In an official blog post on October 28, 2026, Microsoft announced it is opening the Copilot harness through Microsoft Foundry: the same multi-model agent orchestration layer that powers Copilot Code, Copilot Studio, the GitHub Copilot app, the command-line tool, VS Code, and other products. Microsoft's reasoning is blunt: to make good use of different models, you first need a good enough harness, and building and maintaining one yourself is too expensive.
The harness handles everything around the model
By Microsoft's definition, the model supplies the intelligence, while the harness takes care of everything around it: giving the agent instructions, preparing context, providing an execution environment and controls, so it can complete genuinely meaningful tasks. The harness supports open standards such as MCP, agent skills, and agent plugins, so developers can pick a different model for each task without changing the harness or the rest of their stack.
After the opening, developers have two options: download the SDK and run it themselves, or use a fully managed cloud runtime on Foundry. The harness already connects to more than 20 frontier models and supports bringing your own key; Microsoft routes work intelligently to the right model for each task, so developers do not have to track and evaluate every new model themselves. The GitHub team recently ported the harness runtime to Rust, which Microsoft says delivers industry-leading performance while reducing session overhead and time to first token. Because the same harness already serves millions of users inside Microsoft, the company stresses that it has been proven at real scale.
What it saves is orchestration cost, not product judgment
Microsoft takes on the agent loop and sandbox, along with governance and optimization, built in from day one; developers bring the instructions, skills, tools, and model choice. The platform stays open, and developers can keep using other harnesses and frameworks. Note that the blog post gives no pricing details for using the harness on Foundry, so actual costs will depend on the product pages that follow. For a team, the real trade-off is whether to hand the orchestration layer to Microsoft in exchange for less to worry about, or keep building it in-house for full control. Opening the harness only makes the first choice easier; it does not make the decision for the team.
