Azure AI Foundry Agent Service: Updates and Components
The Azure AI Foundry Agent Service (formerly Azure AI Agent Service) is a platform for developing, deploying, and managing sophisticated AI agents. These agents recognize situations, plan steps, acquire knowledge via RAG, and execute actions using tools. Recent updates focus on the new Foundry Project resource type, enhanced tool integration, and General Availability (GA) announcements, providing a streamlined environment for pro-code AI development.
Key Takeaways
The service was renamed to Azure AI Foundry Agent Service.
AI agents use RAG and tools for autonomous action execution.
The new Foundry Project resource simplifies AI service integration.
Development follows a defined code flow: definition, thread, message, run execution.
New tools include SharePoint Search and Bing Custom Search capabilities.
What is the Azure AI Foundry Agent Service and how is it positioned?
The Azure AI Foundry Agent Service is Microsoft's updated platform for building and managing AI agents, recently renamed from the Azure AI Agent Service to reflect its integration into the broader Azure AI Foundry ecosystem. An AI agent is fundamentally defined by its ability to recognize situations based on input, plan necessary steps, acquire external knowledge using Retrieval-Augmented Generation (RAG), and execute actions autonomously or semi-autonomously using integrated tools. Microsoft supports various agent types, including built-in agents like Copilot, third-party app provider agents, and customized enterprise agents developed via low-code or pro-code methods. Recent Build conference updates confirmed General Availability (GA) with SLA applied, alongside updates to tool selection and environment control features.
- Name Change: Transitioned from Azure AI Agent Service to Azure AI Foundry Agent Service.
- Definition of AI Agent: Agents recognize situations, plan steps, acquire knowledge (RAG), and execute tools/actions (autonomous/semi-autonomous).
- Agent Types (Microsoft Context): Includes Built-in Agents (e.g., Copilot), App Provider Agents, and Customized Agents (Pro-Code via Azure AI Foundry or No-Code via Copilot Studio).
- Build Conference Updates: Featured GA Announcement (SLA Applied), Tool/Model Selection Updates, and Environment Control Updates.
How does the new Foundry Project resource type simplify Azure AI development?
The Foundry Project is the new recommended resource type designed to streamline the development of AI services, replacing the legacy Hub-based Project structure. This new resource is based on the Cognitive Services Account and offers a single Foundry API for accessing all necessary AI services, significantly simplifying integration and management for developers. In contrast, the legacy Hub-based Project relied on the Machine Learning Service Workspace, required multiple SDKs, and included additional resources like Storage and Key Vault, leading to increased complexity. Developers can implement Foundry Projects using the Foundry Portal GUI for easy model testing and RAG construction, or leverage the Foundry SDK for robust, scalable production applications.
- New Resource Type (Recommended): Foundry Project, based on Cognitive Services Account, offering a single Foundry API for consolidated AI services.
- Legacy Resource Type: Hub-based Project, based on Machine Learning Service Workspace, requiring multiple SDKs and including additional resources like Storage and Key Vault.
- Implementation Methods: Utilize the Foundry Portal (GUI) for playground, easy model testing, RAG construction, and Index creation, or the Foundry SDK (Code) for production applications.
What is the typical single agent development flow using the Azure AI Foundry Agent Service code?
The single agent development flow focuses on defining the agent's capabilities and managing the conversational thread through a structured sequence of code steps. This process begins with Agent Definition, where developers specify instructions, the chosen LLM model, and the available tools, including knowledge sources (RAG data connection) and action tools (Function/API calls). Next, a Thread is created to serve as the message container, followed by the addition of the User Message, specifying the user's role and input. The core step is Run Execution, where the LLM processes the thread messages to determine the necessary action or tool calls. The run status transitions through several states before the Assistant Message, containing the final response, is added back to the thread as the output.
- Agent Definition (Code): Define instructions (agent's role), Model Specification (LLM choice), and Tools (Knowledge Sources and Action Tools).
- Thread Creation: Establish the message container for the entire conversation history.
- User Message Addition: Input the user's query or request into the thread, specifying the user role.
- Run Execution: The LLM infers actions based on the messages in the thread, determining necessary tool calls.
- Run Status Transition: Tracks the process through Queued (message sent to LLM), InProgress (processing), RequiresAction (tool call determined), and Completed (final response generated) states.
- Assistant Message Addition: The agent's final output is added as a message to the thread.
Which core tools and models are integrated into the Azure AI Foundry Agent Service?
The Azure AI Foundry Agent Service integrates a comprehensive set of tools and diverse models essential for enabling sophisticated agent functionality. Tools are broadly categorized into Knowledge Sources, such as Fabric, Azure AI Search, and the newly added SharePoint Search, which facilitate RAG capabilities. Action Tools include Logic Apps, Azure Functions (for function calling), and External Open APIs, enabling the agent to perform external operations. A key update is the support for Modular Common Protocol (MCP), where the Foundry Agent Service acts as an MCP Client using a Remote MCP Server URL. Furthermore, the service offers flexibility in LLM choice, integrating models from Azure OpenAI Service, Mistral, Llama, and specialized Reasoning Models like O3 mini and O1, allowing developers to select the best fit for their specific task requirements.
- Tools: Includes Knowledge Sources (Fabric, SharePoint Search, Azure AI Search) and Action Tools (Logic Apps, Azure Functions, External Open API).
- MCP Support: The Foundry Agent Service functions as an MCP Client, utilizing a Remote MCP Server URL (June Update).
- Bing Custom Search (New): Allows agents to specify searchable sites, currently available via REST API only.
- Function Calling Implementation: Defined via Docstrings (Explanation, Parameters, Return Structure) and is crucial for the LLM to decide when and how to use a tool.
- Models: Supports Azure OpenAI Service, Mistral, Llama, and specific Reasoning Models (O3 mini, O1).
What are the key cross-cutting concerns for managing and governing Azure AI Foundry Agents?
Managing and governing Azure AI Foundry Agents requires addressing several critical cross-cutting concerns to ensure operational excellence, security, and performance. Agent Evaluation is a necessary step for measuring the effectiveness and reliability of the agents, facilitated by the Azure AI Evaluation SDK. The platform also supports complex architectures through Multi-Agent Orchestration, enabling specialized agents to collaborate seamlessly on intricate tasks. Security and Governance are paramount, addressed through dedicated features like Microsoft Entra Agent ID, which ensures secure identity management and robust access control for all agents. Finally, Environment Control features provide the necessary configuration management capabilities to maintain deployment and operational stability across different stages of the agent lifecycle.
- Agent Evaluation: Utilizes the Azure AI Evaluation SDK for performance measurement and validation.
- Multi-Agent Orchestration: Supports complex interactions and collaboration among multiple specialized agents.
- Security and Governance: Managed via Microsoft Entra Agent ID for secure identity and access control.
- Environment Control: Provides necessary configuration and management capabilities for deployment.
Frequently Asked Questions
What was the previous name of the Azure AI Foundry Agent Service?
The service was formerly known as the Azure AI Agent Service. This renaming reflects its integration into the broader Azure AI Foundry ecosystem, which focuses on providing comprehensive tools for professional, pro-code AI development and management.
How do AI agents acquire external knowledge?
AI agents acquire external knowledge primarily through Retrieval-Augmented Generation (RAG). This process connects the agent to various Knowledge Sources, such as Azure AI Search or SharePoint Search, allowing it to access and utilize proprietary or external data efficiently.
What is the purpose of the new Foundry Project resource?
The Foundry Project is the recommended resource type that simplifies AI service integration. It is based on the Cognitive Services Account and provides a single Foundry API, eliminating the need for multiple SDKs required by the legacy Hub-based Project structure for AI development.
