25/06/2025 • LLMs are a powerful tool, but at what cost?
The Good
LLMs are known to be a terrible research tool, but when used in the way, is the greatest tool anyone can have. Asking ChatGPT “I to write a web API in <language>, what available or libraries exist, or should I write my own?” is a great into learning a new language. With extra prompting, it can a few examples with the libraries of that language, and you can further research on your own to get a good idea about what your into.
Using an LLM to quickly fix a problem I have can work sometimes as . Sometimes I don’t particularly want to research this tiny, niche , and just want to move on to the more important parts of the . Getting an AI to give me rough code to solve the issue is convenient.
The Bad
Using AI as an over-powered autocomplete (with Cursor or other text editors) removes the joy of programming for me. Having an solve problems of moderate complexity and above for me also removes joy.
I enjoy typing and I enjoy solving problems, even if that means I a bit longer to implement some functionality. When an LLM for me, it pauses my thinking and switches my focus on I should hit tab to apply the changes. When I’m writing actually code , why would I want anything to stop my train of?
Most importantly though, I don’t want to depend on an LLM to do my job.
LLMs are not only fallible in terms of the code they generate, but depend on expensive infrastructure. Do I want to offshore my expertise to something that can be shut down at point in time? Having the ability to write code from my brain is I don’t want to loose by getting used to AIs doing it for me.