Google Tests ‘Disco’: Gemini AI Turns Tabs Into Web Apps
Google is exploring a bold new approach to what a web browser can do. Its latest project, Disco, is an experimental tool from Google Labs that uses the Gemini 3 AI model to transform ordinary browser tabs into interactive, task-focused mini-applications called GenTabs. Rather than just displaying links or summaries, Disco analyzes your open tabs to understand your goals and creates app-like interfaces — for example, an itinerary builder from travel research, a meal-planning dashboard from recipe tabs, or an interactive learning module from educational content. Currently, access to Disco is limited to a waitlist only as Google is reviewing the service and collecting user feedback.
What is Disco, and how does it work?
Disco is still not intended to be a direct replacement for Chrome (at least, not for the time being). You can think of it as a sandbox where a generative AI is seamlessly integrated into the user’s browsing experience. When you ask Disco for help or start a project, the Gemini model inspects the tabs you have open, your recent chat history within the Disco session, and the goal you state. From that context, it synthesizes a GenTab — a dynamically generated, interactive mini-app that pulls together relevant content, summarizes, visualizes, or even provides interactive controls to complete the task. You can then refine the GenTab by chatting with the AI using normal language, and it will adapt the app on the fly. The feature emphasizes hands-on interaction: opening new tabs feeds new source material into GenTabs, and your responses guide the app’s evolution.
GenTabs: the functionality of tabs extended with app features
The core idea behind GenTabs is to move beyond passive consumption. Rather than manually going through 12 different webpages and copying the pieces of information into a different tool, Disco is trying to create the tool for you. Some of the examples that Google mentions are: trip planning (an itinerary with maps and timelines automatically created from travel pages), lesson builders (interactive visualizations for classroom topics), meal-planning tools (personalized weekly menus made from recipes and nutrition data), and project dashboards.
Because GenTabs are generated from the actual web pages you have open, the AI always links back to source material — so the mini-app is grounded in the underlying content rather than being pure hallucination. This design attempts to blend convenience with traceability.
Why Gemini 3 matters
Gemini 3 — Google’s most capable large model to date — is the engine powering GenTabs. Google claims Gemini 3 brings improved contextual understanding and code generation abilities, making it better suited to synthesize structured interfaces from disparate web content. In practice, that means the AI can read the content in your tabs, decide how to structure an app (what widgets, lists, maps, charts, etc. to include), and generate the glue code or markup needed to present that interface inside the browser. For Disco, Gemini isn’t just a conversational assistant; it’s a “vibe coder” that tries to transform intent and sources into usable tools. Google’s blog and product pages emphasize this integration and present GenTabs as a first experiment in a larger exploration of AI-driven web tooling.
Real-world examples (how people might actually use it)
To get a sense of Disco’s promise, imagine a few concrete workflows:
Trip planning: You open flight pages, hotel reviews, and a few blog posts about Kyoto. Ask Disco to “plan a 7-day spring trip to Kyoto.” Disco generates a GenTab with an itinerary builder, suggested routes, an embedded map, estimated travel times, and links to the sources for each recommendation.
Meal planning: You gather recipe articles and nutrition pages. Tell Disco, “make a week’s vegetarian meal plan under 2,000 calories a day.” The GenTab returns a daily menu with recipe links, a shopping list aggregated from ingredients across tabs, and portion-adjustment controls.
Study aid: You research the solar system and have multiple reference pages open. Disco can produce an interactive study module that summarizes each planet, offers labeled diagrams, and generates quiz questions for practice.
Project research dashboard: For product research, Disco could compile competitor data into a single, filterable, and exportable interactive comparison table. Google’s Disco aims to reduce tab-juggling friction by having AI create actionable, data-driven interfaces.
Privacy, data use, and safety concerns
A big part of Disco’s reception will come down to how Google handles data. Disco’s GenTabs processes your browsing activity, tabs, and AI inputs to generate its interfaces. Google notes AI chat and browsing data may be collected; users should weigh this before opting in. Obvious questions include what data is stored, for how long, and whether sensitive pages are excluded by default. Google’s experiments trade novel functionality for data governance, refining privacy controls through testing.
How Disco compares to other AI browsers and tools
Disco enters a crowded and fast-moving space where other companies are also embedding generative AI into browsing. For example, OpenAI has been developing ChatGPT’s browsing and its Atlas browser features. But Disco’s approach emphasizes transforming browsing into bespoke apps rather than layering a chat interface over web pages. Google’s key differentiator is generating interactive UIs, not just summaries, provided it ensures reliability and source traceability. Google’s strategic advantage lies in its deep Chrome and web integration, potentially making Disco broadly and uniquely useful.
Limitations and challenges
There are several technical and product challenges that Disco must overcome:
Accuracy and hallucination: While GenTabs link to sources, synthesizing interfaces and extracting structured data reliably from arbitrary webpages remains brittle. Inaccurate extraction would undermine trust.
Scope creep: This invites complexity: how to suggest meaningful apps from tabs without overwhelming the user.
Performance and compatibility: Browser mini-apps must perform efficiently across devices while respecting system resources.
Limited to a macOS waiting list, Disco risks appearing exclusive rather than universally accessible and useful.
Regulation and policy: Regulators will closely scrutinize generative AI browser data practices for potential anticompetitive activities.
Such problems are very grave. Google seems to be responding to this by taking a slow, step-by-step approach through Google Labs, which is a reasonable decision: the company can test different UX patterns and privacy options with users before making a full product launch.
What Google might learn from Disco
Even if Disco never becomes a mainstream browser, it could shape how developers build web products. We could see the emergence of different developer tools, integrations, and accessibility features through context-driven browser interfaces.
It could push web standards to expose structured metadata for more accurate and reliable AI-driven UIs.
Availability and next steps
As of mid-December 2025, Disco is an experimental product in Google Labs and available to interested users via a waitlist; initial downloads are being rolled out on macOS first. Google invites early access sign-ups, testing GenTabs first to learn how people want AI to reshape browsing. Expect more features, refinements to privacy controls, and wider platform support if Disco gains traction.
Final thoughts
Disco champions a bold idea. It rejects the idea of the browser as a passive portal to content. The browser becomes a generative platform, creating tools from the content you are viewing. The tool saves significant time, but without controlling hallucinations, privacy, and usability, it’s just a forgettable demo. Regardless of Disco’s ultimate fate, its arrival signals how quickly the definition of a “browser” is being rewritten. Google’s Disco pioneers a future where AI synthesizes purpose-built experiences from context, moving beyond static web pages.