GitHub Copilot went metered — now govern the fleet

GitHub Copilot stopped being a flat monthly seat and became a meter. That happened June 1, when Copilot moved to usage-based billing on AI credits (1 credit = $0.01). The interesting part is what GitHub shipped next: the governance layer that a metered tool actually needs. Over the past two weeks it added per-user spend budgets, enterprise telemetry export, and bring-your-own-key model access. If your team runs Copilot, the job just changed from "buy seats" to "govern a fleet."

What actually happened

Three concrete changes, all recent and all in the GitHub Changelog. First, spend controls: admins can now set per-user AI credit budgets on cost centers directly in the billing UI (July 7), so every developer in a group inherits a cap instead of you finding out at invoice time. Second, telemetry: enterprise-managed OpenTelemetry export (July 8) lets you mandate where Copilot in VS Code and the CLI sends its OTel data — an approved OTLP collector, with control over whether prompt and response content is captured. Third, model choice: Copilot app support for BYOK (June 23) means you can point agent sessions at your own provider keys — OpenAI, Anthropic, Azure Foundry, Ollama, any OpenAI-compatible endpoint — with keys stored in the local OS keychain.

Why it matters for your business

Metered tools reward the teams that instrument them and punish the ones that don't. A flat seat had a predictable worst case; a meter has a tail. The three features above are exactly the controls you'd demand from any usage-priced infrastructure — a spend cap, a telemetry pipe, and the ability to not be locked to one vendor's hosted models. GitHub built them because customers hit the bill.

The operator move is to treat AI coding like the metered infra it now is. Set cost-center budgets before you scale seats, not after. Wire the OpenTelemetry export into whatever you already use for observability so "which team burns credits, on what" is a dashboard, not a forensics exercise. And use BYOK to keep the model layer a dial you control — the same portability logic we apply to every AI coding tool and dev spend line item.

Sources: GitHub Changelog — per-user budgets for cost centers, GitHub Changelog — enterprise-managed OpenTelemetry export, GitHub Changelog — Copilot app support for BYOK, The GitHub Blog — Copilot moving to usage-based billing.