News · Google's GenTabs turns open tabs into generated web apps

Dec, 114 min to read
Frontend

Google's GenTabs turns open tabs into generated web apps

Inside Disco, a macOS experiment from Google Labs that builds interactive tools from your browsing context using Gemini 3.

What GenTabs actually generates

GenTabs is the first feature inside Disco, a new browsing experiment Google Labs is releasing on macOS behind a waitlist. It is built with Gemini 3, which Google calls its most intelligent model.

The core mechanic is specific: GenTabs reads your open tabs and chat history, infers the task you're working on, and produces an interactive web application to help complete it. The examples Google gives are a weekly meal plan, a trip to see cherry blossoms in Japan, and a lesson about the planets for an elementary schooler.

Users don't write code. They describe the tool they want and refine it in natural language. GenTabs also proposes apps the user hadn't thought to ask for, based on the current task.

The frontend is generated, but sourced

The detail most relevant to frontend work is Google's claim that every generative element ties back to the web and links to its original sources. That is a design constraint, not just a feature: the generated UI is meant to remain traceable to the pages that informed it.

And because every generative element ties back to the web, it always links to the original sources.Montana Labs

This positions the generated app as a view over existing web content rather than a replacement for it. For anyone building similar systems, the interesting engineering question is how you keep dynamically assembled interface elements bound to their provenance while still letting a model restructure them into a new tool.

A browser that treats tabs as input, not just windows

Google frames the problem as tab overload: juggling dozens of open tabs to research a topic or plan a trip. GenTabs' response is to treat those tabs as structured signal about intent rather than as passive documents the user manages by hand.

That is a meaningful shift in where the browsing surface sits. Instead of the browser being a container for pages, Disco makes the browser a context source that a model reads to synthesize a purpose-built frontend. The chat history is folded into that same context.

What the small-cohort rollout signals

Google is explicit that this is early, that not everything will work perfectly, and that it is starting with a small cohort of testers on macOS only. The stated goal is to learn from feedback about what's useful and what needs work.

Google also notes the most compelling ideas from Disco may one day make their way into larger Google products. The specific implication for frontend teams: Google is prototyping a world where interface generation happens per-task and per-user, assembled at runtime from browsing context rather than shipped as fixed pages. Whether it stays a Labs experiment or lands in Chrome, the pattern being tested is generated UI bound to live web sources, and that's the part worth watching.

Find this story relevant to you?

Contact us to find a unique solution

Contact us

Need an AI engineering partner that can actually build?

We help businesses integrate AI, build AI-powered products, automate high-value workflows, and modernize the software systems behind them.

Get in touch

Related reading

More analysis around product delivery, operational AI, and the systems work that makes deployment hold up in reality.

Jul, 134 min to read
Frontend

DNP put ChatGPT Enterprise in front of ten departments and treated the chat window as the interface

Jul, 134 min to read
Frontend

AdventHealth deploys ChatGPT across nine states by treating adoption as the product

Jul, 134 min to read
Frontend

AP+ uses Codex to build behaving payment prototypes, not just clickable screens