I mainly use GitHub copilot as nothing more than a fancy autocomplete. I just program as normal and if it happens to suggest more or less what I was about to type, I accept it.

The completions are usually no more than a few lines.

This speeds up my typing (not that my 60-70wpm is “slow”) and allows me to get to the next bit of thinking, without getting too much in the way, or decreasing my syntax knowledge, or requiring me to put brainpower into checking the code it generates (since it was what I was about to type). And hopefully avoids copyright issues, how can “what I was about to type” be infringing?

Typing speed isn’t usually considered to major bottleneck for programming, but I do think using a LLM this way does actually increase my productivity somewhat. It’s not the typing speed increase itself (hell, I’m dubious there even is a real typing speed increase, reading possible completions takes time. But it does feel faster).

It’s more that my adhd-ass brain had a tendency to get bored while typing and has a tendency to get distracted, either with irrelevant tasks, or I go crazy with Don’t-Repeat-Yourself, wasting way more time creating complex unneeded layers of abstractions.

Using an LLM as a fancy autocomplete helps me short circuit these bad tendencies. The resulting code is less DRY and way more KISS.

https://salmonmode.github.io/2020/08/14/the-harmful-obsessio...

Avoiding copy and paste is the key for me to keeping my syntax memory.

I think it's doing the same for me but tbh, I am ok with that and not trying to fix it. I do not want to go back to the world before claude could knock out all of the tedious parts of programming.

Good riddance to my syntax memory. When am I going to ever need it again? The skill I need now is reviewing and architecture.

I'm forgetting small nuanced details about programming systems that I only occasionally have to access.