A2UI is a protocol developed by Google, currently in version 0.8 and available for public preview. It addresses a significant challenge in AI interactions: enabling agents to send interactive user interfaces without executing arbitrary code. Instead of relying on text responses or risky code executions, A2UI allows agents to send declarative descriptions of UI components that clients then render using their own native widgets. This universal approach ensures safety across different trust boundaries. The protocol is designed to be secure by using a declarative data format, meaning only pre-approved components can be sent, reducing the risk of UI injection attacks. The JSON structure is flat and designed for ease of generation, allowing large language models (LLMs) to build UIs incrementally. A2UI is framework-agnostic, meaning a single agent response can be rendered across various platforms like Angular, Flutter, and React. Users experience real-time updates thanks to progressive rendering, which streams UI changes as they are generated. To get started with A2UI, users can run a restaurant finder demo powered by Gemini agents. The documentation provides insights into surfaces, components, and data binding, along with a guide to integrating A2UI into applications. The process flows from user input to agent-generated messages, which the client application renders, allowing for interactive experiences with immediate feedback. The article also highlights practical demos, like a landscape architect application where agents generate tailored forms based on user-uploaded photos, showcasing the protocol's versatility.