“Satya Nadella believes that for AI agents to be effective, they need to solve for three key components at runtime, outside the core model: memory, entitlements, and the action space.”