Re:MIT Study Finds AI Use Reprograms the Brain, Leading to C...
by rewgs on Thursday September 04, 2025 @01:34PM
(#45129901)
100%. I'm _very_ hesitant to use LLMs _at all_ when programming, as the speed/learning trade-off so far hasn't proved itself to be worth it (especially given that it often _isn't_ faster on balance).
However, one place where LLMs have proved to be incredibly helpful is with build tools, dependency hell, etc. Lately I've been trying to update a decade-old Node/Electron project to modern packages and best-practices, and holy hell I simply was not making any meaningful progress until I turned to Claude.
The JS world simply moves too fast (especially back when this project was written) to make updating a project like that even remotely possible in any reasonable amount of time. I was tearing my hair out for days, but yesterday I was finally able to achieve what I was wanting in a few hours with Claude. I still had to work slowly and methodically, and Claude made more than a few stupid errors along the way, but dealing with the delicate version balancing that coincided with API and import/export style changes, all the changes in the bundler world, etc simply could not have been done without it. It's the first time that I was 100% glad I relied on an LLM and felt that it was precisely the right tool for the job at hand.
However, one place where LLMs have proved to be incredibly helpful is with build tools, dependency hell, etc. Lately I've been trying to update a decade-old Node/Electron project to modern packages and best-practices, and holy hell I simply was not making any meaningful progress until I turned to Claude.
The JS world simply moves too fast (especially back when this project was written) to make updating a project like that even remotely possible in any reasonable amount of time. I was tearing my hair out for days, but yesterday I was finally able to achieve what I was wanting in a few hours with Claude. I still had to work slowly and methodically, and Claude made more than a few stupid errors along the way, but dealing with the delicate version balancing that coincided with API and import/export style changes, all the changes in the bundler world, etc simply could not have been done without it. It's the first time that I was 100% glad I relied on an LLM and felt that it was precisely the right tool for the job at hand.