Microsoft is making a GitHub agentic runtime, or harness, available to developers in Copilot Studio so they can build agents that are able to execute processes with many steps that involve working through ambiguous decision points.
Microsoft said it’s also maintaining availability of two existing harnesses for Copilot Studio users, so they can apply the optimal harness to the required work.
The GitHub Harness gives Copilot Studio the coding and reasoning sophistication associated with Microsoft’s most advanced agents, including Copilot Cowork and GitHub Copilot coding agents.
The harness can be used by agents to develop an execution plan, reason through complex problems, execute agentic loops, use skills, integrate workflows, and connect to tools and agents in other platforms — including Model Context Protocol (MCP) servers — to produce advanced outputs.
Optimized for “long-horizon” work, the new harness uses frontier AI models including Anthropic’s Claude Opus 5 and Fable 5, as well as OpenAI’s GPT-5.6 Sol.
“It’s the same engine behind Copilot Cowork, and it’s made for long-running work: many steps, many systems, and decisions that don’t have an obvious right answer,” said Jason Moore, VP of product for Copilot Studio at Microsoft, in a LinkedIn post.
Microsoft said its internal testing found that, with the GitHub Copilot harness, users realized higher performance and better results in areas including multi-tool use, file analysis, code analysis, and knowledge quality.
The company positioned the three harnesses now available for Copilot Studio as follows, supporting increasing levels of complexity:
- The Copilot Chat harness, which is ideal for customizing chat experiences
- The standard harness, most widely used today, which is optimized for conversational agents with rules-based topics
- The GitHub Copilot harness, which uses the GitHub Copilot SDK to automate complex agentic business processes
“Customizing Copilot Chat, building a conversational agent, and automating an autonomous business process are genuinely different problems, and you should get to pick the foundation that fits,” said Microsoft’s Moore.
The company, as well as Microsoft MVPs commenting on LinkedIn, reminded users that usage-based pricing applies to agents running on the new harness — customers pay for agent usage based on the model they utilize, organizational context and tools, and the runtime used.
Two other noteworthy updates to Copilot Studio rolled out this week:
- The Workflow Designer provides a visual canvas to understand and edit workflows, including adding agent nodes and running workflow evaluations. The company said it will be adding natural language authoring that lets users describe a business goal and assemble the required agents and workflows through a multi-step conversation.
- The agent designer is now more intuitive for authoring, putting the most important tools directly within reach to make authoring faster, while retaining full agent lifecycle management features.
More Copilot Studio Insights:
- Microsoft’s Latest Copilot Studio Updates Improve AI Agent Governance
- How to Pick the Best Pattern for Scalability When Building Agents in Copilot Studio
- Understanding Licensing Options for Agents Built in Copilot Studio
- Microsoft Deepens Integration of Copilot Studio Agents Into Core Workflows
Tom Smith
Analyst
Cloud Wars, Agent and Copilot
Related Posts
