Designing Data and AI Systems That Hold Up in Production

Tech lead Mike Huls discusses the importance of a full-stack, system-level perspective when designing data and AI systems for production. He emphasizes that models are only valuable when embedded in reliable systems with proper data pipelines, APIs, and governance. The conversation covers the hidden complexities of AI agents, which require significant engineering for state management, cost control, and security to be effective beyond simple demos. Huls also advocates for architectural choices that prioritize change, correctness, and privacy, such as using layered architecture and self-hosting models.