Microsoft Agent Framework, now in public preview, and new capabilities in Azure AI Foundry make it simpler for builders to construct, observe, and govern multi-agent techniques.
At this time we’re asserting new capabilities in Azure AI Foundry that make it simpler for builders to construct, observe, and govern multi-agent techniques, whereas serving to organizations shut the belief hole in AI.
As agentic AI adoption accelerates—eight in ten enterprises now use some type of agent-based AI, in keeping with PwC1—the complexity of managing these techniques is rising. Builders face fragmented tooling, and organizations wrestle to make sure brokers behave responsibly. Our newest updates to Azure AI Foundry deal with these challenges head-on.
Introducing Microsoft Agent Framework (public preview)
The Microsoft Agent Framework, now in public preview, is the open-source SDK and runtime that simplifies the orchestration of multi-agent techniques. It converges AutoGen, a former Microsoft Analysis undertaking, and the enterprise-ready foundations of Semantic Kernel right into a unified, commercial-grade framework—bringing cutting-edge analysis to builders.
With Microsoft Agent Framework, builders can:
- Experiment domestically after which deploy to Azure AI Foundry with observability, sturdiness, and compliance in-built.
- Combine any API through OpenAPI, collaborate throughout runtimes with Agent2Agent (A2A), and hook up with instruments dynamically utilizing Mannequin Context Protocol (MCP).
- Use the most recent multi-agent patterns like Magentic One and orchestrate brokers into Workflows.
- Cut back context-switching throughout instruments and platforms.
- Construct multi-agent techniques connecting Azure AI Foundry, Microsoft 365 Copilot, and different agent platforms.
This framework is designed to assist builders keep in move. Based on an trade research2, 50% of builders lose greater than 10 hours per week attributable to inefficiencies like fragmented instruments, highlighting the necessity for options that scale back complexity and enhance the developer expertise.
One group utilizing Microsoft Agent Framework to cut back friction is KPMG. KPMG’s transformation started with KPMG Clara, its cloud-based good audit platform used on each KPMG audit worldwide.
KPMG Clara AI is tightly aligned with the next-generation, open-source Microsoft Agent Framework, constructed on the convergence of Semantic Kernel and AutoGen.
This implies KPMG Clara AI can join specialised brokers to enterprise knowledge and instruments, whereas benefiting from built-in safeguards and an open, extensible developer ecosystem. The framework’s open-source connectors permit brokers in KPMG Clara AI to interoperate not solely with Azure AI Foundry, but additionally with exterior techniques and language fashions, making it doable to scale multi-agent collaboration throughout a worldwide, regulated enterprise.
Foundry Agent Service and Microsoft Agent Framework join our brokers to knowledge and one another, and the governance and observability in Azure AI Foundry present what KPMG companies must be profitable in a regulated trade.
— Sebastian Stöckle, World Head of Audit Innovation and AI at KPMG Worldwide
We invite builders to hitch us in shaping the way forward for agentic AI by contributing code and suggestions to Microsoft Agent Framework.
Multi-agent workflows (non-public preview)
Constructing on Microsoft Agent Framework, we’re extending these capabilities instantly into the cloud with multi-agent workflows in Foundry Agent Service. This new characteristic allows builders to orchestrate refined, multi-step enterprise processes utilizing a structured, stateful workflow layer.
With multi-agent workflows, groups can:
- Coordinate a number of brokers throughout long-running duties with persistent state and context sharing.
- Automate complicated enterprise situations corresponding to buyer onboarding, monetary transaction processing, and provide chain automation.
- Leverage built-in error dealing with, retries, and restoration to enhance reliability at scale.
Workflows could be authored and debugged visually by way of the VS Code Extension or Azure AI Foundry, then deployed, examined, and managed in Foundry alongside current options.
A number of clients are at the moment experimenting with this functionality, and we stay up for increasing to extra clients within the coming weeks.
Multi-agent observability throughout fashionable frameworks with OpenTelemetry contributions
We’re additionally asserting enhancements to multi-agent observability, with contributions to OpenTelemetry that assist standardize tracing and telemetry for agentic techniques.
This offers groups deeper visibility into agent workflows, instrument name invocations, and collaboration—crucial for debugging, optimization, and compliance. We made the above enhancements to OpenTelemetry in collaboration with Outshift, Cisco’s incubation engine.
With the above enhancements, Azure AI Foundry now offers unified observability for brokers constructed with a number of frameworks, together with Microsoft Agent Framework and others like LangChain, LangGraph, and OpenAI Brokers SDK.
Voice Dwell API in Azure AI Foundry now typically obtainable
Mutli-agent workflows are more and more initiated by way of voice inputs and culminate in voice outputs. We’re excited to announce the common availability of Voice Dwell API, which empowers builders and enterprises to construct scalable, production-ready voice AI brokers. Voice Dwell API is a unified, real-time speech-to-speech interface that integrates speech-to-text (STT), generative AI fashions, text-to-speech (TTS), avatar, and conversational enhancement options right into a single, low-latency pipeline.
Organizations corresponding to Capgemini, healow, Astra Tech, and Agora are leveraging Voice Dwell API to construct customer support brokers, academic tutors, HR assistants, and multilingual brokers. Voice Dwell API is remodeling how builders construct voice AI brokers by offering an built-in, scalable, and environment friendly resolution.
Accountable AI capabilities public preview
Constructing on developments in agent observability and framework integration, it’s equally necessary to make sure that AI techniques function responsibly and securely—particularly as they grow to be extra deeply embedded in crucial enterprise workflows.
Based on McKinsey’s 2025 World AI Belief Survey3, the primary barrier to AI adoption is lack of governance and risk-management instruments. That’s why we’re placing the next accountable AI options in public preview within the coming weeks:
These capabilities are constructed into Azure AI Foundry, serving to organizations construct with confidence and adjust to inside and exterior requirements.
Buyer momentum
Azure AI Foundry options are serving to over 70,000 organizations worldwide—from digital natives to enterprise firms—rework AI innovation into sensible outcomes. For instance:
Commerzbank: Commerzbank is piloting Microsoft Agent Framework to energy avatar-driven buyer assist, enabling extra pure, accessible, and compliant buyer interactions.
The brand new Microsoft Agent Framework simplifies coding, reduces efforts and totally helps MCP for agentic options. We’re actually trying ahead to the productive utilization of container-based Azure AI Foundry brokers, which considerably reduces workload in IT operations.
— Gerald Ertl, Managing Director/Head of Digital Banking Options, Commerzbank AG
Citrix: Citrix is exploring how they’ll use agentic AI inside digital desktop infrastructure (VDI) environments to enhance enterprise productiveness and effectivity.
Citrix has at all times embraced versatile methods of working because the chief in safe work. As we transfer right into a world the place agentic AI works side-by-side with us, we’re excited to allow that additionally inside workspaces that our clients already use day by day. Microsoft’s Agent Framework brings a contemporary, developer-first method to constructing brokers. With assist for key APIs and languages, and native adoption of rising protocols for instrument calling and observability, it allows intuitive improvement of brokers on Azure AI Foundry with out compromising developer management. We’re wanting to leverage the framework to ship on our imaginative and prescient – enterprise-scale, production-ready AI brokers for our clients.
— George Tsolis, Distinguished Engineer, Citrix
TCS: Tata Consultancy Providers is actively constructing a multi-agent observe on the Microsoft Agent Framework, with a number of initiatives underway that showcase their strategic funding and technical depth, together with agentic options for finance, IT operations, and retail.
Adopting Microsoft Agent Framework isn’t just a technological development, however a daring step in the direction of reimagining trade worth chains. By harnessing Agentic AI and Frontier fashions, we allow our groups to construct versatile, scalable, enterprise-grade options that rework workflows and ship worth throughout platforms. True management is about empowering innovation, embracing change, and fostering an setting the place agility and collaboration drive excellence.
— Girish Phadke, Head, Microsoft Azure Follow, TCS
Sitecore: Sitecore is growing an answer that permits entrepreneurs to work together seamlessly with the platform by automating duties throughout all the content material provide chain—from creating and managing internet experiences to dealing with digital property—utilizing clever brokers.
By partnering with Microsoft to leverage its new Microsoft Agent Framework, Sitecore can deliver collectively the perfect of each worlds: the flexibleness to energy totally non-deterministic agentic orchestrations and the reliability to run extra deterministic, repeatable brokers. On the similar time, we profit from Microsoft’s enterprise-grade observability and telemetry, guaranteeing that these orchestrations usually are not solely highly effective but additionally safe, measurable, and production-ready.
— Mo Cherif, VP of AI, Sitecore
Elastic: Elasticsearch helps a local connector to Microsoft Agent Framework, enabling builders to seamlessly combine enterprise knowledge into clever brokers and workflows.
Elasticsearch is the context engineering platform and vector database of alternative for organizations to retailer and search their most beneficial operational and enterprise knowledge. With the brand new Microsoft Agent Framework connector, builders can now deliver probably the most related organizational context instantly into clever brokers and multi-agent workflows. This makes it simpler than ever to construct production-ready AI options that mix the reasoning energy of brokers with the velocity and scale of Elasticsearch.
— Steve Kearns, Normal Supervisor Search Options, Elastic
A trusted agent manufacturing facility for builders
Azure AI Foundry is greater than a platform—it’s a trusted agent manufacturing facility for builders and enterprises. Whether or not you’re a CIO trying to scale AI responsibly, a safety architect targeted on governance, or a developer constructing the following technology of clever brokers, Azure AI Foundry offers the instruments, frameworks, and belief you want.
Microsoft stands out within the AI panorama with its dedication to open requirements, interoperability, and accountable AI. The Microsoft Agent Framework, now in public preview, is a unified, enterprise-grade framework that integrates cutting-edge analysis and permits builders to seamlessly orchestrate multi-agent techniques with built-in observability, sturdiness, and compliance.
Not like different options, our framework helps integration with any API through OpenAPI, collaboration throughout runtimes with Agent2Agent (A2A), and dynamic instrument connections utilizing MCP. This ensures builders can scale back context-switching and keep in move, accelerating innovation.
The open-source nature of the framework invitations builders to contribute and form the way forward for agentic AI, making it a very collaborative and forward-thinking platform. With Microsoft, organizations can belief that their AI techniques shall be highly effective, environment friendly, accountable, and safe, addressing the highest limitations to AI adoption recognized in McKinsey’s 2025 World AI Belief Survey.
2 AI adoption is rising, however friction persists.
3 Insights on accountable AI from the World AI Belief Maturity Survey.