[globular-toast]

> To be specific I’m referring to a prompt like “next, add a for loop which iterates over the elements in the array and enumerate an index, then call our function $func by reference for each element.”

When you're fluent in a programming language it's quicker to just type that directly in said programming language...

Instead you're training yourself to be able to say this stuff in English which will never be as powerful.

[cyberpunk]

I hear this sentiment a lot but it doesn’t match my experience. I’m an experienced erlang programmer, i’ve shipped thousands of gen-server and gen-statem over the years and coding models can absolutely wire them up faster than I can type them out… I guess if you’re using lombok in Java or something typing @foo probably doesn’t take all that long.

[globular-toast]

The problem with a lot of these discussions is it's apples and oranges. I'm not experienced with Erlang, but since you say "wiring up" I'm going to assume you're talking about a pretty repetitive task involving putting the right references across several source code files. This is a mechanical process that hasn't annoyed anyone enough yet to write a program to do it. So either Erlang programmers don't care much about their time, or it's a relatively low-frequency thing that doesn't have a big impact overall.

The example I replied to was more the nuts and bolts of programming. It's the thing you're doing 90% of the time. Changes here have a big impact. That's why for almost my entire career I've had expandable "snippets" in my editor to automatically expand, say, "for" to a for loop skeleton where I fill in the variable names. It's like using an electric screwdriver. You don't lose touch with the screws, it just saves you time and effort.

Typing the entire for loop into an LLM in pseudocode actually seems like a regression compared to that. You don't save any typing. But you lose the ability to work independently. You become dependent on a paid subscription and/or powerful hardware just to do what I've been able to do with a keyboard and hardware you can find for free.

It's similar to writing a letter to someone and having it translated to French, but the reader understands English. Why would you do it?

This changes if you go higher level, of course. This is the temptation that LLMs give us. First it's a for loop, then it's an entire class, then entire modules, then it's only a small step to "vibe coding". What we're still figuring out is where this is actually a benefit. Where can we save effort without compromises? I don't think it's typing out code in English, and I don't think it's vibe coding either. Is there something in between? It's too soon to tell.