Google Labs Launches Tabletop, an Experiment Where a Group Prompts One Agent to Build Itineraries, Polls and Bracket Sheets Together

Google Labs has started testing Tabletop, an experiment that reframes the AI assistant from a one-person chat into a shared workspace. Announced on November 11, the product gives a group of people a common online space where they can drop links, files and notes, then summon an agent by prompting @tabletop to turn the pile of material into a working planning tool.

The generated outputs are deliberately everyday: trip itineraries, dinner polls, expense-splitting sheets and tournament brackets, with starting templates available for people who would rather not write a prompt at all. Once a space exists, users share an invite link into a group chat and everyone can edit the tools simultaneously, in a collaboration model the company compares to Google Docs. Further prompts can modify what the agent has already built, so the tool evolves with the conversation instead of being regenerated from scratch.

Google is positioning Tabletop around the scenarios where planning is genuinely a group activity: family trips, amateur leagues, dinner logistics. Roman Nurik, design and engineering lead at Google Labs, presented the experiment as letting participants collaborate on AI-generated tools together through a shared link. The company has not disclosed which model powers the agent, a notable omission for a product whose quality will live or die on how well it reads a group's messy inputs.

Access is narrow for now. Tabletop is rolling out through a waitlist at tabletop.google to users in the United States aged 18 and older, with Google admitting more participants weekly. There is no announced pricing, no timeline for availability outside the US, and no stated cap on how large a shared space or group can get.

The experiment lands in a rapidly crowding category. OpenAI has added multi-user chats to ChatGPT, and the broader industry has been moving AI products from "answer my question" toward "do a piece of work with me," with shared artifacts rather than shared transcripts as the destination. Google's angle is that the workspace itself, not the chat window, is the collaboration surface, which sidesteps the awkwardness of five people taking turns typing to one assistant.

Group AI tools also raise questions that solo assistants mostly avoid: whose inputs does the agent see, what happens to shared files when a member leaves, and whether a group's planning material trains anything downstream. Google has not published answers yet. For now Tabletop is best understood as a probe into whether people actually want to build things with an agent together, and the waitlist numbers will answer that faster than any demo.