Models, Agents & MCP Servers - AI Community Conf Keynote

Deliver as the afternoon keynote at the AI Community Conference in NYC on Jun 6, 2025.

Abstract:
The Azure AI Foundry is the App and Agent Factory. What does that mean? Let's find out in this keynote as we take a tour of what's new in Azure AI with a closer look at key announcements from Microsoft Build 2025. By the end of the talk you should have a good sense for the latest news in Models, Agents & MCP Servers

Transcript

1. Y O U R A I A P P & A G E N T F A C T O R Y M O D E L S , A G E N T S & M C P S E R V E R S Jun 2025 – AI Community Conf NYC Keynote Nitya Narasimhan, PhD Senior AI Advocate @Microsoft https://linkedin.com/in/nityan
2. Want to Build AI Skills? Cultivate an “AI-First” mindset Think - What would an AI Agent Do? Learn – Build your Models & Tools IQ Deliver – Evaluate for AI Quality & Safety
3. What’s New in Azure AI Foundry Motivation = Think · Learn · Deliver 1. Vision: AI App & Agent Factory 2. Models: Catalog to Router 3. Templates: Architecture to App 5. Agents: Catalog to Connected 6. Trust: Observability to Insight Takeway = Platform · Skills · Reuse
4. Azure AI Foundry is your AI App and Agent Factory Dev Tools Unified Platform for End-to-End Development Deployment
5. Awareness Challenge Adoption Challenge Every AI Developer Journey Starts With Model Choice Knowledge Challenge
6. Azure AI Foundry Models: Find the best models for your AI applications • Unified access • Scalable deployment • Enterprise-Ready Streamlined Dev Experience • Catalog = Find Models • Leaderboard = Compare Them • Playground = Try Them Out! Benefits Industry models: Azure OpenAI Model Family Mistral Model Family* Meta Llama Model Family* Cohere Model Family Databricks Model Family Hugging Face Collection NVIDIA Model Family Nixtla Time-Gen1 Model DeepSeek Model Family xAI Grok Model Family Black Forest Labs* Hosted and sold directly by Microsoft Snowflake Model Family Phi SLM Model Family Stability Model Family Bria 2.3 Fast Model Deci Model Family NTT Data Tsuzumi Model Gretel Navigator Model Core 42 Jais Model SDAIA ALLaM Model *coming soon Demo 1 : Model Catalog 11,000+ Frontier & Open Models in Model Catalog
7. Improved dev experience: Use model router for simplified model selection • Less Complexity. No juggling APIs or analyzing compatibility • Less Effort. Pairs query with best model in real time. Key Takeaways • Precision Routing per prompt • Perf Optimized speed, latency • Cost Optimized more savings • Future-proof maintainability Benefits Demo 2 : Model Router Auto-select optimal model for selected task – in real time!
8. What’s New in Azure AI Foundry Motivation = Think · Learn · Deliver 1. Vision: AI App & Agent Factory 2. Models: Catalog to Router 3. Templates: Architecture to App 5. Agents: Catalog to Connected 6. Trust: Observability to Insight Takeway = Platform · Skills · Reuse
9. Design Challenge Deployment Challenge Jumpstart AI Journey with Customizable AI Solutions Provisioning Challenge
10. AI App Templates Azure AI Foundry Public Preview Demo 3 : AI App Templates Key Takeaways • Reusable Base - gives solutions a faster start • High-value Scenarios - curated by Azure AI • Scale to Production - with rich tooling flows • Customizable to you - is open source
11. What’s New in Azure AI Foundry Motivation = Think · Learn · Deliver 1. Vision: AI App & Agent Factory 2. Models: Catalog to Router 3. Templates: Architecture to App 5. Agents: Catalog to Connected 6. Trust: Observability to Insight Takeway = Platform · Skills · Reuse
12. Agentic Design Enterprise Readiness Managed + Multi-Agent with Azure AI Agent Service Ecosystem Connection
13. AI Agent Service (GA) Azure AI Foundry Multi-Agent Orchestration Open & Interoperable Tools Trust & Enterprise Readiness
14. Logic Apps Integrations • 1400+ Azure Logic Apps workflows now accessible as “tools” for agents • Automate complex business processes • Add Actions to trigger agents • Enable task orchestration across systems Azure AI Foundry Agent Service Interoperable Tools
15. Knowledge Integrations • SharePoint as 1P built-in knowledge tool • Support for Microsoft Fabric, Bing Search • Provide agents rich, context-aware insight Azure AI Foundry Agent Service Interoperable Tools
16. Agent Catalog • Growing library of agent code samples • From Microsoft (1P) and partners (3P) • Reusable building blocks for developers • Accelerated agent development for orgs Azure AI Foundry Agent Service Interoperable Tools
17. Multi-agent workflows Give one agent the abilities of another. Declaratively orchestrate multiple agents. Powered by Semantic Kernel Multi-agent orchestration in Azure AI Foundry Service Connected agents
18. Azure AI Foundry supports open protocols Import external agents Assistants API Responses API* Activity protocol* A2A* Agent protocol✧ MCP* Azure AI Foundry Expose your agent externally Assistants API Responses API✧ Activity protocol✧ A2A Agent protocol✧ MCP* * Available with Semantic Kernel ✧ Coming soon
19. Model Context Protocol • MCP Server For Azure • GitHub Copilot – Agent Mode Azure AI Foundry + Tools Open Protocols
20. Model Context Protocol • MCP Client – Access tools in context • Improved developer experience Azure AI Foundry + Tools Open Protocols
21. AI Foundry Extension • Build, test and deploy AI agents without leaving Visual Studio Code • “Open in VS Code” button to import API endpoints and keys with less effort • VS Code support for multi-agent workflows, agent CRUD ops, view agent LLM call traces Visual Studio Code VS Code Extension “Pack”
22. Foundry Local Azure AI Foundry Easily build, test, deploy AI Apps using local AI model Rich collection of models, tools, agents designed for local AI High performance on-device inference fully utilizes silicon on Windows 11 and Mac OS Phi SLM Mistral7 B Meta Llama 3.1 8B DeepSee k R1 Go from Cloud To Edge
23. Foundry Local • Available Runtimes – Mac, PC • On-Device AI – discover & run models • Local Inferencing with Foundry SDK • On-Device Apps – Electron App on Mac Azure AI Agent Service “On-Device AI” Runtimes
24. What’s New in Azure AI Foundry Motivation = Think · Learn · Deliver 1. Vision: AI App & Agent Factory 2. Models: Catalog to Router 3. Templates: Architecture to App 5. Agents: Catalog to Connected 6. Trust: Observability to Insight Takeway = Platform · Skills · Reuse
25. 1. New Models Added 2. Smarter Model System 3. Foundry Agent Service is GA 4. Multi-agent Orchestration 5. Agentic Retrieval 6. Always-on Observability 7. Enterprise-grade identity 8. Trustworthy AI built-in 9. Foundry Local 10. Foundry Labs Innovation Join AMAs in Discord Community Share Feedback in Discussion Forum
26. 1. Advanced Reasoning Models 2. Model Context Protocol 3. Reasoning with SLMs 4. Prompty Template & Tracing 5. AI Toolkit & Extension Pack 6. Azure AI Foundry Labs 7. Hugging Face Models 8. Foundry Local (CLI & SDK) 9. Agent Catalog (Samples) 10. Model Playgrounds 11. Advanced Fine Tuning 12. Evaluations & Observability Model Mondays S2 E01: Jun 16, 2025 S2 E02: Jun 23, 2025 S2 E03: Jun 30, 2025 Registrations Open https://aka.ms/model-mondays/rsvp
27. 1. Intro to Agents & Use Cases 2. AI Agentic Frameworks 3. AI Agentic Design Patterns 4. Tool Use Design Pattern 5. Agentic RAG 6. Build Trustworthy AI Agents 7. Planning Design Pattern 8. Multi-Agent Design Pattern 9. Meta Cognition Pattern 10. AI Agents in Production 11. AI Agents With MCP 12 Lessons that teach you everything you need to build AI Agents
28. 1. Intro to MCP 2. Core Concepts Explained 3. Security in MCP 4. Getting Started with MCP 5. Practical Implementation 6. Advanced Topics in MCP 7. Community Contributions 8. Early Adoption Insights 9. Best Practices 10. Case Studies 9+ Lessons that teach you everything you need to explore MCP