CloudWatch now meters your AI coding agents. Watch the metric.

Every business paying for AI coding tools eventually asks the same question, usually around month four: what did we get for this? On July 20, AWS shipped a dashboard for it. Amazon CloudWatch coding agent insights pulls telemetry from Claude Code, Codex, and GitHub Copilot into the same console as the rest of your operational data. The feature is useful. The metric AWS picked to prove value is the part worth arguing with.

What actually happened

Per the AWS announcement, coding agent insights is built on OpenTelemetry metrics emitted by the agents themselves and presented alongside existing CloudWatch data. For Claude Code it integrates with the Claude apps gateway for AWS, so telemetry flows without additional instrumentation — you configure the gateway to emit to CloudWatch and the dashboard populates. Codex and GitHub Copilot are also supported.

What it surfaces: spending trends, token billing alerts, model cost comparisons, and adoption correlated against commit throughput. AWS frames the questions as which teams would benefit from expanded access, where agents are accelerating delivery, and how to right-size token budgets across departments.

Pricing is standard CloudWatch OpenTelemetry metric ingestion — no separate SKU. Available in all AWS commercial regions except Middle East (UAE), Middle East (Bahrain), and Israel (Tel Aviv).

Why metering your AI coding spend matters for your business

Two things here, and they point in opposite directions.

The good one: token budgets are real money and most teams have no idea where theirs goes. An agentic coding session can burn through more tokens in an afternoon than a month of chat usage. Alerting on that before the invoice arrives is genuinely worth the setup time. If you're already on AWS, this is close to free.

The one we'd push back on: commit throughput is not output. Correlating agent adoption with commits measures how much code got written, and more code is not the goal — working software is. An agent that generates 3,000 lines nobody can maintain will look excellent on this dashboard. If you're going to instrument anything, instrument the thing you actually sell: features shipped, bugs escaped to production, time from request to deploy. Those are harder to graph, which is exactly why the easy metric wins by default.

The quietly important detail is OpenTelemetry. These metrics are an open standard, emitted by the agent, not an AWS proprietary format. Which means the same Claude Code telemetry can go to Grafana, Datadog, Honeycomb, or a Postgres table you own. AWS built a convenient default. You are not required to accept the storage bill that comes with it — and for a shop running three developers, a nightly export into your own database costs nothing and answers the same questions.

Instrument the spend. Choose your own success metric. Keep the data somewhere you can leave.

Sources: AWS.