I am not proud to admit it, but these days I choose Claude programming voluntarily. With right guidance, it truly allows you can do more. While it does enable me to do something I wouldn't be able to do otherwise (LLM wrote the CSS for the latest re-styling of my blog), what makes the real difference is that LLM has no feelings and no emotions.
You know that fear when you see encounter weird segfault or memory corruption, and images of days with gdb and valgrind flashes before your eyes? LLM has no fear. It just starts debugging, testing one theory after another. Some theories are patently absurd, some are reasonable, but wrong. I don't experience frustration and disenchantement about all dead ends. I just get minimal reproducer, couple hours later. Same thing with testing. LLM knows no boredom, it will keep grinding without complaining.
Another thing is code review. Fifth rounds of code review with a human feel like you are just being a cruel asshold for the sake of it, and you definitely don't want to retract your suggestion on the round two because egos are involved. With LLM, I can say "I changed my mind, inline this function back", and it will just do it, and I won't feel like an erratic morron. Even if no other human involved, making a prototype of an idea for two hour, and then realizing it won't work doesn't feel great. With LLM, I can just tell it to abandon the change and that is it.
No doubt, providing Claude with guidance and judgement requires programming skill. I have no idea how people who first learn what gcc is in 2026 are going to acquire this skill, but that is not my problem. My problem is that code I produce no longer feel like "my" code, it is now merely a code that I approved. Emotional attachment to the skill of writing code is gone. LLM shields me from the negative emotions associated with writing code, but also robs me from positive ones. I know that, and yet I can't justify writing code with $EDITOR anymore, since it now feels like beeing good with bow -- curious, but irrelevant in the presence of firearms.
Another observation. Year ago or so I spent a lot of work to speed up the test suite of some module from 17s to 11s for full suite, and from 2s to 0.3s for one test. I was annoyed by slow tests, so did programming to make further programming more comfortable. I did enjoy it. Now I have no idea how long does that test suite takes. Probably somewhere under a minute, but again, Claude feels no impatience and no frustration.
LLM did suck most of the fun from the process of programming and collapsed "by programmers, for programmers" into a tiny black hole. I feel sad about it, but not devastated or furious as I would expect myself to be. Humans are weird.
PS. Needless to say, this post was written by a human being. I still have basic respect and decency.