Copilot Studio tests can burn credits before anyone uses the agent
Reviewed by Helen Jones
A 20 September 2026 community recap of Microsoft 365 Copilot changes says the GitHub Copilot harness in Copilot Studio became generally available on 3 August 2026. Agents on that harness use usage-based billing for all work, whether or not people have Microsoft 365 Copilot licences. Credits are said to be consumed while makers build, preview, and evaluate an agent, not only when it runs in production. Limits can be set on a single agent. Exact credit rates and which agent types sit on the harness still need checking in current Copilot Studio billing documentation, because this is a practitioner summary, not an official Microsoft Learn billing page.
Many teams treated Copilot Studio as a sandbox. If the company already paid for Microsoft 365 Copilot seats, previewing an agent felt like free practice. Spend controls were something you worried about after publish. If this recap is right, the meter starts when a maker hits preview. Informal experiments can create Copilot Credits charges before any colleague uses the agent. Tenant spend controls matter at the first test, not at go-live.
Analysis
Treat this as a trap to avoid, not a cue to start building agents. Pause unofficial Copilot Studio previews until an admin confirms the current billing docs and puts a credit limit on any agent that is already in the tenant.
Source note
Pulse published by Collab365 Spaces, reviewed by Helen Jones on . Cite as "Copilot Studio tests can burn credits before anyone uses the agent", Collab365 Spaces. 1 source referenced.