[motrm]

I wanted to like this but turned it off as I found it too distracting.

As a competent programmer I've already got an idea of what I'm about to write in my head, yet having an immediate prompt suggest a "do you mean this?" turns that train of thought away from what I'm writing and instead to "is the suggestion correct and what I want?" before accepting or ignoring it.

It's not for me, I just want to get my brain waves into the IDE as quickly as I can.

That said I can certainly see how it'd be useful when working with a language you're not super familiar with, perhaps one you just dabble in occasionally, or where you're learning programming in general, or a new language, where these hints could save you a quick search for "for loop syntax in $LANG".

It's very exciting to see what will come out of AI. There's a lot of stuff not relevant to me, but in time far smarter people than I will conjure up ways to make my life easier.

[hu3]

I agree but...

I'm retraining my brain to cooperate more with AI suggestion because they will get larger and more precise too.

It's going to become a force multiplier wether I like it or not, so I don't want to be left behind.

In a near future, I'm afraid that

"this guy still types most of his code, can you believe that?"

will be a popular joke.

[btown]

Yep, I find that Copilot etc. respond very well to pseudocode comments, so much so that I can write a comment about a particularly gnarly bunch of syntax I'd otherwise need to write, and (even in a language I use every day) it takes me out of flow less to have Copilot write that for me. Because Copilot knows how to match brackets, can think of descriptive names for temporary variables, all those things that I don't want to have to worry about. I can keep my eye on how the interactions work together, the temporal flow of data through the application, things like that.

If the price of this is sometimes needing to learn to ignore the gray text before I write that pseudocode comment, well... it's no worse than Clippy, and we all survived that era fine!

[bitwize]

I don't even use bog-standard symbol-level autocomplete. I don't want video game shit happening in my FOV when I'm trying to write code. A big part of it is what you mention: the jarring context shift between typing the thing and checking to see if the completion of the thing is correct.

There is evidence that ADHD sufferers are more distracted by such "helpful" IDE features. But I think we're going to just have to retrain our brains to accommodate them because we'd be leaving productivity on the table otherwise.

[rpigab]

I feel that way too, though in some cases I'd love to be able to summon this on-demand, make the suggestion appear on alt-enter or other context menus. Then I could either accept the suggestion, keep typing the line, continuing to process suggestions until I change lines, or dismiss the suggestions.

[chuckadams]

I find it's really handy for autocompleting the brainless grunt-work parts: e.g. I have a dozen keys from a json response I'm assigning to a DTO, one per line, giving them a default of blank if the key is null/undefined. After I've done a couple, the tab-complete AI is really good about continuing the same pattern for the rest of the keys, I literally just hit <tab> <return> repeatedly til done. It's not doing the thinking for me, just the typing.

[adverbly]

I have the same concerns.

I wonder if they could have a "double tap" setting where you would have to press tab to see the suggestion, and then you could press tab again to accept it. That way, you could "opt in" if you know that you're about to write some brainless boilerplate or if you know you want to ask for help.