Microsoft expands Copilot’s reach in finance workflows

Microsoft has unveiled its Copilot harness, a new service in public preview through Microsoft Foundry, designed to simplify the creation and deployment of autonomous agents. This harness, which brings together orchestration infrastructure from across Microsoft's Copilot products, offers features like sandboxing, reusable skills, MCP integrations, and support for subagents.

While Microsoft manages the agent's execution environment, developers can select a supported model and configure the instructions and tools for their applications. The harness supports several Microsoft Copilot offerings, including Copilot Code, Copilot Studio, GitHub Copilot, the CLI, and VS Code.

Microsoft's harness manages model and tool calls, tracks context, and enforces configured policies, ensuring agents remain within predefined limits. Developers can build and deploy agents using the Azure Developer CLI with a simple command, and keep configurations and supporting files in source control for easier management. However, once an agent is deployed, switching to a different harness is not possible, even in a new version.

While the harness aims to simplify agent creation and deployment, Microsoft retains control over the agent's environment, including CPU and memory allocation and idle time limits. Developers can choose from compatible models deployed through Foundry, but are not able to bring their own models to the managed Copilot harness. For teams seeking more control over their agents, Microsoft offers hosted agents through Foundry, allowing them to build and run agents using their preferred frameworks while Microsoft handles hosting.

The company also announced general availability of long-running resilience for hosted agents, which helps them recover from interruptions during extended tasks. The Copilot harness is currently free until December 1, 2026, with model inference and some tools incurring separate charges thereafter.