I'm working on a commandline application in spirit of tools like grep and jq. It's to filter and edit RetroArch Playlist files in JSON format with file extension .lpl. The program is written in Rust and has many options and features that exceed the simple JSON parsing of jq in example.

All of this is a learning process for me about the Rust language, JSON and RetroArch Playlist format and how the offline Ai tools and models on such low hardware stacks up in real world. I primarily used it to ask occasionally questions, such as what option name it would recommend, or what function description would be good, sometimes asked what it is thinking of a function implementation, or how to do a certain thing in Rust. I look at the code and its reasoning to understand it and if its an improvement. This was incredible helpful to me in learning more and even finding and eliminating bugs. Especially because I was 6 weeks offline and a few weeks after that period.

I also used the Ai to write about 260 unit tests for functions. Off course I checked every single test, reworked it if necesseray. It helped me to find bugs, but also introduced nasty bugs. So using Ai is dangerous, especially beceause It solves problems you start trusting it more. This was an important lesson for me.

I'm currently almost done and write the README and soon will be uploaded on Github and crates.io.