Microsoft has officially brought its Azure AI Foundry Agent Service to general availability, arming developers with a fully managed platform to build, deploy, and scale sophisticated AI agents. The milestone, announced in May 2025, introduces powerful multi-agent orchestration capabilities that enable the automation of complex business processes through interconnected, intelligent agents. This release marks a significant leap in enterprise AI, offering developers a unified toolchain to create agents that can reason, take action, and integrate seamlessly across diverse data sources and cloud environments.

At its core, Azure AI Foundry Agent Service is a managed platform that simplifies the entire lifecycle of AI agents. Developers define an agent’s underlying generative AI model, its instructions, and the tools it can wield—such as APIs or knowledge bases—using either the Azure AI Foundry SDK or the portal. The service handles the heavy lifting of scaling, security, and reliability, allowing teams to focus on logic rather than infrastructure. With GA, what was once a preview promise is now enterprise-ready, complete with SLAs, enhanced security, and production-grade support.

Multi-Agent Orchestration Takes Center Stage

The headliner feature of this release is the introduction of multi-agent orchestration, available in two preview flavors: Connected Agents and Multi-Agent Workflows. These capabilities transform the platform from a single-agent shop into a conductor of agentic symphonies.

Connected Agents enables point-to-point interactions between specialized agents. Think of a customer service scenario where a triage agent delegates a refund request to a finance agent, which in turn verifies account details with a backend systems agent—all without a centralized controller. This modular approach allows for dynamic task delegation and makes it easier to update or replace individual agents without disrupting the entire system.

Multi-Agent Workflows provides a stateful orchestration layer for longer-running, complex processes. Unlike simple request-response chains, workflows maintain context across multiple steps, essential for operations like customer onboarding, insurance claims processing, or supply chain automation. The platform tracks state, handles errors, and ensures that each agent’s output feeds correctly into the next stage, all while giving developers visibility into the overall process.

Both features are currently in preview, signaling Microsoft’s commitment to rapid iteration based on user feedback. Early adopters have already begun testing these capabilities to coordinate fleets of agents that collectively handle tasks far beyond what a single model could manage alone.

Agent2Agent Protocol: The Interoperability Revolution

A key enabler of this multi-agent vision is the newly introduced Agent2Agent (A2A) protocol and API. In an era where no single platform dominates enterprise IT, the ability for agents to communicate across different systems and vendors is critical. The A2A API allows agents built on Azure AI Foundry to seamlessly interact with agents hosted on other platforms, using standardized protocols for discovery, task assignment, and result handoff.

This interoperability extends to the Model Context Protocol (MCP), which Microsoft has embraced to ensure that agents can access external tools and data sources regardless of their origin. By adhering to these open standards, Azure AI Foundry Agent Service positions itself as a neutral orchestrator, capable of integrating agents from competitors like Google Vertex AI or Amazon Bedrock, as well as homegrown solutions built with frameworks like LangChain or CrewAI.

Unified Runtime Merges Semantic Kernel and AutoGen

Under the hood, the service now runs on a unified runtime that merges two of Microsoft’s most prominent AI frameworks: Semantic Kernel and AutoGen. Semantic Kernel excels at connecting AI models with existing codebases and APIs, while AutoGen specializes in multi-agent conversations and autonomous task execution. By combining them, developers gain the best of both worlds: the ability to build agents that can hold multi-turn dialogues, reflect on their own reasoning, and execute code—all within a single, consistent runtime.

This unification also streamlines the development-test-deploy pipeline. Developers can simulate agent behavior locally using the same runtime, then deploy to the cloud with zero modifications. The result is a dramatic reduction in the complexity and risk typically associated with moving from prototype to production.

Deep Integration with Microsoft’s Ecosystem and Beyond

Azure AI Foundry Agent Service comes pre-integrated with a wide array of Microsoft services: Bing for real-time web search, SharePoint for enterprise knowledge retrieval, and Databricks for advanced data analytics, among others. These built-in connectors allow agents to tap into corporate data and external information in seconds, without requiring custom integration work.

Beyond the Microsoft universe, the platform supports connections to hundreds of third-party tools and services through its extensible architecture. Whether an agent needs to pull records from Salesforce, monitor sensors via IoT Hub, or trigger actions in SAP, the framework can accommodate it. This breadth of integration is essential for enterprises that rely on a heterogeneous mix of legacy and modern systems.

AgentOps: Built-in Monitoring and Evaluation

No production-grade AI service is complete without robust observability, and Azure AI Foundry delivers with AgentOps. This suite of monitoring and evaluation tools gives developers deep insights into agent performance, tracking metrics such as task completion rate, response accuracy, and latency. Tracing capabilities illuminate the entire decision-making process, from initial prompt to final action, making it easier to debug failures or optimize for efficiency.

AgentOps also includes evaluation frameworks that measure an agent’s outputs against predefined success criteria, enabling continuous improvement. Teams can set up A/B tests, compare different model configurations, and receive alerts when performance degrades. This data-driven approach is crucial for enterprise deployments where AI reliability directly impacts business outcomes.

What This Means for Developers and the Enterprise

The general availability of Azure AI Foundry Agent Service marks a maturation of Microsoft’s AI platform play. For developers, it lowers the barrier to building intelligent automation, providing a unified interface for model selection, tool chaining, memory management, and now multi-agent coordination. The preview features suggest an ambitious roadmap that will soon make it possible to design entire digital workforces with minimal boilerplate code.

For enterprises, the service addresses three perennial concerns: governance, scalability, and interoperability. With GA, the platform offers compliance certifications, virtual network support, and role-based access controls baked in. Agents can scale from zero to millions of concurrent tasks without manual intervention, and the A2A protocol ensures that investments in Azure AI aren’t siloed but can participate in a broader ecosystem of automation.

Real-world applications are already emerging. Financial institutions are using multi-agent workflows to automate anti-money laundering investigations, coordinating agents that specialize in transaction analysis, customer profiling, and regulatory report generation. Manufacturers are orchestrating agents that monitor equipment sensors, predict failures, and dispatch maintenance teams—all with a single, auditable workflow. As the preview features graduate to GA, expect these use cases to multiply.

Looking Ahead

Microsoft’s release of Azure AI Foundry Agent Service with multi-agent orchestration signals a strategic bet: that the future of enterprise AI lies not in monolithic models but in collaborative, specialized agents working in concert. The integration with the wider Azure ecosystem, plus the embrace of open protocols, gives Microsoft a strong position in the rapidly evolving agentic AI market.

Developers eager to get started can spin up an agent in the Azure AI Foundry portal today. The preview features—Connected Agents and Multi-Agent Workflows—are available in select regions and will expand based on demand. As the community experiments and feedback rolls in, Microsoft plans to refine the orchestration layer, deepen A2A protocol support, and enhance AgentOps with more granular KPIs.

One thing is clear: the era of single-purpose chatbots is giving way to agentic systems that reason, plan, and execute. With this GA release, Microsoft is giving developers the tools to build that future today.