Azure Foundry Agent Service Adds Copilot Harness

Introduction

Microsoft is expanding Azure Foundry Agent Service with new capabilities aimed at teams building enterprise AI agents. The new Microsoft Copilot harness, now in public preview, provides a managed execution layer for agent workflows, while long-running resilience for hosted agents is now generally available.

For IT teams and developers, this matters because it reduces the effort needed to build reliable agents that can coordinate tools, manage context, and recover from interruptions during complex tasks.

What’s new in Foundry Agent Service

Microsoft Copilot harness now in public preview

The Copilot harness gives developers a fully managed way to build agents declaratively in Foundry Agent Service. Instead of assembling the full orchestration stack, teams can focus on business logic, knowledge, and tools.

Key preview capabilities include:

• Reusable skills to package instructions, scripts, and resources into repeatable procedures
• Connected tools for supported Foundry and MCP tools with configured authentication and approvals
• Execution and delegation so agents can work with files, code, and built-in subagents
• Source-controlled agent definitions to manage agent configurations alongside application code

Microsoft manages the agent loop, sandbox, and working context, while customers choose supported Foundry models and define how agents should operate within policy limits.

Long-running resilience for hosted agents is GA

For organizations using hosted agents with their own framework or runtime behavior, Microsoft has made long-running resilience generally available. This helps agents recover from interruptions in workflows that can run for hours or days.

Foundry can restart interrupted work using the original input, but developers still need to design checkpointing and idempotency into their applications. In practice, that means saving intermediate progress and preventing duplicate actions when a workflow resumes.

Why this matters for IT admins and developers

These updates give teams more flexibility in how they build AI agents:

• Use prompt agents for simpler scenarios
• Use the Copilot harness when managed execution, skills, files, and subagents are needed
• Use hosted agents when custom frameworks or runtime control are required

For Azure administrators and platform teams, the managed harness can reduce infrastructure overhead and speed up pilot projects, while hosted agent resilience improves reliability for production-grade automation.

Next steps

• Evaluate whether the Copilot harness preview fits your current Azure Foundry agent projects
• Review hosted agent workflows for checkpointing and duplicate-action prevention
• Test agent definitions in the Foundry portal, Visual Studio Code toolkit, or with the Azure Developer CLI
• Identify which workloads need a managed harness versus full custom hosting

Organizations building AI-driven research, analysis, or workflow automation in Azure should watch this release closely as Microsoft continues to mature Foundry Agent Service.