Having relevant projects is key. My https://llm.datasette.io projects gives me the ideal playground for trying stuff out - any time a new API model comes out I can spin up a new plugin to for LLM, which is a great way to try the model with limited development time (most API plugins are a few dozen lines of code).
I've managed to balance building vs writing a lot better in the past - I lost that balance in November and December, I'm trying to get it back for January.
Oh that's cool. We've been blogging about AI eng recently, but the project is often "try this idea/tool/library in order to write a blog post about it".
Having some kind of standard "I need to integrate this new thing with an existing codebase" makes a great standard project.
I've managed to balance building vs writing a lot better in the past - I lost that balance in November and December, I'm trying to get it back for January.