In this video, I'm going to take you back to 1985, when building AI meant rolling up your sleeves and encoding expertise by hand. I'll tell a personal story from that era—using Prolog, Lisp, and Borland M1 to create rule-based systems that could make decisions in the real world, long before the cloud and GPUs made "intelligence" feel instant. Then we'll jump to 2026, where AI is defined by foundation models, tool-using agents, and systems that learn from enormous datasets rather than just following explicit rules. You'll see what we gained—speed, scale, and the ability to work with messy language and unstructured information—and what we gave up, including some of the determinism and straightforward explainability of classic expert systems. Finally, I'll lay out a practical view of where the industry is headed: the most valuable architectures don't pick sides, they combine modern models with governance, evaluation, and good old-fashioned business logic to deliver outcomes you can trust.