Google Tabletop: Google's test bed for on-demand generative UI
Google has launched Tabletop, a Google Labs experiment that allows entire groups to create interactive tools on demand using natural language. Instead of opening a map, a poll, a spreadsheet and a calendar to plan something together, participants type "create an itinerary for our Santa Fe trip" and the assistant generates, in real time, a set of interactive tabs — a map, voting polls, expense splitting — that everyone can edit and adjust in the same shared space.
Led by Roman Nurik, the lead designer and engineer at Google Labs, Tabletop is the most ambitious generative UI experiment that a technology company has ever released. The concept is not to create an AI that writes text or generates images, but one that builds interface controls — buttons, lists, maps, voting scales — and presents them as functional tools for a specific group. It's the concept of generative UI applied to an everyday situation: planning something with people.
How it works in practice
A Tabletop space has three main elements. The chat, which acts as a thread where the group talks, sends links, files and mentions @tabletop to request changes. The tabs, which are small interactive apps generated by the assistant — they can be polls, checklists, countdown timers, games or voting panels. And the Space Context, which stores the shared state: what the group decided, which links were added, which votes took place.
The flow, as demonstrated by Google, is straightforward. A user starts a space, adds relevant links and files, asks @tabletop to generate something specific (a trip itinerary, a dinner poll, expense splits, tournament brackets) and then shares the invite link so all group members can join simultaneously, much like a Google Doc. The difference is that instead of editable text, each member interacts with interactive tools built specifically for that task.
Google has already showcased several examples: an interactive road trip map to Santa Fe with voted stops, dinner decision wheels, shared budget splits, pickleball tournament brackets, merchandise voting boards and drag-and-drop seating charts — essentially a group coordination toolkit that previously would have required opening five separate applications.
What makes this different
AI-generated interfaces are not a new concept. Models like Gemini can already generate HTML code and interface components. What Tabletop does differently is position AI as tool architect, not content writer. Instead of describing an idea and receiving text or an image, you describe a group intent and receive interactive controls — a voting system that adjusts in real time, a map that updates stops as members vote, an expense splitter that recalculates when someone adds a charge.
This distinction matters because it radically changes the interaction model. Traditional tools require users to learn how to navigate the interface: "where's the poll tab? how do I share this?" With Tabletop, the interface is the answer. You say what the group needs, and the AI builds the control instead of guiding you through menus.
There are profound implications for the business model of platforms like Google Docs and Google Sheets. These tools work because they eliminated the friction of sharing documents, but they still require users to know how to use the document — formulas, formatting, structure. Tabletop removes that layer: the group describes what they need, and the interactive tools appear ready to use. If this works at scale, Google's value shifts from "shared spreadsheets and documents" to "tools that appear when you need them."
Where the experiment is still fragile
Tabletop is explicitly in the experiment phase. Google launched it on waitlist only in the US for users 18 and over, with new tables opening weekly. This controlled rollout is telling: Google knows the tool is still far from reliable enough for everyday use.
There are several unresolved questions. Persistence — how tools are versioned, recovered or kept stable across sessions — is not described in public materials. If a group uses Tabletop to plan something over weeks, what happens to the generated tabs? Do they survive? If so, how?
Permission controls are also a mystery. Google has not detailed collaborator roles, granular access controls, moderation tools or participant limits. In a space that involves voting, expense splitting and file sharing, these details are not secondary — they are fundamental.
Extensibility is practically non-existent right now. No API has been announced, no export format, no source-code access, no external integration framework. This makes sense for an experiment, but it means Tabletop, as it stands, is a walled garden: it works within it, but it does not connect to the rest of the ecosystem.
And the privacy question warrants attention. Unless you opt out, Google may use future Tabletop conversations and data to train its generative AI models. If any member of a space opts out, training is stopped for that space entirely. This consensus rule — where a single participant can block data usage for training — is interesting from a governance perspective, but raises practical questions about what happens with data already processed during the window when training was active.
What this means for the future of collaboration
Tabletop can be seen as a natural evolution of the collaboration tools we have already fragmented into dozens of services: Google Docs for text, Google Maps for routes, Google Forms for polls, Sheets for splits, Discord for chat. Tabletop tries to consolidate all of this in a single space where the AI generates the specific tools for each group task.
There are two possible scenarios. In the first, Tabletop becomes a kind of "coordination operating system" — a generic platform where any group task can be solved by generating the right tool. In that world, Google positions itself as collaboration infrastructure provider, not as owner of isolated tools.
In the second scenario — and this is the one that worries established platforms — Tabletop demonstrates that fixed interfaces are dying. Tools like Trello, Asana, Monday, Notion and even Google Docs operate on a "template" model: the user adapts the tool to their task. Tabletop flips this: the tool adapts to the user's task. If this inversion proves useful in practice, it undercuts the value proposition of an entire SaaS sector.
The question is whether Google will turn Tabletop into a product or keep it as a lab. If the former, the impact on the collaboration tool market could be significant. If the latter, the experiment has already achieved its purpose: showing the world what's possible when the interface is the product, not the means by which it's delivered.
Sources: Google Labs Blog, Sulat, AlphaSignal
✓ Independent sources cross-checked and verified before publishing
