How Microsoft Engineers Build AI: Building and Evaluating Agents

From the creator

🚀 Welcome to Episode 2 of How Microsoft Engineers Build AI! In this episode, go behind the scenes with Microsoft engineers as they dive into how agents are built, evaluated, and deployed — with a practical demo straight from engineering. 💡 Learn how Microsoft thinks about agent architecture, real-world challenges in permissioning and control, shaping agent behavior, handling uncertainty, and evaluation best practices. Whether you're an AI developer, data scientist, or tech enthusiast, this is your front-row seat to how cutting-edge AI systems are engineered at scale. 👇 Chapters 00:00 – Introduction 00:37 – Speaker introduction 01:01 – What did you create? What problem did you solve? 02:04 – Demo 04:11 – Why do we need agents? 06:57 – Where does it fit into the picture? What challenges tend to come up? 13:31 – What about permissioning and control? How do you balance flexibility with security? 16:20 – How do you shape Agent behavior? 21:44 – Uncertainty Threshhold Check: Epistemic uncertainty, aleatoric uncertainty, human fallback, and autonomous action 25:15 – Post-Deployment Feedback Loop 26:54 – What are some evaluation best practices? ✔️Get Started: AI for Developer Hub: https://aka.ms/AIforDeveloper Azure AI Foundry Discord: https://aka.ms/azure-ai-foundry Azure AI Foundry: https://ai.azure.com/ 📌 Let's connect: Emmanuel Barajas Gonzales https://www.linkedin.com/in/adityachallapally/ Aditya Challapally | https://www.linkedin.com/in/vanfalen/ Frank Boucher | https://

Published 2025-07-02 · Watch on YouTube ↗
Availability and playback restrictions are controlled by the original channel.