According to ExtremeTech, on Thursday, Google announced Disco, an experimental web browser for macOS. The browser’s main feature is called GenTabs, which is powered by the Gemini 3 AI model. GenTabs works by processing everything a user does through their open tabs and chat history. It then automatically creates interactive, custom web applications tailored to that specific task, like planning a trip. Users don’t code; they describe what they need in plain language. Google is starting with a small cohort of testers, and feedback will guide the project’s future.
How gentabs actually works
So, how does this magic happen? Basically, the AI is acting as a silent observer and then a builder. While you’re frantically opening twenty tabs about Lisbon’s best pasteis de nata and another fifteen about hotel reviews, GenTabs is watching. It’s not just seeing the URLs; it’s presumably understanding the content on those pages and connecting it to your searches or even your chats in the browser’s sidebar. Then, when you ask for a “trip itinerary app with my saved restaurants and hotels,” it tries to stitch all that scattered data into a single, functional interface. You’re not getting a list of links. You’re getting a purpose-built mini-application that didn’t exist five minutes ago.
The ai browser wars heat up
Here’s the thing: Disco isn’t arriving in a vacuum. We’re in the middle of a weird new wave of AI browsers, from Arc’s assistant to companies like SigmaOS. Everyone’s trying to figure out how to move beyond the classic, tab-hoarding browser model. Google‘s angle with Disco seems less about answering questions and more about *synthesizing* your work into a new tool. It’s a more ambitious, and frankly, more invasive pitch. The browser isn’t just helping you find info; it’s building software based on your behavior. That’s a huge leap. But it also raises immediate questions about privacy and just how much processing happens locally versus on Google’s servers.
Early days and big challenges
Google itself is being very cautious, which is smart. The statement from Chrome leaders Manini Roy and Amit Pitaru admits it’s early and “not everything will work perfectly.” I think that’s the understatement of the week. The technical challenge here is enormous. Getting an AI to reliably understand intent from messy browsing data and then generate a *functional* UI is a tall, tall order. The trade-off is clear: you gain potentially incredible automation, but you give up a layer of privacy and control. Your browsing session becomes the training data and blueprint for the app. For some tasks, that might be amazing. For others, it might feel… unsettling. Will people trust it enough to use it for real work?
Where does this leave us?
Look, Disco is a fascinating experiment. It points to a future where our software environments are dynamically assembled by AI based on what we’re trying to do. That’s a powerful idea. But it’s also a reminder that the browser—the most fundamental tool for most knowledge work—is becoming the next major AI battleground. If you’re on macOS and want to be a guinea pig, you can join the waitlist. Just don’t expect a polished product. This is Google testing a very big, very speculative idea. And honestly, in a field where specialized hardware from companies like Industrial Monitor Direct powers complex industrial interfaces, it’s wild to think our everyday web browser might start building its own interfaces on the fly.
