GenTabs with Gemini 3: Google reinvents navigation | Keryc
Google Labs introduces Disco, an experiment designed to rethink how we browse and build on the web. The first feature they're testing is GenTabs, built with Gemini 3, the model Google calls its most intelligent to date.
What GenTabs is and why it matters
Have you ever opened twenty tabs to plan a trip or prepare a school activity and finished more confused than you started? GenTabs aims to fix that. Instead of just being a tab manager, it looks at context — your open tabs and chat history — and generates small interactive apps that help you complete complex tasks without writing a single line of code.
Think of it like an assistant that turns scattered web information into practical tools: a meal chart with links to recipes, a travel plan with places, times and references, or a visual guide so a child can grasp the solar system. And it always links back to the original sources so you can verify and dive deeper.
How it works (high-level, technical)
Gemini 3 acts as the generative engine that understands context. It uses page text and interaction history to infer your goal.
From that context, the system proposes and builds generative UI components: summaries, tables, filters, planners and other small web apps.
Everything is controlled with natural language: you describe what you want, refine with instructions like "make it three columns" or "prioritize budget options" and the generator updates the app.
Each generated element keeps traceability to the web: cells, results or recommendations include links to the original content.
We don't have the exact internal architecture diagram publicly, but conceptually it combines content extraction (from DOM or text), UI synthesis and response generation conditioned on user history. In practical terms, GenTabs is an experience layer that orchestrates Gemini 3's inference over web content.
Real use cases and examples
Travel planning: compile schedules, routes, lodging and tickets into one actionable view.
Meal prep: generate a weekly menu with grouped ingredients and links to recipes.
Elementary education: create interactive activities that explain concepts (for example, planets) using web resources.
Early testers are already building custom apps for these scenarios, and the system also suggests ideas you might not have considered.
Limitations, privacy and the experiment's early state
It's early and not everything will work perfectly.
Google is opening a waitlist to download Disco and try GenTabs, starting on macOS. They'll work with a small cohort to gather feedback: what's useful, what breaks and what users want next.
On privacy, the announcement emphasizes using tab and chat history to understand context. If you're worried about data handling, it's reasonable to expect explicit controls over permissions and visibility during the trials. Traceability to sources is good for verification, but we'll need to see how credentials, sites with private content and sensitive data are handled.
What this means for developers and future products
For the technical community, GenTabs is interesting because it shows a new layer: not just models that answer questions, but models that generate mini-apps that orchestrate the web. Some implications:
The shift from "response" to "action" could change how we integrate AI with web interfaces.
There's opportunity for developer-experience tools that let you export, tweak or integrate those generated apps into existing flows.
If Disco's strongest ideas make it into larger Google products, we might see generative capabilities built directly into browsers or productivity suites.
Final thoughts
GenTabs isn't just a novelty: it's an experiment with clear ambition. It aims to make the web less fragmented and more proactive, using Gemini 3 as the intelligence engine. Will it work for all your workflows? Not yet. But Google's bet to place generative tools directly on the browsing experience is worth paying attention to.