Most “AI” in production still behaves like a stateless function: useful, but forgetful. This talk shows, concretely, how adding a simple memory layer (write → retrieve → reflect → update) changes behavior without touching the model or tools. You’ll see A/B demos (memory OFF → ON) where the same LLM and the same toolset produce different, better plans because history is now part of the input. We’ll keep it practical: a minimal schema for memories (Identity, Principles, Focus, Signals), where memory lives in the loop.

What you’ll take away
A mental model: edit memory → edit agent (stateful reasoning, not fine-tuning).
A production-ready memory loop you can add to any agent today.
A minimal memory schema that keeps behavior explainable and debuggable.