Microsoft Agent Framework Harness and Hosted Agents Reach General Availability

Agent Framework Moves to Production Runtime

Microsoft has advanced its Agent Framework beyond the SDK stage, making the Agent Harness and Foundry Hosted Agents generally available. This development provides platform teams with a supported production runtime environment for AI agents, enabling them to run and govern agents effectively, rather than solely focusing on their construction.

Addressing Runtime Challenges for AI Agents

The 1.0 general availability of the Agent Framework on April 2, 2026, consolidated Semantic Kernel and AutoGen, moving both predecessors to maintenance mode. The subsequent release of the Agent Harness and Hosted Agents addresses critical runtime questions concerning where agents execute, their access permissions, and how their behavior integrates with existing observability and policy systems.

Key Features of the Agent Harness

The Agent Harness functions as the core runtime for AI models, enabling them to perform multi-step tasks and utilize tools. It includes features such as function invocation, per-call history persistence, context compaction, a todo list with plan and execute modes, file memory, skills, web search, tool approval, and built-in OpenTelemetry. These features are enabled by default but can be individually removed. Opt-in features like shell tooling, file access, background sub-agents, and automatic looping still emit warnings when activated.

Simplified Agent Deployment with Foundry Hosted Agents

Foundry Hosted Agents provide a managed deployment target that bills based on consumption. Developers can supply a chat client, instructions, and tools, and the harness manages the rest through a single API call. This simplifies the process of integrating planning, history persistence, context compaction, approvals, web search, and telemetry into agent operations.