Ask HN: Programmers who don't use autocomplete/LSP, how do you do it?

> and recently copilot

I, for one, have never even used ChatGPT. ¯ \ _ ( ツ ) _ / ¯

I'm a software engineer, pretty highly regarded as very skilled by everyone around me, almost entirely self-taught (programming since age 7), and have worked at 2 household-name companies (and now a spin-off of a third). The only computer science class I've ever taken was a PhD-level introduction to machine learning, in the mid ’90.

I've never seen any task _where the details matter_ that AI appeared to be any good at. For example, I've seen a coworker trying to demo for me how quickly an AI coding assistant could fix a bug in a parser's state machine… but the AI made a mistake, which I spotted immediately.

For most of my projects, the key value I can add is in understanding business needs, translating them to technical requirements, and ensuring sensible choices of data structures, APIs, libraries, etc. Rarely is code generation the rate limiting step.

I do find autocomplete and language servers useful for avoiding typos, but typically there is already some form of compilation, type checking, or linting to ensure valid code.

I have to use VS for work and Intellisense can be so annoying. Too slow (I've already completed the word), or unhelpful. Example: Accessing elements in a 3d vector, having already typed "var." it pops up with the options, x, y, or z, every time. It's not saving me any keypresses, especially after I've already accessed some elements so I know the members of the struct. Also, how about matching to variables or types from the local scope outwards rather than offer something in a system header not present in any preceding code.

So I have it disabled and just search for things (another thing VS can go complete blank on), keeping the relevant file open in another window for reference, which means I also get to read all the comments in the code before using it.

That's at the deepest level. I'd say you can stop one level up: knowing which function serve which purpose, which functions call which functions, which objects maintain which state etc. If you can trace any input/request from the event handler all the way to the database and back, then that's a very good start.

There is some low-key ableism in comments like this. Not all of us have our productivity enhanced by videogame shit happening in our field of vision while we're trying to code, or by software tools attempting to get ahead of us and do or check our work while we're typing. These things are distractions, and the price of distraction, especially for people with certain forms of neurodivergence[0], is very, very steep, nullifying or negating any productivity gains we might otherwise get. So we use our editors with bells and whistles kept to a minimum, and do a bit more up-front thinking in terms of how to structure things and what to name them.

I've worked on very large projects including the entire software stack of a robotics platform and a few enterprise apps using bare Emacs with no autocomplete added. Lack of autocomplete and bulk renaming has never been the productivity bottleneck for me[1]. And I can always tell code by those developers who get a little bit lazy -- or rush through things a little -- and let the IDE do the work, it's marked by things like class and method names that are consistently misspelled everywhere because the programmer typed in the name once (wrongly), and then let autocomplete finish it after typing the first few characters every other time -- never noticing the misspelling.

As for not bothering to write a script -- usually to smooth out my workflow I write a bit of Emacs Lisp to wrap a complicated shell command that does one of the more common tasks such as starting up/shutting down containers for my app, running tests, etc. Emacs Lisp is actually a very powerful environment optimized for this sort of work, even if it's wonky by modern Lisp standards. Then these tasks are just a keystroke away whenever I need them. Great for my ADHD, as I don't have to "stop train of thought, remember shell command(s) to do task, switch to shell, type in command(s), examine output, resume train of thought".

There's even an Emacs command built in for the example you gave: M-x find-grep.

[0] There is actual scientific research backing this up. See: https://arxiv.org/abs/2302.06376

[1] Usually the bottleneck is, I get stuck on some bug and feel the need to bro down and fix it entirely myself, rather than getting a coworker's eyes on it. I find I'm better about this now than I used to be.