Most AI agent demos work once. Then requirements change, you need to swap a model, add a custom tool, or debug why the agent chose path A instead of path B, suddenly you’re fighting the framework instead of building your product. Strands Agents is the open source SDK that Amazon teams used internally to go from months to weeks shipping production agents. In this session, you’ll see how a model-driven agentic loop, reason, act, reflect, lets you build a working agent in a few lines of code, then extend it without rewriting everything. We’ll live-code an agent from scratch, add tools, and show you exactly what’s happening at every step. You’ll leave with a mental model for building agents that survive real traffic, real edge cases, and real users.