Introducing the Microsoft Copilot harness: Unlocking the potential of truly multi-model agentic work - The Official Microsoft Blog
Microsoft is opening up the orchestration layer behind its Copilot products, built by the GitHub team, so developers can run agents on any of 20+ models through Foundry without building and maintaining their own routing and runtime. The pitch is that a shared harness gets improved once and every surface benefits.
- The harness is the layer around the model that handles instructions, context, tools, and execution, and it's what Microsoft says is hard to build and maintain across multiple models.
- It supports 20+ frontier models plus bring-your-own-key, with routing to pick a model per task.
- The runtime was ported to Rust, which the company says cut session overhead and time-to-first-token, and it already runs at the scale of millions of users.
Microsoft is making the Copilot harness available through Microsoft Foundry. The harness is the orchestration layer around a model: it supplies instructions, context, execution environment, and controls so an agent can carry out real work. Microsoft says it was built by GitHub engineers, is already used across Copilot Code, Copilot Studio, the GitHub Copilot app, the CLI, and VS Code, and now becomes available to developers and enterprise customers as a fully managed service. The pitch is that you can pick the right model for each job, from more than 20 frontier models or your own key, without rebuilding your harness or changing the rest of your stack. Support for open standards like MCP, agent skills, and agent plugins is built in.
The reasoning behind the release is that building and maintaining a multi-model harness is hard. A model that leads today may slip next quarter, and keeping a custom router current means ongoing evaluation work. Microsoft's argument is that improvements to the shared harness flow to every product that uses it, so agents built in Foundry get the same upgrades as Microsoft's own surfaces. The article also says the harness routes tasks to different models based on the job, though it gives no detail on how that routing decision is made or how it is evaluated.
On performance, the main concrete claim is that the runtime was ported to Rust, which Microsoft says reduced session overhead and time-to-first-token, though no figures are given. The harness is described as token-efficient and was reportedly the most satisfying option for developers among Microsoft's users, and it is said to run at the scale of millions of users. Under the managed model, Microsoft runs the agent loop and sandbox with governance built in, while you supply instructions, skills, tools, and model choice. The article gives no pricing, no benchmark numbers, and no comparisons to competing harnesses, so the performance and cost claims are essentially untested from the outside.