Microsoft Opens Copilot Agent Harness to External Developers with Lock-in Constraints and Future Pricing
Microsoft launches public preview of managed Copilot agent harness in Foundry, offering sandboxing and MCP integration but imposing model restrictions and post-December usage-based billing.
On October 29, 2026, Microsoft announced that it is opening its internal agent orchestration infrastructure, known as the "Copilot Harness," to external developers through Microsoft Foundry Agent Service in public preview. This move allows developers to leverage Microsoft's proven managed service to build and deploy autonomous agents without constructing their own underlying execution environments.
The managed harness includes built-in sandboxing, reusable skills, Model Context Protocol (MCP) integrations, and support for subagents. Microsoft handles running the agent and managing its execution environment, leaving developers to choose a supported model and configure instructions and tools for their applications. According to The New Stack, the same harness currently supports products such as Copilot Code, Copilot Studio, GitHub Copilot, the CLI, and VS Code.
Despite its capabilities, the service comes with significant technical constraints. First, once the Copilot harness is selected when creating a prompt agent, that choice cannot be changed later, even in a new version, creating potential vendor lock-in. Second, while Microsoft positions its strategy as multi-model, developers can only choose from compatible models deployed through Foundry; bringing their own models (BYO) is not supported with the managed Copilot harness. Additionally, Microsoft controls the CPU and memory allocated to the agent and its idle timeout duration, limiting developer control over the runtime environment.
Regarding costs, the Copilot harness in Foundry Agent Service is free until December 1, 2026, although model inference and some tools and connected services still incur separate charges. After that date, Microsoft will charge $0.0994 per vCPU-hour for compute and $0.0118 per GiB-hour for memory, with rates potentially varying by region and customer agreement. For teams requiring more control, Microsoft suggests using Foundry's hosted agents instead, which allow building agents with custom frameworks while Microsoft handles hosting.