Comment by LVB

Appreciate the thoughts on this. I've been very conflicted about this topic lately. I'm an old Vim programmer, but I've leapt into AI-assistance earnestly. First it was out of curiosity and some amazement at how well they worked. Then, especially with Cursor, it was like, "Holy shit I'm flying through this stuff, basically hitting just tab. It's reading my mind."

But do I feel something weakening in me? Yeah. I don't have nearly the intimate mental model of where things are in code and how it ties together. It's like my brain is optimizing for the next thing that will prompt a useful tab completion, as opposed to tracking the whole. I don't like the feeling, but the actual output doesn't lie. When I switch to a new file and Cursor dumps in a 30 line function, without asking, that is correct based on what I was working as 5 second before, that's almost unquestionably faster. But at what cost...

I'm trying to find the sweet spot in all of this. There are some areas, like writing volumes of HTML/CSS/Tailwind, that I'll happily hand over to an LLM b/c I'm not great at it, I don't care to be, and it is a time compression of probably 50:1. OTOH, I'm thinking about dialing it back a bit on my backend and SQL code, which is a bit nearer and dearer. Simple autocomplete to save some typing is fine, but maybe lighten up on "Is this (20 lines) what you want? No? How about this? Or this? ..." tab-completion mania. I'd like to still be able to land the plane w/o autopilot.