{"id":40636,"date":"2025-05-19T09:00:00","date_gmt":"2025-05-19T16:00:00","guid":{"rendered":""},"modified":"2026-01-05T12:38:23","modified_gmt":"2026-01-05T20:38:23","slug":"azure-ai-foundry-your-ai-app-and-agent-factory","status":"publish","type":"post","link":"https:\/\/azure.microsoft.com\/en-us\/blog\/azure-ai-foundry-your-ai-app-and-agent-factory\/","title":{"rendered":"Azure AI Foundry: Your AI App and agent factory"},"content":{"rendered":"\n

Software development is undergoing a transformation. In the old world, building software was a long relay\u2014weeks to plan, months to build, and quarters to launch. In the new world, AI-powered tools turn ideas into prototypes in hours and production solutions in days. To achieve this, developers need an end-to-end platform that seamlessly connects code, collaboration, and cloud. Microsoft is the only company where all three come together, with Visual Studio Code, GitHub, and Azure forming a unified, AI-native development experience designed to empower every developer to shape the future with AI<\/strong>. We\u2019re delivering a full-stack AI platform with built-in security and trust\u2014giving you everything you need to build and run AI-powered apps and agents, from cloud to edge. <\/p>\n\n\n\n

In just two years, we\u2019ve gone from early foundation models to AI agents capable of complex workflows. Microsoft Azure AI Foundry has already grown to support more than 70,000 customers, processing 100 trillion tokens last quarter, and powering 2 billion daily enterprise search queries. What began as an application layer has now become a full-stack platform for building intelligent agents that deliver real business value.<\/p>\n\n\n\n

\n
Design, customize, and manage AI apps with Azure AI Foundry today <\/a><\/div>\n<\/div>\n\n\n\n

To help these AI agents become even more powerful and handle more complex tasks, they need state-of-the-art models, integrated tools, and built-in governance. Today at Microsoft Build 2025, we\u2019re unveiling 10 major innovations in Azure AI Foundry<\/strong> to make this possible.<\/p>\n\n\n\n

1. New models<\/h2>\n\n\n\n

We\u2019re expanding our catalog with cutting-edge models to give developers more choice. This includes Grok 3<\/strong> from xAI<\/a> available today, Flux Pro 1.1<\/strong> from Black Forest Labs coming soon, and Sora coming soon <\/strong>in preview<\/a> via Azure OpenAI in Foundry Models<\/a>. We now have over 10,000 open-source models from Hugging Face<\/strong><\/a> available in Foundry Models<\/strong>. We support full fine-tuning<\/a> (including LoRA\/QLoRA and DPO), so you can tailor fine-tunable models to your needs. Additionally, we\u2019re rolling out a new developer tier for fine-tuning\u2014no hosting fees<\/strong><\/a>, just a simple way to experiment and evaluate fine-tuning methods without the overhead.<\/a><\/p>\n\n\n\n

\n
Learn more about new models in Azure AI Foundry<\/a><\/div>\n<\/div>\n\n\n\n
\"xAI,\n\t\t\t\n\t\t\t\t\n\t\t\t<\/svg>\n\t\t<\/button><\/figure>\n\n\n\n

2. Smarter model system<\/h2>\n\n\n\n

Choosing the right model for each task is now easier. Our new model router<\/strong><\/a> automatically selects the optimal Azure OpenAI model for your prompt\u2014boosting quality while reducing costs. Starting next month, we\u2019re extending reserved capacity<\/strong><\/a> across Azure OpenAI and select Foundry Models (like models from Black Forest Labs, DeepSeek, Mistral, Meta, and xAI), so you get consistent performance even under heavy load. All these models will be accessible through a unified API and Model Context Protocol (MCP) server for Azure AI Foundry models, making it seamless to go from prototype to production.<\/p>\n\n\n\n

\n
Learn more about model router<\/a><\/div>\n<\/div>\n\n\n\n
\"Model<\/figure>\n\n\n\n

3. Azure AI Foundry Agent Service <\/h2>\n\n\n\n

Azure AI Foundry Agent Service is now generally available<\/a>, empowering you to design, deploy, and scale production-grade AI agents with ease. More than 10,000 organizations<\/strong>\u2014including leaders like Heineken<\/a>, Carvana<\/a>, and Fujitsu<\/a>\u2014have already used Azure AI Foundry to automate complex business processes with their own data and knowledge. This fully managed service<\/a> takes care of infrastructure and orchestration, and it comes with ready-to-use templates, actions, and connectors to more than 1,400 enterprise data sources<\/strong> (such as SharePoint, Microsoft Fabric, and third-party systems), speeding up development of context-aware agents. And with a few clicks, you can deploy your agents into Microsoft 365 (Microsoft Teams and Office apps) or other platforms like Slack and Twilio\u2014bringing agents directly into the tools your employees use every day<\/strong>. Watch the Azure AI Foundry Agent Service demo<\/a>.<\/p>\n\n\n\n

\n
Learn more about Azure AI Foundry Agent Service<\/a><\/div>\n<\/div>\n\n\n\n

4. Multi-agent orchestration<\/h2>\n\n\n\n

Real-world workflows often require multiple agents<\/a> working together. Azure AI Foundry now makes that easy, across any cloud<\/strong>. Agents can call each other as tools (connected agents)<\/strong>, passing tasks between specialized agents to solve complex problems collaboratively. New multi-agent workflows<\/strong> provide a stateful layer to manage context, handle errors, and maintain long-running processes\u2014great for scenarios like financial approvals or supply chain operations. We\u2019ve also implemented open interoperability standards<\/strong>\u2014such as Agent-to-Agent (A2A) communication<\/strong>, to enable different agents to exchange information and coordinate tasks, and the Model Context Protocol (MCP)<\/strong> to enable agents to share and interpret contextual data consistently. These standards ensure agents can collaborate seamlessly across Azure, AWS, Google Cloud, and on-premises environments<\/strong>. Under the hood, we are unifying our Semantic Kernel and AutoGen frameworks<\/strong> to support this seamless agent orchestration.<\/p>\n\n\n\n

And because Azure AI<\/strong> Foundry powers Microsoft Copilot Studio<\/a><\/strong>, you maintain a continuous loop\u2014from model selection and fine-tuning to deploying pre-built agents. For instance, Stanford Medicine<\/a><\/strong> is already using our healthcare agent orchestrator in Azure AI Foundry alongside Microsoft Copilot Studio to streamline tumor-board meetings with custom clinical workflows.<\/p>\n\n\n\n

\n
Learn more about multi-agent orchestration<\/a><\/div>\n<\/div>\n\n\n\n

5. Agentic retrieval<\/h2>\n\n\n\n

We\u2019re making information retrieval smarter to support these advanced agents. Agentic retrieval<\/strong><\/a> <\/strong>in Azure AI Search<\/a> is a new multi-turn query engine designed for complex questions. It uses conversation context and an embedded LLM to break down user queries into sub-queries, runs multiple searches in parallel, and then compiles a comprehensive answer with citations. In early tests, this approach improves answer relevance by up to ~40%<\/strong><\/a> on complex, multi-part questions. Now in public preview, agentic retrieval lets your agents connect to enterprise data more effectively<\/strong> using advanced retrieval for accurate, grounded answers.<\/p>\n\n\n\n

\n
Learn more about agentic retrieval<\/a><\/div>\n<\/div>\n\n\n
\"A\n\t\t\t\n\t\t\t\t\n\t\t\t<\/svg>\n\t\t<\/button><\/figure>\n\n\n\n

6. Always-on observability <\/h2>\n\n\n\n

To hold agents accountable in production, you need visibility into their performance. We\u2019re previewing new Foundry Observability<\/strong><\/a> features that provide end-to-end monitoring and diagnostics. You\u2019ll get built-in metrics on latency, throughput, usage, and quality\u2014plus detailed trace logs<\/strong> of each agent\u2019s reasoning steps and tool calls. During development, our Agents Playground now shows evaluation benchmarks and traces to help you refine prompts and logic. As you move to CI\/CD, we offer integrations with GitHub and Azure DevOps to incorporate tests and guardrails. And in production, a unified dashboard (integrated with Azure Monitor) gives you real-time insight and alerts for your models and agents. Watch the Azure AI Foundry observability demo<\/a>.<\/p>\n\n\n\n

\n
Learn more about Foundry Observability<\/a><\/div>\n<\/div>\n\n\n\n

7. Enterprise-grade identity for agents<\/h2>\n\n\n\n

Microsoft Entra Agent ID<\/a> is the first step in managing agent identities in your organization, giving you full visibility and control over what your AI agents can do. Microsoft Entra Agent ID assigns a unique, first-class identity to every agent you build with Azure AI Foundry or Microsoft Copilot Studio<\/strong>. This means your AI agents get the same identity management as human users\u2014appearing in your Microsoft Entra directory so you can set access controls and permissions for each agent. Soon, security admins will be able to apply Conditional Access policies, multi-factor authentication, and least-privilege roles to agents, and monitor their sign-in activities. If an agent shouldn\u2019t access a resource, it will be blocked just like a regular user would. Watch the enterprise-grade identity for agents demo<\/a>.<\/p>\n\n\n\n

\n
Learn more about Microsoft Entra Agent ID<\/a><\/div>\n<\/div>\n\n\n\n

8. Trustworthy AI built-in<\/h2>\n\n\n\n

Responsible AI is non-negotiable for us and our customers. We\u2019ve added new capabilities to discover, protect, and govern AI systems<\/strong> from the start. Agent Evaluators <\/strong>now automatically check if an agent is following user intent and using tools correctly, flagging issues for developers. An AI Red Teaming Agent<\/strong><\/a> constantly probes your agents for vulnerabilities or biases, so you can fix weaknesses before deployment. Our content filtering has gotten smarter with \u201cSpotlighting\u201d<\/strong><\/a>\u2014an enhancement to Prompt Shields<\/strong> that better detects and mitigates malicious prompt injections <\/strong>(whether from users or incoming data). We\u2019ve also enabled enhanced guardrails<\/strong><\/a> to prevent agents from revealing sensitive information (PII)<\/strong> or straying off approved tasks<\/strong>. On the security<\/a> side, Azure AI Foundry integrates with Microsoft Defender for Cloud<\/strong><\/a> to raise alerts when threats occur. And to help with compliance, we offer out-of-box integration with governance tools<\/strong> like Credo AI<\/strong>, Saidot<\/strong>, and Microsoft Purview<\/strong><\/a> for tracking model performance, fairness, and regulatory requirements. From day one, Azure AI Foundry equips you with the safety, security, and governance<\/strong> tools to build AI you (and your users) can trust.<\/p>\n\n\n\n

\n
Learn more about agents security<\/a><\/div>\n<\/div>\n\n\n
\"A\n\t\t\t\n\t\t\t\t\n\t\t\t<\/svg>\n\t\t<\/button><\/figure>\n\n\n\n

9. Foundry Local<\/h2>\n\n\n\n

Not all AI needs to run in the cloud\u2014sometimes the best solution is at the edge. Foundry Local<\/strong><\/a> is a new runtime for Windows and Mac for AI models and agents.<\/strong> With Foundry Local, you can build cross-platform AI apps that work offline, keep sensitive data locally, and reduce bandwidth costs. This opens scenarios like manufacturing on a factory floor with spotty internet, or field service apps that need AI in remote areas. We are also integrating Foundry with Azure Arc. With the upcoming Azure Arc and Foundry integration, you can manage and update on-device AI deployments centrally. In short, Foundry becomes your AI factory<\/strong>\u2014delivering generative AI where your data is, whether that\u2019s in the cloud or on-premises. Watch the Foundry Local demo<\/a>.<\/p>\n\n\n\n

\n
Learn more about Foundry Local<\/a><\/div>\n<\/div>\n\n\n\n

10. Future innovations for Foundry Labs<\/h2>\n\n\n\n

We\u2019re exploring the next frontiers with Microsoft Research in our Foundry Labs<\/strong><\/a>. One exciting invention is Project Amelie<\/strong>, powered by RD Agent from Microsoft Research, an autonomous agent that can build complete machine learning pipelines from a single prompt. Give it a task like \u201cpredict customer churn from our dataset,\u201d and Amelie will ingest data, train models, and produce a deployable solution\u2014an experiment in AI developing AI.<\/p>\n\n\n

\"A<\/figure>\n\n\n\n

We\u2019re also rethinking how AI agents interact with people and each other: Magentic-UI<\/strong>, now open-source, is an experimental human-centered agent that completes web-based tasks, and TypeAgent<\/strong> is bringing long-term memory to agents, enabling them to retain and recall knowledge over extended periods. In the scientific realm, new AI models like EvoDiff<\/strong> (for generating novel proteins) and BioEmu<\/strong> (for predicting how proteins change shape) are being tested in Azure AI Foundry to accelerate research in biology. These forward-looking projects show how Azure AI Foundry is continually innovating\u2014so you\u2019ll always have access to the latest breakthroughs in AI.<\/p>\n\n\n\n

\n
Check out Foundry Labs<\/a><\/div>\n<\/div>\n\n\n\n

The path forward with Azure AI Foundry<\/h2>\n\n\n\n

Join us at <\/strong>Microsoft Build 2025<\/strong><\/a> to see these new capabilities in action and learn how they can transform your business. We\u2019re excited to work with you\u2014our community of developers and customers\u2014to shape this new era of AI. Together, we\u2019re making AI more accessible, powerful, and trustworthy for everyone. Learn more and get started with Azure AI Foundry<\/a><\/strong>.<\/p>\n\n\n\n

Resources:<\/p>\n\n\n\n