For the past couple of years i have been using LLM chat bots in various different scenarios. most commonly its to write code and speed up development. In the beginning tools like claude code weren't that common, there were some developers who loved it but alot of them felt like it couldn't write code the way it should be 'properly' written and it was simply not good enough. Things may have changed recently.
Now we see that tools like these are widely adopted and there is alot more material online going through tutorials and examples on how to use it. over time i have started to see that the way i use these tools may not be the 'right' way or atleast i have not come across someone else using it the way i do. i tend to prefer setting out a project by handwriting a CLAUDE.md file or a instructions.md file which would tell the agent on what i am working, what i need it to help me with and how it should help me. this includes things like:
"use tmux for running any background tasks or applications..."
"create a change log file which is append only with simple lines of what logic changes you did and why..."
all of this has lead me to think that there is no right way to use these tools, and the way to actually get the most out of them is to just experiment and find what works for you. these tools did not exist a few years ago. everyone is equally guessing; and they will continue to do so.