My submission: I'm a Netscape user from the 90s. I remember the bad old days where the World Wide Web was this battle of standards between a Spyglass Mosaic fork called Microsoft Internet Explorer, and the old incumbent Netscape Navigator.
When Netscape got forked into an open-source project and became Mozilla, they really pulled up their game on the web standards, and when Phoenix^WFirebird^WFirefox was released, it started to make a measurable dint in Microsoft's user numbers.
Apple picked up KHTML/KJS from the KDE project, turned those into Webcore/JavaScriptCore, and eventually they'd release those as Webkit. When Google picked up Webkit and released Chrome, it seemed like Mozilla almost gave up.
We saw a precession of side-project distractions that took away, rather than boosting, the core mission.
Now they're just jumping on the AI wagon in a last-ditch attempt at me-too-ism. No thought as to whether this is really going to help. Yeah, ensuring people don't abuse LLM tools is a nice goal, but it's really a distraction.
The best way to avoid these tools being abused is much like ensuring harmful substances aren't being abused: you keep it out of the community. Radium-dial watches were all the rage 100 years ago, and they haven't been sold for decades because we realised how harmful these things were and put heavy restrictions on the use of such materials. Thus those materials only get used for applications where they are truly useful and not going to cause harm.
Pretending I can "chat" with my computer like it's a human doesn't ensure that my web browser will work when I have to purchase things online, shift money around between bank accounts, interact with the Government or see what work I've been assigned for the week. I don't need a computer to take over the interesting things I do with it, just so that I can settle down and do the boring work.
Doing any real work or creative pursuit is impossible on the Internet if we're stuck back in the 90s with websites effectively "gate keeping" users by requiring specific web browser implementations. We've been there, it was hell. While I've got more trust in Google than I ever had in Microsoft, they're not a company that should be allowed to run the Internet as a dictatorship.
The World Wide Web succeeded because of agreed standards. VBScript and ActiveX controls in web pages died because JavaScript and the Document Object Model provided a better and more open solution that anyone (with sufficient time) could implement.
"AI" does not help this. Translating pages can be done via an extension, and I'd rather such a feature were made an extension that someone installs rather than being baked into the browser. Using a chat bot to navigate the Internet and achieve tasks is a major risk because the web is written in human languages, and human languages are too ambiguous a foundation for the specification of how an application should work. Putting this in the browser does not help achieve safety or privacy, it actively jeopardises it.
The old principle applies: KISS. Keep It Simple Stupid! Firefox is a great standards-compliant and extensible browser that can be adapted to user needs. Thunderbird provides similarly standards-compliant email/groupware client. These are tools that do one thing and one thing well: a demonstration of the Unix ethos at work.
"AI" integration should be via an extension of the user's choice and should not distract from the core mission of delivering functional tools that achieve the core aim of implementing standards. These extensions can be developed by third parties without distracting the development team from delivering truly needed features. Let's focus on what matters.