You picked your AI coding tool — or you decided to skip it. Copilot, Cursor, Claude Code, or just your brain and a terminal. Either way, it felt like choosing your text editor or your coffee order: personal, untraceable, nobody's business.

But here's the thing about personal choices — they stay personal only until someone builds a dashboard. And nobody in management had a dashboard showing how much you use AI, how often you accept its suggestions, or how your adoption compares to the developer sitting two desks over. Until this week.

On April 10, GitHub shipped aggregated cloud agent active user counts through its Copilot usage metrics API — a programmatic interface that lets organizations pull data about tool usage without asking anyone. Three new fields: daily, weekly, and monthly active users for Copilot's cloud agent. That's the autonomous mode where you assign a coding task to @copilot and it works in the cloud, submitting a pull request when done.

But April 10 wasn't isolated. It was the third metrics expansion in eight days — capping a sprint that started back in late March when GitHub quietly added a used_copilot_coding_agent field letting admins see which specific developers triggered agent sessions. That was the groundwork. Here's the escalation:

  • April 2 — Per-user CLI activity landed. Session counts, request counts, token consumption, CLI version — all per developer. They counted your keystrokes.
  • April 6 — Active vs. passive code review tracking. Did you choose Copilot review, or did a repo policy auto-assign it? GitHub's own words: "Measure real engagement, not just coverage." They measured your enthusiasm.
  • April 10 — DAU/WAU/MAU for cloud agents. The classic engagement metrics that every product manager lives and dies by, now applied to developers using AI. They graphed it.

Three updates. Eight days. Each one adds another per-developer data point to organization-level API endpoints — meaning any company with a GitHub Enterprise license can query these numbers programmatically and pipe them into whatever HR analytics or performance dashboards they already run.

GitHub isn't the only one building the observation layer. Cursor's enterprise tier surfaces per-developer AI usage breakdowns. Anthropic's Claude Code exposes session-level cost data for organization admins. OpenAI's Codex launched with usage analytics built into its enterprise offering when Codex-only seats shipped on April 3. The implementations differ, but the pattern converges: every major AI coding tool now generates a paper trail of exactly how much each person uses it.

Now here's where the dashboard meets reality.

I covered the "Debt Behind the AI Boom" study yesterday — 304,000+ verified AI-authored commits across 6,275 repos. The uncomfortable headline: teams where AI-generated code exceeded 40% of total output experienced 20–25% higher rework rates. The metric that makes you look productive on a dashboard — high AI adoption, lots of agent tasks delegated, lots of suggestions accepted — correlates with worse actual outcomes. If you missed that piece, the short version: AI writes bugs fast too.

This is textbook Goodhart's Law: when a measure becomes a target, it ceases to be a good measure. Except now the target has your name on it.

The tradeoff is sharp. Developers who lean heavily on AI agents show up as "high adopters" in the new metrics — exactly the signal a non-technical manager optimizes for. Developers who use AI selectively, rejecting bad suggestions and writing critical code by hand, look like laggards on a spreadsheet they've never seen. And opting out entirely? That's not a personal preference anymore — it's a visible gap in a dataset that your organization's API calls populate every night.

To be clear: GitHub never said these metrics are meant for performance reviews. Their GA announcement on February 27 framed it as helping organizations "track trends, make informed decisions about rollout, and build reports." But the same blog post outlined a roadmap "from tracking adoption to measuring impact." When the data sits behind an API endpoint, the use cases follow — whether the vendor intended them or not.

What started as "here's a helpful autocomplete" now has a number attached to your name. And if you think this stays in coding, think again. Designers using AI mockup tools, PMs using AI spec generators, marketers using AI copy — every platform serving knowledge workers is building the same measurement layer. The infrastructure is already live; it's just waiting for the dashboard.

The voluntary era of AI tool adoption ended not with a company mandate, but with a metrics API. Three updates in eight days. The metric is the mandate now. Choose accordingly.