I’m in group 1 but I do find LLMs to be helpful. I just tend to comb through their output and fix a bunch of issues. Maybe it’s not that much faster that way but it gets over the activation energy of starting a new project/module/ticket.

I do find that they’re pretty much only useful when I already know how I’m going to complete a task. If I can describe the implementation at the stack trace level I’ll do fine with AI. If I’m even a little lost the AI is a total crapshoot.