Today's AI tooling was trained and evolved in a world built for web or server side development, not for mobile clients. As a result, Android developers often over-include information, reuse degraded context, or mix multiple tasks into a single conversation leading to LLMs generating code that doesn't work or fit your use case. In this talk we show why Context matters more than prompting, how to scope Context, avoid Context Rot, reset and reuse context intentionally, and externalize stable knowledge instead of embedding it in prompts. You'll walk away with a concrete mental model and actionable frameworks for building AI workflows to help build your Android applications