Microsoft Copilot Harness Lands in Foundry

Microsoft is productizing the harness—not the model—as the real platform advantage in agent development. That could eliminate huge amounts of orchestration plumbing, but it also makes runtime behavior, governance, and pricing part of Microsoft’s platform moat.

• –Reusable skills, MCP tools, file and code execution, bounded subagents, and source-controlled definitions turn Foundry into a serious workflow runtime.
• –Model choice remains flexible, letting teams change supported models without rebuilding the surrounding orchestration.
• –Microsoft claims the harness already powers Copilot Code, Copilot Studio, the GitHub Copilot app, CLI, and VS Code, giving the preview an unusually broad production pedigree. [Official Microsoft Blog](https://blogs.microsoft.com/blog/2026/10/28/introducing-the-microsoft-copilot-harness-unlocking-the-potential-of-truly-multi-model-agentic-work/)
• –The tradeoff is platform dependence: existing agents cannot switch harnesses after creation during preview, while Microsoft controls the execution environment and policy boundaries.
• –Consumption billing begins December 1, making sandbox efficiency and cost observability important before production adoption. [Microsoft Foundry announcement](https://devblogs.microsoft.com/foundry/microsoft-copilot-harness-in-foundry/)