The world is moving from AI assistants that answer to agents that act — systems that can plan, reason, and collaborate across data, tools, and people. These agents don’t just answer questions; they drive outcomes, automate decisions, and integrate deeply into the fabric of business operations.

For enterprises, this shift brings enormous opportunity — and new technical demands. To realize it, organizations need a platform that combines openness with a powerful control loop to manage, evaluate, and continuously improve agents for developers and IT.

That’s what Foundry Agent Service delivers—designed to help teams build, host, and scale intelligent agents securely and at enterprise scale.

Customer Momentum

Since its general availability at Microsoft Build 2025, Foundry Agent Service has become the foundation for agent-driven transformation across industries. Developers at KPMG, NTT DATA, Fujitsu, YoungWilliams, and Aon are deploying agents that reason over data, integrate with internal systems, and coordinate entire business processes — all within enterprise governance and compliance standards.

BMW is using Foundry Agent Service and the Microsoft Agent Framework to power a multi-agent system that accelerates test-fleet data analysis by 12×, enabling engineers to turn terabytes of telemetry into insights in minutes instead of days. This is helping BMW shorten design cycles, democratize data access, and drive faster innovation

Commerzbank has deployed Ava, an AI-powered agent built on Foundry Agent Service, to handle more than 30,000 monthly customer conversations—resolving 75% autonomously while ensuring every interaction remains secure and compliant.

These real-world deployments show how enterprises trust Foundry Agent Service to run AI agents securely and reliably at scale—backed by the same governance, identity, and reliability standards that define the Microsoft Cloud.

Advancing the Open, Enterprise Agent Platform

At Microsoft Ignite 2025, we’re introducing the next evolution of Foundry Agent Service—the agent-native runtime in Microsoft Foundry. It delivers expanded enterprise capabilities and a unified developer experience across the Foundry portal and SDK, enabling seamless orchestration, deployment, and governance for agents built with any framework or model.

The release adds new interoperability layers, intelligent context systems, managed hosting, and deeper Microsoft 365 integration—anchored in three core pillars: Open by Design, Connected Intelligence, and Enterprise Trust & Reach.

Open by Design

Openness has always been at the heart of Foundry Agent Service. The service is designed to work with any agent framework, model, or tool—built on open standards such as Model Context Protocol (MCP), Agent2Agent (A2A), and OpenAPI for interoperability and developer freedom.

• New Model Support (public preview)

With the arrival of Anthropic Claude in Foundry Models, we’re excited to announce the Claude family – Haiku 4.5, Sonnet 4.5, Opus 4.1 – now available in Foundry Agent Service.

Foundry Agent Service also supports a broader ecosystem of frontier and specialist models from providers including OpenAI, Meta, xAI, and DeepSeek —enabling enterprises to build flexible, enterprise-grade agents optimized for their specific needs.

This multi-model support underscores our commitment to an open, enterprise-ready AI ecosystem—where developers can innovate freely and agents operate with the best intelligence available. Foundry Agent Service also enables fine-tuned models to directly power agents, improving reasoning and task accuracy. With synthetic data generation, teams can create training data from their APIs and business logic, teaching models to reflect enterprise context so agents act with higher precision and reliability.

 

• Microsoft Agent Framework (public preview)

Launched in October, the Microsoft Agent Framework unifies Semantic Kernel and AutoGen into one open-source SDK for designing, testing, and orchestrating intelligent agents across platforms. Built for interoperability, it extends seamlessly to multi-cloud environments and integrates with community-driven research initiatives like Agent Lightning. Updates include:

• Deep integrations with Foundry Agent Service: Shared architecture provides consistent developer experience across environments with built-in durability, security, and interoperability.
• Durable execution: Agents persist and recover automatically across restarts, ensuring reliability for long-running or mission-critical workflows.
• Built-in trust and security: Native Microsoft Purview SDK integration provides end-to-end data protection, automatically classifying and securing prompts, tool outputs, and responses.
• Compatibility with new UI standards and protocols: The framework now supports AG-UI (by CopilotKit) and ChatKit (by OpenAI), enabling rich, chat-based and UI-driven agent experiences through popular front-end protocols. Developers can integrate an agent endpoint with the frontend to synchronize agent state, user interface intents, and user interactions.

“Microsoft Agent Framework's open standards approach delivers key features like built-in observability, checkpoints for long-running processes, and human-in-the-loop controls—all essential for our use case. The framework accelerated our development timeline while maintaining enterprise-grade security and reliability.”  

- Jared Haite, Product Lead, Generative AI, Duck Creek Technologies

“Our experience with Microsoft Agent Framework has been nothing short of transformative. Building agents is remarkably seamless, and the native integration with MCP and A2A servers enables us to move from concept to production faster than ever before. The flexible middleware architecture allows us to tailor every interaction to our clients’ needs, while the smooth migration from Semantic Kernel ensured complete continuity. Together with Microsoft, we’re shaping the next generation of intelligent, agent-powered enterprise solutions.”

– Charlie Doubek, Global VP, Agentic AI Leader, Cloud & Security, NTT DATA

“The new version of Microsoft Agent Framework represents a major step forward: native MCP support, hosted agents in Foundry, and unified observability significantly reduce engineering complexity while strengthening governance and compliance across our multi-tenant banking environment. This empowers us to deliver secure, production-ready AI agents at scale—aligned with the strict requirements of EU banking regulation and the expectations of our customers and employees.”

- Armin Woworsky, Distinguished Engineer, Raiffeisen Bank International

 

• Hosted agents (public preview)

Hosted agents in Foundry Agent Service make it effortless for teams to move from local experimentation to full-scale production. Developers can now deploy custom-code agents built with Microsoft Agent Framework, LangGraph, or other open-source frameworks directly into a fully managed runtime—no container images, Kubernetes clusters, or deployment pipelines required.

Hosted agents are designed to simplify runtime management while preserving developer choice. You bring your code; Foundry Agent Service handles the rest—autoscaling, monitoring, identity integration, and security.

This experience allows teams to operationalize multi-agent systems quickly—whether coordinating specialized copilots across finance, HR, and operations, or orchestrating tool-rich workflows across distributed environments.

 

“Hosted agents in Foundry give Citrix the freedom to build AI solutions using preferred agent frameworks and run them at scale on Azure. With unified security, evaluation, monitoring, and observability via a common management console, we can accelerate delivering enterprise-grade, production-ready AI agents for our customers.”

- George Tsolis, Distinguished Engineer, Citrix

Connected Intelligence

Foundry Agent Service now connect more deeply with data, tools, and agents—empowering systems that reason, collaborate, and act with richer context and continuity.

• Built-in memory (public preview)

Memory is a cornerstone of intelligent agents—turning them from stateless responders into systems that can learn, adapt, and act with continuity.
The new built-in memory in Foundry Agent Service gives developers a scalable, secure way for agents to retain chat history, preferences, and task outcomes across sessions and workflows.
Developers can define custom topics to optimize what agents remember, while Foundry automatically extracts and consolidates key insights for precision and speed. Each memory instance is secured with Entra ID and scoped by user, supporting multi-agent collaboration without data crosstalk—so agents share only what’s relevant while sensitive information stays protected.

 

• Foundry Tools (public preview)

Foundry now introduces Foundry Tools, a unified hub for discovering, testing, and governing the tools that make agents truly intelligent and actionable. With over 1,000 curated Microsoft and partner tools, developers can connect agents to real-time data, business systems, and productivity apps—turning them from responders into active problem-solvers. Enterprises can also create private organizational catalogs for internal APIs, ensuring secure, compliant reuse across teams. Foundry supports open standards like MCP, A2A, and OpenAPI, with flexible authentication through Microsoft Entra Agent Identity, OAuth passthrough, and Managed Identity. Through integration with the AI Gateway, admins can apply Azure Policy and runtime guardrails for every tool call, ensuring governance and observability at scale.

 

• Foundry IQ (public preview)

Now in public preview, Foundry IQ, powered by Azure AI Search, is natively integrated into Foundry Agent Service, giving agents built in Foundry instant access to trusted, high-quality enterprise knowledge. Developers can ground agents on knowledge bases that unify data from sources such as SharePoint, OneLake, Azure Data Lake Storage (ADLS), and the web—without writing custom RAG logic or retrieval pipelines. Each agent call to IQ automatically enforces Purview-based security, ensuring that context stays accurate, up to date, and compliant. With this integration, agents can reason with richer context, deliver more reliable responses, and build trust through grounded, explainable answers—all within the same managed runtime.

 

• Multi-agent workflows (public preview)

Available now in both the Foundry portal and the VS Code extension, multi-agent workflows let developers visually design or programmatically orchestrate complex, multi-step processes that coordinate multiple specialized agents. Built natively into Foundry Agent Service, workflows provide a structured way to define how agents collaborate, share state, recover from errors, and maintain context over long-running operations.

With the visual designer, developers and solution architects can drag, connect, and configure agents, tools, and data flows to quickly model business processes such as onboarding, financial approvals, or IT service automation.

 

“Multi-agent workflows in Foundry Agent Service introduces a powerful and intuitive approach to developing workflow for multiple agents. Workflows help democratize and accelerate multi-agent development.”

- Masahiro Yoshioka, Program Manager, Mazda Motor Corporation

“Multi-agent workflows in Microsoft Foundry are a great approach to orchestrate complex, multi-step processes using specialized agents. This capability could enhance efficiency and transparency across various business functions while supporting security and compliance standards.”

- Gerald Ertl, Managing Director/Head of Digital Banking Solutions, Commerzbank AG

Enterprise Trust & Reach

Foundry Agent Service is extending its enterprise-grade foundation for deployment, security, and scale—so organizations can operationalize agents where people and processes already work, with governance built in.

• One-click publishing to Microsoft 365 and Agent 365 (public preview)

Now in public preview, Foundry Agent Service connects natively with Microsoft 365 and Agent 365, bridging agent development and enterprise deployment. With one-click publishing, developers can instantly deploy agents from Foundry to Microsoft 365 Copilot and Teams Chat—no manual setup or manifest editing required. We have streamlined configuration across Entra ID, Azure Bot Service, and Admin Center into a simple guided flow—all in one portal.

A single build can now reach multiple Microsoft 365 channels, bringing agents directly into the flow of work. Every deployment routes through the Microsoft Admin Center for approval, giving IT administrators unified governance, compliance, and visibility.

At the platform level, Agent 365 extends this integration as the control plane for AI agents—applying the same identity, security, and compliance infrastructure that organizations already trust for human users, adapted for agent operations.

 

“Integrating Microsoft Foundry with Microsoft 365 Copilot fundamentally changed how we deliver AI solutions to our government partners. With Foundry’s one-click publishing to Teams and Copilot, we can take an idea from prototype to production in days instead of weeks—while maintaining the enterprise-grade security and governance our clients expect. It’s a game changer for how public services can adopt AI responsibly and at scale.”

- John Tidwell, CTO, YoungWilliams

 

• Foundry Control Plane (public preview)

Foundry Control Plane now in public preview, lets teams govern any agent in one place, no matter where it is built or deployed. Agents created in Foundry Agent Service appear automatically, and agents built elsewhere register easily. Controls provide consistent guardrails across inputs, outputs, tool calls, and tool responses, covering task adherence, sensitive data detection, groundedness, and prompt injection mitigation.

Observability, now generally available, brings together evaluations, OpenTelemetry based tracing, continuous red teaming, and Azure Monitor insights so teams can understand behavior, spot issues, improve performance, and refine quality.

Security integrations with Microsoft Defender and Microsoft Purview surface risks such as jailbreak attempts, sensitive data exposure, and policy drift, and every agent receives a Microsoft Entra Agent ID for durable identity, lineage, and access governance. Fleet wide operations show health, cost, performance, risk, and policy coverage in one connected view. Upcoming integrations with Palo Alto Networks and Zenity add even more runtime protection, anomaly detection, and continuous policy enforcement across the agent lifecycle.

 

Get started today

Microsoft Foundry brings every part of the agent development lifecycle into one connected experience. From design and testing to deployment and governance, Foundry Agent Service gives developers and enterprises everything they need to move from prototype to production with confidence.

Start building your own intelligent agents today with Microsoft Foundry.

If you’re attending Microsoft Ignite 2025, or watching on-demand content later, be sure to check out these sessions:

• Innovation Session: Build & Manage AI Apps with Your Agent Factory
• AI agents in Azure AI Foundry, ship fast, scale fearlessly
• AI powered automation & multi-agent orchestration in Azure AI Foundry
• AI builder’s guide to agent development in Foundry Agent Service
• AI tools for agents in Azure AI Foundry: Plug, play and reach
• The future of RAG with agentic knowledge retrieval and AI Search

To learn more, visit Microsoft Learn and explore resources including the AI Agents for Beginners, Microsoft Agent Framework, and course materials that help you build and operate agents responsibly.