Generative AI at Scale: LinkedIn’s Production Multi-Agent Systems and Microsoft’s Agentic OS
Session Overview
Generative AI has moved from research labs to production systems serving millions of users, but scaling these systems reliably requires more than powerful models. This keynote draws on real-world experience building and deploying distributed, multi-agent AI systems at LinkedIn—and leading the reimagining of Windows as an agentic platform at Microsoft.
At LinkedIn, Donald Thompson and team built a distributed generative AI platform that powers production agents such as the Hiring Assistant digital worker and enabled 20+ teams to develop agents simultaneously. The program achieved a 75× cost reduction through custom models and a 70% reduction in administrative tasks. At Microsoft, he is helping build Windows as an Agentic OS—platform infrastructure that turns agent capabilities into operating-system primitives with hybrid orchestration, enterprise-grade security, and open standards.
Speaker
Donald Thompson is a Distinguished Engineer in Windows Platform + Developer, where he leads the Agentic OS initiative—architecting the next generation of Windows as an AI-native platform with local, cognitive agents and hybrid AI capabilities. Previously, as Distinguished Engineer at LinkedIn (Microsoft), he oversaw generative AI technology strategy and architected the distributed multi-agent system powering production agents including Hiring Assistant and Digital Workers. He founded Bing’s Knowledge and Reasoning group and created “Satori,” an internet-scale knowledge graph foundational to Microsoft’s AI capabilities. His career spans SQL Server (semantic engine), Windows (Ultra-Mobile PCs), and Microsoft Research (SPOT, Aura). He holds an MSc (with distinction) in Computer Science from the University of Oxford, specializing in AI.