AI Summit NYC 2025 Keynote: From Smart Models to Agent Factories

Keynote Delivered at AI Summit NYC
Dec 10-11 2025
https://newyork.theaisummit.com/

Abstract:
We are seeing a paradigm shift in how AI solutions are being developed - from monolithic AI apps to multi-model, multi-agent architectures that require new end-to-end development workflows and unified platform capabilities to accelerate the path from planning to production. From smart models that automate your model selection to optimize for cost and quality - to agent factories that automate your operations and scale your delivery, with Microsoft Foundry you get a Simple, Trusted, and Powerful platform for your needs,

Keynote slides derived from Microsoft Ignite talk on Microsoft Foundry
See https://ignite.microsoft.com/

Transcript

1. From Smart Models To Agent Factories Nitya Narasimhan, PhD Senior AI Advocate Microsoft AI SUMMIT NYC 2025 Accelerate AI Solution Development With the Microsoft Foundry Platform
2. AI adoption is accelerating and AI agents are at the forefront 1.3B AI agents by 2028 82% of organizations intend to integrate them within 1–3 years 90% of enterprise software engineers will utilize AI code assistants by 2028
3. The new wave: multi-models and agentic AI • Select the best model in a fast-evolving landscape and continue using the best model throughout AI app and agent lifecycle • Lack of streamlined tools for orchestrating AI agents • Complex enterprise workflows with data spread across different sources • Cost management and scalability of generative AI models • Continuously monitor and optimize AI apps and agent performance Challenges 81% of leaders expect agents to be integrated into their company’s AI strategy in next 12-18 months Source: Work Trend Index Annual Report 85% of enterprises are pursuing multi- model strategies, emphasizing the need for rapid testing and deployment of new models to maintain a competitive edge Source: Forrester 70% or more generative AI experiments never make it to production Source: Deloitte
4. Single model Simple POCs Point-in-time ROI Reactive monitoring Human-driven deployment Multiple point-based security solutions Deterministic logic features Manual coding – Building – Testing Multi-model Automated agentic workflows Continuous improvement Next-gen AI platform Ongoing predictive observability Dynamic AI-driven orchestration End-to-end platform-based security approach AI-agentic DevOps Proactive security and governance Adaptive reasoning agents Writing prompts – Creating – Evaluation Shift The way we build AI apps and agents is changing From To
5. AI development fundamentals 1 AI app and agent orchestration 2 Models 3 Knowledge and tools 4 Observability and agent controls 5 Fine-tuning and customization 6 Deployment to local and edge
6. Microsoft Foundry
7. Microsoft Foundry The AI app and agent factory Models Agent Service IQ Tools Machine Learning Control Plane Cloud Edge Security, compliance, and governance
8. Microsoft Foundry Streamline development with native IDE experiences GitHub Visual Studio Visual Studio Code Copilot Studio Models Agent Service IQ Tools Machine Learning Build context-aware and action-oriented agents with 1,400+ pre-built connections and MCP tools Microsoft Bing Microsoft Fabric Microsoft OneLake Control Plane Leverage a complete signals management layer with Microsoft Security integrations Microsoft Agent 365 Microsoft Defender Microsoft Purview Microsoft Entra Security, compliance, and governance
9. Microsoft Foundry The full enterprise package Model policy 99.9% uptime 99% latency SLA for provisioned Data Zone for US and EU Off peak capacity @ 50% for models Priority processing for Azure OpenAI Model router Content safety by default for models RAG through Azure AI Search VNET
10. 80,000+ customers are seeing ROI with Microsoft Foundry 375% growth in transactions ​Reduced hours for healthcare staff 95K auditors adopting Clara Agent​ 75% Requests resolved autonomously
11. Microsoft Foundry Simple Trusted Powerful
12. Microsoft Foundry Simple Trusted Powerful
13. Zava Agents Service agent In-store kiosk agent Marketing agent Customer service agent Weekly planning agent Inventory agent
17. Microsoft Foundry Simple Trusted Powerful
18. Controls Observability Security Fleet-wide Operations 4 Essential Components for Trustworthy AI
20. Controls Keep agents safe and grounded
21. Evaluate and monitor agents for highest quality before & after deployment Observability
22. Integrated with Microsoft Security, protecting every agent at every step Security
23. One dashboard for every agent: health, performance, cost, compliance Fleet-wide Ops
24. Microsoft Foundry Simple Trusted Powerful
25. 1 Models 2 Hosted Agents 3 Developer stack & tools 3 Core Capabilities for Empowering Developers
27. Model router General availability Real-time routing, cost control, and built-in governance
28. Fine Tuning in Foundry General availability Effortless model customization with support for new models, synthetic data, and quick-start training
29. Foundry Local General availability Run, customize, and control AI — right on the device
30. Foundry Agent Service General availability Securely build, connect, and scale agents with workflows, built-in memory, serverless hosting
31. Microsoft Agent Framework Open-source engine for building and orchestrating intelligent AI agents Public Preview
32. Agent Development – Is Different
33. AI Toolkit – In Visual Studio Code
34. AI Agents – In Software Lifecycle Code Verify Plan Test Operate Optimize Deploy Agent mode Coding agent Spark workbench Create issues Brainstorm Spec kit Code quality Code review Autofix Migration Spark runtime Deployment App modernization SRE agent Support Investigate Metrics
35. Accelerate innovation with a complete, integrated, & interoperable AI platform Drive business results with action-oriented, context-aware agents Govern the AI lifecycle with organization-wide observability and controls The App & Agent Factory
36. Join The Community Let’s Build Together!