Designing the agentic era: Experimentation over magic

Many AI agents struggle in practice – not because LLMs lack capability, but because we often overestimate what they can handle without thoughtful design. But as soon as we try to make these systems genuinely useful, it becomes clear that effective outcomes still depend on deliberate, rigorous design. This talk reframes what “design” means when systems behave autonomously and why progress relies on structured experimentation, not perfect plans. We’ll examine the evolving skills needed to build agents that truly perform: strong service design, deliberate decisioning, robust evaluation, and continuous learning.