Google Tabletop Turns Group Requests Into Shared Planning Tools
- Cargo
- Machine Learning
- Shipped
- Shipper
- Duncan Riley
- Readers
- 530
Google’s Tabletop experiment lets groups create and edit shared tools for trips, meals and events by asking an AI agent in a shared space.

Google’s Tabletop experiment puts an AI agent to work on a shared planning task: a group describes what it needs, and the agent builds a tool the group can edit together. That makes the product less like a chatbot taking turns in a conversation and more like a shared workspace where a request can change the artifact everyone is using.
Google Labs began testing Tabletop with examples such as trip itineraries, dinner planning, expense splitting and tournament brackets. A person starts a space, adds the links, files and ideas the group has gathered, then mentions @tabletop with a request. The resulting tool is available for the group to work on at the same time, in a format Roman Nurik, a design and engineering lead at Google Labs, compared to a Google Doc.
From a request to an editable tool
The product’s central idea is that a group shouldn’t need to find, configure and maintain a separate app for every small coordination problem. A request can turn shared material into a purpose-built interface: a schedule for a trip, a way to divide expenses, or a bracket for a tournament. Google also offers starter templates, giving users a starting point when they don’t want to describe a tool from scratch.
Tabletop’s demo of an office party seating chart shows how this could work after the initial build. When a guest is seated next to the boss, someone asks @tabletop to revise the chart. The updated version lets people drag name badges between tables. The point is not just that the agent can generate a first draft; users can ask for changes as the group’s needs shift. Google says tools can be reworked in this way.
That interaction could be useful for tasks with changing constraints, where a fixed form or spreadsheet takes effort to reshape. A group can start with the information it has and make adjustments in the shared tool as plans develop. But the available examples don’t establish how well Tabletop handles complex rules, conflicting requests or mistakes in the information people provide. Those are practical questions for users to assess as access expands.
Shared workspace, shared access
A Tabletop space can be shared by posting an invite link in a group’s existing chat. That keeps the invitation step close to where many plans already begin, while moving the work itself into a common space. The announcement describes simultaneous editing, but doesn’t specify how Tabletop resolves two people changing the same information at once, or what control the person who created a space has over other participants.
Those details matter because a collaborative tool has to do more than produce a useful interface. People need to understand what changed, keep the shared information accurate and recover from unwanted edits. Google’s description highlights how users create and revise tools, but does not explain safeguards, version history or how the agent handles ambiguous instructions.
The experiment sits alongside other efforts to bring AI into group activities. Google previously expanded access to its CC agent so that as many as six household members could share it, as reported in September. OpenAI’s ChatGPT group chats take a different approach: the chatbot participates in a shared conversation rather than producing a separate planning interface. OpenAI said the assistant was trained to judge when to join a group discussion. At launch, replies used GPT-5.1 Auto, which selects a model based on the prompt and the subscription of the person receiving the response; personal memory was excluded from group chats.
Google has not identified which models power Tabletop. That leaves open how the system interprets requests, whether it uses different models for different tasks, and what limitations users should expect. Google’s earlier Disco browser experiment used Gemini 3 to turn a person’s open tabs into small web apps, but that does not establish which model, if any, Tabletop uses.
Access and open questions
For now, Tabletop is available only to people in the United States who are at least 18 and sign up through the waitlist at tabletop.google. Google says it plans to admit more people each week. It has not provided pricing or described plans for access beyond the United States.
The test will show whether groups find it faster to ask for a tool than to adapt a document or an existing app. It will also reveal how much guidance the agent needs, how often users must correct its output, and whether shared editing remains clear as more people contribute. Those details will shape Tabletop’s usefulness for routine planning far more than the range of templates alone.
For developers and teams, the notable product choice is the combination of natural language creation with a shared, editable result. If Tabletop can reliably turn group context into tools that people can adjust without friction, it could make small planning workflows easier to assemble. Its waitlist phase leaves the model, pricing and broader availability unsettled.