AI Code Assistant: Microsoft Agent Framework & Hosted Agents GA

Microsoft has announced the General Availability of its Agent Framework and Foundry Hosted Agents, offering Software Developers a production-ready runtime for creating and deploying sophisticated AI-powered applications, thereby moving beyond just building AI code assistant libraries to managing their execution and governance.

• The Agent Framework has transitioned from an SDK to a fully supported production runtime, reaching 1.0 General Availability.
• This release consolidates previous foundational frameworks like Semantic Kernel and AutoGen, shifting them to maintenance mode.
• The Agent Harness provides essential runtime capabilities, allowing Software Developers to focus on agent logic rather than infrastructure.
• Foundry Hosted Agents offer a managed deployment target, simplifying the governance and scaling of AI agents.

Microsoft Agent Framework Reaches General Availability

On April 2, 2026, Microsoft officially announced the 1.0 General Availability of its Agent Framework. This significant milestone moves the framework beyond its initial SDK phase into a fully supported production runtime. Further announcements at Build 2026, held June 2-3, confirmed the stable release of the Agent Harness, along with crucial connectors for the GitHub Copilot SDK and Claude Agent SDK, and robust multi-agent orchestration patterns. This suite of tools provides Software Developers with a comprehensive platform for building and managing intelligent agents.

The earlier launch of the open-source Agent Framework was noted for consolidating the functionalities of Semantic Kernel and AutoGen. With the 1.0 release, the question of which framework to use for agent construction has been settled, as both predecessors have now been moved into maintenance mode. This strategic consolidation streamlines the development process for Software Developers aiming to create advanced AI solutions.

What Does the Agent Harness Offer for AI Code Assistant Development?

The core of this General Availability announcement for Software Developers is the Agent Harness. As Microsoft principal software engineer Wes Steyn highlighted, a language model on its own can only generate text. The harness acts as the essential runtime that enables models to call tools, execute multi-step tasks, and persist through complex workflows until a job is completed.

This runtime now ships directly with the Agent Framework, freeing development teams from the burden of rebuilding this intricate infrastructure. Key capabilities included by default are function invocation, per-call history persistence, context compaction, a versatile todo list with plan and execute modes, file memory, skills management, web search integration, and tool approval mechanisms. Crucially, built-in OpenTelemetry support is enabled by default, offering critical observability, with each feature individually removable. While shell tooling, file access, background sub-agents, and automatic looping are opt-in, they still provide warnings when enabled, ensuring Software Developers maintain control and awareness over agent behavior.

Simplifying Deployment with Foundry Hosted Agents

Beyond providing a robust build-time solution, the Agent Framework addresses the critical run-time questions that arise in agent deployment: where agents execute, what they are permitted to access, and how their behavior integrates with existing observability and policy systems. This is where Foundry Hosted Agents come into play, serving as Microsoft’s managed deployment target for these advanced AI systems.

Foundry Hosted Agents simplify the operational aspects for platform teams and individual Software Developers by offering a consumption-billed service. Developers supply their chat client, instructions, and custom tools, and the harness handles the rest through a single, streamlined call. This approach effectively wires in planning, history persistence, context compaction, tool approvals, web search, and telemetry, significantly enhancing developer productivity AI by abstracting away complex infrastructure management.

The Unseen Complexity: Why the Harness Matters to Software Developers

The significance of the Agent Harness extends far beyond its apparent function as a runtime. Research from MBZUAI’s VILA-Lab, detailed in an April 2026 paper titled “Dive into Claude Code,” sheds light on the actual composition of advanced AI systems. Their analysis of Claude Code v2.1.88 revealed that approximately 98.4% of the codebase comprised harness infrastructure, including permissions, context management, sandboxing, tool routing, and recovery, with only about 1.6% dedicated to AI decision logic.

While this figure comes with caveats regarding its derivation from a leak-derived bundle, the overarching direction holds true: the majority of an AI system’s complexity lies in its operational infrastructure. This insight underscores why the Microsoft Agent Framework’s robust harness is a pivotal development for Software Developers. It provides a standardized, supported solution for the most challenging aspects of AI code generation and agent orchestration, which are often overlooked when focusing solely on the language model itself. This foundation is essential for building reliable AI tools for developers, including advanced AI debugging tools and sophisticated AI code assistant applications, much like the underlying infrastructure supporting tools such as GitHub Copilot, Cursor, Tabnine, Amazon CodeWhisperer, and Codeium.

Advancing AI Tools for Developers with Governance and Observability

The General Availability of Microsoft’s Agent Framework and Foundry Hosted Agents marks a significant step forward for Software Developers working with AI. By providing a production-ready runtime with built-in governance and observability features, Microsoft is empowering teams to move beyond experimental AI agent development to creating reliable, scalable, and secure AI-powered applications. This framework directly addresses the practical challenges of deploying AI, ensuring that agents can operate effectively within existing enterprise environments.

This release is particularly impactful for the evolution of AI tools for developers, offering a unified approach to agent lifecycle management. It enables greater consistency and control over how AI agents interact with systems and data, fostering an environment where innovation in coding AI can flourish without compromising security or operational integrity. For Software Developers, this means less time spent on foundational infrastructure and more time dedicated to crafting intelligent agent behaviors that drive true developer productivity AI.

Frequently Asked Questions

What is the significance of Microsoft Agent Framework’s General Availability for Software Developers?

The General Availability means the Agent Framework is now a fully supported production runtime, moving beyond an SDK. This provides Software Developers with a robust, pre-built infrastructure for deploying and governing AI agents, rather than just building them.

How does the Agent Framework harness simplify the development of AI code assistant applications?

The Agent Harness acts as the essential runtime for AI models, handling complex tasks like tool invocation, multi-step task execution, history persistence, and context management. This allows Software Developers to concentrate on the agent’s core logic and instructions, significantly boosting developer productivity AI.

What are Foundry Hosted Agents and how do they benefit deployment for AI tools for developers?

Foundry Hosted Agents are Microsoft’s managed deployment target for the Agent Framework, offering a consumption-billed service for running AI agents. They simplify deployment by integrating planning, history, approvals, web search, and telemetry, making it easier for platform teams and Software Developers to manage and scale their AI tools for developers.