This does help and has been around for awhile but I’m still slow af even after working as a dev for 3 years.
I really enjoying coding by hand and figuring out the challenges and architecture of each function. Not to mention the immense feeling of gratification when you spend six hours to find that missing semicolon, comma, or curly brace then it all works. Refactoring massive obscure code into something readable with immediately clear logic, or changing data structures and getting something insane like 100x performance on your biggest bottleneck; all of it feels so good.
However, it seems like this is going to relegated to a hobby in the near future. LLMs are largely better and faster than me plus most, if not all, of my colleagues. I dont like it, but spending the past year working primarily with agentic LLMs i can’t really deny it. It wasn’t this good until relatively recently.
Honestly, having a good LSP that provides context for its suggestions was a huge help for learning how to program in a couple languages. Made sure I actually understood what each part of my code was doing and how pieces fit together. If I’d had claude or copilot, I genuinely don’t think I’d have learned.
Plenty of IDEs have non-LLM predictive text. Common code patterns pop up as a suggestion, and continue to refine as you type.
It’s not the same level of “write a comment and we will spit out something that might work” but it’s more than a few steps above notepad.
This does help and has been around for awhile but I’m still slow af even after working as a dev for 3 years.
I really enjoying coding by hand and figuring out the challenges and architecture of each function. Not to mention the immense feeling of gratification when you spend six hours to find that missing semicolon, comma, or curly brace then it all works. Refactoring massive obscure code into something readable with immediately clear logic, or changing data structures and getting something insane like 100x performance on your biggest bottleneck; all of it feels so good.
However, it seems like this is going to relegated to a hobby in the near future. LLMs are largely better and faster than me plus most, if not all, of my colleagues. I dont like it, but spending the past year working primarily with agentic LLMs i can’t really deny it. It wasn’t this good until relatively recently.
Honestly, having a good LSP that provides context for its suggestions was a huge help for learning how to program in a couple languages. Made sure I actually understood what each part of my code was doing and how pieces fit together. If I’d had claude or copilot, I genuinely don’t think I’d have learned.
I think with frontier models you have to resist the temptation of letting them do too much. If you can, they can be great tools.