GitHub Copilot harness generally available in Copilot Studio

Microsoft has made the GitHub Copilot harness in Copilot Studio generally available, giving the company's agent-building platform access to the coding and reasoning technology behind its Copilot Cowork and GitHub Copilot coding agent products. The capability, previewed over the past two months, is now in production and can be used from the Copilot Studio homepage.

The harness is designed for what Microsoft calls "long-horizon" work: business processes with many steps, multiple sources, and ambiguous decision points. It runs on frontier reasoning models — Opus 5, GPT-5.6 Sol, and Fable 5 — and can plan, run agentic loops, use skills, integrate workflows, and connect to tools and agents on other platforms. Microsoft reports "significant performance and quality gains" in its own testing with real-world business process evaluations, citing improvements in multi-tool use, file analysis, code analysis, and knowledge quality. Those claims come from internal testing, not independent benchmarks.

The structural change is separate from the new capability itself. Copilot Studio now supports three harnesses: the Copilot Chat harness, aligned with Microsoft 365 Copilot Chat; the Standard harness, used by most existing agents for rules-based conversational scenarios; and the new GitHub Copilot harness. Microsoft says existing harnesses remain unchanged, and that different harnesses can be optimized for different outcomes rather than forcing a single approach.

The billing model is where the new harness departs from precedent. Agents running on the GitHub Copilot harness use usage-based billing for all work, regardless of Microsoft 365 Copilot licensing. Costs are based on the models chosen, organizational context and tools added, and runtime used. Certain maker experiences — natural language authoring, evaluations, and testing — also fall under usage-based billing when building with the GitHub Copilot harness. By contrast, Microsoft 365 Copilot licensed users keep fair-use inclusion for Copilot Chat and Standard harness agents, and those harnesses continue on the existing fixed rate card for other usage.

For makers, Microsoft is shipping new authoring tools alongside the harness: an agent designer with more prominent tool access, and a workflow designer offering a visual canvas for editing workflows, adding agent nodes, and running workflow evals. Natural language authoring — describing a business goal and assembling agents and workflows through conversation — is coming soon, and will carry usage-based billing when used with the GitHub Copilot harness.

Cost predictability remains an open question. The fixed rate card model gave organizations a known price per agent; usage-based billing ties spend to model choice, context, and runtime, which is harder to forecast for processes that run long and touch many tools. Microsoft has published a licensing guide for the new billing structure, but the economics of running complex agents at scale will presumably only become clear as production usage accumulates.