In this episode, I talk with Bas Brekelmans, CTO of Copilot Studio. The conversation covers the evolution of conversational AI models, challenges in managing and deploying the latest models, the impact of model evolution on performance and security, and the shift in building agents from specific workflows to evaluating expected outcomes. The discussion also delves into the history and evolution of conversational AI, early models and innovation, challenges of model management, model selection and orchestration, security and guardrails in model orchestration, performance and latency in model orchestration, and the evolution of building agents. The conversation delves into the impact of AI models on engineering discipline, the transition from traditional development to model training, and the evolution of skill sets in natural language understanding. It also explores the role of language proficiency in AI development and the evolution of AI systems and models.TakeawaysEvolution of conversational AI modelsChallenges in managing and deploying the latest modelsImpact of model evolution on performance and securityShift in building agents from specific workflows to evaluating expected outcomes Nudging AI modelsTransition from traditional development to model trainingChapters00:00 Evolution of Building Agents30:06 Automating Reordering and Over-Engineering36:12 Evolution of Engineering Discipline41:33 Transition to Managerial Role in AI Development48:10 Evolution of AI Systems and Models