I feel that for large basic scaffolding or a quick spin up of a framework or application structure is a great use of AI coding tools. Currently getting the tools to hold a lot of code in memory and extrapolate on the functionality tends to lead to a fall off of actually working code in my experience. I also know that using the tools to help solve a problem you are stuck on or something that you can not seem to get right is a good use. To prevent you from spending hours and hours trying to figure out something that a few prompts to an AI tool and some quick testing could figure out. IMHO I feel that current AI tools are great for moving things along in specific situations. Feeling like you're loosing your edge is natural when a tool comes along and shifts the way a job is completed. I feel that if you can understand the code the AI tool writes and you can test and implement the code then it's the same as having a someone else with more expierane write the code.

Sorry but I think you are conflating the efficiency of these tools, and what my true concern here is that I will become a worse developer.

I am not trying to optimise how fast I write code, I want to optimise how well I can do it.

I had made the observation that I increasingly started to "garbage collect" all the small stuff that was basically akin to muscle memory.

To me it's still worth it to keep up some level of "language lawyering" in my head as I had been practicing the craft of it all through daily application for many years. There was IMHO no reason to throw that away over time.

I only use LLMs for new stuff / learning, though even there I'm still conflicted if reading the manual / spec like we used to isn't more efficient in the long run?

Again, retaining of basics / primitives but also maybe better mental modelling by going deep head-first and on my own brain cycles?

TL;DR sharing the concerns; still conflicted in how useful LLMs are for programmers; maybe learning? Maybe interfacing for general utility outside of creative tasks is the main value? IDK...

I feel we are in the same place!

Recently it is just worrying me that yes maybe I fixed a few bugs fast, but I won’t be able to fix them again as I did not learn anything long term.

However if it takes you 3 hours of reading GitHub issues and source to fix something, you remember it for the next 5 years!