Until the last few years, most features added to software I use haven't:

...had functionally nondeterminstic, unpredictable results in response to how I use them.

...written in long-form English text with confidence and no guarantee of factual accuracy.

...coaxed children into codependent pseudo-relationships with ML models or encouraged suicide.

AI isn't a new feature; it's a new category. And the people who don't understand why some of us don't want it everywhere don't understand that distinction, or else are financially motivated to ignore it and gaslight everyone about the categorical boundaries crossed.

I use LLMs and diffusion style image generators... Where I understand the model I've chosen, can control it locally, and have enough tacit knowledge to double check the outputs before I go ahead with something. I don't trust Mozilla to ensure any of those things anymore. They've long since burned that credibility.