I wouldn’t say that I’ve been anti-LLM, but I’ve definitely been skeptical of how useful they actually are. Recently, though, I’ve found that new LLM-assisted coding tools like Codex and Claude Code can be genuinely useful.
This probably isn’t news to most people, but as someone whose primary job isn’t software development (just someone who likes to dabble in coding projects), the change in how I approach those projects has been pretty significant.
Like most people I find myself with no lack of projects and hobbies I'm trying to juggle. Whether that's building out a Home Assistant dashboard, overcomplicating my redundant, highly available, dual WAN OPNsense gateways (this could be a whole post), or scanning my photos and organizing them on my self-hosted Immich instance, it can be hard to dedicate more time to something new.
I’m not going to pretend that my coding projects are revolutionary. They’re mostly little tools that solve problems I have, automate something annoying, or scratch some particular itch.
It’s really the time to a viable product that I find impressive.
Could I spend all day writing a Python script that watches a folder on my NAS and sorts new photos into my preferred folder structure, then package it into a Docker container that lives alongside Immich? Absolutely.
But was I able to explain what I wanted, give the LLM my constraints, iterate on the implementation, and end up with a working and reliable tool in under an hour?
Also yes.
And I think that has been the biggest shift.