One of the possibly obsolete things I enjoy about working with a human junior dev is that they learn and improve. It's nice to feel all this interaction is building something.

It’s common practice to put your preferences and tips/advice into a readme solely for the LLM to consume to learn about what you want it to do.

So you’d set things like code standards (and hopefully enforce them via feedback tools), guides for writing certain architectures, etc. Then when you have the LLM start working it will first read that readme to “learn” how you want it to generally behave.

I’ve found that I typically edit this file as time goes on as a way to add semi-permanent feedback into the system. Even if your context window gets too large when you restart the LLM will start at that readme to prime itself.

That’s the closest analogy I can think of.