Your AI coding tools finally have a bill you can read!
Amazon CloudWatch launched coding agent insights on 20 July. A dashboard that shows engineering leaders what AI coding tools are actually doing across an organisation. It reads OpenTelemetry metrics straight from Claude Code, Codex and GitHub Copilot and puts them next to your existing operational data: total tokens, total cost, active users, sessions, cache hit rate, active hours. It will correlate agent adoption against commit throughput and pull-request velocity, flag which models give the best cost-to-output ratio, and fire alerts before a token budget runs away from you.
What it actually means
For a month this Digest has asked which model you rent. This is the question that comes after, and it's the less glamorous one: can you prove what you're already spending, and on what? Most teams adopted coding agents the way they adopt anything good. One enthusiast, then a team, then a bill nobody owns. The measurement layer arriving is the tell that the experiment phase is closing and somebody upstairs is about to ask for a number. Get ahead of that. But read the dashboard honestly, because it measures the measurable of tokens, sessions, PR velocity and those are inputs, not value. High usage on a team shipping nothing is not a win, and a quiet team with good judgement is not a failure. The genuinely useful move here isn't the ROI slide; it's cost attribution. Knowing which workloads and which models eat your budget is what lets you move the boring bulk to something cheaper and keep the expensive model for the parts that earn it.
20 Julthe day AI coding spend became a dashboard line item
Source: AWS · Sunday 26 July 2026