Google’s Chrome team is quietly running an experiment that treats a browser not as a place for links and tabs but as a factory for tiny, task-focused web apps. The project, called Disco, stitches together open web pages, the user’s prompts, and Google’s Gemini model to generate “GenTabs”: interactive, AI-assembled interfaces — maps, checklists, planners, simple models — built automatically from whatever you’re researching.

The idea is straightforward and unusual at the same time. Instead of summarizing pages into a static block of text or funneling you away from the web into an AI chat, Disco opens relevant tabs and uses them as raw material. Ask for travel ideas for Tokyo and Disco will open research pages, create a planner UI with an itinerary builder and map, and keep that interface synced as you add or remove tabs. Ask about how ankles work and it will pair medical pages with a manipulable foot model. The result looks less like a search and more like an app custom-built around a momentary task.

That shift relies on two new primitives: projects and GenTabs. A project is a container: a chat thread powered by Gemini alongside a constellation of the regular tabs you choose. A GenTab is the AI layer on top — a synthesized workspace that pulls structure from those tabs, your prompts, and the model’s reasoning. Early prototypes simply pasted links into a chat, but the team found that users then stayed inside the conversation and stopped clicking through, which starved the model of current context. By auto-opening tabs and grounding the GenTab in live pages, Google is trying to create a “virtuous cycle”: the more you browse, the richer the AI’s workspace becomes.

Within Google, this started as a hackathon idea, not as an attempt to dethrone Chrome. Parisa Tabriz, who runs the Chrome organization, has framed Disco explicitly as an experiment in Google Labs — a sandbox for probing what happens when browsing and agentic AI are combined rather than a replacement for the company’s long-running flagship browser. That positioning matters: Disco is meant to test whether people can fluidly alternate between conventional browsing and letting an AI remix their research into structured tools such as a meal planner, a study system, or a moving checklist.

If that sounds like marketing copy for an AI productivity tool, there’s a deeper technical bet underneath: GenTabs are not prebuilt components. They’re generated on demand by Gemini (Google’s latest large model), which synthesizes the UI and logic from the content it finds. That’s why the experiment leans on Gemini’s capabilities: Disco’s first demos credit Gemini 3 with the heavy lifting that transforms a pile of pages into a coherent, interactive interface.

That generative approach raises immediate design questions. What is a GenTab, exactly — a document, a spreadsheet, a web app, or some hybrid? Should GenTabs be ephemeral workspaces that vanish when a project closes, or should they be durable, shareable artifacts with URLs and export options? Early testers are already asking for persistence and export features — ways to save a generated itinerary or pull data into Docs or Sheets — and Google hints the answer might be both: throwaway GenTabs for quick tasks and linkable GenTabs that bridge into the rest of Google’s productivity suite.

There are also trust and safety questions. An AI that synthesizes interfaces from multiple web pages must correctly attribute and interpret sources; otherwise, it risks conferring a spurious authority on errors or hallucinations. Google’s prototype behavior — opening pages and trying to keep the AI grounded in live content — is an attempt to mitigate hallucination, but users will still need ways to verify provenance and to understand how a GenTab arrived at a recommendation. The tradeoff is familiar: convenience and immediacy versus transparency and auditability.

Privacy and data handling are another important axis. Google warns that using Disco involves logging AI chats and browsing activity while the experiment runs. That telemetry helps the model stay grounded in current pages and lets engineers iterate, but it also means users adopting Disco are handing sensitive session context to Google’s systems — a nontrivial consideration when the interface can surface medical, financial, or other personal information in a compact app.

Practically, Disco is being rolled out cautiously: it lives in Google Labs and is available to a limited group of testers through a waitlist rather than as a broad Chrome release. That limited exposure is intentional; the team needs to see whether people can form a simple mental model of what Disco does and whether GenTabs fill a real productivity gap or merely add complexity. If adoption remains niche, GenTabs might stay an experimental footnote. If they catch on, Google could have invented not just a new browser feature but a new way to create web apps on demand — interfaces you assemble as you browse, without waiting for a developer to code them.

For web developers and product teams, Disco’s arrival invites reflection. If routine tasks can be quickly scaffolded by an AI that reads and restructures the web into widgets and dashboards, what becomes the role of the one-off developer or the small web app? In one scenario, developers focus on durable APIs and high-quality source pages that generative UIs can consume; in another, Google or others could begin to standardize the building blocks that models use to synthesize interfaces, reshaping front-end design into a mix of content and machine-readable metadata.

For now, Disco is a laboratory. The product’s radicalism is not that it will supplant browsers overnight, but that it asks a new question: what if the web could be something you build as you browse? If GenTabs becomes a natural, shareable way to package research and action, Google will have quietly introduced a new kind of web app — one created on the fly from the web itself. Whether users will prefer these machine-made apps to hand-crafted pages, and whether the ecosystem will accept a model that leans on ephemeral AI workspaces, are the experiments that Disco has been designed to answer.