Microsoft now bills Copilot credits while you build and test workplace agents

Microsoft has switched agents and workflows that run on the GitHub Copilot harness in Copilot Studio to usage-based billing, measured in Copilot Credits. Credits cover model use, tools such as knowledge and MCP, and the harness itself. Unlike the standard harness, which starts charging after you publish, this harness bills from the moment you start building. That includes creating a solution in natural language, previewing, testing, and generating evaluations. Developer and trial environments moved to this model on 1 September 2026. Usage recorded before that date is not a reliable cost estimate. A Microsoft 365 Copilot licence does not cancel this consumption. Administrators allocate credits in the Power Platform admin centre and can attach pay-as-you-go Azure billing.
Until now, many teams treated no-code agent building as free practice. You could paste context, click around, and only worry about cost after something went live. That hid the real work of turning a one-off chat into a workflow with inputs, quality checks, and a human review step. The meter now runs in the sandbox. Casual testing, repeated previews, and evaluation runs spend credits even if nothing is published. The old Copilot licence on a laptop is not a shield. Unscoped experimentation is no longer a cheap way to learn. It is a budget line that rewards a short briefing and a named owner, not another blank chat.
Analysis
Treat this as a trap to avoid, not a reason to chase agents. Ask your Power Platform admin whether the GitHub Copilot harness is enabled, what credit cap sits on trial and developer environments, and pause any unscoped build-and-test until you have a one-page workflow blueprint with inputs, privacy limits, and a human review step.
Source note
Pulse published by Collab365 Spaces, reviewed by Helen Jones on . Cite as "Microsoft now bills Copilot credits while you build and test workplace agents", Collab365 Spaces. 2 sources referenced.