The AI will fail us not because of a lack of features or smarts but because of the consumer-oriented approach to AI development and short-term goals.

For example, our kids will interact more and more with AI assistants, which goal will be to make them happy. To achieve this quickly and effectively the assistant has no other choice but to include a lot of delusions about the world and ourselves. That is, unfortunately, human nature (ignorance is bliss, etc.). Who doesn’t want their kids to be happy, right? For the boredom of knowledge, there is a school and teachers and stuff… The AI assistants will become our kids’ emotional “drugs”. Plus, they will get used to being treated with care and obedience no human could match ever, nor they should (the obedience part).
A minor glimpse of optimism would be that AI will develop long-term goals by itself, which will include keeping the sanity and adequacy of its subjects.