Due to the following reasons.
- The code is my brush. It’s the tool I use solve a problem, express my creativity and construct order from chaos. I love writing code by hand.
- The quality of LLM code is variable, but the quantity is something I personally cannot keep up with.
That leaves me with two choices
- Either I let the LLMs run loose (even with a spec/markdown file), do a cursory glance of the output, close my eyes and push to production.
- Or slow the LLM down to a speed at which I can understand what’s happening and approach it function by function.
I prefer the second option.
- Most of the code I write is intended either for production or for personal learning.
- If it’s production, I care way more about quality than quantity. I want to approach it slowly and methodically.
- If it’s for learning, I don’t see how I can learn the nuances by watching an LLM run in loops.
That being said, LLMs are a great tools and I use them almost daily for a variety of tasks.
- Designing and building this website; it’s fully simple-static content.
- Contextual search. I love asking a complicated technical question and Google’s AI mode summarizing search results to fit the question.
- Code Reviews. I find this extremely valuable since LLMs can apply the entire accrued knowledge of the internet to my code and when I apply a bit of common sense, they are great helpers.
- Occasional copy-paste debugging. I’m not proud of this but there are times when it’s a long week and something trivial is blocking me from progressing. The caffeine has worn off, my blood sugar is low and I’m hanging on adrenaline. So I bite the bullet and paste my error message into a chat in the hopes of a miracle. Works about 50% of the time.