Every group chat has the same weak point. Someone floats a trip idea, twelve people reply with emojis, three useful links get buried under memes, and the one organised friend ends up building a spreadsheet nobody else ever opens. Google Labs reckons an AI agent should be doing that organising instead — and it is now testing the idea in public.

The experiment is called Tabletop. Google introduced it on November 11 in a post by Roman Nurik, design and engineering lead at Google Labs, and it is aimed squarely at messy group decisions: family trips, rec leagues, potlucks, office parties. Instead of one more chat thread, the group gets a single shared space, plus an agent that builds the planning tools inside it.

How a Tabletop space works

The setup is deliberately plain. Whoever creates the space drops in whatever the group has already collected — links, files, half-formed ideas. From there, they prompt @tabletop with what they need, or pick from a set of starter templates, and the agent generates an interactive tool on the spot.

Google's own examples tell you who this is for. Tabletop can spin up a custom trip itinerary, a dinner poll, an expense splitter or a tournament bracket, all generated from the material the group fed it. There is no blank page and no template gallery to scroll through for twenty minutes: you describe the job, and the tool appears.

Once the space is ready, the creator drops an invite link into the group's existing chat, and everyone can jump in and use the tools at the same time. Google compares the experience to working in a Google Doc together — still the simplest mental model most people have for real-time collaboration — and Tabletop borrows it wholesale.

The tools change when the plan does

Here is the part I find most interesting: the generated tools are not frozen. Plans rarely survive first contact with an actual group, so Tabletop lets participants ask the agent to change a rule, reshape a tab or build something new as things evolve.

In one of Google's demos, an office party seating chart turns awkward when a guest ends up next to the boss. A quick request to @tabletop reworks the chart so name badges can be dragged between tables until the seating makes sense again. On the product page, Google shows a simple Pong game being turned into "Pickle Pong" — the ball becomes a bouncing pickle and a live court scoreboard appears, just because someone asked.

The road-trip example stitches several of these together: route mapping, votes on roadside stops, a shared budget split and a spinner wheel that picks where the group eats. None of that is deep software. It is the kind of small, disposable tool a group needs for exactly one weekend — and that disposability is the point. Who is going to build an app for a single potluck? An agent will, apparently.

Google's examples stretch a bit wider than trips and dinners, too. It names exam preparation for study groups, volunteer cleanup events — where the space can add an interactive tide timeline to help pick shifts — and even workplace launch planning. Notice the tilt, though. The headline use cases are life admin shared with friends, family and clubs, not corporate workflow. Tabletop reads less like a rival to your team's project software and more like an attempt to rescue plans that currently die inside chat threads.

Who can try it, and what Google has not said

Access is narrow for now. Tabletop is rolling out in the United States to people aged 18 and over, through a waitlist on the Google Labs site, and Google says new tables open every week as more waitlisted users are let in. There is no announced date for other countries, and Google has not mentioned pricing at all.

It also has not said which AI model does the building. The announcement names no model, gives no technical specs, and does not talk about file limits or integrations. Anyone writing that a specific Gemini model powers Tabletop is guessing, because Google simply has not confirmed it.

Keep expectations where Google puts them, too. This is a Labs experiment, not a finished product. Labs projects can change shape quickly, pick up restrictions, or disappear entirely if the test does not land. Readers outside the U.S. — which is most of us — cannot try it yet, and Google has not given any timeline for wider access.

Agents are starting to build for groups, not just individuals

Tabletop lands in a busy stretch for agent-built software. Cursor recently showed cloud agents that build and host live web apps from a single prompt, while Meta's Muse Ember targets the cheap, high-volume agent work running underneath products like these. The pattern is consistent: describe what you want, and an agent assembles the tool.

What is different here is the audience. Most agent demos still show one person doing one job faster — code shipped quicker, a report drafted sooner. Tabletop points the agent at a group instead: friends and families who do not want another account, another app or a project board, just one place where the plan actually lives and everyone can poke at it.

Whether Tabletop survives Labs is an open question, and Google would probably shrug at it too — that is what the experiment is for. But a group chat that builds its own itinerary, settles its own dinner argument and splits its own bill is a noticeably different pitch from a chatbot that answers questions. If the waitlist opens up beyond the U.S., it is one worth trying with the next trip your group chat keeps failing to organise.

Read the full version of this story on Rohan Tech Official: Google Labs Tests Tabletop — An AI Hub That Builds Interactive Workspaces for Teams