitHub Copilot, and it drops into CloudWatch without any custom instrumentation if you’re already routing through the Claude apps gateway on AWS.
What it tracks
The dashboard pulls OpenTelemetry metrics from whichever coding agents you have configured and presents them alongside your existing CloudWatch data. On the usage side: token consumption, turn-level latency, tool calls, API requests, and tool approval rates. Teams can slice all of it by organization, department, team, cost center, and user. The data is also exportable as CSV and queryable with PromQL for anyone who wants to build their own views.
The practical use cases are mostly about making agent adoption visible at scale. Which teams are getting the most out of expanded access? Where are agents accelerating delivery and where are they slowing things down? What does per-seat token spend actually look like? These are questions that have been hard to answer without building your own pipeline, and CloudWatch Coding Agent Insights is AWS’s answer to that.
You can also set proactive token billing alerts, which matters now that GitHub Copilot, Cursor, and Anthropic have all moved parts of their billing toward usage-based models. Running Claude Opus 5 across a large engineering org adds up quickly, and visibility before the invoice is more useful than surprises after it.
Setup
For Claude Code specifically, the setup runs through the Claude apps gateway for AWS. Once the gateway is emitting telemetry, CloudWatch picks it up automatically. Codex and Copilot have their own integration paths documented in the setup guide. No custom SDK instrumentation is required for any of them.
This sits alongside AWS’s existing work on Bedrock AgentCore and the Strands Agents SDK, which have observability built in at the framework level. Coding Agent Insights targets the developer tooling layer rather than the application agent layer.
Availability
CloudWatch Coding Agent Insights is live in all AWS commercial regions except Middle East (UAE), Middle East (Bahrain), and Israel (Tel Aviv). Standard CloudWatch OpenTelemetry metric ingestion pricing applies.
Full details are in the AWS announcement.