Microsoft Agent Framework Harness and Hosted Agents Reach General Availability

Microsoft’s Agent Framework has moved past SDK-only tooling. At Build 2026 the company put the Agent Harness, GitHub Copilot and Claude Agent SDK connectors, and its orchestration patterns into stable release. The harness and Foundry Hosted Agents have since reached general availability, so the stack now includes a supported runtime for production agents rather than only libraries for prototypes.

On the product side the pieces form a clear split. The Agent Harness is the supported runtime that hosts and governs agent execution. Connectors wire agents to GitHub Copilot and the Claude Agent SDK so those systems sit inside the same control surface. Orchestration patterns ship as stable building blocks for multi-step or multi-agent flows. Foundry Hosted Agents sit on top as a managed path to run agents without standing up the full runtime yourself.

For engineers and builders the change is operational, not cosmetic. An agent is no longer only code you ship from an SDK; it is something you run under a supported harness with connectors and orchestration that Microsoft treats as production surface. That pushes design choices toward lifecycle, governance, and how agents attach to Copilot or Claude paths instead of one-off scripts that never leave a laptop.

The market angle is the move from “build an agent” to “run agents as a governed platform.” Many stacks still stop at SDKs and demos. Microsoft is packaging harness, connectors, orchestration, and hosted agents together so the story is platform operations—control, connectors, and managed hosting—not only model calls in application code.

What to watch next is whether teams adopt the harness and Foundry Hosted Agents as the default path for real workloads, and how heavily the GitHub Copilot and Claude Agent SDK connectors show up in production designs. If those pieces stick, new agent work on this stack will start from runtime and hosting choices first, not from a greenfield SDK sample.