Comment by glenstein

I think the vast majority of attempts to shoehorn ai into the browser are deeply unimaginative and very much garnering a "who ordered that?" reaction. But I suspect we're going to converge on some specific use cases that everyone's going to want and it might just be important to be in the game now as we collectively figure out what they are.

Recently some Ycombinator funded project got highly upvoted on HN, a Chrome based extension that used LLM capabilities to effectively do grease monkey style scripting live in response to human requests. Now that is interesting, and it's a specific application that's actually meaningful and it's not just another AI chat sidebar.

I think it's a matter of workshopping but I bet we're going to be discovering things users actually want that are not yet obvious to us. The example I keep thinking of is non-stupid agent tasking. I wouldn't mind an agent that browsed Amazon for Kindle unlimited hard sci-fi books with critical acclaim. I would be willing to be there's going to be numerous "whybdidn't I think of that" uses cooked up in the next few years.