Generative AI is transforming app experiences, yet most AI agents are confined to plain-text conversational interfaces. What if your AI agent could instantly generate rich, interactive Android UIs on the fly, dynamically tailored to the user's specific intent? Enter the A2UI (Agent-to-UI) protocol: an open-source, streaming JSON specification that empowers agents to natively "speak" in UI instead of markdown. In this session, we will deep-dive into a new `androidx.a2ui` Jetpack library that renders native Jetpack Compose UI for the A2UI protocol. We will explore the layered architecture and APIs of the renderer, and how it maps streaming JSON into Compose's snapshot state achieving highly performant, fine-grained reactivity. We will also learn how the library handles progressive rendering and actively defends against unpredictable LLM hallucinations. We will wrap up with a live coding session showing how to integrate the renderer into your app, how to use the out-of-the-box Material 3 catalog, and how to create and test custom components. Key takeaways and learning points: 1. The A2UI protocol: Understand how AI agents can stream dynamic, interactive UI definitions and data models to a native Android app using JSON. 2. Reactive Compose architecture: Learn how the Jetpack Compose A2UI Renderer leverages Compose's snapshot state system to achieve high-performance, fine-grained UI updates from streamed data. 3. Resilience against AI hallucinations: Discover how the renderer handles unpredictable LLM outputs, including automated schema validation, agent self-correction, and sparse-array Out-Of-Memory (OOM) protection. 4. Extensibility & testing: Gain hands-on knowledge on building agent-driven screens using the pre-built Material 3 catalog, authoring custom components, and testing them using the provided testing APIs.