Wrote my first “one-shot” code project that was completed end-to-end using Google Gemini CLI.
The goal was modest: to create an application that can render the top 3 posts from each RSS/ATOM feed from a list of blogs in an OPML file.
I wrote the specifications for the program in a markdown file named GEMINI.md , and then ran Google Gemini on it with: gemini -p GEMINI.md, and after a few fits, & starts and self-correcting bug fixes, I had a program that did almost everything I wanted out of such a script.
Most of the utility of code writing LLM tools is in using them to do run-of-the-mill stuff, and this was a hit.
It is not all rosy however, Gemini fell flat on its face when I asked it to refactor and add tests to a hobby project (tohray) written in Nim
Code generating LLMs are here to stay, and we have to use them where they are most useful, and know when not to shoehorn vibecoding for way too long.