This is helpful, proving this entire discussion is subjective.

I'm a retired mathematician with a primary research project, and too many tangential projects I fear revisiting; tokens be damned, will they burn all my time? Translate the K&R C computer algebra system that got me tenure to 64-bit modern C. Implement a no syntax macro language to support my Go60 ZMK keyboard. Realize my vision of how interlinear translations should work so I can read Flaubert in the original for an online course this fall. Rejigger my decades-overgrown .bashrc setup and my status, install scripts to manage Bash, Ruby, Lean, Tailscale and my Homebrew setup across four machines. And a waiting queue as these tasks clear.

Fable 5 (with Opus 5 as backup) on Zed with a $200 Max plan has been a sea change for me. Carefully alternating planning and auto modes, I manage all these projects at once using Zed's Threads Sidebar. I'm a virtual CTO taking intense meetings all day, relieved to go cook or run errands when credits stall. Anything I've procrastinated for months is now making steady progress; the translation project I feared taking a month is nearly done with several hours of my attention. My personal IT support is now more advanced and easy to use than I ever imagined possible.

My project creation has been a series of agentic "parenting" steps, so there are years of evolving cultural DNA. I had so hated agent comments that recent agents simply weren't commenting at all, instead recording all context in support documents. We had a "come to Jesus" meeting to discuss what commenting style would benefit both my failing working memory and future agents' token use.

It has taken me two brutal years so far to learn to use AI. AI is a dangerous and powerful Iron Man suit, an extension of our associative minds that is a different experience for each person.

One doesn't ride a surfboard by telling it which way to go. I would surely die surfing a big wave, but my experience with AI doesn't resemble other accounts.