# Microsoft adds a managed Copilot harness to Foundry Agent Service

published: 2026-10-28T17:05:40.431Z
topic: frameworks
source: https://devblogs.microsoft.com/foundry/microsoft-copilot-harness-in-foundry/
url: https://aiagentnews.io/stories/microsoft-adds-a-managed-copilot-harness-to-foundry-agent-service

Microsoft has launched the Copilot harness in public preview within Foundry Agent Service, giving developers a managed execution loop for agents defined declaratively. It also made long-running resilience for hosted agents generally available.

The Copilot harness runs the agent loop and sandbox for agents defined in Foundry Agent Service, so developers supply skills, connected tools and instructions. The preview covers reusable skills, Foundry and Model Context Protocol tools with configured authentication and approvals, file and code execution with built-in subagents, and source-controlled agent definitions. The harness is selected when an agent is created, and existing agents are unchanged. Switching the harness after creation is not yet supported.

Microsoft's example uses the Azure Developer CLI to define a research-brief skill in a SKILL.md file and an azure.yaml file that configures a prompt agent with web search and Code Interpreter. Separately, long-running resilience for hosted agents is now generally available. Foundry restarts interrupted work with its original input, so the application must save progress and prevent duplicate actions.

Billing for the harness begins December 1, 2026, and harness runtime is free until then. From that date, sandbox compute is priced at USD $0.0994 per vCPU-hour and memory at USD $0.0118 per GiB-hour. Model inference and separately billed tools continue to carry their own charges.

## Why it matters if you build agents

Prompt agents can run on a managed loop, skills and sandbox instead of custom orchestration code, but the harness choice is fixed at creation. Teams should plan for the December 1 billing start and check the preview limits for their regions and models. Long-running hosted agents need their own checkpoints and duplicate-action checks, since conversation history alone does not restore work in progress.