Commenting because I'm interested in the answers as well.

Stats say over 90% of engineers use AI to code, but the actual split of manual/AI is harder to say.

Sometimes I do on leetcode type things to keep my skills but today on the job I have run out of my AI allotment and Im not sure what to do.

i didnt use em atall for a long time but since recently i use them almost exclusively, they definitely passed my skills for larger projects...

i code by hand for fun, got few tinkering projects. to learn new things i can tell the thing. sometimes i get them from what the models output. play with it by hand to get a feel for it.

at all? i guess no one, but i think there's still a portion of people writing part of their code by hand, like libs, utils, classes etc that the agent can use directly instead the agent will generate their own piece from the scratch in order to protect maintainability

I do. I have tried using LLMs ( as a bootstrapped solopreneur tried optimizations via free AI ) it worked out ok for fast prototypes but even so had enough issues to not consider it for my non-throw away serious work (including iOS game in production) except in a limited fashion such as to spin up few shallow SwiftUI views , get typical implementation for stuff like IAP , GDPR regs, etc. for all the non-core code which is used by almost every game. This is what I used to look up Stack Overflow etc. pre AI. I do not turn on AI integration in Xcode. Leary of sharing my code base on my proprietary code for the game as well as my other products.

I do, not quite every single day.

how many calculate numbers on pen and papers. Still

Zero usage of LLM (assembly coding, and sometimes plain and simple C).

I cannot access them to test if they would be of any help in my coding use cases.

You tell me once we get some inference access with a web API with public tokens (probably severely rate limited), or with full interop on noscript/basic HTML.

I may have to run locally open weight coding frontier models (slooooooow).

[dead]