Principles for Building Enterprise AI Agents
Session Overview
AI agents are easy to prototype, but deploying them reliably in enterprise environments is far more difficult. Production systems must manage complex tool use, uncertain model behavior, long-running workflows, security constraints, human oversight, and continuous evaluation.
During this talk, Michael will present practical principles for moving AI agents from experimentation into production . He will discuss agent architecture, planning and orchestration, evaluation, observability, reliability, and techniques for improving agent performance over time.
Drawing on experience building agents for cybersecurity and chip design, he will also examine what high-stakes engineering environments teach us about deploying AI systems that must be accurate, auditable, and dependable.
The talk is intended for engineers, technical leaders, and product teams seeking to build enterprise AI agents that deliver measurable value beyond the prototype stage.
Speaker
Michael Albada is a Platform Architect at NVIDIA, where he focuses on accelerating hardware and chip design with agentic AI.
He specializes in the design, evaluation, and production deployment of AI agents and multi-agent systems, with expertise in large language models, reinforcement learning, orchestration, and AI evaluation.
Previously, Michael was a Principal Applied Scientist at Microsoft, where he helped develop the planning, orchestration, and skill-selection systems behind Microsoft Security Copilot. He has also built large-scale machine learning systems at Uber, ServiceNow, and startups.
Michael is the author of Building Applications with AI Agents, published by O'Reilly Media. He holds a Master of Science in Computer Science from Georgia Tech and an MPhil in Public Policy from the University of Cambridge.