Google has started testing a new way for groups to plan things together with AI. Google Tabletop, a Google Labs experiment, gives a group one shared online space and an agent that builds whatever planning tool the group asks for. The news was first reported by SiliconANGLE.
The pitch is simple. Most group planning happens in a chat thread where links, files and half-formed ideas pile up and then get lost. Google Tabletop tries to turn that pile into something usable.
Key Takeaways
- Google Tabletop provides a shared online space for group planning using an AI agent that creates customizable tools.
- Members can mention @tabletop to request changes, allowing real-time editing similar to shared Google Docs.
- Unlike ChatGPT’s focus on conversation, Google Tabletop centers around collaborative output and tool creation.
- The tool may raise questions about privacy, model choice, pricing, and reliability under multiple users editing simultaneously.
- Currently, Google Tabletop tests are limited to U.S. users who sign up through a waitlist, without clear pricing or release details.
Table of contents
How Google Tabletop works
Someone creates a space and fills it with the material the group has gathered: links, files, ideas. From there, any member can mention @tabletop with a request, and the agent produces a tool based on what’s in the space.
Google’s examples include itineraries, dinner polls, expense splitters and tournament brackets. There are also starter templates for people who’d rather not begin with a blank request. Roman Nurik, design and engineering lead at Google Labs, framed the idea in a blog post around everyday coordination, such as a family trip or a potluck, which tends to go better when everyone can contribute from one place.
Sharing happens through an invite link, which the organizer can drop into the group’s existing chat. Once people join, they can edit the tools at the same time, much as they would in a shared Google Doc. That detail matters: the group doesn’t have to move its conversation to a new app, only the planning part.
Tools that change when the plan changes
One of Google’s demos shows why the agent approach could be more than a template library. At an office party, a guest ends up seated next to the boss. A request to @tabletop reworks the seating chart so name badges can be dragged from table to table.
That’s a small example, but it points to the real design bet. Any tool in Google Tabletop can be reworked with a follow-up request, according to the product’s site. A static poll or spreadsheet needs someone to rebuild it by hand when plans shift. Here, the person who spots the problem can ask the agent to change the tool.
Google didn’t say which of its models power Google Tabletop. There’s a hint in its recent history, though. When Labs released its Disco browser experiment last December, it used Gemini 3 to turn a person’s open tabs into small web apps. Tabletop does something similar in spirit, building lightweight software from loose material, but it aims at a group rather than an individual. That’s an inference from the two products, not something Google has confirmed.
Where Google Tabletop fits in the group-AI race
Google Tabletop isn’t Google’s first step toward shared AI. In September, the company let up to six members of a household share the Labs division’s CC agent. Together, the two moves suggest Google is testing how its agents behave when more than one person is giving instructions.
OpenAI got there in a different way. It added group chats to ChatGPT a year ago, and up to 20 people can now talk to the chatbot in a single thread. In that setup, ChatGPT acts as one more participant. OpenAI said at launch that it had trained the model to decide when to chime in, and early replies in those chats came from GPT-5.1 Auto, which picks a model based on the prompt and the subscription of the person being answered. OpenAI also kept personal memory out of group chats altogether.
The contrast is worth noting. ChatGPT’s group feature centers on conversation, with the AI as a member of the discussion. Google Tabletop centers on output: a shared workspace where the agent’s job is to produce something the group can use and edit. Neither approach is clearly better yet, and they may end up serving different moments, such as brainstorming in one case and logistics in the other.
What business leaders should watch
Google Tabletop is aimed at families and friends, not offices, and the party-seating demo is about as corporate as it gets. Still, the underlying pattern deserves attention from anyone thinking about collaboration software. Teams already coordinate offsites, client dinners and event logistics in chat apps, and they often patch together polls and spreadsheets to do it.
A few questions will shape whether this kind of tool moves beyond a Labs experiment:
- Privacy and memory: OpenAI excluded personal memory from group chats. Google hasn’t said how Google Tabletop treats personal data when many people share one space.
- Model choice: Google hasn’t named the models behind it, which makes it harder to judge quality and cost.
- Pricing: The announcement doesn’t mention it.
- Reliability under group editing: Several people asking the agent to change the same tool at once is a hard problem, and the early demos don’t show how conflicts get handled.
Read Next
A few related pieces worth your time:
- Best Inbound Call Center Software by Team Size (2026)
- 8 Best AI Red Teaming Platforms for Application and API Testing
- Nvidia Board Reportedly Weighs Removing CEO Jensen Huang
Access and what comes next
For now, the test is limited. Google Tabletop is open only to U.S. users who are 18 or older and sign up through the waitlist at tabletop.google. Google says it will admit more people each week. The announcement says nothing about pricing or a release outside the U.S.
That makes this an early look rather than a finished product. Still, if you run a team that lives in group chats, Google Tabletop is worth a spot on the watch list, especially to see whether the idea of an agent that builds tools on request holds up once real groups start using it.
Source: SiliconANGLE
