What If AI Understood the Context?

The problem is not what AI can do for the user. The problem is how to convince them to actually use it.

I’m looking at what’s happening in Chromium, and I see movement in the code around browser-AI integration toward cues: small prompts in the address bar that suggest a specific action the user can take with AI. What I find interesting is which scenarios could actually deliver tangible value.

So far, most of this revolves around a few features that have already been integrated into browsers for quite some time: summarization, asking questions about a page, and translation. To me, these are too generic. I’m not convinced that many people use them regularly.

I think one important element is missing: grounding in the user’s working context. I don’t just want to ask a question about the text on the current page. I want information in the context of what I’ve been searching for and working on over the past few days. I don’t just want a summary of a page. I want a summary that answers the question that brought me there in the first place.

Technically, there is nothing particularly difficult about this. It should be feasible if the model is allowed to reason over a broader context. The main challenges are probably elsewhere: privacy, because this would require access to browsing history and search activity, and computational cost. That makes it particularly interesting to see how far local models can take this.

A good starting point seems to be form filling. This scenario requires relatively little context, while people spend a lot of time filling out forms, and AI could remove a noticeable amount of routine work.

Next, I would look at work documents, email, and chats. Here, the idea is less about adding new AI capabilities and more about connecting the browser to the AI features that already exist inside these tools.

Two scenarios are less obvious but potentially more interesting: working with filters in complex data views, and something along the lines of “research this topic for me.” The latter is particularly interesting because it marks a shift from AI that answers questions to AI that takes over part of the research process.

Ultimately, it will be interesting to see which of these scenarios actually take off. The mere existence of a point of entry where the browser proactively suggests a specific action opens up a lot of room for experimentation. Perhaps these experiments are how we eventually move beyond another collection of impressive AI demos and toward tools that people actually want to use.