Google Labs' Tabletop Turns Group Chats Into AI-Built Shared Widgets

Google Labs is testing Tabletop, an invite-only collaborative canvas that turns prompts, links, files, and group ideas into interactive tools. Users can ask the built-in AI agent to create trip maps, dinner polls, expense trackers, tournament brackets, and other shared widgets.

The experiment gives Google a consumer test bed for generative UI, a technique in which a model creates interface elements for a specific task. Tabletop could consolidate coordination now spread across documents, mapping apps, polling services, payment trackers, and group-chat threads.

From prompt to shared widget

Tabletop combines real-time collaboration with an AI agent that builds custom tabs and controls inside a shared space. The basic workflow has four steps:

1. Create a Tabletop space and add links, files, notes, or other source material.
2. Mention @tabletop with a request or select a starter template.
3. Let the agent generate a task-specific widget inside the space.
4. Share an invite link so participants can view and use the result simultaneously.

Google’s product page shows a user requesting a spinning wheel of Santa Fe restaurants. Tabletop responds by creating a “Dinner Decider” tab, while the broader workspace keeps trip stops and other planning details together.

Examples Google highlights

• Road-trip maps with stops that participants can vote on
• Spinning wheels for choosing restaurants
• Live brackets for recreational tournaments
• Drag-and-drop seating charts
• Merchandise voting boards and polls
• Expense splitters and shared budget trackers
• Savings trackers and household chore tables

Google’s generative UI bet

Tabletop treats the interface as generated output. A prompt can produce controls, views, and shared state tailored to the group’s immediate task. In this model, the AI’s core output is an interface that participants continue to use together.

Current examples focus on consumer coordination among friends and family, including trips, meals, potlucks, households, and recreational leagues. Google has not positioned Tabletop as an enterprise project-management service or announced administrative features for larger organizations.

A narrow first rollout

Google Labs’ head of design and development, Roman Nurik, leads the project. He describes Tabletop spaces as link-based collaborative environments where invited participants can work at the same time, similar to a shared Google document.

Eligible users can join the waitlist. Google has not provided a timetable for broader access or availability outside the United States.

The gaps developers will notice

• Model and orchestration: Google has not identified the model that interprets requests or explained how generated interfaces are assembled.
• Persistence: Public materials do not describe how widgets are versioned, recovered, or kept stable across sessions.
• Permissions: Google has not detailed collaborator roles, granular access controls, moderation tools, or participant limits.
• Extensibility: No API, export format, source-code access, or external integration framework has been announced.
• Data handling: The cited product materials do not explain how prompts, uploaded files, links, or collaborator data are retained and used.

A live test for generated interfaces

Tabletop gives developers a concrete example of generative UI inside a multiuser product. Its results can show how well model-generated controls handle latency, shared state, editing, error recovery, permissions, and repeated use after the initial prompt.

The experiment’s broader value will depend on whether its widgets remain predictable and maintainable as plans change. Reliable generated interfaces could let software adapt to narrow group tasks without requiring a separate app or fixed template for each one.