Announcing General Availability of Azure AI Foundry Agent Service

MadsBolaris

Microsoft

May 19, 2025

Powering the Next Generation of Intelligent Agents

Agents are revolutionizing business automation by evolving from simple chatbots into sophisticated, collaborative systems. Fueled by advancements in model reasoning and efficiency, these agents can now handle complex, multi-step processes with speed and accuracy. This marks a shift from isolated tools to dynamic, scalable agent workforces that coordinate tasks, share context, and adapt in real time.

This transformation enables businesses to optimize operations and elevate customer experiences through AI-driven workflows. Instead of relying on single-purpose bots, organizations are deploying ecosystems of specialized agents that can interact, reason, and respond to changing conditions with minimal oversight.

Yet building and scaling these systems is not without challenges. Developers must integrate diverse data sources, manage persistent workflows, enforce enterprise-grade security, and ensure smooth interoperability. Meeting these needs requires a flexible, secure platform that abstracts orchestration complexity while maintaining full control.

The Journey to GA – From Preview to Production

Azure AI Foundry Agent Service was first introduced as a public preview at Ignite 2024, providing developers with a comprehensive platform for building, deploying, and scaling intelligent agents. Since then, over 10,000 customers have used the service to power their AI-driven automation, creating everything from customer support assistants to process optimization bots.

Customers like NTT DATA have leveraged Azure AI Foundry Agent Service to automate routine customer service inquiries, reducing response times and enhancing customer satisfaction, resulting in a 50% faster time to market and improved productivity across functions.

YoungWilliams has taken this further with Priya, an AI assistant that automates complex government program inquiries, supports caseworkers, and ensures compliance with robust security controls. Built to handle programs like the Low Income Home Energy Assistance Program (LIHEAP), Temporary Assistance for Needy Families (TANF), and Supplemental Nutrition Assistance Program (SNAP), Priya has transformed how government agencies engage with citizens, delivering faster responses and better service quality.

These real-world use cases have been instrumental in refining the platform’s capabilities to meet the needs of enterprise-scale deployments.

Today, we’re excited to announce that Azure AI Foundry Agent Service is now generally available. This milestone represents a significant step forward in empowering developers to create enterprise-grade, multi-agent systems with the reliability, security, and flexibility required for mission-critical workloads. With this release, we are introducing several powerful new features and integrations that make it even easier for developers to build and scale their agent solutions.

Multi-Agent Orchestration: Building Digital Workforce

Azure AI Foundry Agent Service now allows developers to design complex, multi-agent workflows that coordinate specialized agents to perform structured, long-running tasks. This orchestration capability is built around two core concepts:

Connected Agents (preview): Connected agents allow for point-to-point interactions, where agents can call other agents as tools to handle specialized tasks. This approach is well-suited for use cases like task delegation, modular processing, or context-specific workflows, where each agent independently contributes to the overall solution.
Multi-Agent Workflows (preview): In contrast, multi-agent workflows provide a structured, stateful orchestration layer that coordinates multiple agents over complex, multi-step processes. These workflows handle context management, error recovery, and long-running durability, making them ideal for scenarios like customer onboarding, financial transaction processing, or supply chain automation, where agents need to maintain context across multiple steps.

To support advanced multi-agent orchestration, Foundry Agent Service integrates directly with the converged runtime for Semantic Kernel and AutoGen. This new client library will combine AutoGen’s dynamic orchestration patterns with the modular, production-grade architecture of Semantic Kernel, exposing a single API surface for defining, chaining, and managing both single-agent and multi-agent workflows. This integration allows teams to build locally, simulate interactions, and deploy to the cloud with consistent behavior across environments.

"The multi-agent workflow in Azure AI Foundry Agent Service provides configurability, seamless deployment, and an intuitive Visual Studio Code interface for creating multi-agent workflows. By incorporating this advanced technology into our AI-powered solutions, we provide exceptional value to our customers." – Girish Phadke, Head, TCS Microsoft Azure Practice

Open and Interoperable Tools: Building a Connected Agent Ecosystem

Agents rarely operate in isolation. They need to access enterprise data, interact with third-party systems, and collaborate with other agents. Azure AI Foundry Agent Service embraces this open, interconnected approach by supporting a wide range of open standards and partner integrations.

Key announcements include:

Logic Apps Integrations: Developers can now leverage 1,400+ Azure Logic Apps workflows as tools for agents in Foundry Agent Service to automate complex business processes and add actions in Azure Logic Apps to trigger agents, enabling seamless task orchestration across systems.
Knowledge Integrations: The platform now includes SharePoint as a first-party, built-in knowledge tool, alongside existing tools like Microsoft Fabric and Bing Search, providing agents with richer, context-aware insights.
Partner Tools: Developers can access a wider range of third-party tools from partners like Auquan, Celonis, InsureMO, LEGALFLY, LexisNexis, MiHCM, Morningstar, and Trademo, enhancing agents' ability to retrieve and process domain-specific knowledge.

Agent Catalog: The agent catalog includes a growing library of agent code samples from both Microsoft and partners like Auquan, Marquee Insights, MiHCM, Saifr, and Sight Machine, providing developers with reusable building blocks to accelerate agent development.

Open Protocols: The new Agent2Agent (A2A) API head from Azure AI Foundry Agent Service enables open-source orchestrators with A2A connectors to seamlessly use Foundry Agent Service agents, facilitating multi-turn conversations between agents without requiring custom integrations. This support also extends to multi-cloud orchestration, allowing developers to connect Foundry agents with agents from platforms like SAP Joule and Google Vertex AI, unlocking more flexible, cross-cloud agent architectures.

“Our aim is to enable productivity for SAP and Microsoft’s many joint customers, with the end goal of true agent-to-agent collaboration,” said Walter Sun, SVP and Global Head of AI, SAP. “Our collaboration on SAP Joule and Microsoft 365 Copilot’s bi-directional integration aims to provide users instant access to information across SAP business applications and Microsoft 365. Together, we’re shaping the future of intelligent work."

Expanded Ecosystem: Azure AI Foundry Agent Service integrates with popular agent orchestration frameworks like Crew AI, LangGraph and LlamaIndex. This integration ensures that agents from Azure AI Foundry Agent Service can be seamlessly called from these frameworks, facilitating multi-turn conversations between agents and enhancing cross-platform interoperability.

"Azure AI Foundry Agent Service works amazingly, and we are able to utilize them in IBM assets and solutions out of the box for chat with document features. This reduces the need to build additional agents and optimizes code development, debugging, and testing for managing our current set of foundational and functional agents." – Rabeela Janorious, Distinguished Engineer and Global CTO, IBM Consulting AI Integration Services

"Twilio is leveraging Microsoft Azure AI Foundry to simplify and accelerate businesses' development of virtual agents. Conversational AI allows businesses to unlock value for their business and their customers by significantly improving the overall customer experience by streamlining interactions, reducing wait times, and efficiently resolving issues. We look forward to expanding our collaboration with Microsoft by adopting new features from Foundry Agent Service to further enhance agents' capabilities." - Andy O'Dower, VP of Product for Voice and Video at Twilio

Trust and Enterprise Readiness: Built for the Enterprise

For businesses to confidently deploy AI agents at scale, robust security, data governance, and AgentOps tools are critical. Azure AI Foundry Agent Service is designed with these needs in mind, providing the controls and safeguards required for mission-critical workloads.

Ensuring Robust Agent Evaluation and Monitoring (AgentOps)

Evaluation Metrics: Built-in evaluation tools enable developers to assess the tool call accuracy, task adherence, efficiency, and responsiveness of agents under real-world conditions. This helps identify gaps and optimize agent behavior proactively.
Built-in Tracing: Gain detailed insights into how agents process requests, including intermediate steps, data flows, and function calls. This tracing helps pinpoint performance bottlenecks and optimize agent workflows.
Monitoring and Reporting: Track agent performance metrics such as response time, error rates, and task completion rates through integrated dashboards. Proactively address issues before they impact users.

Built-In Governance and Safety

Data Isolation: Agents can be configured to access only specific data sources, reducing the risk of unauthorized data exposure.
Content Filtering and Safety Controls: Integrated safety mechanisms prevent harmful or inappropriate content generation.
BYO-Thread Storage: Store conversation data within customer-owned Cosmos DB instances, maintaining complete control over data retention, encryption, and compliance.
New Foundry resource type and Foundry API: Simplifies the management of agents, models, and finetuning under a single Azure resource provider namespace. Allows administrators to oversee resources more effectively while leveraging Azure RBAC, networking, and policies within one cohesive framework.

"At Soul Machines, we believe that the true power of AI lies not just in its intelligence, but in its ability to seamlessly integrate into our daily lives. Azure AI Foundry Agent Service provides a robust and secure foundation, transforming complex AI capabilities into intuitive and engaging conversational experiences." – Shane Blackett, CTO, Soul Machines

Looking Ahead

As we continue to evolve the Azure AI Foundry Agent Service, our vision is to make it the most flexible, interoperable platform for building intelligent agent ecosystems. This includes ongoing work to unify the Semantic Kernel and AutoGen SDKs, integrate containerized agents for managed orchestration, and expand support for external agents. These innovations will further simplify the path from prototyping to production, allowing businesses to rapidly deploy intelligent agents that can operate across hybrid, multi-cloud environments.