I asked my friendly LLM to run every second and dump the delta for each queue into a csv, 10 seconds to write what I wanted, 5 seconds later to run it, then another 10 seconds to reformat it after looking at the output.
It had hardcoded the interface, which is what I told it to do, but I'm happy with it and want to change the interface, so again 5 seconds of typing and it's using argparse to take in a bunch of variables.
That task would have taken me far longer than 30 seconds to do 5 years ago.
Now if only AI can reproduce the intermittent problem with packet ordering I've been chasing down today.
Or even the fact that I was able to start coding in an entirely new ML framework right away without reading any documentation beforehand.
I'm puzzled by the denialism about AI-driven productivity gains in coding. They're blindingly obvious to anyone using AI to code nowadays.
This great comment I saw on another post earlier feels relevant: https://news.ycombinator.com/item?id=46850233
It dropped out a short file which used
from statistics import quantiles
Now maybe that python module isn't reliable, but as it's an idle curiosity I'm happy enough to trust it.
Now maybe I could import a million line spreadsheet and get that data out, but I'd normally tackle this with writing some python, which is what I asked it to do. It was far faster than me, even if I knew the statistics/quantiles module inside out.
This would have taken me an hour previously. It now takes a few minutes at most.
I feel like many AI skeptics are disconnected from reality at this point.
For me, the rise of the TUI agents, emerging ways of working (mostly SDD, and how to manage context well), and the most recent releases of models have pushed me past a threshold, where I now see value in it.