[localghost3000]

I've developed the following methodology with LLM's and "agentic" (what a dumb fucking word...) workflows:

I will use an LLM/agent if

- I need to get a bunch of coding done and I keep getting booked into meetings. I'll give it a task on my todo list and see how it did when I get done with said meeting(s). Maybe 40% of the time it will have done something I'll keep or just need to do a few tweaks to. YMMV though.

- I need to write up a bunch of dumb boilerplatey code. I've got my rules tuned so that it generally gets this kind of thing right.

- I need a stupid one off script or a little application to help me with a specific problem and I don't care about code quality or maintainability.

- Stack overflow replacement.

- I need to do something annoying but well understood. An XML serializer in Java for example.

- Unit tests. I'm questioning if this ones a good idea though outside of maybe doing some of the setup work though. I find I generally come to understand my code better through the exercise of writing up tests. Sometimes you're in a hurry though so...<shrug>

With any of the above, if it doesn't get me close to what I want within 2 or 3 tries, I just back off and do the work. I also avoid building things I don't fully understand. I'm not going to waste 3 hours to save 1 hour of coding.

I will not use an LLM if I need to do anything involving business logic and/or need to solve a novel problem. I also don't bother if I am working with novel tech. You'll get way more usable answers asking about Python then you will asking about Elm.

TL;DR - use your brain. Understand how this tech works, its limitations, AND its strengths.