Google shows how AI agents can reply with maps and lists
The agent can make screens only from parts that the app approved first.
On 29 October 2026, Google and CopilotKit published a guide for developers. CopilotKit is a start-up that makes software to connect agents to apps. The guide shows how an AI agent can reply with screens that work, such as date pickers, maps and booking forms, and not with long text. The screens use A2UI, an open standard from Google. The guide gives no numbers on use.
How it works
The agent sends a flat list of small data messages. It does not send code. The app keeps a catalog of trusted parts and makes them with its own widgets. Google says that the app stops all parts that are not in the catalog. A second open protocol, AG-UI, carries the messages to the app.
The proof
The guide gives no numbers on use. It names no production use of A2UI. The basic catalog has 18 parts. Google says most apps in use make their own catalog. Google says the mode with parts that are declared first is the most popular, but it gives no figures.
What is not known
The guide says nothing about ads. At the other end, an agent can write code that operates in a sandbox. But the screen can be different from the other parts of the app, and the app operates code that a model wrote.
Insight: Google offers the screen layer for agents as an open standard
Google published A2UI as an open standard. Google also opened Android Studio to more than 40 external agents. Vercel Labs made a tool that lets agents read the items on a screen. Open tools at this time read and make screens for agents, and it seems Google wants its format to be the one that apps use. Look for the first apps that add a catalog.
Sources
Posted
