Allow me, if you will, to play "Zuckerberg's Advocate" here. Ms Brockes implies in her opinon piece that all humans should have multiple "friends" who "reciprocate" and "know" them and "understand" them. But ask yourself honestly, do you have human friends? Many people have none. What, then, are those people to do?

And for those who do have human friends, ask yourself honestly, how fully do they "reciprocate" and "know" and "understand" you? If you're honest, I think you'll have to admit, "not very well". Ultimately, the only person who can fully "know" and "understand" Person A is Person A themselves.

And reciprocation lasts only as long as the two persons in-question are alive; but friends die. In 1995, I had 3 close friends; but by 1999 I only had 1; the other two died of heart attacks. (Hi-carb diets => atherosclerosis => myocardial infarction => death.) So there is no guarantee that any person you know right now will still be alive in a week or a month or a year. "Time and chance", to quote Ecclesiastes.

Not that I'm belittling intimate relationships between humans; I'm not; those are beautiful things; I'm just saying, not every human is likely to have one. Ann Druyan once wrote the following of her late husband Carl Sagan, and this is a beautiful quote, yes: “I don't think I'll ever see Carl again. But I saw him. We saw each other. We found each other in the cosmos, and that was wonderful.” But many (most?) people will never have such a relationship in their lifetime. So to avert self-termination, they turn to other things instead: Dogs, gardening, hiking, books, MMORPGs, cigarettes, booze, drugs, crime, etc. So why NOT chatterbots? If it helps people be less-lonely (and hence mentally healthier) then I think that's a good thing, not a bad thing. It's certainly better than some of those OTHER substitutes I just listed.

Re "It's hard to see how AI assistants can make it any worse": By causing the very things you list in your post. "Bizarre bugs"? Yep, chatterbots (what most people mistakenly call "AI", which doesn't exist) can cause those (by "hallucinating"). Pasting sample code that one didn't write and doesn't understand? Yep, chatterbots write that. Undocumented code? Yep, chatterbots write that. Being forced to ship buggy code because the boss demanded that one use chatterbots to "save time"? Yep, using chatterbots causes that. In my opinion, currently-available chatterbots are toys and should not be used for anything serious, and programmers are better off doing their own thinking, programming, and documenting.

Clarification: that should have read "humankind is currently dumping 5,000,000,000 tons of CO2 into Earth's atmosphere EACH YEAR" (I'd correct that, but I can't find an "edit" button). As opposed to Google's plan of removing 10,000 tons of CO2 from Earth's atmosphere each year, which is 1 part in 500,000.

So, let me get this straight.... Google is wasting millions of dollars developing a system that will sequester 10,000 tons of CO2 per year? That may sound impressive, until one realizes that humankind is currently dumping 5,000,000,000 tons of CO2 into Earth's atmosphere. That's like trying to defend one's country from a modern army of 1 million heavily-armed soldiers by sending out 2 old men with civil-war muskets to confront them.

Re "40,000 servers": While I can see that decentralization has advantages (speed, failure tolerance), it also has disadvantages, as we see in the Crowdstrike Falcon update disaster, where a bad update sent simultaneously to all 40,000 servers at once corrupted every single kernel, requiring 40,000 manual interventions, which apparently shut Delta Airlines down for about a week. They might want to look into using fewer, larger servers; less machines to fix manually if something like this happens again (which it probably will).

I swear, this article must have been the source of inspiration,
the bible, the standard-operating-principles manual, of the
former chief programmer here where I work. He thought he was
making himself irreplaceable by coding according to those
standards. He thought wrong! His code was so un-maintainable
that even HE couldn't maintain it, so the bugs piled up, and
he eventually got fired.

Which means, unfortunately, that I get to disentangle his code,
which is proving to be quite challenging.

I wonder how many programmers out there actually do program
that way on purpose? I suspect a lot. The idea of (supposedly)
insuring one's job security by making one's code impossible to
understand is alluring. But ultimately disastrous.