【最新アップデート】Copilot Studio の GitHub Copilot harness が GA ― 新しいライセンス体系と Charge / No Charge の整理Microsoft officially released the GitHub Copilot harness for Copilot Studio on…

Microsoft has moved the GitHub Copilot harness for Copilot Studio to general availability, a step that matters for teams building AI agents because it arrives alongside a revised licensing structure that spells out which capabilities are billed and which are included at no additional cost. According to a post from Microsoft Japan, the harness reached GA on August 3, 2026, following its introduction in June as part of a broader refresh of the Copilot Studio platform.

To understand the significance, it helps to clarify what a harness is in this context. Copilot Studio is Microsoft's low-code environment for building custom copilots and autonomous agents. A harness appears to refer to the orchestration scaffolding that governs how an agent reasons, invokes tools, and assembles responses. The GitHub Copilot harness therefore brings a specific, prebuilt orchestration approach into the Copilot Studio agent-building experience, giving developers a defined configuration to start from rather than wiring one together manually. Reaching GA generally signals that Microsoft considers the feature stable and supported for production use, as opposed to preview status where behavior and terms can still change.

The element most likely to draw attention from organizations is the clarification of the Charge versus No Charge boundaries. Copilot Studio has historically used a consumption-oriented billing model, and questions about what triggers a cost have been a recurring source of confusion for teams planning deployments at scale. By publishing a clearer split between the capabilities that incur charges and those that are provided at no extra cost, Microsoft is giving developers and administrators a more predictable basis for estimating what building and running agents on the harness will cost. The Microsoft Japan post frames this as helping developers grasp agent-building costs more easily, which is a practical concern once pilots expand into wider rollouts.

Licensing clarity of this kind tends to matter as much as the underlying technology. Enterprises evaluating agent platforms often stall not on capability but on budget uncertainty, because unpredictable consumption charges make it difficult to secure internal approval. A published Charge and No Charge breakdown reduces that friction and makes total cost of ownership easier to model before committing. Readers planning to act on this should still consult the official Microsoft licensing documentation for the exact terms, since blog summaries condense details and pricing structures can vary by region, plan, and tenant configuration.

It is worth situating this release within Microsoft's broader agent strategy. Over the past couple of years the company has pushed to unify its Copilot family, spanning Microsoft 365 Copilot for productivity apps, GitHub Copilot for software development, and Copilot Studio for building bespoke agents. Underneath these products, Microsoft has been consolidating tooling around Azure AI Foundry and related orchestration services, positioning Copilot Studio as the accessible, low-code entry point while more code-centric developers work closer to the model and framework layer. The arrival of a GitHub Copilot harness inside Copilot Studio fits this pattern of letting different audiences tap shared capabilities through the surface that suits them.

The naming also reflects an industry-wide shift in how AI systems are described. Terms like harness, orchestration, and agent scaffolding have become common as vendors move beyond single prompt-and-response interactions toward agents that plan, call external tools and data sources, and carry out multi-step tasks. Competing platforms from other providers have introduced comparable concepts, so Microsoft's move can be read as part of a larger trend toward standardizing how agent behavior is packaged and licensed rather than a wholly novel departure.

For teams already invested in Copilot Studio, the practical takeaway is that a supported, generally available option now exists, backed by clearer commercial terms. For those still evaluating, the licensing clarification lowers one of the more persistent barriers to adoption. As always with a fast-moving product area, the specifics of what counts as a chargeable action and how the harness performs across different workloads will become clearer as more organizations deploy it and share results. In the meantime, the combination of GA status and a defined Charge and No Charge model gives developers a firmer footing than the earlier preview and June announcement provided.