Amazon CloudWatch announced Coding Agent Insights on 20 July 2026 — a dashboard purpose-built to answer the question that’s been dogging every engineering leader who greenlit an AI coding tool rollout: is it actually working, and for whom? It pulls OpenTelemetry metrics directly from Claude Code (via the Claude apps gateway, with no extra instrumentation required), OpenAI’s Codex, and GitHub Copilot, and lines them up against CloudWatch’s existing operational data — commit throughput, pull request velocity, deployment frequency.
The practical use case is budget defence. Instead of “our developers say it feels faster,” leaders get answers to specific questions: which teams would benefit from expanded access, where are agents demonstrably accelerating delivery, which models deliver the best cost-to-output ratio, and how token spend should be right-sized across departments. For organisations running Claude Code at scale, a bearer-token setup gets a single developer or small team streaming metrics in minutes; a self-hosted gateway option gives larger orgs centralised identity control and telemetry collection across the whole engineering function.
So what
This is a tell, not just a feature launch. When AWS builds a dedicated observability product for “is this AI coding tool earning its keep,” it means the questions boards and finance teams are asking about AI coding spend have moved past “should we adopt this” into “prove it’s paying off” — and most engineering organisations don’t currently have a straight answer. If your team adopted Claude Code, Cursor, or Copilot in the last year and nobody can tell you which projects it actually accelerated, that’s a governance gap worth closing before the next budget cycle, not after. That’s the kind of AI tooling audit and rollout structure we help clients put in place through our AI-assisted development work — get in touch if you want to know what your own team’s numbers would actually show.