4 ms·
Don't knock it until you try it! AI helps the most with boilerplate work: Those 3 hours of boring, follow-the-design coding can now be implemented in 5-10 minu
by gwbas1c 17d ago
Don't knock it until you try it!
AI helps the most with boilerplate work: Those 3 hours of boring, follow-the-design coding can now be implemented in 5-10 minutes. I then jump into the more interesting work and cleanup.
Other things are easier. For example, saying "Refactor this class to use XYZ serialization library" and having it done in 5 minutes instead of 90 minutes frees me to do the more intellectually stimulating work.
Likewise, when it comes to AI and design, what I can do is type in ideas in a stream-of-conscious way into a prompt, and then it looks at code and comes out with sane text that I can go edit.
BUT: There's interesting things you can do with AI that are harder to do by hand. I ran an experiment where I asked AI to rewrite a method to not use an ORM, which is useful if you need to optimize some DB code. Shopify is having AI write two native versions of their mobile app instead of maintaining a React Native codebase. Lean on the tool to do interesting things that otherwise are too labor intense to do by hand.
---
Don't be like the guy who rejects an IDE, or the guy who only works in assembly.
- lolakutty 17d agoI have tried it. I didn't say LLMs are not useful. It makes some of the boring parts easy and less time consuming. But it does not add it to what makes building something rewarding, on its own. Imagine building LEGO using a robot? Where is the fun in that?
- gwbas1c 16d ago> Imagine building LEGO using a robot? Where is the fun in that? Depends on what you're building!
- gwbas1c 15d agoIf I had more time when I made that comment, I'd clarify a bit: We have a LOT of legos; I could easily see it being more fun to have a robot that we could prompt to find the "needle in a haystack" lego; or when building something big, to follow a pattern.