Comment by s1mplicissimus
2 days ago
It's interesting how LLM enthusiasts will point to problems like IDE, context, model etc. but not the one thing that really matters:
Which problem are you trying to solve?
At this point my assumption is they learned that talking about this question will very quickly reveal that "the great things I use LLMs for" are actually personal throwaway pieces, not to be extended above triviality or maintained over longer than a year. Which, I guess, doesn't make for a great sales pitch.
It's amazing to make small custom apps and scripts, and they're such high quality (compared to what I would half-ass write and never finish/polish them) that they don't end up as "throwaway", I keep using them all the time. The LLM is saving me time to write these small programs, and the small programs boost my productivity.
Often, I will solve a problem in a crappy single-file script, then feed it to Claude and ask to turn it into a proper GUI/TUI/CLI, add CI/CD workflows, a README, etc...
I was very skeptical and reluctant of LLM assisted coding (you can look at my history) until I actually tried it last month. Now I am sold.
At work I need often smaller, short lived scripts to find this or that insight, or to use visualization to render some data and I find LLMs very useful at that.
A non coding topic, but recently I had difficulty articulating a summarized state of a complex project, so I spoke 2 min in the microphone and it gave me a pretty good list of accomplishments, todos and open points.
Some colleagues have found them useful for modernizing dependencies of micro services or to help getting a head start on unit test coverage for web apps. All kinds of grunt work that’s not really complex but just really moves quite some text.
I agree it’s not life changing, but a nice help when needed.
I use it to do all the things that I couldn't be bothered to do before. Generate documentation, dump and transform data for one off analyses, write comprehensive tests, create reports. I don't use it for writing real production code unless the task is very constrained with good test coverage, and when I do it's usually to fix small but tedious bugs that were never going to get prioritized otherwise.