Microsoft Makes Multi-Agent AI More Manageable with New Open Framework

Microsoft is introducing new capabilities to Azure AI Foundry, enabling developers to work more easily with multi-agent systems. The belief is that as organizations become so-called Frontier Firms, teams will need better ways to manage all of the proactive AI agents and ensure they’re working effectively together. Today’s announcements are an effort to address that challenge and help developers build more responsible virtual workers.

Microsoft Agent Framework

First up is a toolkit designed to simplify collaboration between agents and other bots. The Microsoft Agent Framework is an open-source developer SDK that unites AutoGen, the Microsoft Research project that explored how agents communicate, collaborate, and solve problems, with Semantic Kernel, the platform that helps AI apps think, remember, and connect to other systems.

Developers can use the Microsoft Agent Framework to build agents locally with any model before deploying them on Azure AI Foundry. It also enables API connections via OpenAI, and integrations with both Agent2Agent (A2A) and Model Context Protocols (MCP). The Microsoft Agent Framework also supports multi-agent patterns, such as Magentic One, and offers a way to orchestrate agents directly into workflows. Lastly, it enables agents to connect with Azure AI Foundry, Microsoft 365 Copilot, and other agent-centric platforms.

Citing an Atlassian study, Microsoft’s Corporate Vice President of Core AI, Yina Arenas, writes, “50 percent of developers lose more than 10 hours per week due to inefficiencies like fragmented tools, highlighting the need for solutions that reduce complexity and improve the developer experience.”

Microsoft states it will eventually streamline agentic development further by integrating this framework with the Microsoft 365 Agents SDK. Doing so will give developers a straightforward and hassle-free way to build, run, and publish agents across any enterprise environment.

The Microsoft Agent Framework is currently in public preview, allowing developers to try it out and experiment, although customer support may be limited. It’s not recommended for production environments.

Multi-Agent Workflows Come to the Cloud

Microsoft is also extending multi-agent orchestration beyond on-premises or local networks. The company reveals that Foundry Agent Service, the platform for building intelligent agents, now supports multi-agent workflows. When implemented, users will be able to coordinate multiple agents across tasks with persistent state and context sharing. Additionally, they can automate enterprise-level scenarios, such as customer onboarding, financial transaction processing, and supply chain management.

And for that extra peace of mind, error handling, retries, and recovery to improve reliability are all baked in.

Multi-Agent Observability Enhancements

Developers who want better ways to monitor, track, and understand what multiple agents are doing within a system, look no further. Microsoft is working with OpenTelemetry to standardize tracing and telemetry for agentic systems. The company says that now, developers can track more attributes such as quality, performance, safety, and cost.

“This gives teams deeper visibility into agent workflows, tool call invocations, and collaboration—critical for debugging, optimization, and compliance,” Arenas states in a blog post.

Voice Live API Now Generally Available

The final piece of today’s announcement is that every developer can now build production-ready voice AI agents, now that the Voice Live API has launched. This tool offers low-latency, high-quality speech-to-speech interactions, ideal for bots designed for contact centers, automotive assistants, educational companions, virtual tutors, human resource applications, and more.

The Voice Live API was first revealed during Microsoft’s Build conference in May. It went into public preview at the end of July. Microsoft shares that Capgemini, Healow, Astra Tech, and Agora are some of the customers already using the API to build voice-powered agents.

Featured Image: Microsoft CEO Satya Nadella discusses the company's different AI foundries at Build 2025. Credit: Ken Yeung